Datasets:

nkandpa2

/

arxiv-papers

like 1

0802.4254

# Coherent pulsed excitation of degenerate multistate systems: Exact analytic solutions E.S. Kyoseva Department of Physics, Sofia University, James Bourchier 5 blvd., 1164 Sofia, Bulgaria Fachbereich Physik der Universität, 67653 Kaiserslautern, Germany N.V. Vitanov Department of Physics, Sofia University, James Bourchier 5 blvd., 1164 Sofia, Bulgaria Institute of Solid State Physics, Bulgarian Academy of Sciences, Tsarigradsko chaussée 72, 1784 Sofia, Bulgaria ###### Abstract We show that the solution of a multistate system composed of $N$ degenerate lower (ground) states and one upper (excited) state can be reduced by using the Morris-Shore transformation to the solution of a two-state system involving only the excited state and a (bright) superposition of ground states. In addition, there are $N-1$ dark states composed of ground states. We use this decomposition to derive analytical solutions for degenerate extensions of the most popular exactly soluble models: the resonance solution, the Rabi, Landau-Zener, Rosen-Zener, Allen-Eberly and Demkov-Kunike models. We suggest various applications of the multistate solutions, for example, as tools for creating multistate coherent superpositions by generalized resonant $\pi$-pulses. We show that such generalized $\pi$-pulses can occur even when the upper state is far off resonance, at specific detunings, which makes it possible to operate in the degenerate ground-state manifold without populating the (possibly lossy) upper state, even transiently. ###### pacs: 32.80.Bx, 32.80.Qk, 33.80.Be, 32.80.-t, 33.80.-b ## I Introduction The problem of a two-state quantum system driven by a time-dependent pulsed external field plays a central role in quantum physics Shore . First of all, this problem is interesting by itself both physically and mathematically: physically, because the two-state system is the simplest nontrivial system with discrete energy states in quantum mechanics; mathematically, because the Schrödinger equation for two states poses interesting mathematical challenges some of which are exactly soluble. Furthermore, already in the two-state case, important nonclassical phenomena occur, for instance, the famous Rabi oscillations, which often serve as a test for quantum behavior, and also provide a powerful tool for coherent control of quantum dynamics, e.g. by $\pi$ pulses. Finally, in almost all cases (except for a few exactly soluble), the behavior of a multistate quantum system can only be understood by reduction to one or more effective two-state systems, e.g., by adiabatic elimination of weakly coupled states or by using some intrinsic symmetries. Besides the well-known solution for exact resonance, there exist several exactly soluble two-state models, the most widely used being the Rabi Rabi , Landau-Zener LZ , Rosen-Zener RZ , Allen-Eberly AE , Bambini-Berman BB , Demkov-Kunike DK , Carroll-Hioe CH , Demkov Demkov and Nikitin Nikitin models. All these models provide the transition probability between two _nondegenerate_ states. Figure 1: Top: The system studied in this paper. $N$ degenerate (in RWA sense) states $\left|\psi_{1}\right\rangle,\left|\psi_{2}\right\rangle,\ldots,\left|\psi_{N}\right\rangle$ are coupled simultaneously to an upper state $\left|\psi_{N+1}\right\rangle$, possibly off single-photon resonance by a detuning $\Delta(t)$, with Rabi frequencies $\Omega_{n}(t)$ ($n=1,2,\ldots,N$). Bottom: The same system in the Morris-Shore basis. There are $N-1$ uncoupled dark states $\left|\varphi_{1}\right\rangle,\left|\varphi_{2}\right\rangle,\ldots,\left|\varphi_{N-1}\right\rangle$, and a pair of coupled states, a bright state $\left|\varphi_{N}\right\rangle$ and the upper state $\left|\psi_{N+1}\right\rangle$, with the same detuning $\Delta(t)$ as in the original basis and a coupling given by the rms Rabi frequency $\Omega(t)$, Eq. (6). In the present paper, we present the extensions of these exactly soluble models to the case when one of the states is replaced by $N$ degenerate states, as displayed in Fig. 1. By using the Morris-Shore (MS) transformation Morris-Shore we show that the ($N+1$)-state problem can be reduced to an effective two-state problem involving a bright state and the upper, nondegenerate state. If known, the propagator for this subsystem can be used to find the solution for the full ($N+1$)-state system. Such analytic solutions can be very useful in designing general unitary transformations within the $N$-state degenerate manifold, which can be viewed as a _qunit_ for quantum information processing QI . We point out that the same system for $N=3$ has been considered by Unanyan et al tripod and by Kis and Stenholm Kis for general $N$, who have derived the adiabatic solution for pulses generally delayed in time; these schemes extend the well-known technique of stimulated Raman adiabatic passage (STIRAP) (see STIRAP for reviews). Here we derive several _exact_ analytic solutions for pulses _coincident_ in time. This work can therefore be considered as an extension to arbitrary $N$ of an earlier paper Vitanov98 , which treated the case $N=2$. This paper is organised as follows. In Sec. II we describe the system and define the problem. In Sec. III we introduce the MS basis and derive the $\left(N+1\right)$-state propagator in terms of the (presumably known) two- state propagator. In Sec. IV we use this solution to identify various interesting types of population evolutions. In Sec. V we use the analytic solutions for exact resonance and the Rosen-Zener model to propose several applications, for example, creation of maximally coherent superpositions and qunit rotation. In Sec. VI we discuss some aspects of the multistate Landau- Zener and Demkov-Kunike models. Finally, Sec. VII provides a summary of the results. ## II Definition of the problem ### II.1 System Hamiltonian We consider an $\left(N+1\right)$-state system with $N$ degenerate lower (ground) states $\left|\psi_{n}\right\rangle$ $\left(n=1,2,...,N\right)$ and one upper (excited) state $\left|\psi_{N+1}\right\rangle$, as depicted in Fig. 1. The $N$ lower states are coupled via the upper state with pulsed interactions, each pair of which are on two-photon resonance (Fig. 1). The upper state $\left|\psi_{N+1}\right\rangle$ may be off single-photon resonance by some detuning $\Delta(t)$ that, however, must be the same for all fields. In the usual rotating-wave approximation (RWA) the Schrödinger equation of the system reads Shore $i\hbar\frac{d}{dt}\mathbf{C}(t)=\mathsf{H}(t)\mathbf{C}(t),$ (1) where the elements of the $\left(N+1\right)$-dimensional vector $\mathbf{C}(t)$ are the probability amplitudes of the states and the Hamiltonian is given by $\mathsf{H}(t)=\ \frac{\hbar}{2}\begin{bmatrix}0&0&\cdots&0&\Omega_{1}\left(t\right)\\\ 0&0&\cdots&0&\Omega_{2}\left(t\right)\\\ \vdots&\vdots&\ddots&\vdots&\vdots\\\ 0&0&\cdots&0&\Omega_{N}\left(t\right)\\\ \Omega_{1}\left(t\right)&\Omega_{2}\left(t\right)&\cdots&\Omega_{N}\left(t\right)&2\Delta\left(t\right)\end{bmatrix}.$ (2) For the sake of simplicity the Rabi frequences of the couplings between the ground states and the excited state $\Omega_{1}\left(t\right),...,\Omega_{N}\left(t\right)$ are assumed real and positive as the populations do not depend on their signs. The phases of the couplings can easily be incorporated in the description and they can be used to control the inner phases of the created superposition states. Furthermore, the Rabi frequencies are assumed to be pulse-shaped functions with the same time dependence $f(t)$, but possibly with different magnitudes, $\Omega_{n}\left(t\right)=\chi_{n}f(t)\quad\left(n=1,2,...,N\right),$ (3) and hence different pulse areas, $A_{n}=\int_{-\infty}^{\infty}\Omega_{n}(t)dt=\chi_{n}\int_{-\infty}^{\infty}f(t)dt\quad\left(n=1,2,...,N\right).$ (4) ### II.2 Physical implementations The linkage pattern described by the Hamiltonian (2) can be implemented experimentally in laser excitation of atoms or molecules. For example, the $N=3$ case is readily implemented in the $J=1\leftrightarrow J=0$ system coupled by three laser fields with right circular, left circular and linear polarizations, as shown in Fig. 2 (left). These coupling fields can be produced from the same laser by standard optical tools (beam splitters, polarizers, etc.), which greatly facilitates implementation. Moreover, the use of pulses derived from the same laser ensures automatically the two-photon resonance conditions and the condition (3) for the same temporal profile of all pulses. The cases of $N=4-6$ can be realized by adding an additional $J=1$ level to the coupling scheme and appropriately polarized laser pulses, as shown in the right frame of Fig. 2. Figure 2: Examples of physical implementations of the linkage pattern of $N$ degenerate ground states coupled via one upper state, considered in the present paper. Left: $N=3$ degenerate states. Right: $N=4$ degenerate (in the RWA sense) states (dashed arrows indicate two additional possible linkages). ## III General solution ### III.1 Morris-Shore (dark-bright) basis The Hamiltonian (2) has $N-1$ zero eigenvalues and two nonzero ones, $\displaystyle\lambda_{n}$ $\displaystyle=$ $\displaystyle 0\quad(n=1,\ldots,N-1),$ (5a) $\displaystyle~{}\lambda_{\pm}(t)$ $\displaystyle=$ $\displaystyle\frac{1}{2}\left[\Delta\pm\sqrt{\Delta^{2}+\Omega^{2}(t)}\right],$ (5b) where $\Omega(t)=\sqrt{\overset{N}{\underset{n=1}{\sum}}\Omega_{n}^{2}(t)}\equiv\chi f(t)$ (6) is the root-mean-square (rms) Rabi frequency, where $\chi=\sqrt{\overset{N}{\underset{n=1}{\sum}}\chi_{n}^{2}}.$ (7) The set of orthonormalized eigenstates $\left|\varphi_{n}\right\rangle$ $(n=1,2,\ldots,N-1)$ corresponding to the zero eigenvalues can be chosen as $\displaystyle\left|\varphi_{1}\right\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{X_{2}}\left[\chi_{2},-\chi_{1},0,0,\cdots,0\right]^{T},$ (8a) $\displaystyle\left|\varphi_{2}\right\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{X_{2}X_{3}}\left[\chi_{1}\chi_{3},\chi_{2}\chi_{3},-X_{2}^{2},0,\cdots,0\right]^{T},$ (8b) $\displaystyle\left|\varphi_{3}\right\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{X_{3}X_{4}}\left[\chi_{1}\chi_{4},\chi_{2}\chi_{4},\chi_{3}\chi_{4},-X_{3}^{2},0,\cdots,0\right]^{T},$ $\displaystyle\cdots$ $\displaystyle\left|\varphi_{N-1}\right\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{X_{N-1}X_{N}}\left[\chi_{1}\chi_{N},\chi_{2}\chi_{N},\cdots,-X_{N-1}^{2},0\right]^{T},$ (8d) where $X_{n}=\sqrt{\overset{n}{\underset{k=1}{\sum}}\chi_{k}^{2}}\quad\left(n=2,3,...,N\right).$ (9) These eigenstates are dark states, i.e. they do not involve the excited state $\left|\psi_{N+1}\right\rangle$ and, as we shall see, are uncoupled from $\left|\psi_{N+1}\right\rangle$. All dark states are time-independent. We emphasize that the choice (8) of dark states is not unique because any superposition of dark states is a dark state too; hence their choice is a matter of convenience. The Hilbert space is decomposed into two subspaces: an $(N-1)$-dimensional dark subspace comprising the dark states (8) and a two-dimensional subspace orthogonal to the dark subspace. It is convenient to use the Morris-Shore (MS) basis Morris-Shore , which, in addition to the dark states, includes the excited state $\left|\psi_{N+1}\right\rangle\equiv\left|\varphi_{N+1}\right\rangle$ and a bright ground state $\left|\varphi_{N}\right\rangle$. The latter does not have a component of the excited state and is orthogonal to the dark states; these conditions determine it completely (up to an unimportant global phase), $\left|\varphi_{N}\right\rangle=\frac{1}{X_{N}}\left[\chi_{1},\chi_{2},\cdots,\chi_{N},0\right]^{T}.$ (10) We point out that the Morris-Shore basis is _not_ the adiabatic basis because only the dark states are eigenstates of the Hamiltonian, but $\left|\varphi_{N}\right\rangle$ and $\left|\varphi_{N+1}\right\rangle$ are not. In the new, still stationary basis $\left\\{\left|\varphi_{n}\right\rangle\right\\}_{n=1,2,...,N+1}$, the Schrödinger equation reads $i\hbar\frac{d}{dt}\mathbf{B}(t)=\widetilde{\mathsf{H}}(t)\mathbf{B}(t),$ (11) where the original amplitudes $\mathbf{C}(t)$ are connected to the MS amplitudes $\mathbf{B}(t)$ by the time-independent unitary matrix $\mathsf{W}$ composed by the basis vectors $\left|\varphi_{n}\right\rangle$, $\mathsf{W}=\left[\left|\varphi_{1}\right\rangle,\left|\varphi_{2}\right\rangle,~{}\ldots,~{}\left|\varphi_{N+1}\right\rangle\right],$ (12) according to $\mathbf{C}(t)=\mathsf{W}\mathbf{B}(t).$ (13) The transformed Hamiltonian reads $\widetilde{\mathsf{H}}(t)=\mathsf{W}^{{\dagger}}\mathsf{H}(t)\mathsf{W}$, or explicitly, $\widetilde{\mathsf{H}}(t)=\frac{\hbar}{2}\begin{bmatrix}0&0&\cdots&0&0&0\\\ 0&0&\cdots&0&0&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots\\\ 0&0&\cdots&0&0&0\\\ 0&0&\cdots&0&0&\Omega(t)\\\ 0&0&\cdots&0&\Omega(t)&2\Delta(t)\end{bmatrix}.$ (14) We point out that the Hamiltonian of Eq. (2) is a special case of the most general Hamiltonian for which the MS transformation Morris-Shore applies and which includes $N$ degenerate lower states and $M$ degenerate upper states. Hamiltonians of the same type as (2) and related transformations leading to Eq. (14), have appeared in the literature also before the paper by Morris and Shore Morris-Shore , mostly in simplified versions of constant and equal interactions (see e.g. Stenholm and references therein). ### III.2 Solution in the Morris-Shore basis As evident from the first $N-1$ zero rows of $\widetilde{\mathsf{H}}$ the dark states are decoupled from states $\left|\varphi_{N}\right\rangle$ and $\left|\varphi_{N+1}\right\rangle$ and the dark-state amplitudes remain unchanged, $B_{n}(t)=const$ ($n=1,2,\ldots,N-1$). Thus the $\left(N+1\right)$-state problem reduces to a two-state one involving $\left|\varphi_{N}\right\rangle$ and $\left|\varphi_{N+1}\right\rangle$, $i\frac{d}{dt}\left[\begin{array}[]{c}B_{N}\\\ B_{N+1}\end{array}\right]=\frac{1}{2}\begin{bmatrix}0&\Omega\\\ \Omega&2\Delta\end{bmatrix}\left[\begin{array}[]{c}B_{N}\\\ B_{N+1}\end{array}\right].$ (15) The propagator for this two-state system, defined by $\left[\begin{array}[]{c}B_{N}\left(+\infty\right)\\\ B_{N+1}\left(+\infty\right)\end{array}\right]=\mathsf{U}_{MS}^{\left(2\right)}\left[\begin{array}[]{c}B_{N}\left(-\infty\right)\\\ B_{N+1}\left(-\infty\right)\end{array}\right],$ (16) is unitary and can be expressed in terms of the Cayley-Klein parameters as $\mathsf{U}_{MS}^{\left(2\right)}=\begin{bmatrix}a&b\\\ -b^{\ast}&a^{\ast}\end{bmatrix},$ (17) with $\left|b\right|^{2}=1-\left|a\right|^{2}$. Then the transition matrix for the $\left(N+1\right)$-state system in the MS basis reads $\mathsf{U}_{MS}^{\left(N+1\right)}=\begin{bmatrix}1&0&\cdots&0&0&0\\\ 0&1&\cdots&0&0&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots\\\ 0&0&\cdots&1&0&0\\\ 0&0&\cdots&0&a&b\\\ 0&0&\cdots&0&-b^{\ast}&a^{\ast}\end{bmatrix}.$ (18) ### III.3 The solution in the original basis We can find the transition matrix in the original, diabatic basis by using the transformation $\mathsf{U}^{\left(N+1\right)}(\infty,-\infty)=\mathsf{WU}_{MS}^{\left(N+1\right)}(\infty,-\infty)\mathsf{W}^{{\dagger}},$ (19) or explicitly, $\mathsf{U}^{\left(N+1\right)}=\begin{bmatrix}1+\left(a-1\right)\frac{\chi_{1}^{2}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{1}\chi_{2}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{1}\chi_{3}}{\chi^{2}}&\cdots&\left(a-1\right)\frac{\chi_{1}\chi_{N}}{\chi^{2}}&b\frac{\chi_{1}}{\chi}\\\ \left(a-1\right)\frac{\chi_{1}\chi_{2}}{\chi^{2}}&1+\left(a-1\right)\frac{\chi_{2}^{2}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{2}\chi_{3}}{\chi^{2}}&\cdots&\left(a-1\right)\frac{\chi_{2}\chi_{N}}{\chi^{2}}&b\frac{\chi_{2}}{\chi}\\\ \left(a-1\right)\frac{\chi_{1}\chi_{3}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{2}\chi_{3}}{\chi^{2}}&1+\left(a-1\right)\frac{\chi_{3}^{2}}{\chi^{2}}&\cdots&\left(a-1\right)\frac{\chi_{3}\chi_{N}}{\chi^{2}}&b\frac{\chi_{3}}{\chi}\\\ \vdots&\vdots&\vdots&\ddots&\vdots&\vdots\\\ \left(a-1\right)\frac{\chi_{1}\chi_{N}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{2}\chi_{N}}{\chi^{2}}&\left(a-1\right)\frac{\chi_{3}\chi_{N}}{\chi^{2}}&\cdots&1+\left(a-1\right)\frac{\chi_{N}^{2}}{\chi^{2}}&b\frac{\chi_{N}}{\chi}\\\ -b^{*}\frac{\chi_{1}}{\chi}&-b^{*}\frac{\chi_{2}}{\chi}&-b^{*}\frac{\chi_{3}}{\chi}&\cdots&-b^{*}\frac{\chi_{N}}{\chi}&a^{*}\end{bmatrix}.$ (20) The $i$th column of this matrix provides the probability amplitudes for initial conditions $\displaystyle C_{i}(-\infty)$ $\displaystyle=$ $\displaystyle 1,$ (21a) $\displaystyle C_{n}(-\infty)$ $\displaystyle=$ $\displaystyle 0\quad(n\neq i).$ (21b) The initial state $\left|\psi_{i}\right\rangle$ can be one of the degenerate states or the upper state. This general unitary matrix and combinations of such matrices can be used to design techniques for general or special qunit rotations. As evident from Eq. (20) for finding the populations for the initial condition (21) it is sufficient to know only the parameter $a=\left[U_{MS}^{(2)}(\infty,-\infty)\right]_{11}$ because $\left|b\right|^{2}=1-\left|a\right|^{2}$ Vitanov98 . For the sake of simplicity, in the present paper we are interested only in cases when the system starts in a single state and below we shall concentrate on the values of the parameter $a$. In the more general case when the system starts in a coherent superposition of states, Eq. (20) can be used again to derive the solution; then the other Cayley-Klein parameter $b$ is also needed. ## IV Types of population distribution We identify two types of initial conditions: when the system starts in one of the degenerate states $\left|\psi_{i}\right\rangle$ or in the excited state $\left|\psi_{N+1}\right\rangle$, which we shall consider separately. ### IV.1 System initially in a ground state #### IV.1.1 General case When the system is initially in the ground state $\left|\psi_{i}\right\rangle$, Eq. (21), we find from the $i$th column of the propagator (20) that the populations in the end of the evolution are $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle\left|1+\left(a-1\right)\frac{\chi_{i}^{2}}{\chi^{2}}\right|^{2},$ (22a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\left|a-1\right|^{2}\frac{\chi_{i}^{2}\chi_{n}^{2}}{\chi^{4}}\quad(n\neq i,N+1),$ (22b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle\left(1-\left|a\right|^{2}\right)\frac{\chi_{i}^{2}}{\chi^{2}}.$ (22c) Therefore the ratio of the populations of any two degenerate states, different from the initial state $\left|\psi_{i}\right\rangle$, reads $\frac{P_{m}}{P_{n}}=\frac{\chi_{m}^{2}}{\chi_{n}^{2}}\quad(m,n\neq i,N+1).$ (23) Hence these population ratios do not depend on the interaction details but only on the ratios of the corresponding peak Rabi frequencies. For equal Rabi frequencies, $\chi_{1}=\chi_{2}=\cdots=\chi_{N},$ (24) Eqs. (22) reduce to $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle\left|1+\frac{a-1}{N}\right|^{2},$ (25a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{\left|a-1\right|^{2}}{N^{2}}\quad(n\neq i,N+1),$ (25b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle\frac{1-\left|a\right|^{2}}{N}.$ (25c) Thus the populations of all ground states except the initial state $\left|\psi_{i}\right\rangle$ are equal. #### IV.1.2 Special values of $a$ Several values of the propagator parameter $a$ are especially interesting. For $a=0$, which indicates complete population transfer (CPT) in the MS two- state system, Eq. (22) gives $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle\left|1-\frac{\chi_{i}^{2}}{\chi^{2}}\right|^{2},$ (26a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{\chi_{n}^{2}\chi_{i}^{2}}{\chi^{4}}\quad(n\neq i,N+1),$ (26b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle\frac{\chi_{i}^{2}}{\chi^{2}}.$ (26c) For $a=1$, which corresponds to complete population return (CPR) in the MS two-state system, we obtain $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle 1,$ (27a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle 0\quad(n\neq i,N+1),$ (27b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle 0.$ (27c) For $a=-1$, which again corresponds to CPR in the MS two-state system, but with a sign flip in the amplitude, we find $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle\left(1-2\frac{\chi_{i}^{2}}{\chi^{2}}\right)^{2},$ (28a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{4\chi_{n}^{2}\chi_{i}^{2}}{\chi^{4}}\quad(n\neq i,N+1),$ (28b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle 0.$ (28c) It is important to note that although both cases $a=1$ and $a=-1$ lead to CPR in the MS two-state system, they produce very different population distributions in the full $\left(N+1\right)$-state system. The case $a=1$ leads to a trivial result (CPR in the full system), whereas the case $a=-1$ is very interesting because it leads to a population redistribution amongst the ground states with zero population in the upper state; hence this case deserves a special attention. #### IV.1.3 The case $a=-1$ The case of $a=-1$ is particularly important because it allows to create a coherent superposition of all ground states, with no population in the upper state. All ground-state populations in this superposition will be equal, $\displaystyle P_{1}=P_{2}=\cdots=P_{N}=\frac{1}{N},$ (29a) $\displaystyle P_{N+1}=0.$ (29b) if $\displaystyle\chi_{i}=\left(\sqrt{N}\pm 1\right)\chi_{0},$ (30a) $\displaystyle\chi_{n}=\chi_{0}\quad(n\neq i),$ (30b) where $\chi_{0}=\frac{\chi}{\sqrt{2\left(N\pm\sqrt{N}\right)}}.$ (30c) This result does not depend on other interaction details (pulse shape, pulse area, detuning) as long as $a=-1$. For example, for $N=4$ degenerate states, equal populations are obtained when $\chi_{i}=\chi_{n}$ or $\chi_{i}=3\chi_{n}$. We shall discuss later how the condition $a=-1$ can be obtained for several analytically soluble models. Another important particular case is when the initial-state population $P_{i}$ vanishes in the end. This occurs for $\chi_{i}^{2}=\sum_{n\neq i}\chi_{n}^{2}.$ (31) For example, an equal superposition of all lower sublevels except $\left|\psi_{i}\right\rangle$, $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle P_{N+1}=0,$ (32a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{1}{N-1}\quad(n\neq i,N+1),$ (32b) is created for $\displaystyle\chi_{i}$ $\displaystyle=$ $\displaystyle\chi_{0}\sqrt{N-1},$ (33a) $\displaystyle\chi_{n}$ $\displaystyle=$ $\displaystyle\chi_{0}\quad(n\neq i),$ (33b) where $\chi_{0}=\frac{\chi}{\sqrt{2\left(N-1\right)}}.$ (33c) ### IV.2 System initially in the upper state If the system is initially in the excited state $\left|\psi_{N+1}\right\rangle$, at the end of the evolution the populations will be $\displaystyle P_{n}=\left(1-\left|a\right|^{2}\right)\frac{\chi_{n}^{2}}{\chi^{2}}\quad\left(n=1,2,\ldots,N\right),$ (34a) $\displaystyle P_{N+1}=\left|a\right|^{2}.$ (34b) For $a=\pm 1$ at the end of the evolution the system undergoes CPR, as in the MS two-state system. For $a=0$ (CPT in the MS two-state system) the whole population will be in the ground states leaving the excited state empty, $P_{N+1}=0$. If all the couplings are equal, Eq. (24), the ground states will have equal populations, $P_{n}=\frac{1}{N}\quad\left(n=1,2,\ldots,N\right).$ (35) ### IV.3 Discussion In this section we discussed some general features of the population redistribution in the $(N+1)$-state system. There are three particularly interesing results. _First_ , the ratios of the populations of the degenerate states (except the one populated initially) depend only on the ratios of the corresponding Rabi frequencies; hence they can be controlled by changing the corresponding laser intensities alone. The populations values, though, depend on the other interaction details. Moreover, it can easily be seen that the relative phases of the degenerate states can be controlled by the relative laser phases. _Second_ , it is possible to create an equal superposition of all ground states, with zero population in the upper state. This is possible when the system starts in a ground state: then condition (30) is required, along with the CPR condition $a=-1$. Alternatively, an equal superposition can be created when the system starts in the upper state: then condition (24) is required, along with the CPT condition $a=0$. Equal superpositions are important in some applications because they are states with maximal coherence (since the population inversions vanish). _Third_ , it is possible, starting from a ground state, to create a superposition of all other ground states, whereas the initial ground state and the excited state are left unpopulated. This requires $a=-1$ and condition (31). This case has interesting physical implications, which will be discussed in the next section. ## V Applications to exactly soluble models ### V.1 Multistate analytical solutions Model --- Resonance $\Omega(t)=\chi f(t),\quad\Delta(t)=0$ $a=\cos\frac{1}{2}A$ Rabi $\Omega(t)=\chi$ $(\left|t\right|\leqq T),\quad\Delta(t)=\Delta_{0}$ $a=\cos\left(T\sqrt{\chi^{2}+\Delta^{2}}\right)-i\dfrac{\Delta}{\sqrt{\Omega^{2}+\Delta^{2}}}\sin\left(T\sqrt{\chi^{2}+\Delta^{2}}\right)$ Landau-Zener $\Omega(t)=\chi,\quad\Delta(t)=Ct$ $a=\exp\left[-\pi\chi^{2}/4C\right]$ Rosen-Zener $\Omega(t)=\chi$sech$(t/T),\quad\Delta(t)=\Delta_{0}$ $a=\dfrac{\Gamma^{2}\left(\frac{1}{2}+i\delta\right)}{\Gamma\left(\frac{1}{2}+\alpha+i\delta\right)\Gamma\left(\frac{1}{2}-\alpha+i\delta\right)}$ Allen-Eberly $\Omega(t)=\chi$sech$(t/T),\quad\Delta(t)=B\tanh(t/T)$ $a=\dfrac{\cos\left(\pi\sqrt{\alpha^{2}-\beta^{2}}\right)}{\cosh\left(\pi\beta\right)}$ Demkov-Kunike $\Omega(t)=\chi$sech$(t/T),\quad\Delta(t)=\Delta_{0}+B\tanh(t/T)$ $a=\dfrac{\Gamma\left(\frac{1}{2}+i(\delta+\beta)\right)\Gamma\left(\frac{1}{2}+i(\delta-\beta)\right)}{\Gamma\left(\frac{1}{2}+\sqrt{\alpha^{2}-\beta^{2}}+i\delta\right)\Gamma\left(\frac{1}{2}-\sqrt{\alpha^{2}-\beta^{2}}+i\delta\right)}$ Table 1: Values of the Cayley-Klein parameter $a=\left[U_{MS}^{(2)}(\infty,-\infty)\right]_{11}$ for several exactly soluble models. Here $\Gamma(z)$ is the Gamma function and $\alpha=\frac{1}{2}\chi T$, $\beta=\frac{1}{2}BT$, $\delta=\frac{1}{2}\Delta_{0}T$, are scaled dimensionless parameters, which are assumed positive without loss of generality. The values of the propagator parameter $a=\left[U_{MS}^{(2)}(\infty,-\infty)\right]_{11}$ for the most popular analytically exactly soluble models are listed in Table 1. Equation (20), supplied with these values, provides several exact multistate analytical solutions, which generalize the respective two-state solutions. Among these solutions, the resonance case is the simplest and most important one, which will receive a special attention below. It will be followed by a detailed discussion of the Rosen-Zener (RZ) model, which can be seen as an extension of the resonance solution to nonzero detuning for a special pulse shape (hyperbolic secant). Both the resonance and the RZ model allow for the parameter $a$ to obtain the important values $0,\pm 1$. The Rabi model can also be used to illustrate the interesting cases of population distribution associated with these values of $a$ but its rectangular pulse shape is less attractive (and also less realistic) than the beautiful sech-shape of the pulse in the RZ model. The Landau-Zener (LZ) and Allen-Eberly (AE) models are of level-crossing type, i.e. the detuning crosses resonance, $\Delta(0)=0$. For these models in the adiabatic limit the transition probability approaches unity, that is $a\rightarrow 0$. The parameter $a$ is always nonnegative, i.e. the most interesting value in the present context, $a=-1$, is unreachable. Nevertheless, because of the popularity and the importance of the LZ model, and because the present multistate LZ solution supplements other multistate LZ solutions, we discuss this solution in detail in Sec. VI. The Demkov-Kunike (DK) model is a very versatile model, which combines and generalizes the RZ and AE models. Indeed, as seen in Table 1, the DK model reduces to the RZ model for $B=0$ and to the AE model for $\Delta_{0}=0$. For the DK model, the parameter $a$ can be equal to the most interesting value of $-1$ only when $B=0$, i.e. only in the RZ limit. Therefore, we shall only consider the RZ model below, and leave the AE and DK models to readers interested in other aspects of the analytic multistate solutions presented here. ### V.2 Exact resonance In the case of exact resonance, $\Delta=0,$ (36) the elements of the evolution matrix for the MS two-state system for any pulse shape of $\Omega(t)$ are $\displaystyle a$ $\displaystyle=$ $\displaystyle\cos\frac{A}{2},~{}~{}$ (37a) $\displaystyle b$ $\displaystyle=$ $\displaystyle-i\sin\frac{A}{2},$ (37b) where $A$ is the rms pulse area defined as $A=\int_{-\infty}^{\infty}\Omega(t^{\prime})dt^{\prime}.$ (38) In the important case of $N=3$ we have $\mathsf{U}_{d}^{\left(4\right)}=\begin{bmatrix}1-2\frac{\chi_{1}^{2}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-2\frac{\chi_{1}\chi_{2}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-2\frac{\chi_{1}\chi_{3}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-i\frac{\chi_{1}}{\chi}\sin\frac{1}{2}A\\\ -2\frac{\chi_{1}\chi_{2}}{\chi^{2}}\sin^{2}\frac{1}{4}A&1-2\frac{\chi_{2}^{2}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-2\frac{\chi_{2}\chi_{3}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-i\frac{\chi_{2}}{\chi}\sin\frac{1}{2}A\\\ -2\frac{\chi_{1}\chi_{3}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-2\frac{\chi_{2}\chi_{3}}{\chi^{2}}\sin^{2}\frac{1}{4}A&1-2\frac{\chi_{3}^{2}}{\chi^{2}}\sin^{2}\frac{1}{4}A&-i\frac{\chi_{3}}{\chi}\sin\frac{1}{2}A\\\ -i\frac{\chi_{1}}{\chi}\sin\frac{1}{2}A&-i\frac{\chi_{2}}{\chi}\sin\frac{1}{2}A&-i\frac{\chi_{3}}{\chi}\sin\frac{1}{2}A&\cos\frac{1}{2}A\end{bmatrix}.$ (39) We have $a=0,\pm 1$ for the following pulse areas, $\displaystyle a=0:\quad$ $\displaystyle A=\left(2l+1\right)\pi,$ (40a) $\displaystyle a=1:\quad$ $\displaystyle A=4l\pi,$ (40b) $\displaystyle a=-1:\quad$ $\displaystyle A=2\left(2l+1\right)\pi.$ (40c) where $l=0,1,2,...$. The pulse areas for the three important cases discussed in Sec. IV.3 are easily calculated. An equal superposition of all $N$ ground states is created when starting from the excited state and all individual pulse areas are equal to (see Sec. IV.2) $A_{n}=\frac{\left(2l+1\right)\pi}{\sqrt{N}}\quad\left(n=1,2,\ldots,N\right),$ (41) where $l=0,1,2,...$. An equal superposition of all $N$ ground states is created also when starting from one ground state $\left|\psi_{i}\right\rangle$ and the pulse areas are [see Eq. (30)] $\displaystyle A_{i}$ $\displaystyle=$ $\displaystyle\sqrt{2\frac{\sqrt{N}\pm 1}{\sqrt{N}}}\left(2l+1\right)\pi,$ (42a) $\displaystyle A_{n}$ $\displaystyle=$ $\displaystyle\sqrt{\frac{2}{N\pm\sqrt{N}}}\left(2l+1\right)\pi\quad\left(n\neq i\right),$ (42b) where $l=0,1,2,...$ The other interesting case when the system starts in one ground state $\left|\psi_{i}\right\rangle$ and ends up in an equal superposition of all other ground states is realised for pulse areas [see Eq. (33)] $\displaystyle A_{i}$ $\displaystyle=$ $\displaystyle\sqrt{2}\left(2l+1\right)\pi,$ (43a) $\displaystyle A_{n}$ $\displaystyle=$ $\displaystyle\sqrt{\frac{2}{N-1}}\left(2l+1\right)\pi\quad\left(n\neq i\right),$ (43b) where $l=0,1,2,...$. ### V.3 Multistate Rosen-Zener model Equation (20) and the value of the parameter $a$ in Table 1 represent the multistate RZ solution in the degenerate two-level system. It is easy to show that $\left|a\right|^{2}=1-\frac{\sin^{2}\left(\frac{1}{2}\pi\chi T\right)}{\cosh^{2}\left(\frac{1}{2}\pi\Delta_{0}T\right)},$ (44) where we have used the reflection formula $\Gamma(\frac{1}{2}+z)\Gamma(\frac{1}{2}-z)=\pi/\cos\pi z$ AS . Hence in this model $\left|a\right|=1$ for $\alpha=\frac{1}{2}\chi T=l$ ($l=0,1,2,...$). The phase of $a$, however, depends on the detuning $\Delta_{0}$ Vitanov98 ; we use this to an advantage to select values of $\Delta_{0}$ for which $a=-1$. For $\alpha=l$ we find Vitanov98 $a=(-1)^{n}\prod_{k=0}^{n-1}\frac{2l+1-i\Delta_{0}T}{2l+1+i\Delta_{0}T}\quad(\alpha=l),$ (45) where the recurrence relation $\Gamma(z+1)=z\Gamma(z)$ AS has been used. Thus, the equation $a=-1$ reduces to an algebraic equation for $\Delta_{0}$, which has $l$ real solutions Vitanov98 . The first few values of $\chi$ and $\Delta_{0}$ for which $a=-1$ are shown in Table 2. As the table shows, $\Delta_{0}=0$ is a solution for odd $\alpha=\frac{1}{2}\chi T$ but not for even $\alpha$, in agreement with the conclusions in Sec. V.2. Moreover, the $a=-1$ solutions do not depend on the number of degenerate states $N$. $\chi T$ | $\Delta_{0}T$ | | | | | ---|---|---|---|---|---|--- 2 | 0 | | | | | 4 | $\pm 1.732$ | | | | | 6 | 0 | $\pm 4.796$ | | | | 8 | $\pm 1.113$ | $\pm 9.207$ | | | | 10 | 0 | $\pm 2.756$ | $\pm 14.913$ | | | 12 | $\pm 0.943$ | $\pm 4.936$ | $\pm 21.903$ | | | 14 | 0 | $\pm 2.243$ | $\pm 7.595$ | $\pm 30.171$ | | 16 | $\pm 0.855$ | $\pm 3.916$ | $\pm 10.708$ | $\pm 39.715$ | | 18 | 0 | $\pm 1.988$ | $\pm 5.907$ | $\pm 14.265$ | $\pm 50.534$ | 20 | $\pm 0.799$ | $\pm 3.418$ | $\pm 8.195$ | $\pm 18.260$ | $\pm 62.627$ | 22 | 0 | $\pm 1.830$ | $\pm 5.098$ | $\pm 10.766$ | $\pm 22.687$ | $\pm 75.993$ 24 | $\pm 0.759$ | $\pm 3.113$ | $\pm 7.006$ | $\pm 13.613$ | $\pm 27.545$ | $\pm 90.634$ 26 | 0 | $\pm 1.719$ | $\pm 4.606$ | $\pm 9.130$ | $\pm 16.729$ | $\pm 32.833$ | | $\pm 106.549$ | | | | 28 | $\pm 0.728$ | $\pm 2.901$ | $\pm 6.289$ | $\pm 11.461$ | $\pm 20.113$ | $\pm 38.548$ | | $\pm 123.736$ | | | | 30 | 0 | $\pm 1.636$ | $\pm 4.268$ | $\pm 8.150$ | $\pm 13.994$ | $\pm 23.760$ | | $\pm 44.690$ | $\pm 142.198$ | | | Table 2: Some approximate solutions of the equation $a(\Delta_{0})=-1$ for the RZ model, where $a$ is given in Table 1, for various even integer values of $\chi T$. In the present context the RZ model is interesting for it shows that one can create superpositions within the ground-state manifold even when the excited state is off resonance by a considerable detuning ($\Delta_{0}\gg 1/T$), for which the transition probability in the MS two-state system is virtually zero, i.e. $\left|a\right|\approx 1$. This fact allows us, for specific detunings, to essentially contain the _transient_ dynamics within the ground states; in contrast, in the resonance case the excited state can get significant transient population, $P_{N+1}(t)=\sin^{2}\frac{1}{2}A(t)$, although it vanishes in the end. Figure 3 displays the populations against the detuning $\Delta_{0}$ for a hyperbolic-secant pulse with $\chi T=18$ for couplings chosen to satisfy Eqs. (30) (upper frame) and (33) (lower frame). In both cases we have $\left|a\right|=1$ [see Eq. (44)], which leaves the excited state unpopulated in the end. For several special values of the detuning $\Delta_{0}$, as predicted in Table 2, we have $a=-1$. For these values, an equal superposition of all degenerate states including the initially populated state $\left|\psi_{1}\right\rangle$ is created in the upper frame, and an equal superposition of all degenerate states except $\left|\psi_{1}\right\rangle$ is created in the lower frame. Figure 3: (Color online) Populations vs the detuning $\Delta_{0}$ for $N=3$ lower states and $\chi T=18$. The coupling strengths $\chi_{n}$ are given by Eqs. (30) in the upper frame and Eqs. (33) in the lower frame. The system is initially in state $\left|\psi_{1}\right\rangle$. Figure 4 shows the final populations versus the rms pulse area $A=\pi\chi T$ for $N=3$ degenerate lower states for couplings chosen to satisfy Eqs. (30) (upper frame) and (33) (lower frame). As follows from Table 2, an equal superposition of all degenerate states is created for rms pulse area $A=18\pi$; this is indeed seen in the figure in the upper frame. For the same value of $A$ in the lower frame an equal superposition is created of all degenerate states except the initially populated state $\left|\psi_{1}\right\rangle$. In both frames, there are other values of $A$ for which the same superpositions are apparently created; a closer examination (not shown) reveals that for these other values of the rms pulse area the created superposition has almost, but not exactly, equal components. Figure 4: (Color online) Final populations versus the rms pulse area $\chi$ for $N=3$ degenerate lower states and detuning $\Delta T=50.534$. The coupling strengths $\chi_{n}$ are given by Eqs. (30) in the upper frame and Eqs. (33) in the lower frame. The system is initially in state $\left|\psi_{1}\right\rangle$. Figure 5 displays the time evolution of the populations for $N=3$ degenerate lower states and rms pulse area of $18\pi$, and two detunings: $\Delta=0$ in the upper frame and $\Delta T=50.534$ in the lower frame. For these pairs of areas and detunings, Figs. 3 and 4 have already demonstrated that an equal superposition of all degenerate states is created. Figure 5 shows that the evolution towards such a superposition can be dramatically different on and off resonance. Indeed, for $\Delta=0$ (upper frame) the nondegenerate upper state receives considerable transient population, which would lead to significant losses if this state can decay on the time scale of the pulsed interaction. In strong contrast, off resonance this undesired population is greatly reduced (lower frame), and still the desired equal superposition of the degenerate states emerges in the end. We have verified numerically that for larger detunings this transient population continues to decrease, e.g. for $\Delta T=142.198$ and $\Omega T=30$ it is less than 1%. Figure 5: (Color online) Populations versus time for $N=3$ lower states and rms Rabi frequency $\chi T=18$. The coupling strengths $\chi_{n}$ are given by Eqs. (30). The detuning is $\Delta=0$ in the upper frame and $\Delta T=50.534$ in the lower frame. The system is initially in state $\left|\psi_{1}\right\rangle$. To conclude this section we point out that one can create any desired superposition, with arbitrary unequal populations, in very much the same manner, on or off resonance, by appropriately chosing the individual couplings, while still maintaning particular values of the overall rms pulse area. Tuning on resonance gives the advantage of smaller pulse area required, whereas tuning off resonance (with larger pulse area) provides the advantage of greatly reducing the transient population of the possibly lossy common upper state. ## VI Multistate Landau-Zener model As seen in Table 1 the propagator parameter $a$ for the LZ model, $a=\exp\left(-\pi\chi^{2}/4C\right)$, cannot be equal to 0 or 1 or $-1$, but may approach 0 or 1 arbitrarily closely. However, it is always positive and cannot approach the value of $-1$; hence the LZ model is unsuitable for unitary operations within the degenerate manifold, in contrast to the resonance and RZ models discussed above. Still, the present multistate LZ solution represents an interesting and important addition to the available LZ solutions (see CI transitions and references therein). #### VI.0.1 The Demkov-Osherov model The present multistate LZ model complements the Demkov-Osherov (DO) model DO , wherein a slanted energy crosses $N$ parallel _nondegenerate_ energies. In the DO model, the exact probabilities $P_{n\rightarrow m}$ have the same form — products of LZ probabilities for transition or no-transition applied at the relevant crossings — as what would be obtained by naive multiplication of LZ probabilities while moving across the grid of crossings from $\left|\psi_{n}\right\rangle$ to $\left|\psi_{m}\right\rangle$, without accounting for phases and interferences. For example, if the states $\left|\psi_{n}\right\rangle$ ($n=1,2,\ldots,N$) are labeled such that their energies increase with the index $n$, and if the slope of the slanted energy of state $\left|\psi_{N+1}\right\rangle$ is positive, the transition probabilities in the DO model are $\displaystyle P_{n\rightarrow m}$ $\displaystyle=$ $\displaystyle p_{n}q_{n+1}q_{n+2}\cdots q_{m-1}p_{m}\quad(n<m),$ (46a) $\displaystyle P_{n\rightarrow m}$ $\displaystyle=$ $\displaystyle 0\quad(n>m),$ (46b) $\displaystyle P_{n\rightarrow n}$ $\displaystyle=$ $\displaystyle q_{n},$ (46c) $\displaystyle P_{n\rightarrow N+1}$ $\displaystyle=$ $\displaystyle p_{n}q_{n+1}q_{n+2}\cdots q_{N},$ (46d) $\displaystyle P_{N+1\rightarrow n}$ $\displaystyle=$ $\displaystyle q_{1}q_{2}\cdots q_{n-1}p_{n},$ (46e) $\displaystyle P_{N+1\rightarrow N+1}$ $\displaystyle=$ $\displaystyle q_{1}q_{2}\cdots q_{N},$ (46f) where $q_{n}=\exp\left(-\pi\chi_{n}^{2}/2C\right)$ is the no-transition probability and $p_{n}=1-q_{n}$ is the transition probability between states $\left|\psi_{N+1}\right\rangle$ and $\left|\psi_{n}\right\rangle$ at the crossing of their energies. #### VI.0.2 The degenerate case The present multistate LZ solution provides the transition probabilities for the special case when all parallel energies are degenerate, which cannot be obtained from the DO model. ##### The propagator The elements of the transition matrix for our $\left(N+1\right)$-state degenerate LZ problem are readily found from Eq. (20) to be $\displaystyle U_{m,n}=-\frac{\chi_{n}\chi_{m}}{\chi^{2}}\left(1-e^{-\Lambda}\right)\quad\left(m,n=1,\ldots,N;\text{ }m\neq n\right),$ (47a) $\displaystyle U_{n,n}=1-\frac{\chi_{n}^{2}}{\chi^{2}}\left(1-e^{-\Lambda}\right)\quad\left(n=1,\ldots,N\right),$ (47b) $\displaystyle U_{n,N+1}=\frac{\chi_{n}}{\chi}b\quad\left(n=1,\ldots,N\right),$ (47c) $\displaystyle U_{N+1,n}=-\frac{\chi_{n}}{\chi}b^{\ast}\quad\left(n=1,\ldots,N\right),$ (47d) $\displaystyle U_{N+1,N+1}=e^{-\Lambda},$ (47e) with $\Lambda=\pi\chi^{2}/4C$ and $\left|b\right|^{2}=1-e^{-2\Lambda}$. ##### System initially in the nondegenerate state When the system begins initially in the nondegenerate state $\left|\psi_{N+1}\right\rangle$, with the tilted energy, the system ends in a coherent superposition of all states with populations $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{\chi_{n}^{2}}{\chi^{2}}\left(1-e^{-2\Lambda}\right)\quad\left(n=1,\ldots,N\right),$ (48a) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle e^{-2\Lambda}.$ (48b) In the adiabatic limit $\Lambda\gg 1$ the population is distributed among the degenerate states according to their couplings, whereas the initially populated state $\left|\psi_{N+1}\right\rangle$ is almost depleted, $P_{N+1}\approx 0$. For equal couplings, all degenerate-state populations will be equal, $P_{n}\approx 1/N$. In the opposite, diabatic limit $\Lambda\ll 1$ the population remains in state $\left|\psi_{N+1}\right\rangle$ with almost no population in the degenerate states. ##### System initially in a degenerate state When the system is initially in an arbitrary degenerate state $\left|\psi_{i}\right\rangle$, at the end of the evolution the populations are $\displaystyle P_{i}$ $\displaystyle=$ $\displaystyle\left[1-\frac{\chi_{i}^{2}}{\chi^{2}}\left(1-e^{-\Lambda}\right)\right]^{2},$ (49a) $\displaystyle P_{n}$ $\displaystyle=$ $\displaystyle\frac{\chi_{n}^{2}\chi_{i}^{2}}{\chi^{4}}\left(1-e^{-\Lambda}\right)^{2}\quad\left(n=1,\ldots,N;n\neq i\right),$ (49b) $\displaystyle P_{N+1}$ $\displaystyle=$ $\displaystyle\frac{\chi_{i}^{2}}{\chi^{2}}\left(1-e^{-2\Lambda}\right).$ (49c) In the adiabatic limit $\Lambda\gg 1$ and for equal couplings, the populations will be $\displaystyle P_{i}$ $\displaystyle\approx$ $\displaystyle\left(1-\frac{1}{N}\right)^{2},$ (50a) $\displaystyle P_{n}$ $\displaystyle\approx$ $\displaystyle\frac{1}{N^{2}}\quad\left(n=1,\ldots,N;n\neq i\right),$ (50b) $\displaystyle P_{N+1}$ $\displaystyle\approx$ $\displaystyle\frac{1}{N}.$ (50c) Obviously, Eqs. (48) and (49) cannot be reduced to the DO solution (46), which implies that the non-degeneracy assumption in the DO model is essential. Figure 6: (Color online) Populations for the degenerate LZ model vs the LZ parameter $\Lambda=\pi\chi^{2}/4C$ for $N=3$ degenerate states and equal couplings. The system is supposed to start in one of the degenerate states $\left|\psi_{i}\right\rangle$. The arrows on the right point the adiabatic values (50). Figure 6 shows the transition probability for the multistate LZ model plotted against the LZ parameter $\Lambda=\pi\chi^{2}/4C$. As $\Lambda$ increases the populations approach their steady adiabatic values (50). Different coherent superpositions can be created by choosing appropriate values for the couplings $\chi_{n}$. ## VII Conclusions In this paper we have described a procedure for deriving analytical solutions for a multistate system composed of $N$ degenerate lower states coupled via a nondegenerate upper state with pulsed interactions of the same temporal dependence but possibly with different peak amplitudes. The multistate resonance and Rosen-Zener solutions have been discussed in some detail because they allow one to find special values of parameters, termed generalised $\pi$ pulses, for which various types of population transfer can occur, for example, creation of maximally coherent superpositions. The RZ solution is particularly useful because it allows to prescribe appropriately detuned pulsed fields for which the dynamics can be essentially contained within the degenerate-state space, without populating the upper state even transiently, thus avoiding possible losses from this state via spontaneous emission, ionization, etc. We have analyzed in some detail also the multistate Landau-Zener model, which complements the Demkov-Osherov model in the case of degenerate energies. The presented analytical solutions and general properties have a significant potential for manipulation of multistate quantum bits in quantum information processing, for example, in designing arbitrary unitary gates. ###### Acknowledgements. This work has been supported by the European Union’s Transfer of Knowledge project CAMEL (Grant No. MTKD-CT-2004-014427) and the Alexander von Humboldt Foundation. ESK acknowledges support from the EU Marie Curie Training Site project No. HPMT-CT-2001-00294. ## References * (1) B.W. Shore, _The Theory of Coherent Atomic Excitation_ (Wiley, New York, 1990). * (2) I.I. Rabi, Phys. Rev. 51, 652 (1937). * (3) L.D. Landau, Physik Z. Sowjetunion 2, 46 (1932); C. Zener, Proc. R. Soc. Lond. Ser. A 137, 696 (1932). * (4) N. Rosen and C. Zener, Phys. Rev. 40, 502 (1932). * (5) L. Allen and J. H. Eberly, _Optical Resonance and Two-Level Atoms_ (Dover, New York, 1987); F.T. Hioe, Phys. Rev. A 30, 2100 (1984). * (6) A. Bambini and P.R. Berman, Phys. Rev. A 23, 2496 (1981). * (7) Yu.N. Demkov and M. Kunike, Vestn. Leningr. Univ. Fiz. Khim. 16, 39 (1969); see also F.T. Hioe and C.E. Carroll, Phys. Rev. A 32, 1541 (1985); J. Zakrzewski, Phys. Rev. A 32, 3748 (1985); K.-A. Suominen and B.M. Garraway, Phys. Rev. A 45, 374 (1992). * (8) C.E. Carroll and F.T. Hioe, J. Phys. A: Math. Gen. 19, 3579 (1986). * (9) Yu.N. Demkov, Sov.Phys.-JETP 18, 138 (1964); N.V. Vitanov, J. Phys. B 26, L53 (1993), erratum _ibid._ 26, 2085 (1993). * (10) E.E. Nikitin, Opt. Spectrosc. 13, 431 (1962); Discuss. Faraday Soc. 33, 14 (1962); Adv. Quantum Chem. 5, 135 (1970); N.V. Vitanov, J. Phys. B 27, 1791 (1994). * (11) J.R. Morris and B.W. Shore, Phys. Rev. A 27, 906 (1983). * (12) C.P. Williams and S.H. Clearwater, _Explorations in Quantum Computing_ , (Springer-Verlag, Berlin, 1997); A. Steane, Rep. Prog. Phys. 61, 117 (1998); M.A. Nielsen and I.L. Chuang, _Quantum Computation and Quantum Information_ (Cambridge University Press, Cambridge, 2000). * (13) R.G. Unanyan, M. Fleischhauer, B.W. Shore, and K. Bergmann, Opt. Commun. 155, 144 (1998); H. Theuer, R.G. Unanyan, C. Habscheid, K. Klein and K. Bergmann, Optics Express 4, 77 (1999). * (14) Z. Kis and S. Stenholm, Phys. Rev. A 64, 63406 (2001). * (15) K. Bergmann, H. Theuer, and B.W. Shore, Rev. Mod. Phys. 70, 1003 (1998); N.V. Vitanov, T. Halfmann, B.W. Shore, and K. Bergmann, Ann. Rev. Phys. Chem. 52, 763 (2001); N.V. Vitanov, M. Fleischhauer, B.W. Shore and K. Bergmann, Adv. At. Mol. Opt. Phys. 46, 55 (2001). * (16) S. Stenholm, in _Frontiers of Laser Spectroscopy_ , Les Houches Summer School Session XXVII, edited by R. Bailian, S. Haroche and S. Liberman (Amsterdam, North Holland, 1975), p. 399; R. Lefebvre and J. Savolainen, J. Chem. Phys. 60, 2509 (1974); M. Bixon and J. Jortner, J. Chem. Phys. 48, 715 (1968). * (17) N.V. Vitanov, J. Phys. B 33, 2333 (2000). * (18) M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions (Dover, New York, 1964). * (19) A.A. Rangelov, J. Piilo, and N.V. Vitanov, Phys. Rev. A 72, 053404 (2005). * (20) Y.N. Demkov and V.I. Osherov, Zh. Eksp. Teor. Fiz. 53, 1589 (1967) [Sov. Phys. JETP 26, 916 (1968)]; Y.N. Demkov and V.N. Ostrovsky, J. Phys. B 28, 403 (1995).

arxiv-papers

2024-09-04T02:48:54.042024

0802.4331

# A note on evaluations of multiple zeta values Shuichi Muneta ###### Abstract Multiple zeta values (MZVs) with certain repeated arguments or certain sums of cyclically generated MZVs are evaluated as rational multiple of powers of $\pi^{2}$. In this paper, we give a short and simple proof of the remarkable evaluations of MZVs established by D. Borman and D. M. Bradley. ## 1 Introduction The multiple zeta value (MZV) is defined by the convergent series $\zeta(k_{1},k_{2},\ldots,k_{n}):=\sum_{m_{1}>m_{2}>\cdots>m_{n}>0}\frac{1}{m_{1}^{k_{1}}m_{2}^{k_{2}}\cdots m_{n}^{k_{n}}},$ where $k_{1},k_{2},\ldots,k_{n}$ are positive integers and $k_{1}\geq 2$. The remarkable property of MZVs is that MZVs are evaluated for some special arguments as rational multiple of powers of $\pi^{2}$. For example, the following evaluations were proven by many authors ([BBB], [H1], [Z]): $\zeta(\\{2\\}_{m})=\frac{\pi^{2m}}{(2m+1)!}\quad(m\in\mathbb{Z}_{>0})$ where $\\{2\\}_{m}$ denotes the $m$-tuple $(2,2,\ldots,2)$. In [Z], D. Zagier conjectured the following evaluations: $\zeta(\\{3,1\\}_{n})=\frac{2\pi^{4n}}{(4n+2)!}\quad(n\in\mathbb{Z}_{>0}).$ These evaluations were proved by J. M. Borwein, D. M. Bradley, D. J. Broadhurst and P. Lison$\mathrm{\check{e}}$k ([BBBL1], [BBBL2]). In addition, D. Bowman and D. M. Bradley proved the following theorem which contained these results: ###### Theorem 1 ([BB]). For non-negative integers $m$, $n$, we have $\displaystyle\sum_{\begin{subarray}{c}j_{0}+j_{1}+\cdots+j_{2n}=m\\\ j_{0},j_{1},\ldots,j_{2n}\geq 0\end{subarray}}\zeta(\\{2\\}_{j_{0}},3,\\{2\\}_{j_{1}},1,\\{2\\}_{j_{2}},\ldots,\\{2\\}_{j_{2n-2}},3,\\{2\\}_{j_{2n-1}},1,\\{2\\}_{j_{2n}})$ $\displaystyle\qquad\qquad=\binom{m+2n}{m}\frac{\pi^{2m+4n}}{(2n+1)\cdot(2m+4n+1)!}.\qquad\qquad\qquad\qquad$ In this article, we provide a short and simple proof of Theorem 1 which refines the proof of Theorem 5.1 in [BB]. ## 2 Algebraic setup We summarize the algebraic setup of MZVs introduced by Hoffman (cf. [H2], [IKZ]). Let $\mathfrak{H}=\mathbb{Q}\left\langle x,y\right\rangle$ be the noncommutative polynomial ring in two indeterminates $x$, $y$ and $\mathfrak{H}^{1}$ and $\mathfrak{H}^{0}$ its subrings $\mathbb{Q}+\mathfrak{H}y$ and $\mathbb{Q}+x\mathfrak{H}y$. We set $z_{k}=x^{k-1}y$ $(k=1,2,3,\ldots)$. Then $\mathfrak{H}^{1}$ is freely generated by $\\{z_{k}\\}_{k\geq 1}$. We define the $\mathbb{Q}$-linear map (called evaluation map) $Z:\mathfrak{H}^{0}\longrightarrow\mathbb{R}$ by $Z(1)=1\;\;\mathrm{and}\;\;Z(z_{k_{1}}z_{k_{2}}\cdots z_{k_{n}})=\zeta(k_{1},k_{2},\ldots,k_{n}).$ We next define the shuffle product sh on $\mathfrak{H}$ inductively by $\displaystyle 1\xrm\mbox{sh}\,\xiirm w$ $\displaystyle=$ $\displaystyle w\xrm\mbox{sh}\,\xiirm 1\;=\;w,$ $\displaystyle u_{1}w_{1}\xrm\mbox{sh}\,\xiirm u_{2}w_{2}$ $\displaystyle=$ $\displaystyle u_{1}(w_{1}\xrm\mbox{sh}\,\xiirm u_{2}w_{2})+u_{2}(u_{1}w_{1}\xrm\mbox{sh}\,\xiirm w_{2})$ ($u_{1},u_{2}\in\\{x,y\\}$ and $w$, $w_{1}$, $w_{2}$ are words in $\mathfrak{H}$), together with $\mathbb{Q}$-bilinearity. The shuffle product sh is commutative and associative. For this product, we have $Z(w_{1}\xrm\mbox{sh}\,\xiirm w_{2})=Z(w_{1})Z(w_{2})$ for any $w_{1},w_{2}\in\mathfrak{H}^{0}$. We also define the shuffle product $\,\widetilde{\xrm\mbox{sh}\,\xiirm}$ on $\mathbb{Q}\left\langle z_{1},z_{2},\ldots\right\rangle$ inductively by $\displaystyle 1\,\widetilde{\xrm\mbox{sh}\,\xiirm}w$ $\displaystyle=$ $\displaystyle w\,\widetilde{\xrm\mbox{sh}\,\xiirm}1\;=\;w,$ $\displaystyle u_{1}w_{1}\,\widetilde{\xrm\mbox{sh}\,\xiirm}u_{2}w_{2}$ $\displaystyle=$ $\displaystyle u_{1}(w_{1}\,\widetilde{\xrm\mbox{sh}\,\xiirm}u_{2}w_{2})+u_{2}(u_{1}w_{1}\,\widetilde{\xrm\mbox{sh}\,\xiirm}w_{2})$ ($u_{1},u_{2}\in\\{z_{k}\\}_{k\geq 1}$ and $w$, $w_{1}$, $w_{2}$ are words in $\mathbb{Q}\left\langle z_{1},z_{2},\ldots\right\rangle$), together with $\mathbb{Q}$-bilinearity. For example, we have $\displaystyle z_{m}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{n}$ $\displaystyle=z_{m}z_{n}+z_{n}z_{m},$ $\displaystyle z_{m}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{n}z_{l}$ $\displaystyle=z_{m}z_{n}z_{l}+z_{n}z_{m}z_{l}+z_{n}z_{l}z_{m}.$ Then Theorem 1 can be restated as follows: $Z\left(z_{2}^{m}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{n}\right)=\binom{m+2n}{m}\frac{\pi^{2m+4n}}{(2n+1)\cdot(2m+4n+1)!}\quad\big{(}m,n\in\mathbb{Z}_{\geq 0}\big{)}.$ ## 3 Proof of Theorem 1 We restate Proposition 4.1 and Proposition 4.2 of [BB] by using $\,\widetilde{\xrm\mbox{sh}\,\xiirm}$ and prove them by induction. ###### Proposition 2. For integers $n$, $N$ which satisfy $0\leq n\leq N$, we have $\displaystyle z_{2}^{n}\xrm\mbox{sh}\,\xiirm z_{2}^{N}$ $\displaystyle=\sum_{k=0}^{n}4^{k}\binom{N+n-2k}{n-k}\left\\{z_{2}^{N+n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\right\\},$ (1) $\displaystyle z_{1}z_{2}^{n}\xrm\mbox{sh}\,\xiirm z_{1}z_{2}^{N}$ $\displaystyle=2\sum_{k=0}^{n}4^{k}\binom{N+n-2k}{n-k}z_{1}\left\\{z_{2}^{N+n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{1}(z_{3}z_{1})^{k}\right\\}.$ (2) ###### Proof. We prove identities (1) and (2) simultaneously by induction on $n$. [Step 1] The case $n=0$ of (1) is clear. We can easily prove the case $n=0$ of (2) by induction on $N$. [Step 2] Suppose that (1) and (2) have been proven for $n-1$. We prove (1) for $n$ by induction on $N$. $\displaystyle z_{2}^{n}\xrm\mbox{sh}\,\xiirm z_{2}^{n}$ $\displaystyle=2xy\\{(xy)^{n-1}\xrm\mbox{sh}\,\xiirm(xy)^{n}\\}+2x^{2}\\{y(xy)^{n-1}\xrm\mbox{sh}\,\xiirm y(xy)^{n-1}\\}$ $\displaystyle=2\sum_{k=0}^{n-1}4^{k}\binom{2n-1-2k}{n-1-k}z_{2}\\{z_{2}^{2n-1-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+\sum_{k=0}^{n-1}4^{k+1}\binom{2n-2-2k}{n-1-k}z_{3}\\{z_{2}^{2n-2-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{1}(z_{3}z_{1})^{k}\\}$ $\displaystyle=\sum_{k=0}^{n-1}4^{k}\binom{2n-2k}{n-k}z_{2}\\{z_{2}^{2n-1-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+\sum_{k=1}^{n}4^{k}\binom{2n-2k}{n-k}z_{3}\\{z_{2}^{2n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{1}(z_{3}z_{1})^{k-1}\\}$ $\displaystyle=\binom{2n}{n}z_{2}^{2n}+\sum_{k=1}^{n-1}4^{k}\binom{2n-2k}{n-k}\\{z_{2}^{2n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}+4^{n}(z_{3}z_{1})^{n}$ $\displaystyle=\sum_{k=0}^{n}4^{k}\binom{2n-2k}{n-k}\\{z_{2}^{2n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}.$ Hence (1) is true for $N=n$. Suppose that the case $N-1$ of (1) has been proven. (We may assume that $N-1\geq n$ in the following calculation.) $\displaystyle z_{2}^{n}\xrm\mbox{sh}\,\xiirm z_{2}^{N}$ $\displaystyle=xy\\{(xy)^{n-1}\xrm\mbox{sh}\,\xiirm(xy)^{N}\\}+2x^{2}\\{y(xy)^{n-1}\xrm\mbox{sh}\,\xiirm y(xy)^{N-1}\\}$ $\displaystyle\quad+xy\\{(xy)^{n}\xrm\mbox{sh}\,\xiirm(xy)^{N-1}\\}$ $\displaystyle=\sum_{k=0}^{n-1}4^{k}\binom{N+n-1-2k}{n-1-k}z_{2}\\{z_{2}^{N+n-1-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+\sum_{k=0}^{n-1}4^{k+1}\binom{N+n-2-2k}{n-1-k}z_{3}\\{z_{2}^{N+n-2-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{1}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+\sum_{k=0}^{n}4^{k}\binom{N+n-1-2k}{n-k}z_{2}\\{z_{2}^{N+n-1-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle=\sum_{k=0}^{n-1}4^{k}\binom{N+n-2k}{n-k}z_{2}\\{z_{2}^{N+n-1-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+\sum_{k=1}^{n}4^{k}\binom{N+n-2k}{n-k}z_{3}\\{z_{2}^{N+n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}z_{1}(z_{3}z_{1})^{k-1}\\}$ $\displaystyle\quad+4^{n}z_{2}\\{z_{2}^{N-n-1}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{n}\\}$ $\displaystyle=\binom{N+n}{n}z_{2}^{N+n}+\sum_{k=1}^{n-1}4^{k}\binom{N+n-2k}{n-k}\\{z_{2}^{N+n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}$ $\displaystyle\quad+4^{n}\\{z_{2}^{N-n}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{n}\\}$ $\displaystyle=\sum_{k=0}^{n}4^{k}\binom{N+n-2k}{n-k}\\{z_{2}^{N+n-2k}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{k}\\}.$ Hence (1) is true for $N$. We can prove (2) for $n$ by induction on $N$ with using (1) for $n$. ∎ Before proceeding the proof of Theorem 1, we prove a key identity. Comparing coefficients of $(x+1)^{2m+4n+2}=(x^{2}+2x+1)^{m+2n+1}$, we have $\binom{2m+4n+2}{2n+1}=\sum_{k=0}^{n}2^{2k+1}\frac{(m+2n+1)!}{(n-k)!(2k+1)!(m+n-k)!}.$ We can transform this identity as follows: $\displaystyle\frac{1}{(2n+1)!}\frac{1}{(2m+2n+1)!}$ $\displaystyle=\sum_{k=0}^{n}4^{k}\binom{m+2n-2k}{n-k}\binom{m+2n}{2k}\frac{1}{(2k+1)\cdot(2m+4n+1)!}.$ (3) ###### Proof of Theorem 1. We prove Theorem 1 by induction on $n$. The case $n=0$ is well known as has been mentioned in Section 1. Suppose that the assertion has been proven up to $n-1$. Putting $N=m+n$ in (1), we have $\displaystyle 4^{n}Z\left(z_{2}^{m}\,\widetilde{\xrm\mbox{sh}\,\xiirm}(z_{3}z_{1})^{n}\right)$ $\displaystyle=\frac{\pi^{2n}}{(2n+1)!}\frac{\pi^{2m+2n}}{(2m+2n+1)!}-\sum_{k=0}^{n-1}4^{k}\binom{m+2n-2k}{n-k}\binom{m+2n}{2k}\frac{\pi^{2m+4n}}{(2k+1)\cdot(2m+4n+1)!}$ $\displaystyle\stackrel{{\scriptstyle(\ref{eq:3})}}{{=}}4^{n}\binom{m+2n}{m}\frac{\pi^{2m+4n}}{(2n+1)\cdot(2m+4n+1)!}.$ This completes the proof of Theorem 1. ∎ ## References * [BB] D. Bowman, D. Bradley, The algebra and combinatorics of shuffles and multiple zeta values, J. Combin. Theory Ser. A 97 (2002), 43–61. * [BBB] J. M. Borwein, D. M. Bradley, and D. J. Broadhurst, Evaluations of k-fold Euler/Zagier sums: A compendium of results for arbitrary k, Electron. J. Combin. 4, No. 2 (1997). * [BBBL1] J. M. Borwein, D. M. Bradley, D. J. Broadhurst and P. Lison$\mathrm{\check{e}}$k, Combinatorial aspects of multiple zeta values, Electron. J. Combin. $5$, No. 1 (1998). * [BBBL2] J. M. Borwein, D. M. Bradley, D. J. Broadhurst and P. Lison$\mathrm{\check{e}}$k, Special values of multiple polylogarithm, Trans. Amer. Math. Soc. $353$, No.3 (2001), 907–941. * [IKZ] K. Ihara, M. Kaneko, D. Zagier, Derivation and double shuffle relations for multiple zeta values, Compos. Math. 142 (2006), 307–338. * [H1] M. Hoffman, Multiple harmonic series, Pacific J. Math. 152 (1992), 275–290. * [H2] M. Hoffman, The algebra of multiple harmonic series, J. Algebra 194 (1997), 477–495. * [Z] D. Zagier, Values of zeta functions and their applications, First European Congress of Mathematics, Vol. II, Birkh$\mathrm{\ddot{a}}$user, Boston, 1994, pp. 497–512. Graduate School of Mathematics, Kyushu University Fukuoka 812-8581, Japan E-mail address: muneta@math.kyushu-u.ac.jp

arxiv-papers

2024-09-04T02:48:54.048052

0802.4427

# Numerical Relativity meets Data Analysis: Spinning Binary Black Hole Case Deirdre Shoemaker, Birjoo Vaishnav, Ian Hinder and Frank Herrmann Center for Gravitational Wave Physics, Penn State, University Park, PA 16802 ###### Abstract We present a study of the gravitational waveforms from a series of spinning, equal-mass black hole binaries focusing on the harmonic content of the waves and the contribution of the individual harmonics to the signal-to-noise ratio. The gravitational waves were produced from two series of evolutions with black holes of initial spins equal in magnitude and anti-aligned with each other. In one series the magnitude of the spin is varied; while in the second, the initial angle between the black-hole spins and the orbital angular momentum varies. We also conduct a preliminary investigation into using these waveforms as templates for detecting spinning binary black holes. Since these runs are relativity short, containing about two to three orbits, merger and ringdown, we limit our study to systems of total mass $\geq 50M_{\odot}$. This choice ensures that our waveforms are present in the ground-based detector band without needing addition gravitational wave cycles. We find that while the mode contribution to the signal-to-noise ratio varies with the initial angle, the total mass of the system caused greater variations in the match. ## 1 Introduction Gravitational waves produced during the coalescence of compact objects such as black holes, are one of the most promising sources for detection by interferometric detectors such as LIGO, VIRGO, GEO and TAMA [1]. For these ground based-detectors, where any gravitational-wave signals may be deeply buried in detector noise, the matched filtering technique [2] is the detection strategy of choice. Matched filtering is optimal when accurate representations of the expected signal are used. For low mass compact object binaries that means templates built from well-known analytic methods such as the post- Newtonian (PN) approximation [3]. When the total mass is larger than approximately $50M_{\odot}$ [4], the merger of the binary black hole (BBH) system will not only be present in the sensitivity band of ground-based detectors but also generate the strongest signal. Quantifying the last orbits and merger of a BBH system has long been the purview of numerical relativity. A new direction in the efforts to generate accurate templates is the inclusion of waveforms produced by numerical relativity. The waveforms from numerical relativity may be used in data analysis methods involving searches for inspiraling sources in several ways. For instance, they can be used in creating hybrid templates for detection, validating and extending the mass range of the current search strategy and may be implemented as templates themselves especially for the higher mass range [5, 6, 7, 8, 9, 10, 11]. In a previous paper [9] (paper I), we studied waveforms from a series of numerical evolutions of equal-mass, spinning BBH mergers, which we label the A-series. In the A-series, the black holes had spins that were equal in magnitude with one spin aligned and the other anti-aligned with the orbital angular momentum, $J_{orb}$. We found that using only the dominant harmonic of the radiation, a common practice in numerical relativity and data analysis, caused a complete degeneracy of the this spin space with respect to detection. In other words, a template of a non-spinning BBH waveform faithfully matched the signal of any anti-aligned spinning waveform when only the dominant modes were used. However, retaining non-dominant harmonics of non-negligible amplitude broke the degeneracy to some degree, most notably for spins $J/M^{2}\geq 0.6$. In this paper, we further our previous study in two ways. First we include a new series of BBH mergers, the B-series, in which the initial black-hole spins are still equal in magnitude and anti-aligned with each other but the initial angle the black-hole spins make with the orbital angular momentum, $\vartheta$, is allowed to vary. Second, for both the A and B series of data we conduct a new study of the contributions of the individual modes to the signal-to-noise ratio. In addition, for the B-series we also calculate the dependence of the minimax matches between the dominant mode and the full waveform on the initial spin orientation. ## 2 The Waveforms The binary black hole coalescence problem can be divided into three phases called the inspiral, merger and ringdown. Due to the early accessibility of PN and other analytic based approaches, most of the work in setting up detection schemes has been done with these analytic approaches. These methods are well suited for the inspiral phase of the coalescence, only breaking down at some yet-to-be-determined point within a few orbits of the merger. Fortunately, the signal resulting from the inspiral is in the ground-based frequency band for systems of total mass less than approximately $50M_{\odot}$. As the mass increases, the inspiral lowers in frequency, and the detectable signal contains more merger and ringdown. Now that numerical relativity is producing the waveforms for the final orbits and the merger phase of the coalescence [12, 13, 14] we can investigate the inclusion of the merger regime in detection strategies. We study two sets of waveforms both the result of evolutions conducted by the PSU numerical relativity group. The A-series was published in [15] with initial black-hole spins covering the set $a={0.0,0.2,0.4,0.6,0.8}$. The B-series was published in [16] and generalizes the A-series with variation of the initial angle that one of the anti-aligned spins makes with the axis, $\vartheta$, at a fixed magnitude of spins, $a=0.6$. When $\vartheta=0$, we recover the a=0.6 waveform of A-series. When $\vartheta=\pi/2$, the spin- directions lie in the plane of the orbit and are in the“superkick” [17] configuration in which the maximum gravitational recoil from the BBH mergers has been found. Since these waveforms were originally produced to study the gravitational recoil imparted to the final black hole after an asymmetric collision, only two to three orbits were evolved (the merger phase dominates the recoil). The number of orbits is set by the initial orbital frequency for a given total mass. In order to place the numerical waveforms firmly in the frequency band of the detector, we use the initial LIGO noise curve [18] and we only investigate masses larger than $50M_{\odot}$ when calculating matches between waveforms. The total mass sets the frequency at which the signal enters the band. For example, the cutoff frequency for a binary system of $50M_{\odot}$ is $0.02/M$ or approximately $80$Hz and about $40$Hz for $100M_{\odot}$. The waveforms were extracted from the numerical evolution of the spacetime in terms of the Newman-Penrose scalar, $\Psi_{4}(t,x,y,z)$, which is expanded into angular modes via ${}_{-2}Y_{\ell m}(\theta,\phi)$, the spin-weighted $s=-2$ spherical harmonics, by extraction on a sphere. The dominant mode for the quasi-circular orbits is the quadrupole mode ($\ell=m=2$). The angles $\theta$ and $\phi$ correspond to the inclination and azimuthal angles between the source and detector in the source frame. When $\theta=0$, the observer is directly above the orbital plane of the binary and sees primarily the $\ell=m=2$ mode. As $\theta$ increases, the waveforms are a mixture of modes. We truncate the infinite series of modes at $\ell\leq 4$ because we do not extract all the modes from the simulations and modes of $\ell>4$ were zero within our numerical error. As more complicated configurations are evolved, more modes will need to be accurately extracted from the codes. ## 3 Faithfulness The multipolar analysis of BBH waveforms produced by numerical relativity has been pursued for both unequal mass and spinning BBH configurations [19, 20, 21]. In paper I, we found that using only the dominant mode in comparing waveforms, tantamount to choosing an inclination angle with the detector in the source frame of $\theta=0$, resulted in a degeneracy of the A-series parameter space. In this paper, we focus our attention on how different initial configurations, in this case $a$ and $\vartheta$, result in different mode contributions to the signal-to-noise ratio. To build intuition about what parameters might be important to the template space of black-hole mergers, we further calculate the overlap between pairs of our waveforms. We perform a preliminary analysis on how faithful $\ell=m=2$ waveforms of various parameters would be in matching with waveforms at random inclination angle. We keep the total mass for each template fixed and vary the inclination angle of the detector, $\theta$, the spin $a$, and initial angle $\vartheta$ when appropriate. The minimax match is given by [22, 23] $M\equiv\max_{t_{0}}\min_{\Phi}\frac{\langle h_{1}|h_{2}\rangle}{\sqrt{\langle h_{1}|h_{1}\rangle\langle h_{2}|h_{2}\rangle}}\,,$ (1) where $\langle h_{1}|h_{2}\rangle=4\,\mbox{Re}\int_{f_{\mathrm{min}}}^{f_{\mathrm{max}}}\frac{\tilde{h}_{1}(f)\tilde{h}^{*}_{2}(f)}{S_{h}(f)}\,df.$ (2) The Fourier transform of the strain, $h(t)$, is given by $\tilde{h}_{+}(f)=\mathcal{F}(\mathrm{Re}(\Psi_{4}))(f)/(-4\pi^{2}f^{2})\,$ where $\Psi_{4}(t)=\frac{d^{2}}{dt^{2}}(h_{+}(t)-ih_{\times}(t))\,,$ and $\mathcal{F}$ is a Fast Fourier Transform. The signal-to-noise ratio, $\rho$, is given by $\rho=\left[4\int_{f_{\mathrm{min}}}^{f_{\mathrm{max}}}\frac{|\tilde{h}(f)|^{2}}{S_{h}(f)}\right]^{1/2}\,.$ (3) The variable $S_{h}(f)$ denotes the noise spectrum for which we use the initial LIGO noise curve. The domain $[f_{\mathrm{min}},f_{\mathrm{max}}]$ is determined by the detector bandwidth and the masses of our signal. The masses are set such that the overlaps will not change significantly if we were to add the inspiral portion of the signal because the A and B series of waveforms have orbital frequencies that increase almost monotonically. Owing to this, the gravitational wave frequency also increases monotonically with time implying that extending the signal back in time will not change the spectrum in the merger band. When precession is significant this will no longer be true and the inspiral will likely contribute to the signal at higher frequencies. A-series: In Fig. 1 we plot the match versus spin at different inclination angles, $\theta$, for a given total mass of $100M_{\odot}$. This plot first appeared in paper I and is included here for reference. The figure shows that as the spin increases, the match between a waveform of $\theta=0$ and one of non-zero $\theta$ decreases. This indicates that the non-dominant modes are important both for distinguishing between different spinning waveforms and in making a detection. This is most notable for the $a=0.8$ case. Figure 1: The minimax match versus $a$ for three values of $\theta$, ${\pi/4,\pi/3,\pi/2}$ for the case of total mass $100M_{\odot}$. The value of $\theta=0$ is not included in the plot, since this is the waveform of comparison and the match is one by definition. In order to have a qualitative understanding of why the match behaves as in Fig. 1, we study the $\rho$ per mode for the A-series. This is only a qualitative estimate because the relative fraction of modes present in the signal will depend on the relative spin-weighted spherical harmonics values at the particular angle. In practice the error induced by ignoring the mixed terms that are important in constructing $\rho(\theta,\phi)$ from the modes is less than 20%, as the relative overlaps of the significant modes from both the A and B-series of data are of this order. Since the $\rho$ of the $\ell=m=2$ mode is much larger than the $\rho$ of the other modes, we plot the ratio, $\rho(\ell,m)/\rho(2,2)$ in Fig. 2. The upper left plot corresponds to a system of mass $50M_{\odot}$, the upper right to $100M_{\odot}$, lower left to $200M_{\odot}$ and lower right to $300M_{\odot}$. Figure 2: A series of plots are shown with the ratio of $\rho$ per mode to the $\rho$ of the $\ell=m=2$ mode versus the initial spin of the black holes. This is done for series-A. Each plot refers to the calculation for a different total mass of the binary. Starting from the upper left and moving right and then down, we have $50M_{\odot}$, $100M_{\odot}$, $200M_{\odot}$, and $300M_{\odot}$ on the lower right. Across all the masses sampled, the ratio of the $\rho$ for each mode grows with increasing spin. This is especially true for the $m=1$ and $m=3$ modes which are suppressed at low $a$. The $m=2$ and $m=4$ increase slightly with $a$. While the $\ell=2$, $m=1$ mode is the next mode dominant mode after the $\ell=m=2$ mode for the high-spin regime at low masses, the $\ell=m=4$ mode is secondary for the entire $a$ range at higher masses. For low spins, the waveform is entirely dominated by the $m=2$ modes. The linear-like growth of the odd-$m$ modes with spin magnitude is expected from PN expressions like Eq(1)-(4) in [21]. B-series: The minimax match versus the initial angle for the B-series is presented in Fig. 3. Each line represents a choice of total mass, with $50M_{\odot}$ the top most line and $300M_{\odot}$ the bottommost. The match was computed by setting one waveform to $\theta=0$ and the other to $\theta=\pi/2.4$. Figure 3: The minimax match versus initial angle $\vartheta$ for the B-series. Each curve represents a particular choice of total mass, $50M_{\odot}$ at the top with each successively lower line a higher mass. To compute the match, we used one waveform with $\theta=0$ and the other at $\theta=\pi/2.4$ radians for a given $a$ and $\vartheta$. We find that the variation of the match across initial angle for the given spin of $a=0.6$ does not change more than about $2\%$. The variation amongst different total masses is more dramatic, dropping down below a match of 0.9 for most in the angles at a mass of $300M_{\odot}$. At that large mass range, the ringdown is contributing significantly to the signal, and differences in the modes, like the $\ell=m=4$ mode, begin to make important contributions. These BBH configurations settle down to a final black hole with a spin of $a=0.62$. In Fig. 4, we once again investigate a qualitative interpretation of the match through the $\rho$ as plotted versus the initial angle, $\vartheta$ for each mode. The upper left plot corresponds to a system of mass $50M_{\odot}$, the upper right to $100M_{\odot}$, lower left to $200M_{\odot}$ and lower right to $300M_{\odot}$. Figure 4: A set of plots is shown with the ratio of $\rho$ per mode to the $\rho$ of the $\ell=m=2$ mode versus the initial angle for series-B. Each plot is the ratio computed for a different total mass of the binary. Starting from the upper left and moving right and then lower left and right, we have $50M_{\odot}$, $100M_{\odot}$, $200M_{\odot}$, and $300M_{\odot}$ on the lower right. Across the mass scales sampled, as $\vartheta$ increases, the signal in odd-$m$ modes decrease. In the non-precessing case, the strength of the odd- modes is expected to vary with the z-component of the spin [21]. Since in this series of runs, the spins precess about the z-axis and $\vartheta$ remains nearly constant, the relative strength of the modes show similar trends as the non-precessing case. At higher masses, where the ringdown dominates the signal, the $\ell=m=4$ mode contributes a large portion of the $\rho$ of the total signal, over 20% for a BBH of $300M_{\odot}$. This is in part due to the enhancement of the ringdown signal which occurs when the frequency of the higher modes lies around the detector’s sweet spot. It is interesting to note that in the ”superkick” the spread of the modes is reduced in $\rho$ compared with the parallel configuration at $\vartheta=0$. The decrease of the odd $m$ modes is expected from the PN expressions. For example, the $\ell=2$, $m=1$ will be suppressed when the spins lie in the orbital plane for equal-mass black holes as discussed in [21]. These waveforms are the solution to the BBH coalescence as expressed by general relativity with errors arising from several sources, see [24]. In paper I, we analyzed the effects of resolution on the matched filtering technique and found that for the resolutions used to compute the waveforms studied in the A-series, the largest error would be $\pm 0.02$ in the match, although that is only for the $a=0.8$ case, and is typically smaller. The waveforms in the B-series have comparable errors, i.e. the resolutions, wave extraction and other numerical techniques were the same in computing both series of waveforms as discussed in [9, 15, 16]. For reference, the typical resolution on the finest grids were $h=M/35.2$ where $M$ is the total mass. ## 4 Discussion and Conclusion In this paper, we investigated the contribution of individual modes to $\rho$ from the last orbits, merger and ringdown of an equal-mass, spinning binary black coalescence. In the A-series, the spins were kept parallel/anti-parallel to the direction of the orbital angular momentum, but the magnitude of the spins varied. In the B-series, the magnitude was kept fixed to $a=0.6$, but the initial angle the spins make with the orbital angular momentum varied. In paper I, we investigated the match between a waveform from the A-series containing only the $\ell=m=2$ mode and a waveform of a sum of modes. There we found strong dependence on the match with spin, with the $\ell=m=2$ waveform failing to match to spinning waveforms especially for spins equal to and greater than $a=0.6$. We did a similar study here for the B-series, comparing two waveforms of $a=0.6$ at various $\vartheta$. We found, despite the variation of the $\rho$ versus $\vartheta$, the match had a much greater dependence on mass than the initial angle. The inclusion of modes was much more important to templates of higher mass, where the merger and ringdown dominate the signal, than at lower masses. This importance will be more evident in matches using unequal-mass and spinning waveforms with larger spin as well as waveforms with more cycles. To qualitatively understand the matches, we conducted a multipolar analysis of the modes in each waveform and calculated the ratio of the $\rho$ of each mode versus $\ell=m=2$. For the A-series, the $\rho$ per mode increased as the magnitude of the spins increased at every total mass as seen in Fig. 2. The odd-$m$ modes increased from almost no contribution at low spins to a $10\%$ contribution at larger spins. At a given spin, the $\ell=2,m=1$ mode dominated the $\rho$ at low mass, but the $\ell=m=4$ mode’s ratio to $\ell=m=2$ grew with increasing mass. For the B-series, Fig. 4, we found that the diversity of contributing modes decreases with increasing angle, except for the $\ell=m=4$ and $\ell=3$, $m=2$ modes which remain relatively constant across $\vartheta$ for a given mass. As in the variation with $a$, at low total binary mass, the secondary signal is the $\ell=2$, $m=1$ mode, but at higher masses the $\ell=m=4$ and the $\ell=m=2$ modes are stronger. As anticipated, the $\ell=2$, $m=1$ mode decreases to zero as the initial angle moves to lie parallel to the orbital plane. For both the series of runs, the variation of the signal in different modes is consistent with the expectation from PN [21]. ## 5 Acknowledgments We thank The Center for Gravitational Wave Physics is supported by the NSF under cooperative agreement PHY-0114375. Support for this work was also provided by NSF grants PHY-0653443 and PHY-0653303. ## References * [1] Scientific Collaboration and TAMA Collaboration. Joint ligo and tama300 search for gravitational waves from inspiralling neutron star binaries. Phys. Rev. D, 73:102002, 2006. * [2] L.A. Wainstein and V.D. Zubakov. Prentice-Hall: Englewood Cliffs, 1962. * [3] Luc Blanchet. Gravitational radiation from post-newtonian sources and inspiralling compact binaries. Living Reviews in Relativity, 9(4), 2006. * [4] E. Flanagan and S. Hughes. Measuring gravitational waves from binary black hole coalescences: I. signal to noise for inspiral, merger, and ringdown. Phys. Rev. D, 57:4535–4565, 1998. * [5] A. Buonanno, G. B. Cook, and F. Pretorius. Phys. Rev. D, 75:124018, 2007. * [6] T. Baumgarte, P. Brady, J. Creighton, L. Lehner, F. Pretorius, and R. DeVoe. preprint (gr-qc/0612100), 2006. * [7] Y. Pan, A. Buonanno, J. G. Baker, J. Centrella, B. J. Kelly, S. T. McWilliams, F. Pretorius, and J. R. van Meter. Phys. Rev., D77:024014, 2008. * [8] P. Ajith, S. Babak, Y. Chen, M. Hewitson, B. Krishnan, J. T. Whelan, B. Bruegmann, P. Diener, J. Gonzalez, M. Hannam, S. Husa, M. Koppitz, D. Pollney, L. Rezzolla, L. Santamaria, A. M. Sintes, U. Sperhake, and J. Thornburg. Classical and Quantum Gravity, 24:689, 2007. * [9] B. Vaishnav, I. Hinder, F. Herrmann, and D. Shoemaker. Phys. Rev. D, 76:084020, 2007. * [10] A. Buonanno, Y. Pan, J. G. Baker, J. Centrella, B. J. Kelly, S. T. McWilliams, and J. R. van Meter. preprint (arXiv.org:0706.3732), 2007. * [11] P. Ajith, S. Babak, Y. Chen, M. Hewitson, B. Krishnan, A. M. Sintes, J. T. Whelan, B. Bruegmann, P. Diener, N. Dorband, J. Gonzalez, M. Hannam, S. Husa, D. Pollney, L. Rezzolla, L. Santamaria, U. Sperhake, and J. Thornburg. preprint (arXiv.org:0710.2335), 2007. * [12] F. Pretorius. Phys. Rev. Lett., 95:121101, 2005. * [13] M. Campanelli, C. O. Lousto, P. Marronetti, and Y. Zlochower. Phys. Rev. Lett., 96:111101, 2006. * [14] J. G. Baker, J. Centrella, D. Choi, M. Koppitz, and J. van Meter. Phys. Rev. Lett., 96:111102, 2006. * [15] F. Herrmann, I. Hinder, D. Shoemaker, P. Laguna, and R. A. Matzner. Ap. J., 661:430–436, 2007. * [16] F. Herrmann, I. Hinder, D. M. Shoemaker, P. Laguna, and R. A. Matzner. Phys. Rev. D, 76(8):084032, 2007. * [17] J. A. Gonzalez, M. D. Hannam, U. Sperhake, B. Brugmann, and S. Husa. Phys. Rev. Lett., 98:231101, 2007. * [18] Albert Lazzarini and Rainer Weiss. LIGO Science Requirements Document. Technical Report E950018-02-E, 1996. * [19] J. D. Schnittman, A. Buonanno, J. R. van Meter, J. G. Baker, W. D. Boggs, J. Centrella, B. J. Kelly, and S. T. McWilliams. preprint (arXiv.org:0707.0301), 2007. * [20] E. Berti, V. Cardoso, J. A. Gonzalez, U. Sperhake, M. Hannam, S. Husa, and B. Bruegmann. Inspiral, merger and ringdown of unequal mass black hole binaries: a multipolar analysis. Phys. Rev. D, 76(6):064034, 2007. * [21] E. Berti, V. Cardoso, J. A. Gonzalez, U. Sperhake, and B. Bruegmann. preprint (arXiv.org:0711.1097), 2007. * [22] B. Owen. Phys. Rev. D, 53:6749–6761, 1996. * [23] T. Damour, B. R. Iyer, and B. S. Sathyaprakash. Phys. Rev. D, 57:885, 1998. * [24] Michael Boyle et al. High-accuracy comparison of numerical relativity simulations with post-Newtonian expansions. Phys. Rev., D76:124038, 2007.

arxiv-papers

2024-09-04T02:48:54.052549

0803.0040

# Minimal distance transformations between links and polymers: Principles and examples Ali R. Mohazab† and Steven S. Plotkin†***e-mail: steve@physics.ubc.ca † Department of Physics and Astronomy, University of British Columbia, 6224 Agricultural Road, Vancouver, BC V6T1Z1, Canada ###### Abstract The calculation of Euclidean distance between points is generalized to one- dimensional objects such as strings or polymers. Necessary and sufficient conditions for the minimal transformation between two polymer configurations are derived. Transformations consist of piecewise rotations and translations subject to Weierstrass-Erdmann corner conditions. Numerous examples are given for the special cases of one and two links. The transition to a large number of links is investigated, where the distance converges to the polymer length times the mean root square distance (MRSD) between polymer configurations, assuming curvature and non-crossing constraints can be neglected. Applications of this metric to protein folding are investigated. Potential applications are also discussed for structural alignment problems such as pharmacophore identification, and inverse kinematic problems in motor learning and control. ###### pacs: 02.30.Xx, 45.10.Db, 45.40.-f , 45.40.Ln, 46.25.Cc , 64.70.Nd , 64.70.km , 82.39.Rt , 87.10.-e , 87.10.Ed , 87.15.A- , 87.15.Cc , 87.15.hp ††: J. Phys.: Condens. Matter ## 1 Introduction The standard variational definition of distance can be generalized to higher dimensional objects such as strings or membranes. In a previous paper [1], one of us has introduced the formalism for this calculation. Consider first zero- dimensional objects (points). The distance between two points $A$ and $B$ is defined through a transformation that takes $A$ to $B$, an object of dimension one higher than the points themselves (here one-dimensional). The transformation minimizing the arc-length travelled between $A$ and $B$ gives the scalar distance $\mathcal{D}^{\ast}$. The differential increment of arc- length may be defined as either $\sqrt{1+(dy/dx)^{2}+(dz/dx)^{2}}dx$, or without the assumption that $y,z$ are functions of $x$, parametrically. To be specific, introduce a “time” parameter $t$ such that $0\leq t\leq T$, and ${\bf r}(0)={\bf r}_{\mbox{\tiny{A}}}$, ${\bf r}(T)={\bf r}_{\mbox{\tiny{B}}}$, and ${\bf r}(t)=(x(t),y(t),z(t))$. The distance between ${\bf r}_{\mbox{\tiny{A}}}$ and ${\bf r}_{\mbox{\tiny{B}}}$ can be found variationally [2]: $\displaystyle\mathcal{D}^{\ast}$ $\displaystyle=$ $\displaystyle\mathcal{D}\left[{\bf r}^{\ast}(t)\right]\mbox{where ${\bf r}^{\ast}(t)$ satisfies}$ (1b) $\displaystyle\delta\int_{0}^{T}\\!\\!\\!dt\>\left(g_{\mu\nu}\dot{x}^{\mu}(t)\dot{x}^{\nu}(t)\right)^{1/2}=0\>.$ or $\displaystyle\delta\int_{0}^{T}\\!\\!\\!dt\>\sqrt{\dot{{\bf r}}^{2}}=0\;\;\;\;\mbox{(Euclidean metric)}$ (1c) Here we have let $\dot{x}\equiv dx/dt$, and $\dot{{\bf r}}\equiv d{\bf r}/dt$. The boundary conditions on the extremal path are ${\bf r}^{\ast}(0)={\bf r}_{\mbox{\tiny{A}}}$ and ${\bf r}^{\ast}(T)={\bf r}_{\mbox{\tiny{B}}}$. Taking the functional derivative in eq. (1c) gives Euler-Lagrange (EL) equations for the Lagrangian $\mathcal{L}=\sqrt{\dot{{\bf r}}^{2}}$: $\displaystyle\frac{d}{dt}\left(\frac{\partial\mathcal{L}}{\partial\dot{{\bf r}}}\right)$ $\displaystyle=$ $\displaystyle 0$ $\displaystyle\mbox{or}\;\;\;\;\dot{\hat{{\bf v}}}$ $\displaystyle=$ $\displaystyle 0$ (2) with $\hat{{\bf v}}$ the unit vector in the direction of the velocity. Since the derivative of a unit vector is always orthogonal to that vector, equation (2) says that the direction of the velocity cannot change, and therefore straight line motion results. Applying the boundary conditions gives $\hat{{\bf v}}=({\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}})/\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$. However, any function ${\bf v}(t)=\left|v_{o}(t)\right|\hat{{\bf v}}$ satisfying the boundary conditions is a solution, so long as $\int_{0}^{T}\\!\\!dt\>\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$. The solution is reparameterization-invariant. Then the extremal functional ${\bf r}^{\ast}(t)$ is given by ${\bf r}^{\ast}(t)={\bf r}_{\mbox{\tiny{A}}}+\frac{{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}}{\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|}\int_{0}^{t}\\!\\!dt\>\left|v_{o}(t)\right|$ (3) and the distance by ${\cal D}^{\ast}=\int_{0}^{T}\\!\\!dt\>\sqrt{\dot{{\bf r}}^{\ast^{2}}}=\int_{0}^{T}\\!\\!dt\>\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$ (4) which represents the diagonal of a hypercube, as expected. At this point we could fix the parameterization by choosing $\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|/T$ (constant speed), for example. The extremal transformation (3) is also a minimum. In section 2.4 we will give the sufficient conditions for an extremum to be a (local) minimum, where we will return to this example. The above idea can be generalized to space curves, surfaces, or higher dimensional manifolds [1]. The distance is defined through the transformation between the objects that minimizes the cumulative amount of arc-length travelled by all parts of the manifold. ## 2 Distance for polymers or strings Describing the transformation ${\bf r}(s,t)$ between two space curves ${\bf r}_{\mbox{\tiny A}}(s)$ and ${\bf r}_{\mbox{\tiny{B}}}(s)$ requires two scalar parameters: $s$ the arc-length along the space curve, and $t$ the “time” as in the above zero-dimensional case measuring progress during the transformation. The boundary conditions are then ${\bf r}(s,0)={\bf r}_{\mbox{\tiny A}}(s)$ and ${\bf r}(s,T)={\bf r}_{\mbox{\tiny{B}}}(s)$. The minimal transformation ${\bf r}^{\ast}(s,t)$ is an object of dimension one higher than $A$ or $B$, i.e. it yields a distance that is two-dimensional. The distance $\mathcal{D}^{\ast}=\mathcal{D}[{\bf r}^{\ast}(s,t)]$, where the functional $\mathcal{D}[{\bf r}]$ is given by $\mathcal{D}[{\bf r}]=\int_{0}^{L}\\!\\!\\!ds\\!\\!\int_{0}^{T}\\!\\!\\!dt\>\sqrt{\dot{{\bf r}}^{2}}\>.$ (5) Here we have used the shorthand ${\bf r}\equiv{\bf r}(s,t)=(x(s,t),y(s,t),z(s,t))$ (a 3-vector), and $\dot{{\bf r}}\equiv\partial{\bf r}/\partial t$. It has been shown previously that the problem of distance does not map to a simple soap film, nor to the minimal area of a world-sheet (which corresponds to the action of a classical relativistic string) [1]. Formulated as above, the string can contract and expand arbitrarily in order to minimize the distance travelled. The transforming object is akin to a rubber band, and all points on ${\bf r}_{\mbox{\tiny A}}(s)$ will move in straight lines to their partner points on ${\bf r}_{\mbox{\tiny{B}}}(s)$ to minimize the distance. It is worth mentioning that protein chains for example only change their length by about one percent at biological temperatures. To accurately represent the transformation of a non-extensible string, a Lagrange multiplier $\lambda(s,t)$ must be introduced into the effective Lagrangian, weighting the constraint: $\sqrt{{\bf r}^{\prime 2}}=1\>,$ (6) where ${\bf r}^{\prime}\equiv\partial{\bf r}/\partial s$. Under this constraint, points along the string can no longer move independently of each other, but must always be a fixed (infinitesimal) distance apart. The tangent vector $\hat{{\bf t}}={\bf r}^{\prime}$ is now a unit vector, and the total length of the string is $L=\int_{0}^{L}\\!\\!ds\>\sqrt{{\bf r}^{\prime 2}}=\int_{0}^{L}\\!\\!ds$. Consider the minimal distance transformation between two configurations ${\bf r}_{\mbox{\tiny{A}}}(s)$ and ${\bf r}_{\mbox{\tiny{B}}}(s)$ of an ideal polymer of length $L$. Let us derive the EL equations for this case. From equations (5) and (6), the effective action is $\displaystyle{\cal D}$ $\displaystyle=$ $\displaystyle\int_{0}^{L}\\!\\!\int_{0}^{T}\\!\\!\\!ds\,dt\>\mathcal{L}\left(\dot{{\bf r}},{\bf r}^{\prime}\right)$ (7a) $\displaystyle\mbox{where}\;\;\;\;\mathcal{L}$ $\displaystyle=$ $\displaystyle\sqrt{\dot{{\bf r}}^{2}}-\lambda\left(\sqrt{{\bf r}^{\prime 2}}-1\right)$ (7b) and the Lagrange multiplier $\lambda\equiv\lambda(s,t)$ is a function of both $s$ and $t$. The extrema of the distance functional ${\cal D}$ in (7a) are found from $\delta{\cal D}=0$. Taking the functional derivative gives EL equations [1]: $\dot{\hat{{\bf v}}}=\lambda\boldsymbol{\kappa}+\lambda^{\prime}\hat{{\bf t}}\>.$ (8) where $\hat{{\bf v}}$ is the unit velocity vector, $\hat{{\bf t}}$ is the unit tangent vector, and $\boldsymbol{\kappa}$ is the curvature vector. In eq. (8) we see explicitly that if the non-extensibility constraint is set to zero ($\lambda=0$), all points on ${\bf r}_{\mbox{\tiny{A}}}(s)$ move in straight lines to ${\bf r}_{\mbox{\tiny{B}}}(s)$. ### 2.1 Discrete chains To make the problem more amenable to solution, we can discretize the spatial variables while letting the time variable remain continuous, i.e. we implement the method of lines to solve eq. (8). Rather than directly discretizing eq. (8) however, it is more natural to consider a discretized chain as shown in figure 1 from the outset, and to calculate the EL equations for this system. This recipe then gives the same result as properly discretizing eq. (8). For the discretized chain, the constraint in eq. (6) becomes $|\Delta{\bf r}|=\Delta s=L/(N-1)$, giving the length of each link. As the number of beads $N\rightarrow\infty$ the system approaches a continuous chain. For finite $N$, the Lagrangian becomes a function of the positions and velocities $\left\\{{\bf r}_{i},\dot{{\bf r}}_{i}\right\\}$ of all beads $i$, $1\leq i\leq N+1$. We use the shorthand notation $\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})$. | ---|--- a | b Figure 1: Continuum (a) and discretized (b) polymer chain. The EL equation for the continuum polymer is a nonlinear (vector) PDE, while the EL equations for the discretized polymer are a set of nonlinear ODEs. This recipe yields the distance metric for an ideal, freely-jointed chain, which has no non-local interactions and no curvature constraints. While this approximation is often used as a first step, real chains may behave quite differently for several reasons. In many cases, the configuration which is an energetic minimum is a straight line, or a single conformation dictated by the chemistry of the polymeric bonds. At finite temperature, energy in the bath induces conformational fluctuations. Real polymers also cannot cross themselves, and because of their stereochemistry also take up volume. We leave these interesting features for later analysis. Equation (6) for the discretized chain becomes $N$ constraint equations added to the effective Lagrangian: $\sum_{i=1}^{N}\hat{\lambda}_{i,i+1}\left(\sqrt{({\bf r}_{i+1}-{\bf r}_{i})^{2}}-\Delta s\right)$ where each $\hat{\lambda}_{i,i+1}\equiv\hat{\lambda}_{i,i+1}(t)$ is a function of $t$, and $\hat{\lambda}_{N,N+1}=0$. Letting $\lambda\equiv 2\hat{\lambda}\,\Delta s$ and ${\bf r}_{i+1/i}\equiv{\bf r}_{i+1}-{\bf r}_{i}$ we rewrite this strictly for convenience as $\sum\frac{\lambda_{i,i+1}}{2}\>\left({{\bf r}_{i+1/i}^{2}\over\Delta s^{2}}-1\right)\>.$ We next convert to dimensionless variables by letting ${\bf r}=(\Delta s)\hat{{\bf r}}$. To simplify the notation, from here on we simply refer to $\hat{{\bf r}}$ as ${\bf r}$. The distance for the discretized chain becomes ${\cal D}[{\bf r}_{i},\dot{{\bf r}}_{i}]=\Delta s^{2}\int_{0}^{T}\\!\\!\\!dt\>\mathcal{L}\left({\bf r}_{i},\dot{{\bf r}}_{i}\right)$ (9) with effective Lagrangian $\mathcal{L}\left({\bf r}_{i},\dot{{\bf r}}_{i}\right)=\sum_{i=1}^{N}\left(\sqrt{\dot{{\bf r}}_{i}^{2}}-{\lambda_{i,i+1}\over 2}\left({\bf r}_{i+1/i}^{2}-1\right)\right)\>.$ (10) The derivatives $\dot{{\bf r}}$ and ${\bf r}_{i+1/i}$ are raised to different powers in (10), however so long as ${\bf r}_{i+1/i}$ satisfies the constraint $\left|{\bf r}_{i+1/i}\right|=1$, the EL equations for ${\bf r}_{i}(t)$ will be the same whether the constraint $\sqrt{{\bf r}_{i+1/i}^{2}}=1$ or ${\bf r}_{i+1/i}^{2}=1$ is used. The reparameterization invariance present for point particles (c.f. section 1) is still present for beads on the chain, but the parameterization of arclength along the chain is taken to be fixed by the discretization. ### 2.2 General variation of the distance functional For reasons that will become clear as we progress, we consider the general variation of the functional ${\cal D}$, allowing for broken extremals. That is, we allow the curves describing the particle trajectories to be non-smooth in principle at one or more points in time. Consider the case of one such point at time $t_{1}$. The distance can be written as ${\cal D}=\int_{0}^{t_{1}}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})+\int_{t_{1}}^{T}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})$ (11) The space-trajectories of the particles must be continuous at time $t_{1}$, so ${\bf r}_{i}(t_{1}-\epsilon)={\bf r}_{i}(t_{1}+\epsilon)$, or in shorthand: ${\bf r}_{i}\left(t_{1}^{-}\right)={\bf r}_{i}\left(t_{1}^{+}\right)\>.$ (12) Let ${\bf r}_{i}(t)$ and $\tilde{{\bf r}}_{i}(t)$ be two neighboring trajectories from ${\bf r}_{i}(0)={\bf r}_{\mbox{\tiny{A}}i}$ to ${\bf r}_{i}(T)={\bf r}_{\mbox{\tiny{B}}i}$ (see figure 2). Neighboring curves will differ by the first order quantity ${\bf h}_{i}(t)=\tilde{{\bf r}}_{i}(t)-{\bf r}_{i}(t)$. The fixed boundary conditions at $t=0,T$ dictate that ${\bf h}_{i}(0)={\bf h}_{i}(T)=0$. The difference in distance between the two trajectories is $\displaystyle\Delta{\cal D}$ $\displaystyle=$ $\displaystyle{\cal D}[{\bf r}_{i}+{\bf h}_{i}]-{\cal D}[{\bf r}_{i}]$ (13) $\displaystyle=$ $\displaystyle\int_{0}^{t_{1}+\delta t_{1}}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i}+{\bf h}_{i},\dot{{\bf r}}_{i}+\dot{{\bf h}}_{i})-\int_{0}^{t_{1}}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})+\int_{t_{1}+\delta t_{1}}^{T}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i}+{\bf h}_{i},\dot{{\bf r}}_{i}+\dot{{\bf h}}_{i})-\int_{t_{1}}^{T}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})$ Figure 2: General variations of a functional with fixed end points allow for broken extremals. In the text we derive the extra “corner” conditions for a piecewise continuous path to still be extremal for our distance functional. Taylor expanding the Lagrangian to first order in ${\bf h}_{i}$:‡**footnotetext: ‡ We use the notation $F_{{\bf r}}\equiv\partial F/\partial{\bf r}$, $F_{\dot{{\bf r}}}\equiv\partial F/\partial\dot{{\bf r}}$. $\mathcal{L}\approx\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})+\sum_{i=1}^{N}\left(\mathcal{L}_{{\bf r}_{i}}\cdot{\bf h}_{i}+\mathcal{L}_{\dot{{\bf r}}_{i}}\cdot\dot{{\bf h}}_{i}\right)$ and integrating by parts using the fixed boundary conditions at $t=0,T$, the difference in distance up to first order in ${\bf h}_{i}$ is $\displaystyle\Delta{\cal D}$ $\displaystyle\approx$ $\displaystyle\int_{0}^{t_{1}}\\!\\!\\!dt\>\sum_{i}\left(\mathcal{L}_{{\bf r}_{i}}-{d\over dt}\mathcal{L}_{\dot{{\bf r}}_{i}}\right)\cdot{\bf h}_{i}+\int_{t_{1}}^{T}\\!\\!\\!dt\>\sum_{i}\left(\mathcal{L}_{{\bf r}_{i}}-{d\over dt}\mathcal{L}_{\dot{{\bf r}}_{i}}\right)\cdot{\bf h}_{i}$ (14) $\displaystyle+$ $\displaystyle\mathcal{L}(t_{1}^{-})\delta t_{1}-\mathcal{L}(t_{1}^{+})\delta t_{1}+\sum_{i}\left.\mathcal{L}_{\dot{{\bf r}}_{i}}\cdot{\bf h}_{i}\right|_{t_{1}^{-}}-\sum_{i}\left.\mathcal{L}_{\dot{{\bf r}}_{i}}\cdot{\bf h}_{i}\right|_{t_{1}^{+}}$ with the shorthand $\mathcal{L}(t)\equiv\mathcal{L}({\bf r}_{i}(t),\dot{{\bf r}}_{i}(t))$. ### 2.3 Conditions for an extremum The variation $\delta{\cal D}$ differs from $\Delta{\cal D}$ above only by second order terms. Then for the transformation from $\\{{\bf r}_{\mbox{\tiny{A}}i}\\}$ to $\\{{\bf r}_{\mbox{\tiny{B}}i}\\}$ to be an extremum, $\delta{\cal D}=0$. Thus, the EL equations (in the top line of eq. (14)) must vanish in each regime $[0,t_{1})$, $(t_{1},T]$. Using the form of the Lagrangian in eq. (10), the EL equations become: $\displaystyle\dot{\hat{{\bf v}}}_{1}+\lambda_{12}\,{\bf r}_{2/1}=0$ (15a) $\displaystyle\dot{\hat{{\bf v}}}_{2}-\lambda_{12}\,{\bf r}_{2/1}+\lambda_{23}\,{\bf r}_{3/2}=0$ (15c) $\displaystyle\hskip 28.45274pt\vdots$ $\displaystyle\dot{\hat{{\bf v}}}_{N}-\lambda_{N-1,N}\,{\bf r}_{N/(N-1)}=0$ According to equation (14) there are additional conditions for the transformation to be an extremum. To find these first note that up to first order (see figure 2) $\displaystyle{\bf h}_{i}(t_{1})\approx\delta{\bf r}_{i}(t_{1})-\dot{{\bf r}}_{i}(t_{1})\,\delta t_{1}\>.$ (16) Then the first variation in the distance is $\displaystyle\delta{\cal D}$ $\displaystyle=$ $\displaystyle\left[\left.\left(\mathcal{L}-\sum_{i}\dot{{\bf r}}_{i}\cdot\mathcal{L}_{\dot{{\bf r}}_{i}}\right)\right|_{t_{1}^{-}}-\left.\left(\mathcal{L}-\sum_{i}\dot{{\bf r}}_{i}\cdot\mathcal{L}_{\dot{{\bf r}}_{i}}\right)\right|_{t_{1}^{+}}\right]\delta t_{1}$ (17) $\displaystyle+$ $\displaystyle\sum_{i}\left[\left.\mathcal{L}_{\dot{{\bf r}}_{i}}\right|_{t_{1}^{-}}-\left.\mathcal{L}_{\dot{{\bf r}}_{i}}\right|_{t_{1}^{+}}\right]\cdot\delta{\bf r}_{i}(t_{1})$ which must vanish at an extremum. Because the variations $\delta{\bf r}_{i}$ and $\delta t_{1}$ are all independent, the terms in square brackets in equation (17) must vanish. Writing these expressions in terms of the conjugate momenta ${\bf p}_{i}=\mathcal{L}_{\dot{{\bf r}}_{i}}$ and Hamiltonian $\mathcal{H}=\sum_{i}\dot{{\bf r}}_{i}\cdot{\bf p}_{i}-\mathcal{L}$ gives the conditions: $\displaystyle\left.{\bf p}_{i}\right|_{{}_{t_{1}^{-}}}=\left.{\bf p}_{i}\right|_{{}_{t_{1}^{+}}}$ (18a) $\displaystyle\left.\mathcal{H}\right|_{{}_{t_{1}^{-}}}=\left.\mathcal{H}\right|_{{}_{t_{1}^{+}}}$ (18b) These conditions are called the Weierstrass-Erdmann conditions or corner conditions in the calculus of variations [2]. According to the Lagrangian in equation (10), the Hamiltonian is given by $\mathcal{H}=-\sum_{i=1}^{N}{\lambda_{i,i+1}\over 2}\left({\bf r}_{i+1/i}^{2}-1\right)$ which is identically zero, so corner condition (18b) provides no further information. The conjugate momenta according to (10) are given by ${\bf p}_{i}={\dot{{\bf r}}_{i}\over\left|\dot{{\bf r}}_{i}\right|}=\hat{{\bf v}}_{i}\>.$ (19) Therefore, according to corner condition (18a), extremal trajectories cannot suddenly change direction: each ${\bf r}_{i}(t)$ follows a smooth path continuous up to first derivatives in the spatial coordinates. The fact that one corner condition provided no information due to the vanishing of the Hamiltonian is related to our choice of parameterization in formulating the problem. For example, in the case of the distance of the single point particle mentioned in the introduction, the Lagrangian may be defined either through independent variable $x$ as $\mathcal{L}^{(x)}=\sqrt{1+y^{\prime 2}+z^{\prime 2}}$ (with e.g. $y^{\prime}=dy/dx$), or parametrically through independent variable $t$ as $\mathcal{L}^{(t)}=\sqrt{\dot{{\bf r}}^{2}}$. The conjugate momenta are then either $\mathcal{L}^{(x)}_{y^{\prime}}=y^{\prime}/\sqrt{1+y^{\prime 2}+z^{\prime 2}}$ and $\mathcal{L}^{(x)}_{z^{\prime}}=z^{\prime}/\sqrt{1+y^{\prime 2}+z^{\prime 2}}$, or $\mathcal{L}^{(t)}_{\dot{{\bf r}}}=\dot{{\bf r}}/|\dot{{\bf r}}|\equiv\hat{{\bf v}}$. The Hamiltonia are either $\mathcal{H}^{(x)}=1/\sqrt{1+y^{\prime 2}+z^{\prime 2}}$ or $\mathcal{H}^{(t)}=\mathcal{L}^{(t)}-\dot{{\bf r}}\cdot(\dot{{\bf r}}/|\dot{{\bf r}}|)=0$. The corner conditions can be shown to be equivalent for both choices of independent variable: for $\mathcal{L}^{(t)}$ they give $\hat{{\bf v}}(t_{1}^{-})=\hat{{\bf v}}(t_{1}^{+})$, so that the direction of the tangent to the curve cannot have a discontinuity. Together, the Hamiltonian and two conjugate momenta for $\mathcal{L}^{(x)}$ can be interpreted as components of the unit tangent vector to the curve, i.e. $\hat{{\bf t}}(x)=(\hat{{\bf i}}+y^{\prime}\hat{{\bf j}}+z^{\prime}\hat{{\bf k}})/\sqrt{1+y^{\prime 2}+z^{\prime 2}}$, and so once again the corner conditions enforce a continuous tangent vector, here $\hat{{\bf t}}(x_{1}^{-})=\hat{{\bf t}}(x_{1}^{+})$. #### 2.3.1 Boundary conditions In the continuum limit, the boundary conditions on ${\bf r}(s,t)$ are ${\bf r}(s,0)={\bf r}_{\mbox{\tiny{A}}}(s)$, ${\bf r}(s,T)={\bf r}_{\mbox{\tiny{B}}}(s)$ where ${\bf r}_{\mbox{\tiny{A}}}$ and ${\bf r}_{\mbox{\tiny{B}}}$ are the two configurations of the polymer. For discrete chains, these boundary conditions become $\displaystyle\left\\{{\bf r}_{i}(0)\right\\}$ $\displaystyle=$ $\displaystyle\\{{\bf r}_{i}^{\left(A\right)}\\}$ (20a) $\displaystyle\left\\{{\bf r}_{i}(T)\right\\}$ $\displaystyle=$ $\displaystyle\\{{\bf r}_{i}^{\left(B\right)}\\}\>.$ (20b) There are also boundary conditions that hold for the end points of the chain at all times. From equations (15a, 15c) we see that there are three solutions for the end points of the chain: 1) If $\lambda\neq 0$, purely rotational motion results. This can be seen by taking the dot product of eq. (15a) with ${\bf v}_{1}$, which yields $\lambda_{12}{\bf v}_{1}\cdot{\bf r}_{2/1}=0$, so the velocity of the end point is orthogonal to the link. The rotation must be about a point that is internal to the link, i.e. on the line between points $1$ and $2$ for end point $1$. This can be seen straightforwardly for the case of one link by removing point $3$ from equations (15a) and (15c). Then the accelerations $\dot{\hat{{\bf v}}}_{i}$ must be in opposite directions. This can only occur if rotation is about a point on the line between points $1$ and $2$. 2) If $\lambda=0$, $\dot{\hat{{\bf v}}}_{i}=0$, and straight-line motion of the end point results. 3) Writing out the time-derivative in (15a) yields ${\bf v}_{1}^{2}\dot{{\bf v}}_{1}-\left({\bf v}_{1}\cdot\dot{{\bf v}}_{1}\right){\bf v}_{1}=-\lambda_{12}\left|{\bf v}_{1}\right|^{3}{\bf r}_{2/1}$ (21) which has the trivial solution ${\bf v}_{1}=0$. The end point can be at rest, while other parts of the chain move. ### 2.4 Sufficient conditions for a minimum For a transformation to be minimal, it is necessary, but not sufficient, that it be an extremum. We now derive the sufficient conditions for a given transformation to minimize the functional (9). We describe the formalism in some detail because it is not typically taught to physicists- for further reading see for example reference [2]. This section can be read independently of the others, and might be skipped on first reading. According to Sylvester’s criterion, a quadratic form $\sum_{ij}A_{ij}x_{i}x_{j}$ is positive definite if and only if all descending principle minors of the matrix $\|A_{ij}\|$ are positive, i.e. $A_{11}>0\>,\hskip 14.22636pt\begin{vmatrix}A_{11}&A_{12}\\\ A_{21}&A_{22}\end{vmatrix}>0\>,\quad\begin{vmatrix}A_{11}&A_{12}&A_{13}\\\ A_{21}&A_{22}&A_{23}\\\ A_{31}&A_{32}&A_{33}\end{vmatrix}>0\>,\quad\ldots\quad,\mbox{det}\|A_{ij}\|>0\>,$ (22) and a function $F$ of ${\bf x}\equiv(x_{1},x_{2},\ldots,x_{n})$ has a minimum at ${\bf x}^{\star}$ if the Jacobian matrix $\|\partial^{2}F/\partial x_{i}\partial x_{j}\|$ is positive definite at the position of the extremum (where $\partial F/\partial x_{i}=0$). For a function to be a minimum of a given functional, it must satisfy similar sufficient conditions. Consider again the difference in distance between two trajectories in (9)‡**footnotetext: ‡ We ignore corner conditions for purposes of the derivation. It can be shown that they do not modify the result.. Taylor expanding the Lagrangian to second order in ${\bf h}_{i}$: $\displaystyle\Delta{\cal D}$ $\displaystyle=$ $\displaystyle{\cal D}\left[{\bf r}_{i}+{\bf h}_{i}\right]-{\cal D}\left[{\bf r}_{i}\right]$ (23) $\displaystyle=$ $\displaystyle\int_{0}^{T}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i}+{\bf h}_{i},\dot{{\bf r}}_{i}+\dot{{\bf h}}_{i})-\int_{0}^{T}\\!\\!\\!dt\>\mathcal{L}({\bf r}_{i},\dot{{\bf r}}_{i})$ $\displaystyle\approx$ $\displaystyle\int_{0}^{T}\\!\\!\\!dt\>\left[\sum_{i=1}^{N}\left(\mathcal{L}_{{\bf r}_{i}}\cdot{\bf h}_{i}+\mathcal{L}_{\dot{{\bf r}}_{i}}\cdot\dot{{\bf h}}_{i}\right)+{1\over 2}\sum_{i,j}^{3N}\left(\mathcal{L}_{x_{i}x_{j}}h_{i}h_{j}+2\mathcal{L}_{x_{i}\dot{x}_{j}}h_{i}\dot{h}_{j}+\mathcal{L}_{\dot{x}_{i}\dot{x}_{j}}\dot{h}_{i}\dot{h}_{j}\right)\right]$ At an extremum, the first order term in (23) is zero, and $\Delta{\cal D}\approx\delta^{2}{\cal D}$, the second variation. For the extremum to be a minimum, $\delta^{2}{\cal D}>0$. From eq. (10), the matrix $\|\mathcal{L}_{x_{i}\dot{x}_{j}}\|=\|0\|$. Assuming $\|\mathcal{L}_{x_{i}\dot{x}_{j}}\|$ is in general a symmetric matrix, i.e. $\mathcal{L}_{x_{i}\dot{x}_{j}}=\mathcal{L}_{x_{j}\dot{x}_{i}}$, the second term in the quadratic form of (23) may be integrated by parts to give: $\delta^{2}{\cal D}={1\over 2}\int_{0}^{T}\\!\\!\\!dt\>\left[\langle\dot{h}|\mathrm{P}\dot{h}\rangle+\langle h|\mathrm{Q}h\rangle\right]\>,$ (24) where we have let $|h\rangle$ denote the vector $(h_{1},h_{2},\ldots,h_{3N})$, and used the shorthand $\mathrm{P}$ and $\mathrm{Q}$ for the matrices: $\displaystyle\mathrm{P}(t)=\|\mathrm{P}_{ij}\|=\|\mathcal{L}_{\dot{x}_{i}\dot{x}_{j}}\|$ $\displaystyle\mathrm{Q}(t)=\|\mathrm{Q}_{ij}\|=\left(\|\mathcal{L}_{x_{i}x_{j}}\|-{d\over dx}\|\mathcal{L}_{x_{i}\dot{x}_{j}}\|\right)\>.$ (25) From (10) the explicit form for these matrices may be calculated. $\mathrm{P}$ is block diagonal: $\mathrm{P}=\begin{bmatrix}I_{ij}^{\left(1\right)}&0&\cdots&0\\\ 0&I_{ij}^{\left(2\right)}&\cdots&0\\\ \vdots&\vdots&\ddots&\vdots\\\ 0&0&\cdots&I_{ij}^{\left(N\right)}\\\ \end{bmatrix}$ (26) with each block matrix having elements $\|\mathrm{I}_{ij}^{\left(J\right)}\|={1\over{\left|\dot{{\bf r}}^{\,J}\right|}^{3}}\left(\delta_{ij}{\left.\dot{{\bf r}}^{\left(J\right)}\right.}^{2}-\dot{x}^{\,\left(J\right)}_{i}\dot{x}^{\,\left(J\right)}_{j}\right)={1\over{\left|\dot{{\bf r}}^{\,J}\right|}^{3}}\begin{bmatrix}\dot{y}^{2}+\dot{z}^{2}&-\dot{x}\dot{y}&-\dot{x}\dot{z}\\\ -\dot{x}\dot{y}&\dot{x}^{2}+\dot{z}^{2}&-\dot{y}\dot{z}\\\ -\dot{x}\dot{z}&-\dot{y}\dot{z}&\dot{x}^{2}+\dot{y}^{2}\\\ \end{bmatrix}_{\mbox{\footnotesize{(particle $J$)}}}$ (27) Interestingly the numerator of (27) has the form of an inertia tensor for a point particle in velocity-space. The matrix $\mathrm{Q}$ is block tri- diagonal, because the spatial derivatives in (2.4) couple each bead to its two neighbors. Using indices $I,J$ to enumerate beads and $i,j$ to enumerate $x,y,z$ components for each bead: $\displaystyle\|\mathrm{Q}_{IJ,ij}\|$ $\displaystyle=\delta_{ij}\left[\lambda_{J-1,J}\left(\delta_{IJ}-\delta_{I,J-1}\right)+\lambda_{J,J+1}\left(\delta_{IJ}-\delta_{I,J+1}\right)\right]$ or $\displaystyle\mathrm{Q}$ $\displaystyle=\begin{bmatrix}\lambda_{12}\mathbf{1}&-\lambda_{12}\mathbf{1}&0&0&\cdots&\\\ -\lambda_{12}\mathbf{1}&\left(\lambda_{12}+\lambda_{23}\right)\mathbf{1}&-\lambda_{23}\mathbf{1}&0&\cdots&\\\ 0&-\lambda_{23}\mathbf{1}&\left(\lambda_{23}+\lambda_{34}\right)\mathbf{1}&-\lambda_{34}\mathbf{1}&&\\\ \vdots&\ddots&\ddots&\ddots&&\\\ &&&&&\\\ &&&&-\lambda_{N-1,N}\mathbf{1}&\lambda_{N-1,N}\mathbf{1}\\\ \end{bmatrix}$ (28) For the transformation ${{\bf r}^{\ast}(t)}$ to be a minimum of ${\cal D}[{\bf r}]$, the functional (24) must be positive definite for all $|h\rangle$. To derive the conditions for this, we can temporarily ignore the fact that (24) arose from the second variation of (9), and treat (24) as a new functional of the function $|h(t)\rangle=|h_{1}(t),\ldots h_{3N}(t)\rangle$. We then ask what $|h(t)\rangle$ extremizes (24). If $\delta^{2}{\cal D}>0$ we expect that the only extremal solution would be the trivial one: $|h(t)\rangle=|0\rangle$, at least for small variations of the $h_{i}(t)$. That is, changing the transformation $\\{{\bf r}^{\ast}_{i}(t)\\}$ from that which extremized (9) to a neighboring transformation $\\{{\bf r}^{\ast}_{i}(t)+{\bf h}_{i}(t)\\}$ would increase the distance travelled. The system of $3N$ EL equations for $|h\rangle$ from (24) is $-{d\over dt}|\mathrm{P}\dot{h}\rangle+|\mathrm{Q}h\rangle=|0\rangle$ (29) with boundary conditions $|h(0)\rangle=|h(T)\rangle=|0\rangle\>.$ (30) Equation (29) is referred to as the Jacobi equation in the calculus of variations. First note that if $|h\rangle$ satisfies the system of equations in (29) as well as the boundary conditions (30), then integration by parts gives $\delta^{2}{\cal D}=\int_{0}^{T}\\!\\!\\!dt\>\left(\langle\dot{h}|\mathrm{P}\dot{h}\rangle+\langle h|\mathrm{Q}h\rangle\right)=\int_{0}^{T}\\!\\!\\!dt\>\langle h|-{d\over dt}\left(\mathrm{P}\dot{h}\right)+\mathrm{Q}h\rangle=0\>.$ (31) This means that for $\delta^{2}{\cal D}$ to be $>0$, any nontrivial $|h(t)\rangle$ which satisfies the boundary conditions must not itself be an extremal solution of the Jacobi equation, otherwise solutions $|{\bf r}^{\ast}(t)\rangle$ perturbed by any constant times $|h(t)\rangle$ are themselves extremals. One may think of this by analogy as the necessity for the absence of any “Goldstone modes”, where excitations by various $C|h(t)\rangle$ would lead to a family of curves with zero cost in action, and thus zero effective restoring force, between them. Alternatively we can ask what equation ${\bf h}\equiv|h\rangle$ must satisfy if the EL equations are satisfied for both $\mathcal{L}({\bf r},\dot{{\bf r}})$ and the neighboring extremal $\mathcal{L}({\bf r}+{\bf h},\dot{{\bf r}}+\dot{{\bf h}})$. Taylor expanding $\mathcal{L}({\bf r}+{\bf h},\dot{{\bf r}}+\dot{{\bf h}})$ in $\mathcal{L}_{{\bf r}}({\bf r}+{\bf h},\dot{{\bf r}}+\dot{{\bf h}})-{d\over dt}\mathcal{L}_{\dot{{\bf r}}}({\bf r}+{\bf h},\dot{{\bf r}}+\dot{{\bf h}})=0$ gives $-{d\over dt}\left(\mathcal{L}_{\dot{{\bf r}}\dot{{\bf r}}}\cdot\dot{{\bf h}}\right)+\left(\mathcal{L}_{{\bf r}{\bf r}}-{d\over dt}\mathcal{L}_{{\bf r}\dot{{\bf r}}}\right)\cdot{\bf h}=0$ which is exactly Jacobi’s equation (29) with definitions (2.4). From here on, it is much simpler to elucidate the central concepts for sufficient conditions using the case of a single scalar function $h(t)$. The analysis can be generalized to the multi-dimensional case with a bit more effort, but the conclusions are essentially the same and so they will simply be stated along with the conclusions for the ’1-D’ case. For further details see [2]. We write equation (24) in 1-D as: ${1\over 2}\int_{0}^{T}\\!\\!\\!dt\>\left(P\dot{h}^{2}+Qh^{2}\right)$ (32) It was realized originally by Legendre that the integral could be brought to simpler form by adding zero to it in the form of a total derivative. Since $\int_{0}^{T}\\!\\!\\!dt\>{d\over dt}\left(w(t)h^{2}\right)=0$ for any $w(t)$ so long as $h(t)$ satisfies the boundary conditions (30), we can add it to the integral in (32) and seek a function $w(t)$ such that the expression $\delta^{2}{\cal D}={1\over 2}\int_{0}^{T}\\!\\!\\!dt\>\left(P\dot{h}^{2}+2wh\dot{h}+\left(Q+\dot{w}\right)h^{2}\right)$ may be written as a perfect square. This yields the differential equation $P\left(Q+\dot{w}\right)=w^{2}$ (33) for $w(t)$, and second variation $\delta^{2}{\cal D}[h]={1\over 2}\int_{0}^{T}\\!\\!\\!dt\>P\left(\dot{h}+{w\over P}h\right)^{2}\>.$ (34) Therefore a necessary condition for a minimum is for $P>0$. The analogous condition in the multi-dimensional case is for the matrix $\|\mathrm{P}\|$ to be positive definite. If the differential term $\dot{h}+{w\over P}h$ in (34) were equal to zero for some $h(t)$, the boundary condition $h(0)=0$ would then imply $\dot{h}(0)=0$ and thus $h(t)=0$ for all $t$ by the uniqueness theorem as applied to this first order differential equation. Therefore the functional (34) is positive definite if, and only if, 1.) $P>0$ , 2.) A solution for eq. (33) exists for the whole interval $[0,T]$. In general, there is no guarantee of condition (2) even if condition (1) is valid. For example if $P=1$, $Q=-1$, (33) has solution $w(t)=\tan(t+c)$, which has no finite solution if $|T|>\pi$. †**footnotetext: † Because reparameterization invariance in our problem, the value of $T$ is adjustable, however precisely because of this invariance, $\det\|\mathrm{P}\|=0$ and so is no longer positive definite. We discuss this problem and its resolution below. If (33) has a pole at say $\tilde{t}$, then for the integral (34) to remain finite, $h(\tilde{t})\rightarrow 0$. This point is said to be conjugate to the point $t_{o}=0$, i.e. it is a conjugate point. Moreover, equation (33) is a Riccati equation, which may be brought to linear form by the transformation $w(t)=-P\dot{H}/H$, with $H(t)$ an unknown function. Substitution in (33) gives $-{d\over dt}\left(P\dot{H}\right)+QH=0$ (35) which is precisely equation (29)- the Jacobi equation for $h(t)$. This means that for equation (33) to have a solution on $[0,T]$, $H(t)$, as given by the solution to (35), must have no roots on $[0,T]$. But because equation (35) holds for $h(t)$ as well, $h(t)$ must have no roots (conjugate points) on $[0,T]$. Because $h(0)=h(T)=0$, the only way to extremize (32) is to satisfy eq. (35) with the trivial solution $h(t)=0$. If $h(t)\neq 0$ for $0<t<T$ then it would mean that there was a conjugate point at $\tilde{t}=T$. In the multi-dimensional case an extremal $|h\rangle$ is one of $3N$ vectors satisfying equations (29), i.e. $|h^{(\alpha)}\rangle=|h_{1}^{(\alpha)}\ldots h_{3N}^{(\alpha)}\rangle$, $1\leq\alpha\leq 3N$. A conjugate point is defined as a point where the determinant vanishes: $\det\begin{vmatrix}h_{1}^{\left(1\right)}\left(t\right)&\cdots&h_{1}^{\left(3N\right)}\left(t\right)\\\ \vdots&&\vdots\\\ h_{1}^{\left(3N\right)}\left(t\right)&\cdots&h_{3N}^{\left(3N\right)}\left(t\right)\end{vmatrix}=0$ The conditions for a transformation to be minimal are then: 1.) The transformation $|{\bf r}^{\ast}(t)\rangle=\\{{\bf r}^{\ast}_{i}(t)\\}$ is extremal, 2.) Along $|{\bf r}^{\ast}(t)\rangle$, the matrix $\mathrm{P}(t)=\mathcal{L}_{\dot{x}_{i}\dot{x}_{j}}$ is positive definite, and 3.) The interval $[0,T]$ contains no conjugate points to $t=0$. The above ideas can be made clear with a few examples below. #### 2.4.1 Distance between points From the effective Lagrangian $\mathcal{L}=\sqrt{\dot{{\bf r}}^{2}}$, $\mathrm{P}=\|\mathcal{L}_{\dot{x}_{i}\dot{x}_{j}}\|$ is given in equation (27), which has determinant $\det\mathrm{P}=0$, and so is not positive definite. This is due to our choice of parameterization. If we break symmetry by choosing one spatial direction as the independent variable, $\mathcal{L}\left(x,y^{\prime},z^{\prime}\right)=\sqrt{1+y^{\prime 2}+z^{\prime 2}}$ (with e.g. $y^{\prime}\equiv dy/dx$ and $x_{0}\leq x\leq x_{1}$). Then $\mathrm{P}={1\over{\left(1+y^{\prime 2}+z^{\prime 2}\right)^{3/2}}}\begin{pmatrix}1+z^{\prime 2}&-y^{\prime}z^{\prime}\\\ -y^{\prime}z^{\prime}&1+y^{\prime 2}\end{pmatrix}$ with positive definite determinant $\det\|\mathrm{P}\|=\left(1+y^{\prime 2}+z^{\prime 2}\right)^{-1/2}>0$ for any trajectory. From eq (2.4), $\|\mathrm{Q}(t)\|=\|0\|$. Along the extremal, where $y(x)=ax+y_{0}$, $z(x)=bx+z_{0}$, equation (29) gives $\mathrm{P}\cdot{\bf h}^{\prime}={\bf c}$, with ${\bf c}$ a constant vector and $\mathrm{P}$ a positive definite matrix of constant values with respect to $x$. Solving this first-order equation gives straight line solutions for ${\bf h}(x)$. Because ${\bf h}(x_{0})=0$, there can be no conjugate points, and because ${\bf h}(x_{1})=0$, the only solution to (29) is the trivial one, and the extremum is a minimum. #### 2.4.2 Geodesics on the surface of a sphere Taking the azimuthal angle $\phi$ as the independent variable, and polar angle $\theta(\phi)$ as the dependent variable, the arc-length on the surface of a unit sphere may be written as ${\cal D}[\theta]=\int_{\phi_{0}}^{\phi_{1}}\\!\\!\\!d\phi\>\sqrt{\theta^{\prime 2}+\sin^{2}\theta}\>.$ (36) The EL equations give the extremal trajectory as $\cos\theta=A\sin\theta\cos\phi+B\sin\theta\sin\phi$ with $A,B$ constants. This is the equation of a plane $z=Ax+By$, which intersects the surface of the sphere to make a great circle. The scalar $P=\mathcal{L}_{\theta^{\prime}\theta^{\prime}}=\sin^{2}\theta/\left(\theta^{\prime 2}+\sin^{2}\theta\right)^{3/2}$ which is always positive. To simplify the problem, let $\phi_{0}=0$, and $\theta(\phi_{0})=\theta(\phi_{1})=\pi/2$, so the great circle lies in the $z=0$ plane. Along this extremal $P$ is constant and equal to $1$, while $Q=-1$. The second variation, eq. (32), is then $(1/2)\int_{0}^{\phi_{1}}\\!\\!\\!d\phi\>\left(h^{\prime 2}-h^{2}\right)$. The corresponding Jacobi equation, $h^{\prime\prime}+h=0$, must not have a root between $[0,\phi_{1}]$. The nontrivial solution to the Jacobi equation satisfying the initial condition $h(0)=0$ is $h(\phi)=C\sin\phi$, which has a conjugate point at $\phi=\pi$. Thus for the extremal curve to be minimal, $\phi_{1}$ must be $<\pi$, the location of the opposite pole on the sphere. If $\phi_{1}<\pi$, there is no extremal solution for $h(\phi)$ other than the trivial one which satisfies the boundary conditions. It is instructive to look at the arc-length under sinusoidal variations around the extremal path which satisfy the boundary conditions $h(0)=h(\phi_{1})=0$, so that $\theta(\phi)=\pi/2+h(\phi)=\pi/2+\epsilon\sin\left(\pi\phi/\phi_{1}\right)$. Inserting this into eq (36) above and expanding to second order in $\epsilon$, we see that first order terms in $\epsilon$ vanish, and the difference in distance from the extremal path is $\Delta{\cal D}=\left(\epsilon^{2}/4\phi_{1}\right)\left(\pi^{2}-\phi_{1}^{2}\right)$. For $\phi_{1}<\pi$ this is always greater than zero indicating the extremal is a minimum. For $\phi_{1}>\pi$ this is always less than zero indicating the extremal is a maximum with respect to these perturbations: the length may be shortened. When $\phi_{1}=\pi$, $\Delta{\cal D}=0$ to second order. When $h(\phi)$ represents the difference between great circles $\Delta{\cal D}$ is precisely zero. #### 2.4.3 Harmonic oscillator It is not widely appreciated that the classical action for a simple harmonic oscillator is not always a minimum, and indeed in many cases can be a maximum with respect to some perturbations. The action for a harmonic oscillator with given spring constant is proportional to $S[x]=\int_{0}^{T}\\!\\!\\!dt\>{1\over 2}(\dot{x}^{2}-x^{2})$, which has EL equation $\ddot{x}+x=0$. Taking the specific initial conditions $x(0)=1$, $\dot{x}(0)=0$, the extremal solution is $x(t)=\cos t$. The scalar $P(t)=\mathcal{L}_{\dot{x}\dot{x}}=1$, which is always positive and satisfies the necessary conditions for a minimum. The scalar $Q=\mathcal{L}_{xx}-{d\over dt}\mathcal{L}_{x\dot{x}}=-1$. The second variation $\delta^{2}S[h]={1\over 2}\int_{0}^{T}\\!\\!\\!dt\>(\dot{h}^{2}-h^{2})$, which has Jacobi equation $\ddot{h}+h=0$. This is the same Jacobi equation as that for geodesics on a sphere, so the sufficient conditions will parallel those above. The boundary condition $h(0)=0$ gives $h(t)=A\sin t$, with conjugate points at $t=n\pi$, $n=1,2,\ldots$. This means that the action is a minimum only so long as $T<\pi$, i.e. a half-period. If we let $x(t)$ be the extremal solution plus a $\sin$ perturbation satisfying the Jacobi equation at the conjugate points: $x(t)=\cos t+\epsilon\sin t$, then the difference in action from the extremal path becomes $\Delta S=(\epsilon^{2}/4T)(\pi^{2}-T^{2})$. This result is exact because the action for the oscillator is quadratic (as opposed to the action for geodesics). When $T<\pi$, $\Delta S>0$ indicating the extremal is a minimum. When $T$ is larger than a half-period, $\Delta S<0$ and the extremal trajectory is a maximum (with respect to half-wavelength sinusoidal perturbations), and when $T=\pi$, the end point is the conjugate point and $\Delta S=0$. We discuss sufficient conditions further below in the context of minimal transformations for links. ## 3 Single Links In the limit of one link, equations (15a-15c) reduce to: $\displaystyle\dot{\hat{{\bf v}}}_{\mbox{\tiny{A}}}+\lambda\,{\bf r}_{\mbox{\tiny{B/A}}}=0$ $\displaystyle\dot{\hat{{\bf v}}}_{\mbox{\tiny{B}}}-\lambda\,{\bf r}_{\mbox{\tiny{B/A}}}=0$ (37) where we have let $A$ represent point $1$, $B$ point $2$, and $\lambda\equiv\lambda_{12}$. The link has length $1$ in our dimensionless formulation, so the vector ${\bf r}_{\mbox{\tiny{B/A}}}$ could also have been written as a unit vector $\hat{{\bf r}}_{\mbox{\tiny{B/A}}}$. Both points $A$ and $B$ are end points and satisfy the boundary conditions of section 2.3.1. This means that points $A$ and $B$ move by either pure rotation, straight-line translation, or remain at rest. The initial and final conditions may be written ${\bf r}_{\mbox{\tiny{A}}}(0)={\bf A}$, ${\bf r}_{\mbox{\tiny{B}}}(0)={\bf B}$, ${\bf r}_{\mbox{\tiny{A}}}(T)={\bf A}^{\prime}$, ${\bf r}_{\mbox{\tiny{B}}}(T)={\bf B}^{\prime}$. The link in our problem has direction, so $A$ must transform to $A^{\prime}$ and $B$ to $B^{\prime}$. We will often use arrowheads in figures to denote this direction. ### 3.1 Straight line transformations As a first example, consider the two links shown in figure 3a. The four points $A,B,A^{\prime},B^{\prime}$ need not lie in a plane (see for example fig 3b). Let angle $\angle BAA^{\prime}\equiv a$ be obtuse. We draw straight lines from $A$ to $A^{\prime}$ and $B$ to $B^{\prime}$, and ask whether such a transformation is possible. We can thus derive the following rule: $\bullet$ For a straight line transformation to exist between two links, opposite angles of the quadrilateral made by $\overline{AB}$, $\overline{A^{\prime}B^{\prime}}$, $\overline{AA^{\prime}}$, $\overline{BB^{\prime}}$ must be obtuse. | | ---|---|--- a | b | c Figure 3: Possible (a,b) and impossible (c) straight line transformations between links $AB$ and $A^{\prime}B^{\prime}$. Figure $b$ shows a straight line transformation where the initial and final states do not lie in the same plane. In the text we derive the conditions for the possibility of a straight line transformation between links. Let the length that point $A$ travels be $x_{\mbox{\tiny A}}$, i.e. we imagine the point $A^{\prime}$ and the distance $x_{\mbox{\tiny A}}=|AA^{\prime}|$ to be variable. The length $r_{\mbox{\tiny B}}$ that point $B$ travels is then a function of $x_{\mbox{\tiny A}}$ and the original angle $a$, $r_{\mbox{\tiny B}}(x_{\mbox{\tiny A}},a)$. We can now find conditions on the angle $b\equiv\angle BB^{\prime}A^{\prime}$ such that the transformation is possible. After some distance $x_{\mbox{\tiny A}}$ travelled by point $A$, the length of the line from $B$ to $A^{\prime}$ is $\displaystyle\overline{BA^{\prime}}$ $\displaystyle=$ $\displaystyle x_{\mbox{\tiny A}}^{2}+1-2x_{\mbox{\tiny A}}\cos a$ $\displaystyle=$ $\displaystyle r_{\mbox{\tiny B}}^{2}+1-2r_{\mbox{\tiny B}}\cos b$ so that $r_{\mbox{\tiny B}}(x_{\mbox{\tiny A}},a)=\cos b\pm\sqrt{\cos^{2}b+f(x_{\mbox{\tiny A}},a)}$ with $f(x_{\mbox{\tiny A}},a)=x_{\mbox{\tiny A}}^{2}-2x_{\mbox{\tiny A}}\cos a$. Since $a$ is obtuse, $f>0$ when $x_{\mbox{\tiny A}}>0$, and so the positive root must be taken for $r_{\mbox{\tiny B}}$ to positive. When $x_{\mbox{\tiny A}}=0$, $f(0,a)=0$, and $r_{\mbox{\tiny B}}(0,a)=\cos b+\left|\cos b\right|=0$ Therefore $b$ must also be an obtuse angle. If two opposite angles are obtuse, then the other two angles must be acute. This concludes the proof that the above conditions are sufficient. An additional proof that they are necessary is given in A. We readily see that figure 3A is one pair of a larger set of straight line transformations that can continue until one or both of the obtuse angles reaches $90^{\circ}$. This collection forms a “bow tie” of admissible configurations, as in figure 3.1. Note that straight lines in the quadrilateral may cross as in the transformation from $A,B$ to $A^{\prime},B^{\prime}$ in figure 3.1. Trivial translations of the link without any concurrent rotation are a special case of general straight line transformations. | ---|--- A | B Figure 4: (A) An example of a set of link configurations connected by a straight-line transformation. The link rotates clockwise as it translates to allow the end points to move in straight lines. The translation can proceed no farther than the end points $AB$ and $A^{\prime}B^{\prime}$, which have link vectors $\overrightarrow{AB}$ or $\overrightarrow{A^{\prime}B^{\prime}}$ that are perpendicular to one or the other of the vectors $\hat{{\bf v}}_{\mbox{\tiny A}}$ or $\hat{{\bf v}}_{\mbox{\tiny B}}$. The totality of states thus connected forms a “bowtie”. (B) A bowtie where the terminal states $AB$ and $A^{\prime}B^{\prime}$ happen to cross each other. ### 3.2 Piece-wise extremal transformations: transformations with rotations An immediate question is the nature of the transformation between $AB$ and $A^{\prime}B^{\prime}$ in figure 3C, where opposite angles of the quadrilateral are not obtuse. Recall our link has direction so $A$ cannot transform to $B^{\prime}$. Then direct straight-line solution is not possible due to the constraint of constant link length. The only remaining solution is for the link to rotate as part of the transformation. Consider first the rotation of link $AB$. The EL equations (3) allow for pure rotations about $A$, $B$, or a common center along the link. Likewise for link $A^{\prime}B^{\prime}$. The rotation can occur from either link $AB$ (fig 5a) or link $A^{\prime}B^{\prime}$ (fig 5b). After the link rotates to a critical angle, it can then travel in a straight line. The extremals are broken in that they involve matching up a piece consisting of pure rotation with a piece consisting of pure translation of the end points of the link. Where the pieces match they must satisfy the corner conditions (18a, 18b). This means that the end points cannot suddenly change direction, a situation which is only satisfied by a straight line trajectory that lies tangent to the circle of rotation. From figure 3.1, we see that a straight line transformation exists only when an angle between a link and one of the straight line trajectories reaches $\pi/2$. The critical angle that link $AB$ must rotate is then determined by the point where a line drawn from $B^{\prime}$ is just tangent to the unit sphere centered at point $A$, point $B_{1}$ in figure 5a. There is generally a different critical angle if the rotation occurs at link $A^{\prime}B^{\prime}$ as in fig 5B. It is shown in B that in general the critical angle is determined by drawing the tangent to a circle or sphere about one of the link ends. a | | b | ---|---|---|--- | | | c | | d | | | | e | | | | | | Figure 5: Transformations between two links involving broken extremals consisting of rotation and translation. $(b)$ is the global minimum, with shortest distance travelled during the transformation. $(a)$, $(c)$, and $(d)$ are local minima. $(e)$ is extremal, but not minimal as the trajectory of arc $\stackrel{{\scriptstyle\frown}}{{B^{\prime}B_{1}}}$ passes through a conjugate point. If the rotation was about a common center, we see that one or another of the link ends would violate a corner condition, so the rotation must be about one of the link ends. According to eqs. (26) and (27), the matrix $\mathrm{P}$ has a determinant of zero due to the parametric formulation in the problem and so is not positive definite. To show that the transformations in fig. 5a,b are indeed minimal, we need to then express the problem in non-parametric form. To do this, let the independent variable be the angle $\theta$ of the link with the vertical. Then the displacement $x$ along the line $\overline{AA^{\prime}}$ is the unknown function of $\theta$ to be determined by minimizing the total arc length travelled. This distance can be written as ${\cal D}[x]=\int_{\theta_{0}}^{\theta_{1}}\\!\\!\\!d\theta\>\left(\sqrt{x^{\prime 2}+2x^{\prime}\cos\theta+1}+\sqrt{x^{\prime 2}}\right)$ In this formulation, the scalar quantity $\mathrm{P}(\theta)=\mathcal{L}_{x^{\prime}x^{\prime}}$ becomes $\mathrm{P}(\theta)={\sin^{2}\theta\over\left(x^{\prime 2}+2x^{\prime}\cos\theta+1\right)^{3/2}}$ which is always $>0$ except for the isolated point $\theta=0$, in particular it is positive along the extremal trajectory which is necessary for a minimum. So we conclude that the transformation with the smaller angle of rotation in fig 5b is here the global minimum, and the other transformation (fig 5a) is a local minimum. Figure 5e is also an extremal trajectory, satisfying corner conditions, and with positive definite $P$. However it is not a local minimum because the trajectory passes through a conjugate point (denoted by point $CP$, where the dotted line along $\overline{A^{\prime}B^{\prime}}$ meets the great circle about $A^{\prime}$). According to the results in section 2.4.2, if the extremal trajectory (a great circle) traverses an angle larger than $\pi$ radians, it passes through a conjugate point and thus becomes unstable to long-wavelength perturbations. Transformations involving rotations about points $B$ or $B^{\prime}$ in figure 5 both have conjugate points and so are not minimal. The transformation in fig. 5c does not pass through a conjugate point and so is in fact another local minimum. The part of the extremum along the straight line section of the trajectory has no conjugate points as discussed above. ### 3.3 Systematically exploring transformations by varying link positions We can investigate what happens to the minimal transformation when one of the link positions or angles is varied with respect to the other. Let us start by putting the two links head to tail as shown in figure 6A. The distance between them is $2$ by simple translation of link end points. We can now increase the angle between the two vectors by rotating the right link for example, as in figures 6B-H. So long as the angle between the two vectors is less than $90^{\circ}$, one link may slide along another and the distance is unchanged (figs 6A-C). This is a special case of the transformations shown in figure 3.1 (compare for example figure 6B with the middle three unlabelled links in that figure). Beyond $90^{\circ}$ however, the transformation must include rotation. Fig 6D has an angle of $150^{\circ}$. The minimal transformation first rotates, for example with the tail of the horizontal black arrow fixed, and the head tracing out the blue arc, until the critical angle is reached, where a straight line made from the final arrowhead (at the top of the figure) is just tangent to the circle made by the blue arc. This state is indicated by a red link in figure 6D. The link then translates to its reciprocal position at the opposite end of the bowtie, denoted by a second red link (c.f. also figure 3.1B). At this point the arrowhead has completed the transformation. Finally the tail rotates into its final position. The total distance travelled is slightly larger than $2$. When the angle between the vectors is $120^{\circ}$ as shown in 6E, the transformation consists of pure rotations. Taking the initial state to be the horizontal black vector, the link first rotates about its fixed tail, the head tracing out the blue arc, until the link reaches the state shown in red, where the position of the arrowhead has reached its final end point. Then the link rotates about its head until the position of the tail reaches the final state. When the angle between the links is larger than $120^{\circ}$ as shown in figs 6F-G, the transformation must involve rotation about an internal point along the link. Let points $A$ and $B$ denote the tail and head of the link respectively. If an infinitesimal rotation $\Delta\theta$ occurs about an internal point $P$, the increment in distance travelled is $\Delta{\cal D}=|{\bf r}_{\mbox{\tiny{B/P}}}|\Delta\theta+|{\bf r}_{\mbox{\tiny{B/A}}}|\Delta\theta=\Delta\theta$ which is independent of the position of the instantaneous center of rotation (ICR). This means that there are an infinity of transformations all giving the same distance, depending on the time-dependence of the ICR. Two simple alternatives with only two discrete positions of ICR are shown in figures 6F,G. Specifically, in figure 6F, the horizontal black vector first rotates about its tail to the red configuration, which is a mirror image of the final black vector. Then rotation is about an internal point determined by the intercept of the red vector with the final black vector, with end points tracing out the green arcs. In figure 6G the two ICRs are both internal and determined by the intercepts of the initial and final states with the red vector shown. Figure 6H depicts the transformation for overlapping, opposite pointing vectors. Rotation can now only occur about one point in the center of the vectors. a | | b | | c | ---|---|---|---|---|--- | | | | | d | | e | | f | | | | | | g | | h | | | | | | | | Figure 6: Successive transformations between two links made by rotating a link so that there is a progressively larger angle between the links as vectors (or smaller angle made between them as lines). The two boundary conditions (the initial and final conditions) are shown as black links, and an intermediate state is shown as a red link or links. The arcs traced out by the end points are shown in blue or green, while straight line motions when they are not along the links themselves are shown in grey. The distance travelled over the course of the transformation is given below each figure. Figure 7 illustrates what happens when one of the links is translated with respect to another, starting from two different scenarios shown in 7A and 7E. In 7A, the tail of the vertical link is displaced $(1/3,-1/3)$ with respect to the tail of the horizontal link. The minimal transformation is a pure rotation by $\pi/2$. In figure 7B, the tail of the vertical link is now displaced to $(2/3,-1/3)$. Pure rotations again give a distance of $\pi/2$. Rotation about a point on the horizontal link that is equidistant from both arrowheads transforms the initial arrowhead to the final (red intermediate state). Then rotation of the tail about the arrowhead transforms to the final state. In figure 7C, the minimal transformation first involves a translation by sliding the arrowhead along the vertical, until the arrowheads overlap (red intermediate state). The tail end of the link then rotates into place. In figure 7D, straight lines from the end points will not satisfy the obtuse condition in section 3.2, so the transformation must involve rotations. Here a straight line transformation takes the link almost to the final state. It then must undergo a small rotation to complete the transformation. Seen in reverse, the vertical arrow must rotate to a critical angle determined by the criterion in section 3.2, before the link can finish the transformation by pure translation. Figure 7E is figure 6F once again. The final condition (the tilted link) will be systematically changed by translating it vertically away from the horizontal link (which we choose arbitrarily as the initial configuration). In figure 7F the tilted link is translated a distance $1/3$ vertically. The transformation can be achieved by rotating the horizontal link about a point equidistant from both arrowheads, to the red intermediate configuration. The link then rotates about the arrowhead into the final configuration. The distance is still the angle rotated for the reasons mentioned above in the context of figures 6F-G, $\theta=(150/180)\pi$, which is unchanged from 7E. In fact, so long as the arrowhead can be reached by rotation (the translated distance is less than $d$ where $d$ is the solution to $d^{2}+d+1-\sqrt{3}=0$ for this angle), then the distance will be unchanged. The transformation at the critical distance is shown in figure 7G. The rotations now occur about the end-points: the tail and head of the link. In figure 7H the translated distance is now equal to $1$. The transformation first consists of a rotation about the tail to a critical angle (blue arc and red intermediate state), then a translation much like that in figure 3.1 (grey straight lines between red intermediate states), and finally a rotation about the head (green arc) to the final configuration. | a | | e ---|---|---|--- | | | | b | | f | | | | c | | g | | | | d | | h | | | Figure 7: Successive transformations between two links made by translating one link with respect to the other. In (a-d) the initial and final configurations are perpendicular, while in (e-h) they are at an angle of $150^{\circ}$ to each other. Note the distances in (e-g) are all the same, even though the end points of the links are at varying distances from each other. ## 4 2-link chains We now consider the next simplest case of $2$ links ($3$ beads). The Lagrangian now reads: $\mathcal{L}({\bf r}_{1},{\bf r}_{2},{\bf r}_{3},\dot{{\bf r}}_{1},\dot{{\bf r}}_{2},\dot{{\bf r}}_{3})=\sqrt{\dot{{\bf r}}_{1}^{2}}+\sqrt{\dot{{\bf r}}_{2}^{2}}+\sqrt{\dot{{\bf r}}_{3}^{2}}-{1\over 2}\lambda_{12}\left(\left({\bf r}_{2}-{\bf r}_{1}\right)^{2}-1\right)-{1\over 2}\lambda_{23}\left(\left({\bf r}_{3}-{\bf r}_{2}\right)^{2}-1\right)$ (38) which has EL equations (c.f. eq.s 15a-15c) ‡**footnotetext: ‡ The links have length $1$ in our dimensionless formulation, so the vectors ${\bf r}_{\mbox{\tiny{B/A}}}$ and ${\bf r}_{\mbox{\tiny{C/B}}}$ could also have been written as unit vectors $\hat{{\bf r}}_{\mbox{\tiny{B/A}}}$ and $\hat{{\bf r}}_{\mbox{\tiny{C/B}}}$. : $\displaystyle\dot{\hat{{\bf v}}}_{\mbox{\tiny{A}}}+\lambda_{\mbox{\tiny{AB}}}\,{\bf r}_{\mbox{\tiny{B/A}}}=0$ (39a) $\displaystyle\dot{\hat{{\bf v}}}_{\mbox{\tiny{B}}}-\lambda_{\mbox{\tiny{AB}}}\,{\bf r}_{\mbox{\tiny{B/A}}}+\lambda_{\mbox{\tiny{BC}}}\,{\bf r}_{\mbox{\tiny{C/B}}}=0$ (39b) $\displaystyle\dot{\hat{{\bf v}}}_{\mbox{\tiny{C}}}-\lambda_{\mbox{\tiny{BC}}}\,{\bf r}_{\mbox{\tiny{C/B}}}=0\>.$ (39c) The corner conditions (18a), (19) imply $\hat{{\bf v}}_{i}\left(t^{-}\right)=\hat{{\bf v}}_{i}\left(t^{+}\right)$ so the direction of motion cannot suddenly change, unless along one part of the extremal the velocity of point $i$ is zero (the point is at rest), where its direction $\hat{{\bf v}}$ is then undefined. The boundary conditions described in section 2.3.1 hold as well, so the end points can either be at rest, move in straight lines, or purely rotate. This gives $3\times 3=9$ possible scenarios to investigate here, many of which can readily be ruled out. For example consider the states in figure 8a. Because $A$ and $A^{\prime}$ are in the same position, rotation and translation of $A$ are ruled out and point $A$ remains at rest, leaving $3$ scenarios for the other end point $C$. However since $C$ and $C^{\prime}$ are at different positions and $ABC$ are along a straight line, $C$ cannot remain at rest initially, leaving either translation or rotation for point $C$. a | | b | ---|---|---|--- | | | c | | d | | | | Figure 8: (a) Initial and final states for a chain of two links. The transformation in (b) is non-extremal because it violates a corner condition at $C^{\prime\prime}$. (c) and (d) are degenerate minima- rotations occurring about $B^{\prime}$ or $B$ both have the same length. Intermediate states is shown in red have opposite convexity in (c) and (d). Suppose $C$ translates towards $C^{\prime}$ as in figure 8b. Then $\dot{\hat{{\bf v}}}_{\mbox{\tiny{C}}}=0$ and from (39c,39b) $\lambda_{\mbox{\tiny{BC}}}=0$ and $\dot{\hat{{\bf v}}}_{\mbox{\tiny{B}}}=\lambda_{\mbox{\tiny{AB}}}\,{\bf r}_{\mbox{\tiny{B/A}}}$. $B$ cannot move in a straight line without moving point $A$, so $\lambda_{\mbox{\tiny{AB}}}\neq 0$ and thus $B$ must rotate about point $A$. The transformation then proceeds as in figure 8b until $B$ reaches $B^{\prime}$ and $C$ reaches $C^{\prime\prime}$. Then however if $C^{\prime\prime}$ were to rotate to $C^{\prime}$, the trajectory would violate corner conditions at point $C^{\prime\prime}$. Therefore the direction of translation of $C$ must not be directly to $C^{\prime}$ but must be tangential to the arc $\stackrel{{\scriptstyle\frown}}{{C^{\prime}C^{\prime\prime}}}$ as in figure 8c. The reverse of this transformation is allowable as well, as can be seen by swapping the labels $ABC\rightarrow A^{\prime}B^{\prime}C^{\prime}$. Here $C$ first rotates to the critical angle $\theta$ shown in fig 8d and then translates to $C^{\prime}$. In fact one can see that links $BC$ and $B^{\prime}C^{\prime}$ along with lines $\overline{BB^{\prime}}$ and $\overline{CC^{\prime}}$ form a quadrilateral as in figure 5, with the same consequences for rotation to a critical angle. For the links in fig 8 the situation is symmetric so rotation can occur at the beginning or end of the transformation. Figure 9a shows an example with this symmetry broken, so that the distance is different depending where the rotation occurs, as in figures 5a,b. In this case, the transformation in fig 9c has the minimal distance, and that in fig 9b is subminimal. Extensions of the transformation in figure 9 to large numbers of links were explored in [1]. | | ---|---|--- a | b | c Figure 9: (a) Initial and final states for a polymer of $2$ links. The angle between $\overline{AB}$ and $\overline{A^{\prime}B^{\prime}}$ is $\pi/4$. The minimal transformations in (b) and (c) are now no longer degenerate. (c) is the global minimum. ### 4.1 Transformations involving a change in convexity Transformations between configurations with opposite convexity involve motion out of the plane, even if the initial and final states lie in the plane. If the transformation is constrained to lie in plane, the trajectories of some points will be non-monotonic- those points must move farther away from their final positions before approaching them. We illustrate these ideas with some examples below. Figure 10: A transformation between two states of opposite convexity: $ABC$ has convexity down and right, while $A^{\prime}B^{\prime}C^{\prime}$ has convexity up and left. There is no extremal transformation in the plane that can connect them, without some apparent violation of corner conditions. Consider the initial and final states in figure 10. We again imagine $B$ rotating to $B^{\prime}$. If C were to translate to $C^{\prime}$ one would have the intermediate configuration $A^{\prime}B^{\prime\prime}C^{\prime}$. Now $C^{\prime}$ and $A^{\prime}$ must remain at rest to satisfy corner conditions. Then the only way to finish the transformation is for $B^{\prime\prime}$ to rotate about the axis $\overline{A^{\prime}C^{\prime}}$, however then the trajectory of $B$ violates corner conditions and so is not extremal. In C we take up the issue of minimal transformations for this case when the links are constrained to lie in a plane. We thus seek a point $B^{\prime\prime}$ and resulting trajectory $\overrightarrow{BB^{\prime\prime}B^{\prime}}$ such that arc $\stackrel{{\scriptstyle\frown}}{{BB^{\prime\prime}}}$ satisfies corner conditions with arc $\stackrel{{\scriptstyle\frown}}{{B^{\prime\prime}B^{\prime}}}$. One solution is to effectively place $B^{\prime\prime}$ at position $B^{\prime}$ by considering the boundary condition with $C$ at rest (and $A$ at rest). Then $B$ rotates to $B^{\prime}$ about axis $\overline{AC}$, and the trajectory of $B$ lies on a circle defined by the intercept of two unit spheres centered at $A$ and $C$. The sphere about $A$ is drawn in figure 11 as a visual aid. Along arc $\stackrel{{\scriptstyle\frown}}{{BB^{\prime}}}$ both $\lambda_{\mbox{\tiny{AB}}}\neq 0$ and $\lambda_{\mbox{\tiny{BC}}}\neq 0$. Once in configuration $A^{\prime}B^{\prime}C$, $C$ can then undergo rotation about $B^{\prime}$ to $C^{\prime}$, with $A^{\prime}$ and $B^{\prime}$ stationary. The transformation in 11a is a local minimum in distance, however it is not the global minimum. A shorter distance transformation can be seen by considering the reverse transformation. Imagine $A^{\prime}$ and $C^{\prime}$ stationary while $B^{\prime}$ rotates about axis $\overline{A^{\prime}C^{\prime}}$ in figure 11b. This rotation of $B^{\prime}$ follows a circular trajectory defined by the intercept of two unit spheres centered at $A^{\prime}$ and $C^{\prime}$. The rotation occurs until point $B^{\prime\prime}$, which is the point where above circle is tangent to a great circle on the unit sphere about $A$ and passing through $B$. The arc $\stackrel{{\scriptstyle\frown}}{{BB^{\prime\prime}}}$ is a great circle because this is a geodesic for point $B$ given $A$ is fixed, which follows from the Euler equations (39b,39c) when $\lambda_{\mbox{\tiny{BC}}}=0$. The great circle is defined by the plane containing the points $A$, $B$, and $B^{\prime\prime}$. The angle between the (variable) vector $\overrightarrow{BC}$ of link $BC$ and the tangent the the arc $\stackrel{{\scriptstyle\frown}}{{B^{\prime}B^{\prime\prime}}}$ is always $\pi/2$, so once the corner condition is met, point $C$ on link $BC$ can move in straight line motion from $C^{\prime}$ to $C$ while $B$ moves on the great circle from $B^{\prime\prime}$ to $B$. That is, the quadrilaterial criterion of section 3.1 is met for $\Box BB^{\prime\prime}C^{\prime}C$. To find point $B^{\prime\prime}$, let its position be ${\bf r}_{\mbox{\tiny{B''}}}=(x_{o},y(x_{o}),z(x_{o}))$. The great circle is defined by the plane passing through the points $A$, $B$, and $B^{\prime\prime}$. This plane has normal ${\bf n}\equiv\overrightarrow{AB}\times\overrightarrow{AB^{\prime\prime}}=(1,0,0)\times(x_{o},y(x_{o}),z(x_{o}))=(0,-z(x_{o}),y(x_{o}))$. At the point $B^{\prime\prime}$ the normal is orthogonal to the tangent vector of the circle defined by rotation about the $AC^{\prime}$ axis. This tangent vector is $\hat{{\bf t}}=\partial{\bf r}/\partial s=x_{s}(1,y_{x},z_{x})$ by the chain rule. At $B^{\prime\prime}$, $\hat{{\bf t}}\cdot{\bf n}=0$, or $-z(x_{o})y_{x}(x_{o})+y(x_{o})z_{x}(x_{o})=0$ (40) The functions $y(x)$ and $z(x)$ are defined by the intercept of two unit spheres centered at $(0,0,0)$ and $(1/\sqrt{2},1+1/\sqrt{2},0)$, giving $\displaystyle y(x)$ $\displaystyle=1-{\sqrt{2}\over{2+\sqrt{2}}}x$ $\displaystyle z(x)$ $\displaystyle=\sqrt{1-x^{2}-y(x)^{2}}\>.$ (41) Together (40) and (41) give $\displaystyle{\bf r}_{\mbox{\tiny{B''}}}=\begin{pmatrix}\sqrt{2}-1\\\ 2(\sqrt{2}-1)\\\ \sqrt{2(5\sqrt{2}-7)}\end{pmatrix}$ The distance travelled along arc $\stackrel{{\scriptstyle\frown}}{{BB^{\prime\prime}}}$ is $\theta_{\mbox{\tiny{BB''}}}$, where $\cos\theta_{\mbox{\tiny{BB''}}}=x_{o}=\sqrt{2}-1$. The distance travelled along arc $\stackrel{{\scriptstyle\frown}}{{B^{\prime\prime}B^{\prime}}}$ can similarly be shown to be $r\theta_{\mbox{\tiny{B''B'}}}=\sin(\pi/8)\cos^{-1}(2\sqrt{2}-3)$. Adding the distance $\overline{CC^{\prime}}$, the total (minimal) distance is thus ${\cal D}=2.576$. There is of course a degenerate solution to the above with $z\rightarrow-z$. | ---|--- a | b Figure 11: Subminimal (a) and minimal (b) transformations for the boundary conditions in figure 10. The distances for each transformation are approximately $3.007L^{2}$ and $2.576L^{2}$ respectively. Transformation (a) proceeds from $ABC$ by first rotating $B$ to $B^{\prime}$ about axis $\overline{AC}$, then rotating $C$ about point $B^{\prime}$. Transformation (b) proceeds from $ABC$ by simultaneously translating $C$ to $C^{\prime}$ while rotating $B$ about $A$ on a great circle to point $B^{\prime\prime}$. Finally point $B$ rotates from $B^{\prime\prime}$ to $B^{\prime}$ about axis $A^{\prime}C^{\prime}$. ### 4.2 Transformations with initial and final states in 3-D We now give a representative example where the initial and final configurations do not lie in the same plane, as shown in figure 12. Because $\overline{AB}\perp\overline{AA^{\prime}}$ and $\overline{BC}\perp\overline{CC^{\prime}}$, neither $A$ nor $C$ will rotate about $B$ as part of the transformation. Nor can $ABC$ simultaneously translate directly to $A^{\prime}B^{\prime}C^{\prime}$, because for example quadrilateral $\Box AA^{\prime}B^{\prime}B$ does not satisfy the rule of opposite angles $\geq\pi/2$, so link $AB$ cannot slide (translate) to $A^{\prime}B^{\prime}$. This leaves $3$ options for the initial stages of the transformation: 1.) $A$ translates, $B$ rotates, $C$ remains fixed. $B$ then rotates about $C$ in the $CBB^{\prime}$ plane. The initial direction of motion of $B$ is then $\hat{{\bf v}}_{\mbox{\tiny B}}=(-\hat{{\bf i}}+\hat{{\bf k}})/\sqrt{2}$, however then $\hat{{\bf v}}_{\mbox{\tiny A}}$ can only move backward to preserve link length ($\hat{{\bf v}}_{\mbox{\tiny A}}=-\hat{{\bf k}}$), similar to figure 15. This rules out case (1). 2.) $A$ remains fixed, $B$ rotates, $C$ remains fixed. $B$ then rotates towards $B^{\prime}$ about axis $\overline{AC}$ until it reaches a critical angle where line $\overline{B^{\prime\prime}B^{\prime}}$ is tangent to its circular trajectory (see fig. 12a). At this point the quadrilateral $\Box B^{\prime\prime}CC^{\prime}B^{\prime}$ does not have opposite obtuse angles, so a straight line transformation to $A^{\prime}B^{\prime}C^{\prime}$ is not possible. It is possible to transform to a configuration $A^{\prime}B^{\prime}C^{\prime\prime}$, where $C^{\prime\prime}$ is at position $(1,1,1)$ and angle $\angle B^{\prime}C^{\prime\prime}C=\pi/2$, so that $\hat{{\bf v}}_{\mbox{\tiny C}}=\hat{{\bf k}}$. Then the transformation is completed by a $\pi/2$ rotation of $C^{\prime\prime}$ about $B^{\prime}$. This transformation is subminimal. 3.) $A$ remains fixed, $B$ rotates, $C$ translates. In this case, $B$ rotates toward $B^{\prime}$ in the $BAB^{\prime}$ plane, while $C$ translates to $C^{\prime}$, until the state $AB^{\prime\prime}C^{\prime\prime}$ is reached (see fig. 12b). State $AB^{\prime\prime}C^{\prime\prime}$ can be found as follows. Because the rotation of $B$ is about the axis $(0,-1/\sqrt{2},1/\sqrt{2})$, the position $\overrightarrow{AB^{\prime\prime}}$ of $B^{\prime\prime}$ after rotation of the (critical) angle $\theta$ is $(\cos\theta,\sin\theta/\sqrt{2},\sin\theta/\sqrt{2})$. This angle is then determined by the condition $\overrightarrow{AB^{\prime\prime}}\cdot\overrightarrow{B^{\prime\prime}B^{\prime}}=0$, where $\overrightarrow{B^{\prime\prime}B^{\prime}}=\overrightarrow{AB^{\prime}}-\overrightarrow{AB^{\prime\prime}}$. The solution to this condition is simply $\theta=\pi/4$. The location of $C^{\prime\prime}$ is then determined from the condition that the link length from $B^{\prime\prime}$ to $C^{\prime\prime}$ is one: $|\overrightarrow{B^{\prime\prime}C^{\prime\prime}}|=1$, where $\overrightarrow{B^{\prime\prime}C^{\prime\prime}}=\overrightarrow{AB^{\prime\prime}}+t\overrightarrow{CC^{\prime}}$. Solving this condition for $t$ gives the position of $C^{\prime\prime}$ as $({3+\sqrt{2}\over 5},1,{2(2-\sqrt{2})\over 5})$. At this point the quadrilateral $\Box B^{\prime}B^{\prime\prime}C^{\prime\prime}C^{\prime}$ has opposite obtuse angles, and quadrilateral $\Box AB^{\prime\prime}B^{\prime}A^{\prime}$ has opposite angles $=\pi/2$, so it is in a bowtie configuration as in the end point configurations in figure 3.1. Therefore all points $AB^{\prime\prime}C^{\prime\prime}$ can translate from this intermediate state to their final positions $A^{\prime}B^{\prime}C^{\prime}$. The total distance travelled is $\theta+|AA^{\prime}|+|CC^{\prime}|+|B^{\prime\prime}B^{\prime}|$ or ${\cal D}=2+\pi/4+\sqrt{5}\approx 5.022$. The reverse of this transformation is also possible, where point $B^{\prime}$ rotates about $A^{\prime}$ in the plane $B^{\prime}AB$, while $C^{\prime}$ translates along $\overrightarrow{C^{\prime}C}$. Inspection reveals the distance covered is the same as the forward transformation. | ---|--- a | b Figure 12: (a) Subminimal transformation and (b) minimal transformations between $ABC$ and $A^{\prime}B^{\prime}C^{\prime}$ (see text) . ## 5 Limit of large link number From the transformation discussed in section 4.1, we see that if both $\angle ABC$ and $\angle A^{\prime}B^{\prime}C^{\prime}$ were $\pi/2$ as in figure 13a, then the transformations in figures 11a and 11b become degenerate, having distance ${\cal D}=\pi/\sqrt{2}$. The transformation is completed by a single rotation about axis $\overline{13}$. | ---|--- a | b | c | d Figure 13: Examples of transformations between initial and final states of opposite convexity, for increasing numbers of links. (a) illustrates the transformation for $N=2$ links. (b) $N=4$ and initial and final state form an octagon. (c,d) $N=6$ and initial and final states form a dodecagon. (c) top view. (d) view in perspective. Rotations are shown as solid color lines (either green or blue). Translations are shown as dashed lines. The grey dashed lines underneath $\overline{3^{\prime\prime}3^{\prime}}$ in (b) and $\overline{4^{\prime\prime}4^{\prime}}$ in (d) are shown only to illustrate that those lines are above the plane. We can now examine the effect of increasing the link number. Let the number of links increase to $4$, and let us preserve the symmetry that is present about the horizontal axis in fig 13a, so the initial and final states become an octagon (figure 13b). In the limit $N\rightarrow\infty$, the figure becomes a circle. If we separated the links in figure 13a by some distance in the $y$ direction (perpendicular to axis $\overline{13}$), then the minimal transformation involves the same rotation of $2$ about axis $\overline{13}$ until a critical angle $\theta_{c}$, after which all three points $123$ can translate in straight lines to $1^{\prime}2^{\prime}3^{\prime}$. In the same fashion, the minimal transformation for the octagonal transformation in fig 13b involves a rotation of point $3$ out of the plane about axis $\overline{24}$ to a critical angle $\theta_{c}$ at which the point is located at position $3^{\prime\prime}$. Once this critical angle is reached, point $3$ translates in a straight line from $3^{\prime\prime}$ to $3^{\prime}$. Because points $1$ and $5$ are stationary to satisfy corner conditions, points $2$ and $4$ must move in great circles about points $1$ and $5$. However points $2$ and $4$ cannot finish the transformation by moving on great circles. At the configuration $1^{\prime}2^{\prime\prime}3^{\prime}4^{\prime\prime}5^{\prime}$ in figure 13b, point $3$ has finished the transformation, but points $2$ and $4$ have not. To satisfy corner conditions at the points $2^{\prime\prime}$ and $4^{\prime\prime}$, the great circles must be out of plane as well. At points $2^{\prime\prime}$ and $4^{\prime\prime}$, the transformation finishes with rotations about axes $\overline{1^{\prime}3^{\prime}}$ and $\overline{3^{\prime}5^{\prime}}$. The total distance ${\cal D}\approx 7.93$. Of course the time reverse of this transformation (equivalent to swapping primed and unprimed labels) is also a minimal transformation, as is the transformation obtained by reflection about the $z=0$ plane. Now consider increasing the chain to $6$ links, so the combination of ${\bf r}_{i}(0)$ and ${\bf r}_{i}(T)$ becomes a dodecagon (12-sided polygon, see figures 13c-d). As before the midpoint vertex (here ${\bf r}_{4}$) must rotate out of the plane about axis $\overline{35}$ to a critical angle $\theta_{c}$ before translating in a straight line to ${\bf r}_{4^{\prime}}$. This critical angle is where $\overrightarrow{34^{\prime\prime}}\cdot\overrightarrow{4^{\prime\prime}4^{\prime}}=\overrightarrow{54^{\prime\prime}}\cdot\overrightarrow{4^{\prime\prime}4^{\prime}}=0$. The quadrilaterals $\Box 22^{\prime}3^{\prime}3$ and $\Box 655^{\prime}6^{\prime}$ are of the type in figure 5, so point $3$ must rotate about ${\bf r}_{2}(0)$ to a critical angle where $\overrightarrow{23^{\prime\prime}}\cdot\overrightarrow{3^{\prime\prime}3^{\prime}}=0$, and likewise for point $5$. While point $3$ rotates to its critical angle, point $4$ translates along line $\overline{4^{\prime\prime}4^{\prime}}$. Points ${\bf r}_{1}(0)$ and ${\bf r}_{7}(0)$ overlap with ${\bf r}_{1}(T)$ and ${\bf r}_{7}(T)$ and so remain fixed to satisfy corner conditions. After point $3$ has reached its critical angle, it can translate along $\overline{3^{\prime\prime}3^{\prime}}$ as point $2$ rotates about ${\bf r}_{1}$. However to satisfy corner conditions at point $2^{\prime\prime}$, the rotation cannot remain in the $x-y$ plane. Point ${\bf r}_{2^{\prime\prime}}$ is determined as the point where $\hat{{\bf t}}\cdot{\bf n}_{plane}=0$, where $\hat{{\bf t}}$ is the tangent to the arc $\stackrel{{\scriptstyle\frown}}{{22^{\prime\prime}}}$ defined by rotation about axis $\overline{13^{\prime}}$, and ${\bf n}_{plane}$ is the normal to the plane $122^{\prime\prime}$, i.e. ${\bf r}_{2/1}\times{\bf r}_{2^{\prime\prime}/1}$. The same process holds for point $6$. These critical points and some intermediate states for the transformation are shown in figure 13d. The total distance covered by the transformation is ${\cal D}\approx 16.3$. It is sensible to consider the total length of chain as fixed to say $L=1$, and to let the link length $ds_{\mbox{\tiny N}}$ for the chain of $N$ links be determined by $Nds_{\mbox{\tiny N}}=L$. Because distances scale as $ds_{\mbox{\tiny N}}^{2}$, the $N=2,4,6$ cases have ${\cal D}_{2}\approx 0.555L^{2}$, ${\cal D}_{4}\approx 0.496L^{2}$, ${\cal D}_{6}\approx 0.445L^{2}$. Note that this distance decreases with increasing number of links: the constraints on the motion of the various beads during the transformation are relaxed as the number of links is increased. We can then imagine resting a piece of string on a table in the shape of a semi-circular arc, and then asking how one can move this string to a facing semicircle of opposite convexity. So long as the string has some non-zero persistence length $\ell_{\mbox{\tiny P}}$, the transformation of minimal distance must involve lifting the string off of the table to change its local convexity. The vertical height the string must be lifted (see fig 13d) is of order $\sim\sin(\pi\ell_{\mbox{\tiny P}}/L)\sim\ell_{\mbox{\tiny P}}/L$, which goes to zero for an infinitely long chain. As the number of links $N\rightarrow\infty$, some simplifications emerge. In particular the contribution to the total distance due to rotations becomes negligible, and the translational component dominates. To see this note that the distance due to straight line motion scales as: ${\cal D}(\mbox{st. line})\sim ds\,NL\sim L^{2}$ while the distance travelled during rotations scales as ${\cal D}(\mbox{rot.})\sim ds\,N(\overline{\theta_{c}}ds)\sim L^{2}/N$ where we assume the worst case scenario where an extensive number of links must rotate before translating. Because translation dominates the distance as $N\rightarrow\infty$, the distance travelled converges to $L$ times the mean root square distance (MRSD), i.e. $\displaystyle{\cal D}_{\infty}$ $\displaystyle\rightarrow ds\,\sum_{i=1}^{N+1}\left|{\bf r}_{i}(T)-{\bf r}_{i}(0)\right|$ $\displaystyle=L{1\over N}\sum_{i}\sqrt{\left({\bf r}_{\mbox{\tiny{B}}i}-{\bf r}_{\mbox{\tiny{A}}i}\right)^{2}}$ $\displaystyle=L\left(\mbox{MRSD}\right)$ (42) The MRSD for the examples in figures 13b,d are $0.394\,L$ and $0.400\,L$ respectively, which are both less than the actual distances travelled (in units of $L$). In the limit $N\rightarrow\infty$, where the polygon becomes a circle, the distance converges to ${\cal D}_{\infty}=4L^{2}/\pi^{2}\approx 0.4053L^{2}$. For large $N$ systems then, it is a good first approximation to use $MRSD$ for the distance. The MRSD is always less than the root mean square distance (RMSD), except in special cases when they are equal. To see this, we can apply Hölder’s inequality $\sum_{k=1}^{N}\left(g_{k}\right)^{\alpha}\,\left(h_{k}\right)^{\beta}\leq\left(\sum_{k=1}^{N}g_{k}\right)^{\alpha}\left(\sum_{k=1}^{N}h_{k}\right)^{\beta}$ where $g_{k},h_{k}\geq 0$, $\alpha,\beta\geq 0$, and $\alpha+\beta=1$. With the specific identifications $g_{k}=({\bf r}_{\mbox{\tiny{B}}k}-{\bf r}_{\mbox{\tiny{A}}k})^{2}\equiv\Delta{\bf r}_{k}^{2}$, $h_{k}=1$, and $\alpha=\beta=1/2$, we have directly ${1\over N}\sum_{k}\sqrt{\Delta{\bf r}_{k}^{2}}\>\leq\sqrt{{1\over N}\sum_{k}\Delta{\bf r}_{k}^{2}}$ For example the RMSD for the circle configuration discussed above is $\sqrt{2}L/\pi\approx 0.4502L$, which is greater than the MRSD. The fact that the distance converges for large $N$ to MRSD rather than RMSD suggests that RMSD may not be the best metric for determining similarity between molecular structures, although it is ubiquitously used. This fact warrants future investigation- it has implications in research areas from structural alignment based pharmacophore identification [3, 4, 5] to protein structure and function prediction [6, 7]. It was shown in [1] that chains with persistence length characterized by some radius of curvature $R$ have extensive corrections to the MRSD-derived minimal distance, which do not vanish as $N\rightarrow\infty$, but remain so long as $R/L$ is nonzero. Likewise, chains that cannot cross themselves have nonlocal EL equations and extensive corrections to the minimal distance. Nevertheless, it is worthwhile to investigate some more complex polymers with MRSD as an approximate distance metric. We pursue this in the next section. ### 5.1 MRSD as a metric for protein folding Here we examine the use of MRSD as a metric or order parameter for protein folding. To this end we adopt an unfrustrated $C_{\alpha}$ model of segment $84-140$ of src tyrosine-protein kinease (src-SH3), by applying a Gō-like Hamiltonian [8, 9, 10] to an off–lattice coarse-grained representation of the src-SH3 native structure (pdb 1fmk). Amino acids are represented as single beads centered at their $C_{\alpha}$ positions. The Gō-like energy of a protein configuration $\alpha$ is given by the following Hamiltonian, which we will explain term by term: $\displaystyle\mathcal{H}(\alpha|N)$ $\displaystyle=$ $\displaystyle k_{r}\sum_{bonds}\left(r_{\alpha}-r_{N}\right)^{2}+k_{\theta}\sum_{triples}\left(\theta_{\alpha}-\theta_{N}\right)^{2}$ (43) $\displaystyle+$ $\displaystyle\sum_{n=1,3}k_{\phi}^{\left(n\right)}\sum_{quads}\left[1-\cos\left(n\times(\phi_{\alpha}-\phi_{N})\right)\right]$ $\displaystyle+$ $\displaystyle\epsilon_{\mbox{\tiny N}}\sum_{j\geq i+3}\left[6\left(\frac{\sigma_{ij}}{r_{ij}}\right)^{10}-5\left(\frac{\sigma_{ij}}{r_{ij}}\right)^{12}\right]+\epsilon_{\mbox{\tiny NN}}\sum_{j\geq i+3}\left(\frac{\sigma_{ij}}{r_{ij}}\right)^{12}\>.$ Adjacent beads are strung together into a polymer through harmonic bond interactions that preserve native bond distances between consecutive $C_{\alpha}$ residues. Here $r_{\alpha}$ and $r_{N}$ represent the distances between two subsequent residues in configurations $\alpha$ and the native state $N$. As with other parameters in the Hamiltonian, the distances $r_{N}$ are based on the pdb structure and may vary pair to pair. The angles $\theta_{N}$ represent the angles formed by three subsequent $C_{\alpha}$ residues in the pdb structure, and the angles $\phi_{N}$ represent the dihedral angles defined by four subsequent residues. The dihedral potential consists of a sum of two terms, one with period $2\pi$ and another with $2\pi/3$, which give cis and trans conformations for angles between successive planes of three amino acids, with a global dihedral potential minimum at $\phi_{N}\in[-\pi,\pi]$. The parameters $k_{r}$, $k_{\theta}$, and $k_{\phi}$, are taken to accurately describe the energetics of the protein backbone: we used the values $k_{r}=50\>\mbox{kcal/mol}$, $k_{\theta}=20\>\mbox{kcal/mol}$, $k_{\phi}^{(1)}=1\>\mbox{kcal/mol}$ and $k_{\phi}^{(3)}=0.5\>\mbox{kcal/mol}$ for molecular dynamics (MD) simulations using the AMBER software package. For MD simulations using LAMMPS, we had used slightly different values: $k_{r}=80\>\mbox{kcal/mol}$, $k_{\theta}=16\>\mbox{kcal/mol}$, $k_{\phi}^{(1)}=0.8\>\mbox{kcal/mol}$ and $k_{\phi}^{(3)}=0.4\>\mbox{kcal/mol}$. The last line in equation (43) deals with non-local interactions, both native and non-native. If two amino acids are separated by $3$ more along the chain ($|i-j|\geq 3$), and have one or more pairs of heavy atoms within a cut-off distance of $r_{c}=4.8$ Å in the pdb structure, the amino acids are said to have a native contact. Then the respective coarse-grained $C_{\alpha}$ residues are given a Lennard-Jones-like 10-12 potential of depth $\epsilon_{\mbox{\tiny N}}=-0.6\>\mbox{kcal/mol}$ ($-0.8\>\mbox{kcal/mol}$ for LAMMPS simulations) and a position of the potential minimum equal to the distance of the $C_{\alpha}$ atoms in the pdb structure. That is, $\sigma_{ij}$ is taken equal to native distance between $C_{\alpha}$ residues $i$ and $j$ if $i$–$j$ have a native contact. If two amino acids are not in contact, their respective $C_{\alpha}$ residues sterically repel each other ($\epsilon_{\mbox{\tiny NN}}=+0.6\>\mbox{kcal/mol}$). Thus $\epsilon_{\mbox{\tiny NN}}=0$ if $i$-$j$ is a native residue pair, while $\epsilon_{\mbox{\tiny N}}=0$ if $i$-$j$ is a non-native pair. For non-native residue pairs, $\sigma_{ij}=4\>\mbox{Angstroms}$. In an arbitrary configuration $\alpha$, two $C_{\alpha}$ residues $i$ and $j$ are considered to have formed a native contact if they have a distance $r_{ij}\leq 1.2\sigma_{ij}$. The results do not strongly depend on the specific value of this cutoff. The fraction of native contacts present in the particular configuration $\alpha$ is then defined as $Q$ (or $Q_{\alpha}$). The MRSD of configuration $\alpha$ is found by aligning this configuration to the native structure, by minimizing MRSD over $3$ translational and $3$ rotational degrees of freedom. Constant temperature molecular dynamics simulations were run for this system using both AMBER and LAMMPS simulation packages. The probability for the system to have given values of $Q$ and $MRSD$ within $(Q,Q+\Delta Q)$ and $(MRSD,MRSD+\Delta MRSD)$ is proportional to the exponential of the free energy $F(Q,MRSD)$. Thus the free energy can be directly obtained by sampling, binning, and taking the logarithm: $\displaystyle F(Q_{1},MRSD_{1})-F(Q_{2},MRSD_{2})=-k_{\mbox{\tiny B}}T\log\left({p(Q_{1},MRSD_{1})\over p(Q_{2},MRSD_{2})}\right)$ (44) with $F(1,0)=E_{\mbox{\tiny N}}$, the energy of the native structure. Figure 14 shows the free energy surfaces obtained using the above recipe, for the AMBER (fig 14a) and LAMMPS (fig 14b) molecular dynamics routines. The temperature is taken to be the transition or folding temperature $T_{\mbox{\tiny F}}$, where the unfolded and folded free energies are equal. Notice that $F(Q)$ is comparable for both as it should be, moreover $F(MRSD)$ is as well. However the free energy surface plotted as a function of both $Q$ and $MRSD$ shows a marked difference. In addition to a native minimum, the LAMMPS routine has an additional minimum at $Q\approx 0.95$ and $MRSD\approx 8.4$. The conformational states in this bin are closely related, with an average MRSD between them of $1.8$Å. We can take the most representative state in this bin as that which has a minimum MRSD from all the others in the bin (at $Q\approx.95$, $MRSD\approx 8.4$): $\leavevmode{\raisebox{-4.30554pt}{$\stackrel{{\scriptstyle\mbox{min}}}{{\mbox{\scriptsize{i}}}}$}}\left(\sum_{j\neq i}^{{}^{\prime}}MRSD_{ij}/\sum_{j\neq i}^{{}^{\prime}}\right)\approx 1.6$Å. Inspection reveals that this state is a mirror image of the pdb structure (see fig 14b): If we reflect this structure about one plane, and subsequently align this reflected structure to the pdb one, the MRSD is only $1.1$Å. The discrepancy in free energy surfaces corresponding to the presence of a low energy mirror-image structure arises because the COMPASS class 2 dihedral potentials in the LAMMPS algorithm do not ascribe a sign to the angle $\phi$, so the full range $[-\pi,\pi]$, is projected onto $[0,\pi]$. This gives the set of actual dihedral angles $\\{\phi_{i}+\pi\\}$ the same energy as the set $\\{\phi_{i}\\}$, so that the dihedral potentials have two minima rather than one, and thus a protein chain of the opposite chirality (a mirror image) is allowed and has the same energy as the pdb structure. We found that the CHARMM and harmonic dihedral styles do not have this problem, however they have less versatile function forms, so that we favored modifying the COMPASS dihedrals to define $\phi$ over its full range. a --- b Figure 14: Free energy surfaces for the folding of Gō-model src-SH3 using two molecular dynamics simulation packages, AMBER (a) and LAMMPS (b). The contour plots give $F(Q,MRSD)$. The projections $F(Q)$ and $F(MRSD)$ are also shown on each side. The COMPASS class 2 dihedral potential in LAMMPS allows for a mirror image of the folded structure (red color structure in inset) that is not immediately evident from the $F(Q)$ or $F(MRSD)$ surfaces. Future implementations of LAMMPS using COMPASS dihedrals for biomolecular simulations must then correct for dihedral angles defined on the interval $[-\pi,\pi]$. ## 6 Conclusions Analogously to the distance between two points, the distance between two finite length space curves is a variational problem, and may be calculated by minimizing a functional of $2$ independent variables $s$ and $t$, where $s$ is the arc-length along the chain, and $t$ is the ’elapsed time’ during the transformation. We derived the Euler-Lagrange (EL) equation giving the solution to this problem, which is a vector partial differential equation, with extremal solution ${\bf r}^{\ast}(s,t)$. We also derived the sufficient conditions for the extremal solution to be a minimum, through the Jacobi equation. Once the minimal transformation ${\bf r}^{\ast}(s,t)$ is known, the distance ${\cal D}^{\ast}\equiv{\cal D}[{\bf r}^{\ast}]$ follows. We provided a general recipe for the solution to the EL equation using the method of lines. The resulting $N+1$ EL equations for the discretized chain are ODEs that can be interpreted geometrically and solved for minimal solutions. Solutions consist generally of rotations and translations pieced together so the direction of velocity of any link end point does not suddenly change (the Weierstrass-Erdmann corner conditions). We explored the minimal transformations for the simplest polymers, consisting of $1$ or $2$ links, in depth. For transformations between $2$ links, convexity becomes an issue (the analog to the direction of the radius of curvature for a continuous string). For example, even if the initial and final states lie in the same plane, if the convexities of these states are of opposite sign the transformation must pass through intermediate states that are out of the plane. Similarly, given a semicircular piece of string lying on a table, to move it to a semicircle of opposite convexity using the minimal amount of motion, the string must be lifted off the table. The study of minimal transformations between small numbers of links has applications to the inverse kinematic problem in robotics and movement control. In the inverse kinematic problem, one is given the initial and final positions of the end-effector (the hand of the robot), and asked for the functional form of the joint variables for all intermediate states. Generally there is no unique solution until some optimization functional is introduced, such as minimizing the time rate of change of acceleration (the jerk), torque, or muscle tension (see the review [11] and references therein). The minimal distance transformation would be relevant if one sought the fastest transformation between initial and final states, without explicit regard to mechanical limitations. The indeterminate intermediate points can be handled variationally as a free boundary value problem. In the limit of a large number of links, some simplifications emerge. For chains without curvature or non-crossing constraints, the distance converges to $L$ times the mean root square distance ($MRSD$) of the initial and final conformations. So for example the distance between 2 strings of length $L$ forming the top and bottom halves of a circle respectively is $4L^{2}/\pi^{2}$, the distance between horizontal and vertical straight lines of length $L$ which touch at one end is $L^{2}/\sqrt{2}$, and the distance to fold a straight line upon itself (to form a hairpin) is $L^{2}/4$. The fact that for large $N$ the distance (over $L$) converges to MRSD rather than RMSD suggests that RMSD may not be the best metric for determining similarity between molecular structures, although it is ubiquitously used. Adopting MRSD may lead to improvements in structural alignment algorithms. The MRSD was investigated as an approximate metric for protein folding. Free energy surfaces for folding were constructed for two simulation packages, AMBER and LAMMPS. It was found that including MRSD as an order parameter uncovered discrepancies between the two molecular dynamics algorithms. Because dihedral angles in LAMMPS (at least in COMPASS class 2 style) are only defined on $[0,\pi]$, the potential admits a mirror image structure degenerate in energy with the native structure. This is easily remedied and should not be interpreted as a deficiency in the LAMMPS simulation package so long as one is aware of it. It should be mentioned that the mirror-image structure would also have been seen had RMSD been used as an additional order parameter. It will be important for future studies to address the effects of persistence length and non-crossing on the distance between biopolymer conformations [1]. Also important is the role of entropy of paths or transformations in describing the accessibility of a particular biomolecular structure. Along these lines it will be interesting to investigate whether the distance can be a predictor of folding kinetics, or proximity to the native structure. It is also an interesting question to ask whether the actual dynamics between polymer configuraitons resembles the minimal transformation, after a suitable averaging over trajectories. This question is linked with the role of the entropy of transformations described above. It is also related to the problem of finding the dominant pathway for a chemical reaction [12], which has recently been applied to the problem of protein folding [13]. We have focused here on the question of geometrical distance for complex systems, which can be separated from the calculation of quantities such as reaction paths that depend intrinsically on energetics, i.e. on the specific Hamiltonian of the system. Quantifying the relationship between geometrical distance and the dominant reaction path is an interesting future question worthy of investigation. The notion of distance and corresponding optimal transformation for a system with many degrees of freedom is fundamental to a diverse array of research subjects. Hence we saw potential applications for this metric in areas ranging from drug design to robotics. It is not clear at present how useful the calculation of the true Euclidean distance between high-dimensional objects will be for practical applications, but we are optimistic. ## 7 Acknowledgements We are thankful to Shirin Hadizadeh, Mike Prentiss, and Peter Wolynes for helpful discussions. S.S.P. gratefully acknowledges support from the Natural Sciences and Engineering Research Council and the A. P. Sloan Foundation. ## Appendix A Necessary conditions for straight line transformations It was shown in section 3.1 that to have straight line transformations between links, it is sufficient to have facing obtuse angles on opposite sides of the the quadrilateral defined by the transformation as shown in figure 3A. We now show that it is a necessary condition as well, i.e. we show that a slide in the correct direction is not possible in the absence of obtuse angles. Figure 15: Without loss of generality assume that the link is initially along the z axis. The paths travelled by the link ends are shown in the figure. Note that the end point trajectories of $A$ and $B$ are in 3D space so the paths travelled by $A$ and $B$ need not cross or lie in the same plane. Let the unit vector along $A$’s path be $\hat{{\bf v}}_{\mbox{\tiny A}}$ and the unit vector along $B$’s path be $\hat{{\bf v}}_{\mbox{\tiny B}}$. Because the angles that the path of A and the path of B make with the link are acute, the z-component of $\hat{{\bf v}}_{\mbox{\tiny B}}$ ($\equiv z_{\mbox{\tiny{B}}}$) is negative and the z-component of $\hat{{\bf v}}_{\mbox{\tiny A}}$ ($z_{\mbox{\tiny{A}}}$) is positive. One can write $\hat{{\bf v}}_{\mbox{\tiny A}}$ and $\hat{{\bf v}}_{\mbox{\tiny B}}$ as $\displaystyle\hat{{\bf v}}_{\mbox{\tiny A}}$ $\displaystyle=$ $\displaystyle\boldsymbol{\rho}_{\mbox{\tiny A}}+z_{\mbox{\tiny{A}}}\hat{{\bf z}}$ $\displaystyle\hat{{\bf v}}_{\mbox{\tiny B}}$ $\displaystyle=$ $\displaystyle\boldsymbol{\rho}_{\mbox{\tiny B}}+z_{\mbox{\tiny{B}}}\hat{{\bf z}}$ where $\boldsymbol{\rho}_{\mbox{\tiny A}}$ and $\boldsymbol{\rho}_{\mbox{\tiny B}}$ are vectors in xy plane and $z_{\mbox{\tiny{A}}}>0$ and $z_{\mbox{\tiny{B}}}<0$. Let ${\bf r}_{\mbox{\tiny{A}}}(t)$ and ${\bf r}_{\mbox{\tiny{B}}}(t)$ denote the positions of the A and B ends at time $t$: $\displaystyle{\bf r}_{\mbox{\tiny{A}}}$ $\displaystyle=$ $\displaystyle t\hat{{\bf v}}_{\mbox{\tiny A}}$ $\displaystyle{\bf r}_{\mbox{\tiny{B}}}$ $\displaystyle=$ $\displaystyle g(t)\hat{{\bf v}}_{\mbox{\tiny B}}+\hat{{\bf z}}$ The rigid link constraint dictates that $({\bf r}_{\mbox{\tiny{A}}}-{\bf r}_{\mbox{\tiny{B}}})\cdot({\bf r}_{\mbox{\tiny{A}}}-{\bf r}_{\mbox{\tiny{B}}})=1$ which translates to: $g^{2}+2g\left(z_{\mbox{\tiny{B}}}-t\,(c+z_{\mbox{\tiny{A}}}\,z_{\mbox{\tiny{B}}})\right)-2tz_{\mbox{\tiny{A}}}+t^{2}+1=1$ with $c=\boldsymbol{\rho}_{\mbox{\tiny A}}\cdot\boldsymbol{\rho}_{\mbox{\tiny B}}$. Solving for $g$ as a function of $t$, keeping in mind that $g(0)=0$: $g(t)=-\left(z_{\mbox{\tiny{B}}}-t\,\left(c+z_{\mbox{\tiny{A}}}\,z_{\mbox{\tiny{B}}}\right)\right)+\sqrt{\left(z_{\mbox{\tiny{B}}}-t\,\left(c+z_{\mbox{\tiny{A}}}\,z_{\mbox{\tiny{B}}}\right)\right)^{2}-t^{2}+2tz_{\mbox{\tiny{A}}}}\>.$ Now if $g^{\prime}(t)>0$ it means that the B-end of the link is travelling in the assumed direction, and if $g^{\prime}(t)<0$ it means that B-end is travelling in the opposite direction (which means that the angle is not acute anymore). Writing $g^{\prime}(0)$ we get: $g^{\prime}(0)={{2\,z_{\mbox{\tiny{B}}}\,c+2\,z_{\mbox{\tiny{A}}}\,z_{\mbox{\tiny{B}}}^{2}-2\,z_{\mbox{\tiny{A}}}}\over{2\,\left|z_{\mbox{\tiny{B}}}\right|}}+c+z_{\mbox{\tiny{A}}}z_{\mbox{\tiny{B}}}={-z_{\mbox{\tiny{A}}}\over|z_{\mbox{\tiny{B}}}|}<0\>.$ Thus point $B$ can only travel in the opposite direction from what was assumed, which in turn means an all-acute slide is not possible. We conclude that the condition of “facing obtuse angles” is necessary and sufficient for transformations consisting only of pure translations. ## Appendix B Critical angles The concept of critical angle was first introduced in 3.2. In order for a straight-line slide of both ends to be possible, at some stage during the transformation the link needs to rotate about one of the ends, with the other end being stationary. In principle the rotation can be about either of the two ends and it can happen at the beginning or the end of the transformation. The conditions on the critical angle or orientation can be readily derived from the broken extremal conditions. It was seen from 18a and 19, the non-trivial corner conditions read: $\hat{\bf v}_{i}|_{{}_{+}}=\hat{\bf v}_{i}|_{{}_{-}}\>.$ (B.45) We know that the path travelled by the moving bead during the rotation is circular and the path that is travelled during the slide part is a straight line. Broken extremal condition forces these two paths to be patched smoothly, which means that the straight-line path should be tangent to the circle. In the 3D case, for the broken extremal condition to be satisfied, the straight line slide path and the circular rotation path should lie in the same plane. For example in figure 5 where $B$ is rotating about $A$ initially to $B_{1}$ and then slides to $B^{\prime}$, the rotation has to be in the plane formed by the three points $ABB^{\prime}$. Matching the directions of velocity as in (B.45) does not itself mean that a link can subsequently slide in a straight line, however at the tangent point, the tangent line to the circle is perpendicular to the radius, hence one satisfies this second condition as well. Below we derive an analytical expression for the critical angle for a particular case of single link problem, as an example and illustration of the discussed concepts. Furthermore the particular example will be used later in C to introduce minimal transformations in $2$ dimensions. Consider the single link action with the particular parametrization $s=s(\theta)$, as discussed in section 3.2: $\int(\sqrt{\dot{s}^{2}+1+2\dot{s}cos\theta}+\sqrt{\dot{s}^{2}})\ d\theta.$ (B.46) where $s\equiv\overrightarrow{A(\theta)A}$ is the (signed) distance of $A$-end from its initial position, and $\theta$ is the angle between the link and the horizontal line (see figure 16). The Euler Lagrange equation of motion reads: ${d\over d\theta}({\dot{s}\over\sqrt{\dot{s}^{2}}}+{\dot{s}+cos\theta\over\sqrt{\dot{s}^{2}+1+2\dot{s}cos\theta}})=0$ (B.47) We consider a transformation which is not (necessarily) a minimum: $s=a\cos\theta-\sin\theta+b$ (B.48) with $a$ and $b$ parameters to be determined. Such a transformation in fact forces the two ends to travel on a straight line (right from the beginning) , but the $A$ side may in fact retreat and then move forward. We call such a transformation a “hyperextended transformation”. A sample transformation of this kind is shown in figure 16. The parameters $a$ and $b$ in (B.48) can be tuned to meet the boundary conditions (see below). Figure 16: Transformation in which both ends stay on a linear track In fact it is seen that point $A$ on the link retreats backwards until it reaches some critical angle, which is when link $\overline{AB}$ makes an angle ${\pi\over 2}$ with the straight line $\overline{BB^{\prime}}$ that point $B$ travels on. Subsequently $A$ then moves forward towards $A^{\prime}$. Assume that $\theta$ runs from $\theta_{1}$ to $\theta_{2}$, where $0<\theta_{2}<\pi/2$. For simplicity assume that both these angles are between $0$ and ${\pi\over 2}$. The boundary conditions dictate that: $\displaystyle s(\theta_{1})$ $\displaystyle=$ $\displaystyle 0$ (B.49) $\displaystyle s(\theta_{2})$ $\displaystyle=$ $\displaystyle l$ (B.50) where $l$ is the distance between $A$ and $A^{\prime}$. $a$ and $b$ can be explicitly solved to give: $\displaystyle a$ $\displaystyle=$ $\displaystyle{{-\sin\theta_{2}+\sin\theta_{1}-l}\over{\cos\theta_{1}-\cos\theta_{2}}}$ (B.51) $\displaystyle b$ $\displaystyle=$ $\displaystyle-{{\cos\theta_{1}\,\left(-\sin\theta_{2}-l\right)+\sin\theta_{1}\,\cos\theta_{2}}\over{\cos\theta_{1}-\cos\theta_{2}}}$ (B.52) For our purposes we only need to note that the critical angle occurs when $\dot{s}\equiv{ds\over d\theta}$ becomes zero, that is when $A$ stops going backward and starts moving forward: $\dot{s}=-a\sin\theta-\cos\theta=0$ (B.53) where $a$ is given in B.51. We can now ask what should $\theta_{1}$ be so that there is no need for the link to go backward, i.e. it moves forward from the beginning and the transformation is monotonic. Equations (B.53) and (B.51) give: $\cos\theta+{{-\sin\theta_{2}+\sin\theta-l}\over{\cos\theta-\cos\theta_{2}}}\sin\theta=0$ (B.54) Figure 17: Geometric proof for critical angle condition For pedagogical reasons we prove condition (B.54) using analytic geometry as well. Looking at figure 17 we have the following: $\displaystyle g^{2}+l_{1}^{2}$ $\displaystyle=$ $\displaystyle 1$ (B.55) $\displaystyle g^{2}+l_{2}^{2}$ $\displaystyle=$ $\displaystyle a^{2}$ (B.56) $\displaystyle{g\over a}$ $\displaystyle=$ $\displaystyle{1\over l+l_{1}+l_{2}}$ (B.57) We can solve $g=\sqrt{1-l_{1}^{2}}$ and $a=\sqrt{1-l_{1}^{2}+l_{2}^{2}}$ from the first two equations and substitute in the third equation to give: $l={\sqrt{1-l_{1}^{2}+l_{2}^{2}}\over\sqrt{1-l_{1}^{2}}}-l_{1}-l_{2}$ (B.58) On the other hand based on our results for $g$ and $a$ we have: $\displaystyle\sin\theta_{1}$ $\displaystyle=$ $\displaystyle{\sqrt{1-l_{1}^{2}}\over\sqrt{1-l_{1}^{2}+l_{2}^{2}}}$ (B.59) $\displaystyle\cos\theta_{1}$ $\displaystyle=$ $\displaystyle{l_{2}\over\sqrt{1-l_{1}^{2}+l_{2}^{2}}}$ (B.60) $\displaystyle\sin\theta_{2}$ $\displaystyle=$ $\displaystyle l_{1}$ (B.61) $\displaystyle\cos\theta_{2}$ $\displaystyle=$ $\displaystyle\sqrt{1-l_{1}^{2}}$ (B.62) Substitution of eqns (B.59-B.62) in equation (B.54) gives equation (B.58) after some simplification. For the particular case that we have discussed, the proposed transformation is in fact a minimal solution if $\theta_{1}$ is greater than the critical angle, because in that case a simple slide would be possible. If $\theta_{1}$ is less than the critical angle a locally minimum solution as we know is pure rotation to the critical angle and then straight line slide. Pure rotation has a nice geometric interpretation in our parametrization. it corresponds to the null solution $s=0$. Since at the critical angle $\dot{s}=0$ we see that $s=0$ will be smoothly patched with $s=a\cos\theta-sin\theta+b$, as mandated by the corner conditions in equation (18a). Figure 18: A minimal transformation in $s(\theta)$ parametrization. The horizontal segment corresponds to pure rotation and the curved section corresponds to slide on straight paths. Here the corner conditions demand that the derivative $\dot{s}$ be continuous at the critical angle. ## Appendix C Minimal transformations in 2 dimensions It was seen in section 4.1 that for the case of two links when one is confined to moving in a plane, satisfying the constant link length constraints and corner conditions do not seem to lead to solutions which are extremal. However given the additional constraint that the links must lie in a plane, there must be one or a set of minimal transformations. We need to look at other forms of transformations, namely compound straight line transformations. We will elaborate on the idea starting with single links. The hyper extended solution that was discussed previously in B can be considered as a very special example of compound straight line transformation. These are transformations that are made strictly from straight line paths with no pure rotation. A more general transformation is shown figure C beside the old transformation. Figure 19: The previous hyper extended solution is shown along with a more general compound straight-line transformation, where $\overrightarrow{AA^{\prime\prime}}$ travels in some general direction. Length of each line segment is written beside it. For the hyper extended solution the value of $AA^{\prime\prime}$ is multiplied by two because the path is travelled twice. Note that the corners do not technically violate the corner conditions because the speed of “A” bead is zero at the corner point in any parametrization that can simultaneously describe $A$ motion and $B$ motion: Since at the corner point, the link makes an angle of 90 degrees with the path that B travels, the speed of B at the critical angle in infinitely larger than the speed of A. In fact one sees that we have an instantaneous pure rotation about $A$-bead, when it is at the corner point. $\hat{v}_{a}$ is not clearly defined at the corners, and everywhere else (when the speed of the bead(s) is not zero), the two beads are travelling on a straight line. The two solutions depicted in the figure come from two different parametrization of the most general form of the action and result in different distances. But each of them is a local minimum once the direction of $\overrightarrow{AA^{\prime\prime}}$ is picked, and these local minima have different values for the distance. We can then ask about the best position to put the corner point, to minimize the distance travelled in the compound straight line transformation, with respect to other compound straight line transformations. We assume the corner occurs on one side and we take it to be the “A” side. Note that at the corner, the link makes a $90^{\circ}$ angle with the B-bead path $\overline{BB^{\prime}}$, meaning that the distance from the corner point to B path is always the length of the link, i.e. unity here. Also note that the total distance that the “A”-bead travels is the distance from the initial point $A$ to the corner point $A^{\prime\prime}$, plus the distance from $A^{\prime\prime}$ to the final position $A^{\prime}$. The locus of points with equal sum of distances from two points $A$ and $A^{\prime}$ defines an ellipse with foci at $A$ and $A^{\prime}$. Moreover the length of the major axis of the ellipse equals the sum of the distances from the foci. Thus the smaller the major axis of the ellipse with foci $A$ and $A^{\prime}$, the smaller the total distance travelled by the “A”-bead. Moreover $A^{\prime\prime}$ should sit on a line parallel to B-path at a distance of $1$ from the B-path line $\overline{BB^{\prime}}$. So in seeking the shortest distance travelled the $A$ end of the link, we seek the point $A^{\prime\prime}$ such that it lies on an ellipse with foci $A$ and $A^{\prime}$, the ellipse shares at least one point with a line parallel to $\overline{BB^{\prime}}$ and distance $1$ away from it, and lastly that the ellipse has the smallest possible major axis (see figure 20). So the ellipse giving the minimal distance is tangent to the parallel line, and $A^{\prime\prime}$ is the tangent point. This is illustrated in figure 20. Figure 20: Optimal Compound Straight Line Transformation This solution can be straightforwardly extended to 2 links, as depicted in figure 21. Consider then the example in figure 13a, where the links are no longer allowed to move out of the plane (see figure 22). Here ${\bf r}_{A}={\bf r}_{A^{\prime}}$ and ${\bf r}_{C}={\bf r}_{C^{\prime}}$ and the above ellipses turns into a circles centered at $A$ and $C$. The circles have radii $1-1/\sqrt{2}$, so that the perpendicular distance from line $\overline{BB^{\prime}}$ to the farthest point on the circle is $1$ and a fully extended intermediate state is allowed. Figure 21: An optimal compound Straight line solution for 2 link. For this particular class of solutions, the problem is divided into to disjoint problems (one for each link) and solved separately. Figure 22: Minimal transformation restricted to 2 dimensions, for 2 links of opposite convexity which form opposite sides of a square. References ## References * [1] Plotkin, S. S. (2007). Generalization of distance to higher dimensional objects. Proc. Natl Acad. Sci. USA 104, 14899–14904. * [2] Gelfand, I. M & Fomin, S. V. (2000) Calculus of Variations. (Dover). * [3] J. Greene, S. Kahn, H. S. P. S & Teig, S. (1994). Chemical Function Queries for 3D Database Search. J. Chem. Inf. Comput. Sci. 34, 1297–1308. * [4] Lemmen, C & Lengauer, T. (2000). Computational methods for the structural alignment of molecules. J. Comput. Aided Mol. Des. 14, 215–231. * [5] Y. Patel, V. J. Gillet, G. B & Leach, A. R. (2002). A comparison of the pharmacophore identification programs: Catalyst, DISCO and GASP. J. Comput. Aided Mol. Des. 16, 653–681. * [6] Gerstein, M & Levitt, M. (1998). Comprehensive assessment of automatic structural alignment against a manual standard. Protein Science 7, 445–456. * [7] Baker, D & Sali, A. (2001). Protein Structure Prediction and Structural Genomics. Science 294, 93–96. * [8] Ueda, Y, Taketomi, H, & Gō, N. (1975). Studies on protein folding, unfolding and fluctuations by computer simulation. I. The effects of specific amino acid sequence represented by specific inter-unit interactions. Int. J. Peptide Res. 7, 445–459. * [9] Shea, J & Brooks III, C. (2001). From folding theories to folding proteins: A review and assessment of simulation studies of protein folding and unfolding. Ann. Rev. Phys. Chem. 52, 499–535. * [10] Clementi, C & Plotkin, S. S. (2004). The effects of nonnative interactions on protein folding rates: Theory and simulation. Protein Science 13, 1750–1766. * [11] Kawato, M. (1996) in Advances in Motor Learning and Control, ed. Zelaznik, H. N. (Human Kinetics), pp. 225–259. * [12] Onsager, L & Machlup, S. (1953). Fluctuations and irreversible processes. Phys Rev 91, 1505–1512. * [13] Sega, M, Faccioli, P, Pederiva, F, Garberoglio, G, & Orland, H. (2007). Quantitative Protein Dynamics from Dominant Folding Pathways. Phys. Rev. Lett. 99, 118102. .

arxiv-papers

2024-09-04T02:48:54.061386

0803.0074

# A codimension two CR singular submanifold that is formally equivalent to a symmetric quadric Xiaojun Huang111 Supported in part by NSF-0500626 and Wanke Yin ###### Abstract Let $M\subset\mathbb{C}^{n+1}$ ($n\geq 2$) be a real analytic submanifold defined by an equation of the form: $w=|z|^{2}+O(|z|^{3})$, where we use $(z,w)\in\mathbb{C}^{n}\times\mathbb{C}$ for the coordinates of $\mathbb{C}^{n+1}$. We first derive a pseudo-normal form for $M$ near $0$. We then use it to prove that $(M,0)$ is holomorphically equivalent to the quadric $(M_{\infty}:w=|z|^{2},\ 0)$ if and only if it can be formally transformed to $(M_{\infty},0)$. We also use it to give a necessary and sufficient condition when $(M,0)$ can be formally flattened. The result is due to Moser for the case of $n=1$. ## 1 Introduction Let $M\subset\mathbb{C}^{n+1}$ ($n\geq 1$) be a submanifold. For a point $p\in M$, we define $CR(p)$ to be the CR dimension of $M$ at $p$, namely, the complex dimension of the space $T^{(0,1)}_{p}M$. A point $p\in M$ is called a CR point if $CR(q)=CR(p)$ for $q(\in M)\approx p$. Otherwise, $p$ is called a CR singular point of $M$. The local equivalence problem in Several Complex Variables is to find a complete set of holomorphic invariants of $M$ near a fixed point $p\in M$. The investigation normally has quite different nature in terms of whether $p$ being a CR point or a CR singular point. The CR case was first considered by Poincaré and Cartan. A complete set of invariants in the strongly pseudoconvex hypersurface case was give by Chern-Moser in [CM] (see the survey paper of Baouendi-Ebenfelt-Rothschild [BER1] and the lecture notes of the first author [Hu1] for many references along these lines). The study for the CR singular points first appeared in the paper of Bishop [Bis]. Further investigations on the precise holomorphic structure of $M$ near a non- degenerate CR singular point, in the critical dimensional case of $dim_{{\mathbb{R}}}M=n+1$, can be found in the work of Moser-Webster [MW] and in the work of Moser [Mos], Gong [Gon1-2], Huang-Yin [HY], etc. (The reader can find many references in [Hu1] on this matter.) Recently, there appeared several papers, in which CR singular points in the non-critical dimensional case were considered (see [Sto], [DTZ], [Cof1-2], to name a few). In [Sto], among other things, Stolovitch introduced a set of generalized Bishop invariants for a non-degnerate general CR singular point, and established some of the results of Moser-Webster [MW] to the case of $dim_{{\mathbb{R}}}M>dim_{{\mathbb{C}}}{{\mathbb{C}}}^{n+1}$. In [DTZ], Dolbeault-Tomassini-Zaitsev introduced the concept of the elliptic flat CR singular points and studied global filling property by complex analytic varieties for a class of compact submanifold of real codimension two in ${{\mathbb{C}}}^{n+1}$ with exactly two elliptic flat CR singular points. In this paper, we study the local holomorphic structure of a manifold $M$ near a CR singular point $p$, for which we can find a local holomorphic change of coordinates such that in the new coordinates system, $p=0$ and $M$ near $p$ is defined by an equation of the form: $w=|z|^{2}+O(|z|^{3})$. Here we use $(z,w)\in{{\mathbb{C}}}^{n}\times{\mathbb{C}}$ for the coordinates of ${{\mathbb{C}}}^{n+1}$. Such a non-degenerate CR singular point has an intriguing nature that its quadric model has the largest possible symmetry. We will first derive a pseudo-normal form for $M$ near $p$ (see Theorem 2.3). As expected, the holomorphic structure of $M$ near $p$ is influenced not only by the nature of the CR singularity, but also by the fact that $(M,p)$ partially inherits the property of strongly pseudoconvex CR structures for $n>1$. Unfortunately, as in the case of $n=1$ first considered by Moser [Mos], our pseudo-normal form is still subject to the simplification of the complicated infinite dimensional formal automorphism group of the quadric $aut_{0}(M_{\infty})$, where $M_{\infty}$ is defined by $w=|z|^{2}$. Thus, our pseudo-normal form can not be used to solve the local equivalence problem. However, with the rapid iteration procedure, we will show in $\S 4$ that if all higher order terms in our pseudonormal form vanish, then $M$ is biholomorphically equivalent to the model $M_{\infty}$. Namely, we have the following: Theorem 1: Let $M\subset\mathbb{C}^{n+1}$ ($n\geq 1$) be a real analytic submanifold defined by an equation of the form: $w=|z|^{2}+O(|z|^{3})$. Then $(M,0)$ is holomorphically equivalent to the quadric $(M_{\infty},0)$ if and only if it can be formally transformed to $(M_{\infty},0)$. One of the differences of our consideration here from the case of $n=1$ is that a generic $(M,0)$ can not be formally mapped into the Levi-flat hypersurface $Im(w)=0$. As another application of the pseudo-normal form to be obtained in $\S 2$, we will give a necessary and sufficient condition when $(M,0)$ can be formally flattened (see Theorem 3.5). Theorem 1, in the case of $n=1$, is due to Moser [Mos]. Indeed, our proof of Theorem 1 uses the approach of Moser in [Mos] and Gong in [Gon2], which is based on the rapidly convergent power series method. Convergence results along the lines of Theorem 1 near other type of CR singular points can be found in the earlier papers of Gong [Gon1] and Stolovitch [Sto]. The papers of Coffman [Cof1-2] also contain the rapid convergence arguments in the setting of other CR singular cases. ## 2 A formal pseudo-normal form We use $(z,w)=(z_{1},\cdots,z_{n},w)$ for the coordinates in $\mathbb{C}^{n+1}$ with $n\geq 2$ in all that follows. We first recall some notation and definitions already discussed in the previous papers of Stolovitch [Sto] and Dolbeault-Tomassini-Zaitsev [DTZ]. Let $(M,0)$ be a formal submanifold of codimenion two in ${\mathbb{C}}^{n+1}$ with $0\in M$ as a CR singular point and $T^{(1,0)}_{0}M=\\{w=0\\}$. Then, $M$ can be defined by a formal equation of the form: $w=q(z,\overline{z})+o(|z|^{2}),$ (2.1) where $q(z,\overline{z})$ is a quadratic polynomial in $(z,\overline{z})$. We say that $0\in M$ is a not-completely-degenerate CR singular point if there is no change of coordinates in which we can make $q\equiv 0.$ Following Dolbeault-Tomassini-Zaitsev, we further say that $0$ is a not-completely- degenerate flat CR singular point if we can make $q$ real-valued after a linear change of variables. Assume that $0$ is a not-completely-degenerate flat CR singular point with $q(z,z)=A(z,\overline{z})+B(z,\overline{z})\in{\mathbb{R}}$ for each $z$. Here $A(z,\overline{z})=\sum_{\alpha,\beta=1}^{n}a_{\alpha\overline{\beta}}z_{\alpha}\overline{z_{\beta}},\ B(z,\overline{z})=2Re(\sum_{\alpha,\beta=1}^{n}b_{\alpha\beta}z_{a}z_{\beta}).$ Then the assumption that $A(z,\overline{z})$ is definite is independent of the choice of the coordinates system. Suppose that $A$ is definite. Then making use of the classical Takagi theorem, one can find a linear change of coordinates in $(z,w)$ such that in the new coordinates, in the defining equation for $(M,0)$ of the form in (2.1), one has that $q(z,\overline{z})=\sum_{\alpha=1}^{n}\\{|z_{\alpha}|^{2}+\lambda_{\alpha}(z_{\alpha}^{2}+\overline{z_{\alpha}}^{2})\\},$ where $0\leq\lambda_{\alpha}<\infty$ with $0\leq\lambda_{1}\leq\cdots\leq\lambda_{n}<\infty$. In terms of Stolovitch, we call $\\{\lambda_{1},\cdots,\lambda_{n}\\}$ the set of generalized Bishop invariants. When $0\leq\lambda_{\alpha}<1/2$ for all $\alpha$, we say that $0$ is an elliptic flat CR singular point of $M$. Notice that $0\in M$ is an elliptic flat CR singular point if and only if in a certain defining equation of $M$ of the form as in (2.1), we can make $q(z,\overline{z})>0$ for $z\not=0$. (Hence the definition coincides with the notion of elliptic flat Complex points in [DTZ].) When $\lambda_{\alpha}>1/2$ for all $\alpha$, we say $0\in M$ is a hyperbolic flat CR singular point. Notice that, in the other case, we can always find a two dimensional linear subspace of ${\mathbb{C}}^{n+1}$ whose intersection with $M$ has a parabolic complex tangent at $0$. For a more general related notion on ellipticity and hyperbolicity, we refer the reader to the paper of Stolovitch [Sto]. In terms of the terminology above, the manifold in Theorem 1 has vanishing generalized Bishop invariants at the CR singular point. In [Gon1] [Sto], one finds the study on the related convergence problem in the other situations, where, among other non-degeneracy conditions, all the generalized Bishop invariants are assumed to be non-zero. The method studying CR singular points with vanishing Bishop invariants is different from that used in the non- vanishing Bishop invariants case (see [MW] [Mos] [Gon2] [Sto] [HY]). We now return to the manifolds with only vanishing generalized Bishop invariants. Let $E(z,\bar{z})$ $\left(\mbox{respectively},\ f(z,w)\right)$ be a formal power series in $(z,\bar{z})\left(\mbox{respectively, in}\ (z,w)\right)$ without constant term. We say $Ord\left(E(z,\bar{z})\right)\geq k$ if $E(tz,t\bar{z})=O(t^{k})$. Similarly, we say $Ord_{wt}\left(f(z,w)\right)\geq k$ if $f(tz,t^{2}w)=O(t^{k})$. Set the weight of $z,\bar{z}$ to be 1 and that of $w$ to be 2. For a polynomial $h(z,w)$, we define its weighted degree, denoted by $deg_{wt}h$, to be the degree counted in terms the weighted system just given. Write $E^{(t)}(z,\bar{z})$ and $f^{(t)}(z,w)$ for the sum of monomials with weighted degree $t$ in the expansion of $E$ and $f$ at $0$, respectively. Write $u_{k}=\sum_{i=1}^{k}|z_{i}|^{2}$ for $1\leq k\leq n$ and $v_{k}=\sum_{i=1}^{k-1}|z_{i}|^{2}-|z_{k}|^{2}$ for $2\leq k\leq n$. We also write $u=u_{n}=|z|^{2}$. In what follows, we make a convention that the sum $\sum_{p=j}^{l}a_{p}$ is defined to be $0$ if $j>l$. We start with the following elementary algebraic lemma: Lemma 2.1: $Span_{{\mathbb{C}}}\\{|z_{1}|^{2},\cdots,|z_{n}|^{2}\\}=Span\\{u,v_{2},\cdots,v_{n}\\}$. Moreover, for each index $i$ with $1\leq i\leq n$, $|z_{i}|^{2}$ can be uniquely expressed as the following linear combination of $u,\ v_{2},\cdots,\ v_{n}$: $\left\\{\begin{array}[]{l}|z_{1}|^{2}=2^{1-n}\left(u+\sum\limits_{h=2}^{n}2^{n-h}v_{h}\right),\\\ |z_{i}|^{2}=2^{-(n+1-i)}\left(u+\sum\limits_{h=i+1}^{n}2^{n-h}v_{h}-2^{n-i}v_{i}\right)\ \mbox{for}\ 2\leq i\leq n.\end{array}\right.$ (2.2) Proof of Lemma 2.1: By a direct computation, we have $\begin{array}[]{l}2^{1-n}\left(u+\sum\limits_{h=2}^{n}2^{n-h}v_{h}\right)=2^{1-n}\left(\sum\limits_{i=1}^{n}|z_{i}|^{2}+\sum\limits_{h=2}^{n}2^{n-h}(\sum\limits_{i=1}^{h-1}|z_{i}|^{2}-|z_{h}|^{2})\right)\\\ \hskip 28.45274pt=2^{1-n}\left((1+\sum\limits_{h=2}^{n}2^{n-h})|z_{1}|^{2}+\sum\limits_{j=2}^{n-1}(1+\sum\limits_{h=j+1}^{n}2^{n-h}-2^{n-j})|z_{j}|^{2}\right)\\\ \hskip 28.45274pt=2^{1-n}(2^{n-1}|z_{1}|^{2})=|z_{1}|^{2};\\\ 2^{-(n+1-i)}\left(u+\sum\limits_{h=i+1}^{n}2^{n-h}v_{h}-2^{n-i}v_{i}\right)\\\ \hskip 28.45274pt=2^{-(n+1-i)}\left(\sum\limits_{i=1}^{n}|z_{i}|^{2}+\sum\limits_{h=i+1}^{n}2^{n-h}(\sum\limits_{j=1}^{h-1}|z_{j}|^{2}-|z_{h}|^{2})-2^{n-i}(\sum\limits_{j=1}^{i-1}|z_{j}|^{2}-|z_{i}|^{2})\right)\\\ \hskip 28.45274pt=2^{-(n+1-i)}\left(\sum\limits_{j=1}^{i-1}(1+\sum\limits_{h=i+1}^{n}2^{n-h}-2^{n-i})|z_{j}|^{2}+(1+\sum\limits_{h=i+1}^{n}2^{n-h}+2^{n-i})|z_{i}|^{2}\right.\\\ \hskip 36.98866pt\left.+\sum\limits_{j=i+1}^{n}(1+\sum\limits_{h=j+1}^{n}2^{n-h}-2^{n-j})|z_{j}|^{2}\right)=|z_{i}|^{2},\ \hbox{for}\ i\geq 2.\end{array}$ Hence, we see that $\hbox{span}_{{\mathbb{C}}}\\{|z_{1}|^{2},\cdots,|z_{n}|^{2}\\}=\hbox{span}_{{\mathbb{C}}}\\{u,v_{2},\cdots,v_{n}\\}.$ The uniqueness assertion in the lemma now is obvious. For a formal (or holomorphic) transformation $f(z,w)$ of $({{\mathbb{C}}}^{n},0)$ to itself, we write $\left\\{\begin{array}[]{l}f(z,w)=\left(f_{1}(z,w),\cdots,f_{n}(z,w)\right),\\\ f_{k}(z,w)=\sum_{(i_{1},\cdots,i_{n})}f_{k,(I)}(w)z^{I},\ I=(i_{1},\cdots,i_{n})\ \hbox{and}\ z^{I}=z_{1}^{i_{1}}\cdots z_{n}^{i_{n}}.\end{array}\right.$ (2.3) Let $E(z,\bar{z})$ be a formal power series with $E(0)=0$. We next prove the following: Lemma 2.2: $E(z,\bar{z})$ has the following expansion: $E(z,\bar{z})=\sum_{\\{i_{k}\cdot j_{k}=0,\ k=1\cdots,n\\}}E_{(I,J)}(u,v_{2},\cdots,v_{n})z^{I}{\overline{z}}^{J}=\sum_{\\{i_{k}\cdot j_{k}=0,\ k=1,\cdots,n\\}}E_{(I,J)}^{(K)}z^{I}{\overline{z}}^{J}u^{k_{1}}{v_{2}}^{k_{2}}\cdots{v_{n}}^{k_{n}}.$ (2.4) Here and in what follows, we write $I=(i_{1},\cdots,i_{n})$, $J=(j_{1},\cdots,j_{n})$, $K=(k_{1},\cdots,k_{n})$, $z^{I}=z_{1}^{i_{1}}\cdots z_{n}^{i_{n}}$ and ${\overline{z}}^{J}=\overline{z}_{1}^{j_{1}}\cdots\overline{z}_{n}^{j_{n}}$. Moreover, the coefficients $E_{(I,J)}^{(K)}$ are uniquely determined by $E$. Proof of Lemma 2.2: Since $\\{|z_{i}|^{2}\\}_{i=1}^{n}$ and $\\{u,v_{2},\cdots,v_{n}\\}$ are the unique linear combinations of each other by Lemma 2.1, one sees the existence of the expansion in (2.4). Also, to complete the proof of Lemma 2.3, it suffices for us to prove the following statement: $\sum\limits_{(I,J,K)\in A(N,N^{*})}E_{(I,J)}^{(K)}z^{I}{\overline{z}}^{J}|z_{1}|^{2k_{1}}\cdots|z_{n}|^{2k_{n}}=0\ \mbox{if and only if }\ E_{(I,J)}^{(K)}\equiv 0.$ Here, we define $A(N,N^{*})=\\{(I,J,K)\in\mathbb{Z}^{n}\times\mathbb{Z}^{n}\times\mathbb{Z}^{n},\ i_{l}\cdot j_{l}=0,\ i_{l},j_{l},k_{l}\geq 0\ \mbox{for}\ 1\leq l\leq n,\ \sum_{l=1}^{n}(i_{l}+k_{l})=N,\ \sum_{l=1}^{n}(j_{l}+k_{l})=N^{*}\\}$. Let $P=(p_{1},\cdots,p_{n})$ and $Q=(q_{1},\cdots,q_{n})$ with $p_{1},\cdots,p_{n},q_{1},\cdots,q_{n}$ non-negative integers be such that $|P|=N,|Q|=N^{*}$. We define $A(N,N^{*};P,Q)=\\{(I,J,K)\in A(N,N^{*}):\ \ i_{l}\cdot j_{l}=0,\ i_{l},j_{l},k_{l}\geq 0,i_{l}+k_{l}=p_{l},\ j_{l}+k_{l}=q_{l},\ \mbox{for}\ 1\leq l\leq n\\}.$ Now, suppose that $\sum\limits_{(I,J,K)\in A(N,N^{*})}E_{(I,J)}^{(K)}z^{I}{\overline{z}}^{J}|z_{1}|^{2k_{1}}\cdots|z_{n}|^{2k_{n}}=0$. We then get $\sum\limits_{(I,J,K)\in A(N,N^{*};P,Q)}E_{(I,J)}^{(K)}\equiv 0,\ \hbox{for each}\ P,\ Q\ \hbox{with}\ |P|=N,\ |Q|=N^{*}.$ We next claim that there is at most one element in $A(N,N^{*};P,Q)$. Indeed, $(I,J,K)\in A(N,N^{*};P,Q)$ if and only if $i_{l}+k_{l}=p_{l},\ j_{l}+k_{l}=q_{l},\ i_{l}\cdot j_{l}=0$, for $1\leq l\leq n.$ Now, if $i_{l}=0$, then $k_{l}=p_{l}$. Since $j_{l}=q_{l}-p_{l}\geq 0$, thus this happens only when $q_{l}\geq p_{l}$. If $j_{l}=0$, then $k_{l}=q_{l}$. Since $i_{l}=p_{l}-q_{l}\geq 0$, we see that this can only happen when $p_{l}\geq q_{l}$. Hence, we see that $i_{l},j_{l}$ are uniquely determined by $p_{l}$ and $q_{l}$ when $p_{l}\not=q_{l}$. When $p_{l}=q_{l}$, it is easy to see that $i_{l}=j_{l}=0,\ k_{l}=q_{l}=p_{l}$. We thus conclude the argument for the claim. This completes the proof of Lemma 2.3. We now let $M\subset\mathbb{C}^{n+1}$ be a formal submanifold defined by: $w=|z|^{2}+E(z,\bar{z})$ (2.5) where $E$ is a formal power series in $(z,\bar{z})$ with $Ord(E)\geq 3$. We will subject (2.5) to the following formal power series transformation in $(z,w)$: $\left\\{\begin{array}[]{ll}z^{\prime}=F=z+f(z,w)&\ Ord_{wt}(f)\geq 2\\\ w^{\prime}=G=w+g(z,w)&\ Ord_{wt}(g)\geq 3.\end{array}\right.$ (2.6) Write $e_{j}\in{{\mathbb{Z}}}^{n}$ for the vector whose component is $1$ at the $j^{\hbox{th}}$-position and is $0$ elsewhere. We next give a formal pseudo-normal form for $(M,0)$ in the following theorem: Theorem 2.3: There exits a unique formal transformation of the form in (2.6) with the normalization $\begin{cases}f_{i,(0)}(u)=0,\ 1\leq i\leq n;\\\ f_{i,(e_{j})}(u)=0\ \mbox{for}\ 1\leq j<i\leq n;\\\ \ f_{1,(e_{1})}(u)=0,\ \hbox{Im}\left(f_{i,(e_{i})}(u)\right)=0\ \mbox{for}\ 2\leq i\leq n,\end{cases}$ (2.7) that transforms M to a formal submanifold defined in the following pseudo- normal form: $w^{\prime}=|z^{\prime}|^{2}+\varphi(z^{\prime},\overline{z^{\prime}}).$ (2.8) Here $\varphi=O(|z^{\prime}|^{3})$ and in the following unique expansion of $\varphi$, $\varphi=\sum_{i_{l}\cdot j_{l}=0,l=1,\cdots,n}\varphi_{(I,J)}z^{I}{\overline{z}}^{J}=\sum_{i_{k}\cdot j_{k}=0,\ k=1,\cdots,n}\varphi_{(I,J)}^{(K)}z^{I}{\overline{z}}^{J}u^{k_{1}}{v_{2}}^{k_{2}}\cdots{v_{n}}^{k_{n}}.$ (2.9) we have, for any $k\geq 0,\ l\geq 1$, $\tau\geq 2$, the following normalization condition: $\begin{cases}\varphi_{(0,0)}^{(\tau e_{1})}=0\ ;\\\ Re(\varphi_{(0,0)}^{(le_{1}+e_{i})})=0,\ \mbox{for}\ 2\leq i\leq n\ ;\\\ \varphi_{(e_{i},e_{j})}^{(le_{1})}=0,\ \mbox{for}\ i>j\ ;\\\ \varphi_{(I,0)}^{(le_{1})}=\varphi_{(0,I)}^{(le_{1})}=\varphi_{(0,I)}^{(ke_{1}+e_{j})}=0,\ \mbox{for}\ |I|\geq 1;\\\ \varphi_{(I,e_{h})}^{(ke_{1})}=0,\ \mbox{for}\ h\geq 1,|I|\geq 2,i_{h}=0;\\\ \varphi^{(0)}_{(0,I)}=\overline{\varphi^{(0)}_{(I,0)}},|I|>2.\end{cases}$ (2.10) Proof of Theorem 2.3: We need to prove that the following equation, with unknowns in $(f,g,\ \varphi)$, can be uniquely solved under the normalization conditions in (2.7) and (2.10): $\begin{array}[]{l}w+g(z,w)=\sum\limits_{i=1}^{n}\big{(}z_{i}+f_{i}(z,w)\big{)}\left(\bar{z_{i}}+\overline{f_{i}(z,w)}\right)+\varphi\left(z+f(z,w),\overline{z}+\overline{f(z,w)}\right).\end{array}$ (2.11) Collecting terms of degree $t$ in the above equation, we obtain for each $t\geq 3$ the following: $\begin{array}[]{ll}E^{(t)}(z,\bar{z})+g^{(t)}(z,u)=2Re\sum\limits_{i=1}^{n}\Big{(}\overline{z_{i}}f_{i}^{(t-1)}(z,u)\Big{)}+\varphi^{(t)}(z,\bar{z})+I^{(t)}(z,\bar{z}),\end{array}$ (2.12) where $I^{(t)}(z,\bar{z})$ is a homogeneous polynomial of degree $t$ depending only on $g^{(\sigma)}$, $f^{(\sigma-1)}$, $\varphi^{(\sigma)}$ for $\sigma<t$. Thus, by an induction argument, we need only to uniquely solve the following equation under the above given normalization: $\Gamma(z,\bar{z})+g(z,u)=2Re\left(\sum\limits_{i=1}^{n}\left(\overline{z_{i}}f_{i}(z,u)\right)\right)+\varphi(z,\bar{z}).$ (2.13) Indeed, if we can uniquely solve (2.13), then, we can start with (2.12) with $t=3$ and $\Gamma=E^{(3)}$. We then get $(F^{(2)},G^{(3)}).$ Now, we transform $M$ by $H_{2}=(z,w)+(F^{(2)},G^{(3)})$. Then the new manifold is normalized up to weighted order $3$. Let $H=(F,G)=(z+O_{wt}(3),w+O_{wt}(4))$ be a normalized map and consider (2.12) with $t=4$. We can then uniquely determine $(F^{(3)},G^{(4)})$. Transforming the manifold by the map $H_{2}=(z,w)+(F^{(3)},G^{(4)})$, we get one which is normalized up to order $4$. Now, by an induction, we can prove the existence part of Theorem 2.3. The uniqueness part of the Theorem follows also from the unique solvability of (2.13). Expand $\Gamma$, $\varphi$ as in (2.4) and (2.9) and expand $f$, $g$ as in (2.3). Making use of Lemma 2.2 and comparing the coefficients in (2.13) of $z^{I}\overline{z}^{J}$ with $i_{l}\cdot j_{l}=0,\ l=1,\cdots,n$, we get the following system: $\displaystyle z^{0}\overline{z}^{0}:$ $\displaystyle\ \ \ -g_{(0)}+\sum\limits_{i=1}^{n}2Re\left(|z_{i}|^{2}f_{i,(e_{i})}\right)+\varphi_{(0,0)}=\Gamma_{(0,0)};$ (2.14) $\displaystyle z_{j},\ \overline{z_{j}}:$ $\displaystyle\ \ \ \begin{cases}-g_{(e_{j})}+\overline{f_{j,(0)}}+\sum\limits_{i=1}^{n}|z_{i}|^{2}f_{i,(e_{i}+e_{j})}+\varphi_{(e_{j},0)}=\Gamma_{(e_{j},0)}\\\ f_{j,(0)}+\sum\limits_{i=1}^{n}|z_{i}|^{2}\overline{f_{i,(e_{i}+e_{j})}}+\varphi_{(0,e_{j})}=\Gamma_{(0,e_{j})}\end{cases}\hskip 5.0pt\mbox{for}\ 1\leq j\leq n;$ (2.15) $\displaystyle z_{i}\overline{z_{j}}:$ $\displaystyle\ \ \ \begin{cases}f_{j,(e_{i})}+\overline{f_{i,(e_{j})}}+\varphi_{(e_{i},e_{j})}=\Gamma_{(e_{i},e_{j})}\\\ \overline{f_{j,(e_{i})}}+f_{i,(e_{j})}+\varphi_{(e_{j},e_{i})}=\Gamma_{(e_{j},e_{i})}\end{cases}\hskip 5.0pt\mbox{for}\ i\neq j;$ (2.16) $\displaystyle z_{i}\overline{z}^{J},\ z^{J}\overline{z_{i}}:$ $\displaystyle\ \ \ \begin{cases}\overline{f_{i,(J)}}+\varphi_{(e_{i},J)}=\Gamma_{(e_{i},J)}\\\ f_{i,(J)}+\varphi_{(J,e_{i})}=\Gamma_{(J,e_{i})}\end{cases}\hskip 5.0pt\mbox{for}\ |J|\geq 2,j_{i}=0;$ (2.17) $\displaystyle z^{I},\ \overline{z}^{I}:$ $\displaystyle\ \ \ \begin{cases}-g_{(I)}+\sum\limits_{i=1}^{n}\left(|z_{i}|^{2}f_{i,(I+e_{i})}\right)+\varphi_{(I,0)}=\Gamma_{(I,0)}\\\ \sum\limits_{i=1}^{n}\left(|z_{i}|^{2}\overline{f_{i,(I+e_{i})}}\right)+\varphi_{(0,I)}=\Gamma_{(0,I)}\end{cases}\hskip 5.0pt\mbox{for}\ |I|\geq 2;$ (2.18) $\displaystyle z^{I}\overline{z}^{J}:$ $\displaystyle\ \ \ \varphi_{(I,J)}=\Gamma_{(I,J)}\ \hbox{for}\ |I|,|J|\geq 2,\ i_{l}\cdot j_{l}=0,\ l=1,\cdots,n.$ (2.19) Here we demonstrate in details how the system (2.18) is uniquely solved. The others are done similarly (and, in fact, more easily). We first substitute (2.2) to (2.18) and then collect coefficients of the zeroth order term, linear terms and higher order terms in $v_{2},\cdots,v_{n}$, respectively, while taking u as a parameter. We obtain, by Lemma 2.2, the following: $\displaystyle\sum_{k}\Gamma_{(I,0)}^{(ke_{1})}u^{k}=-g_{(I)}(u)+2^{1-n}uf_{1,(I+e_{1})}+\sum_{i=2}^{n}2^{i-1-n}uf_{i,(I+e_{i})}+\sum_{k}\varphi_{(I,0)}^{(ke_{1})}u^{k};$ (2.20) $\displaystyle\sum_{k}\Gamma_{(I,0)}^{(ke_{1}+e_{j})}u^{k}=2^{1-j}f_{1,(I+e_{1})}+\sum_{i=2}^{j-1}2^{i-1-j}f_{i,(I+e_{i})}-2^{-1}f_{j,(I+e_{j})}+\sum_{k}\varphi_{(I,0)}^{(ke_{1}+e_{j})}u^{k},\ j\geq 2;$ (2.21) $\displaystyle\varphi_{(I,0)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(I,0)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},k_{2}+\cdots+k_{n}\geq 2;$ (2.22) $\displaystyle\sum_{k}\Gamma_{(0,I)}^{(ke_{1})}u^{k}=2^{1-n}u\overline{f_{1,(I+e_{1})}}+\sum_{i=2}^{n}2^{i-1-n}u\overline{f_{i,(I+e_{i})}}+\sum_{k}\varphi_{(0,I)}^{(ke_{1})}u^{k};$ (2.23) $\displaystyle\sum_{k}\Gamma_{(0,I)}^{(ke_{1}+e_{j})}u^{k}=2^{1-j}\overline{f_{1,(I+e_{1})}}+\sum_{i=2}^{j-1}2^{i-1-j}\overline{f_{i,(I+e_{i})}}-2^{-1}\overline{f_{j,(I+e_{j})}}+\sum_{k}\varphi_{(0,I)}^{(ke_{1}+e_{j})}u^{k},\ j\geq 2;\ $ (2.24) $\displaystyle\varphi_{(0,I)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(0,I)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ k_{2}+\cdots+k_{n}\geq 2.$ (2.25) Using the normalization in $\varphi$ and letting $u=0$ in (2.20) (2.23), we get $\Gamma^{(0)}_{(I,0)}=-g_{(I)}(0)+\varphi^{(0)}_{(I,0)}$ and $\Gamma^{(0)}_{(0,I)}=\varphi^{(0)}_{(0,I)}$. By the normalization $\varphi^{(0)}_{(I,0)}=\overline{\varphi^{(0)}_{(0,I)}}$, we get $\varphi^{(0)}_{(I,0)}=\overline{\Gamma^{(0)}_{(I,0)}}$ and $\displaystyle g_{(I)}(0)=\overline{\Gamma^{(0)}_{(0,I)}}-\Gamma^{(0)}_{(I,0)}.$ (2.26) Sum up (2.24) with $j=2,\cdots,n$ and then add it to (2.23). By the the normaliztaion condition $\varphi_{(0,I)}^{(le_{1})}=\varphi_{(I,0)}^{(ke_{+}e_{j})}=0$ for $k\geq 0,l\geq 1$, we obtain the following: $\displaystyle f_{1,(I+e_{1})}(u)=\sum_{k\geq 1}\overline{\Gamma_{(0,I)}^{(ke_{1})}}u^{k-1}+\sum_{k\geq 0}\sum_{j=2}^{n}\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{j})}}u^{k}.$ (2.27) Subtracting the complex conjugate of (2.24) from (2.21), we obtain $\displaystyle\varphi_{(I,0)}^{(ke_{1}+e_{j})}=\Gamma_{(I,0)}^{(ke_{1}+e_{j})}-\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{j})}},\ j\geq 2,\ k\geq 0.$ (2.28) From (2.20) and (2.24), we can similarly get $\displaystyle g_{(I)}(u)=\sum_{k=0}^{\infty}\left(\overline{\Gamma_{(0,I)}^{(ke_{1})}}-\Gamma_{(I,0)}^{(ke_{1})}\right)u^{k},\ \ |I|\geq 2.$ (2.29) Back to the equation (2.24), we can inductively get: $\displaystyle f_{j,(I+e_{j})}(u)=\sum_{k\geq 1}\overline{\Gamma_{(0,I)}^{(ke_{1})}}u^{k-1}+\sum_{k\geq 0}\left(\sum_{i=0}^{n-j-1}\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{n-i)}}}-\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{j})}}\right)u^{k}\ \mbox{for}\ 2\leq j\leq n.$ (2.30) Similarly, we get from (2.14) the following $\displaystyle g_{(0)}(u)=\sum_{k\geq 2}\left(-\Gamma_{(0,0)}^{(ke_{1})}u^{k}\right)-Re\left(\sum_{k\geq 1;\ j=2,\cdots,n}\Gamma_{(0,0)}^{(ke_{1}+e_{j})}u^{k+1}\right);$ (2.31) $\displaystyle f_{h,(e_{h})}(u)=\frac{1}{2}\sum_{k\geq 1}\left(-\sum_{j=2}^{h-1}Re(\Gamma_{(0,0)}^{(ke_{1}+e_{j})}u^{k})-2Re(\Gamma_{(0,0)}^{(ke_{1}+e_{h})}u^{k})\right),\ h\geq 2;$ (2.32) $\displaystyle\varphi_{(0,0)}=\Gamma_{(0,0)}-\sum\limits_{k\geq 2}\Gamma_{(0,0)}^{(ke_{1})}u^{k}-Re(\sum\limits_{k\geq 1,j=2,\cdots,n}\Gamma_{(0,0)}^{(ke_{1}+e_{j})}u^{k}v_{j}).$ (2.33) From (2.15), we obtain the following: $\displaystyle f_{1,(e_{1}+e_{j})}(u)=\sum_{k\geq 1}\overline{\Gamma_{(0,e_{j})}^{(ke_{1})}}u^{k-1}+\sum_{k\geq 0}\sum_{i=2}^{n}\overline{\Gamma_{(0,e_{j})}^{(ke_{1}+e_{i})}}u^{k};$ (2.34) $\displaystyle f_{i,(e_{j}+e_{i})}(u)=\sum_{k\geq 1}\overline{\Gamma_{(0,e_{j})}^{(ke_{1})}}u^{k-1}+\sum_{k\geq 0}\left(\sum_{l=0}^{n-i-1}\overline{\Gamma_{(0,e_{j})}^{(ke_{1}+e_{n-l)}}}-\overline{\Gamma_{(0,e_{j})}^{(ke_{1}+e_{i})}}\right)u^{k}\ \mbox{for}\ 2\leq i\leq n;$ (2.35) $\displaystyle g_{(e_{j})}(u)=\sum_{k=1}^{\infty}\left(\overline{\Gamma_{(0,e_{j})}^{(ke_{1})}}-\Gamma_{(e_{j},0)}^{(ke_{1})}\right)u^{k};$ (2.36) $\displaystyle\varphi_{(e_{j},0)}^{(ke_{1}+e_{l})}=\Gamma_{(e_{j},0)}^{(ke_{1}+e_{l})}-\overline{\Gamma_{(0,e_{j})}^{(ke_{1}+e_{l})}},\ l\geq 2,\ k\geq 0;$ (2.37) $\displaystyle\varphi_{(e_{j},0)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(e_{j},0)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ k_{2}+\cdots+k_{n}\geq 2;$ (2.38) $\displaystyle\varphi_{(0,e_{j})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(0,e_{j})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ k_{2}+\cdots+k_{n}\geq 2.$ (2.39) From (2.16), we get $\displaystyle f_{i,(e_{j})}(u)=\sum_{k=1}^{\infty}{\Gamma_{(e_{j},e_{i})}^{(ke_{1})}}u^{k},\ i<j;$ (2.40) $\displaystyle\varphi^{ke_{1}}_{(e_{i},e_{j})}=\Gamma^{(ke_{1})}_{(e_{i},e_{j})}-\overline{\Gamma^{(ke_{1})}_{(e_{j},e_{i})}},\ i<j,\ k\geq 1;$ (2.41) $\displaystyle\varphi_{(e_{i},e_{j})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(e_{i},e_{j})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ \rm{for}\ k_{2}+\cdots+k_{n}\geq 1.$ (2.42) From (2.17), we obtain $\displaystyle f_{i,(J)}(u)=\sum_{k\geq 0}{\Gamma_{(J,e_{i})}^{(ke_{1})}}u^{k},\ \ 1\leq i\leq n,$ (2.43) $\displaystyle\varphi^{(ke_{1})}_{(e_{i},J)}=\Gamma_{(e_{i},J)}^{(ke_{1})}-\overline{\Gamma_{(J,e_{i})}^{(ke_{1})}},\ \ 1\leq i\leq n,k\geq 0,$ (2.44) $\displaystyle\varphi_{(J,e_{i})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(J,e_{i})}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ \rm{for}\ k_{2}+\cdots+k_{n}\geq 1,$ (2.45) $\displaystyle\varphi_{(e_{i},J)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})}=\Gamma_{(e_{i},J)}^{(k_{1}e_{1}+k_{2}e_{2}+\cdots+k_{n}e_{n})},\ \rm{for}\ k_{2}+\cdots+k_{n}\geq 1.$ (2.46) where $|J|\geq 2$ and $j_{i}=0$. Summarizing the solutions just obtained, we have the following formula: (One can also directly verify that they are indeed the solutions of (2.13) with the normalization conditions given in (2.7) and (2.10)) $\begin{array}[]{cll}F_{1}(z,u)&=&z_{1}+f_{1}(z,u)=z_{1}+\sum\limits_{k\geq 0,j_{1}=0,|J|\geq 1}z^{J}\Gamma_{(J,e_{1})}^{(ke_{1})}u^{k}+\sum\limits_{|I|\geq 1}z^{I+e_{1}}S^{(1)}_{I},\\\ F_{h}(z,u)&=&z_{h}+f_{h}(z,u)=z_{h}+\frac{1}{2}z_{h}\sum\limits_{k\geq 1}\big{(}-\sum\limits_{j=2}^{h-1}Re(\Gamma_{(0,0)}^{(ke_{1}+e_{j})}u^{k})-2Re(\Gamma_{(0,0)}^{(ke_{1}+e_{h})}u^{k})\big{)}\\\ &&+\sum\limits_{k\geq 1,i>h}z_{i}\Gamma_{(e_{i},e_{h})}^{(ke_{1})}u^{k}+\sum\limits_{k\geq 0,j_{h}=0,|J|\geq 2}z^{J}\Gamma_{(J,e_{h})}^{(ke_{1})}u^{k}+\sum\limits_{|I|\geq 1}z^{I+e_{h}}S^{(h)}_{I},\ \ \ n\geq h\geq 2,\\\ G(z,u)&=&u+g(z,u)=u+\left(-\sum\limits_{k\geq 2}\Gamma_{(0,0)}^{(ke_{1})}u^{k}-Re(\sum\limits_{k\geq 1,j=2,\cdots,n}\Gamma_{(0,0)}^{(ke_{1}+e_{j})}u^{k+1})\right)\\\ &&+\sum\limits_{k\geq 0,|I|\geq 1}z^{I}u^{k}\left(\overline{\Gamma_{(0,I)}^{(ke_{1})}}-\Gamma_{(I,0)}^{(ke_{1})}\right),\\\ \varphi&=&\Gamma(z,\bar{z})+g(z,u)-2Re\left(\sum\limits_{i=1}^{n}\left(\overline{z_{i}}f_{i}(z,u)\right)\right),\end{array}$ (2.47) where $\left\\{\begin{array}[]{l}S^{(1)}_{I}=\sum\limits_{k\geq 1}\overline{\Gamma_{(0,I)}^{(ke_{1})}}u^{k-1}+\sum\limits_{k\geq 0}\sum\limits_{i=2}^{n}\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{i})}}u^{k},\\\ S^{(h)}_{I}=\sum\limits_{k\geq 1}\overline{\Gamma_{(0,I)}^{(ke_{1})}}u^{k-1}+\left(\sum\limits_{k\geq 0}\sum\limits_{i=0}^{n-h-1}\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{n-i)}}}u^{k}\right)-\sum\limits_{k\geq 0}\overline{\Gamma_{(0,I)}^{(ke_{1}+e_{h})}}u^{k},\ \mbox{for}\ 2\leq h\leq n.\end{array}\right.$ (2.48) This completes the proof of Theorem 2.3. Let $(M,0)$ be as in (2.5). We say that $(M^{*},0)$ is a formal pseudo-normal form for $(M,0)$ if $(M^{*},0)$ is formally equivalent to $(M,0)$ and $M^{*}$ is defined by $w=|z|^{2}+\varphi$ with $\varphi$ satisfying the normalizations in (2.13). We notice that pseudo-normal forms of $(M,0)$ are not unique. Furthermore, we have the following observations: Remark 2.4: (A).The pseudo-normal form obtained in Theorem 2.3 contains information reflecting both the singular CR structure and partial strongly pseudoconvex CR structure at the point under study. For instance, the following submanifold in ${{\mathbb{C}}}^{3}$ is given in a pseudo-normal form: $M:\ w=|z|^{2}+2Re\sum_{j_{1}+j_{2}\geq 3}\left(a_{j_{1}j_{2}}z_{1}^{j_{1}}z_{2}^{j_{2}}\right)+\sum_{j_{1}\geq 2,j_{2}\geq 2}b_{j_{1}\overline{j_{2}}}z_{1}^{j_{1}}\overline{z_{2}}^{j_{2}}.$ (2.49) Here the harmonic terms $Re\sum_{j_{1}+j_{2}\geq 3}\left(a_{j_{1}j_{2}}z_{1}^{j_{1}}z_{2}^{j_{2}}\right)$ are presented due to the nature of CR singularity of $M$ at $0$, which may be compared with the Moser pseudo-normal form in [Mos] in the pure CR singularity setting. Typical mixed terms like $\sum_{j_{1}\geq 2,j_{2}\geq 2}b_{j_{1}\overline{j_{2}}}z_{1}^{j_{1}}\overline{z_{2}}^{j_{2}}$ are associated with the partial CR structure near $0$, which can be compared with the Chern-Moser normal form in the pure CR setting [CM]. (B). Suppose that $M$ is defined by a formal equation of the form: $w=|z|^{2}+E(z,\overline{z})$ with $Ord(E)\geq 3$ and $\overline{E(z,\overline{z})}=E(z,\overline{z})$. In the normalized map $H(z,w)=(F(z,w),G(z,w))$ transforming $M$ into its normal form in Theorem 2.3, the $w$-component $G(z,u)$ is only a function in $u$ and is formally real- valued, by the formula in (2.47). This is due to the fact that the $\Gamma$ in (2.47) obtained from each induction stage in the process of the proof of Theorem 2.3 is formally real-valued. Hence, the $\varphi$ in the pseudo- normalization of $M$ obtained in Theorem 2.3 is also formally real-valued. However, fundamentally different from the two dimensional case, this is no longer true for a general $M$. Indeed, we will see in Theorem 3.5 that $M$ can be formally flattened if and only if its pseudo-normal form is given by a formal real-valued function. ## 3 Normalization of holomorphic maps by automorphisms of the quadric In this section, we first compute the isotropic automorphism group of the model space $M_{\infty}\subset\mathbb{C}^{n+1}$ defined by the equation: $w=\sum_{i=1}^{n}|z_{i}|^{2}$. Write $Aut_{0}(M_{\infty})$ for the set of biholomorphic self-maps of $(M_{\infty},0)$. We have the following: Proposition 3.1: $Aut_{0}(M_{\infty})$ consists of the transformations given in the following (3.1) or (3.2) : $\left\\{\begin{array}[]{l}z^{\prime}=b(w)\frac{wa(w)-\frac{\langle z,\bar{a}(w)\rangle}{\langle a(w),\bar{a}(w)\rangle}a(w)+\sqrt{1-wa(w)\bar{a}(w)}\left(z-\frac{\langle z,\bar{a}(w)\rangle}{\langle a(w),\bar{a}(w)\rangle}a(w)\right)}{1-\langle z,\bar{a}(w)\rangle}U(w)\\\ w^{\prime}=b(w)\bar{b}(w)w\end{array}\right.$ (3.1) $(z^{\prime},w^{\prime})=\left(b(w)zU(w),b(w)\bar{b}(w)w\right).$ (3.2) where $a=(a_{1},\cdots,a_{n})$, $\sum_{j=1}^{n}a_{j}(0)\bar{a_{j}}(0)<1$, $\langle z,\bar{a}\rangle=\sum\limits_{i=1}^{n}\bar{a_{i}}z_{i}$, $b(0)\neq 0$, $a(0)\neq 0$, $U(Re(w))$ is a unitary matrix and $a(w),b(w),U(w)$ are holomorphic in $w$. Proof of Proposition 3.1: Write $w=x+\sqrt{-1}y$. Let $(F,G)\in Aut_{0}(M_{\infty})$. Then $Im(G(z,|z|^{2}))\equiv 0$ for $z\approx 0$. Since $M_{\infty}$ bounds a family of balls near 0 defined by $B_{r}=\left\\{(z,w)\in\mathbb{C}^{n+1}:w=x+\sqrt{-1}y,y=0,x=r^{2}\geq|z^{2}|\right\\}.$ We see that $Im(G(z,x))\equiv 0$ for $z\approx 0$ and $x(\in{{\mathbb{R}}})\approx 0$. Therefore, $G(z,w)=G(w)=cw+o(w)$ $(c>0)$ is independent of $z$ and takes real value when $w=x$ is real. Now $F(z,r^{2})$ must be a biholomorphic map from $|z|^{2}<r^{2}$ to $|z|^{2}<G(r^{2})$ for any $r>0$. Using the explicit expression for automorphisms of the unit ball (see [Rud]), we obtain either: $F(z,r^{2})=\sqrt{G(r^{2})}\frac{a(r)-\frac{\langle{z\over r},\bar{a}(r)\rangle}{\langle a(r),\bar{a}(r)\rangle}a(r)+v\left({z\over r}-\frac{\langle{z\over r},\bar{a}(r)\rangle}{\langle a(r),\bar{a}(r)\rangle}a(r)\right)}{1-\langle{z\over r},\bar{a}(r)\rangle}U(r)\\\ $ (3.3) where $U(r)$ is a unitary matrix and $v=\sqrt{1-a(r)\bar{a}(r)},\ a\not=0$, or we have $F(z,r^{2})=\sqrt{G(r^{2})}({z\over r})U(r).$ (3.4) Write $G(x)=xb(x)\overline{b}(x)$ with $b(0)\not=0$ and $b(w)$ holomorphic in $w$. In the case of (3.4), $F(z,x)=b(x)zU(r)e^{\sqrt{-1}\theta(x)}$ is real analytic, where $\theta(x)$ is real-valued real analytic function in $x$ . Hence, $b(x)U(r)e^{\sqrt{-1}\theta(x)}$ is the Jacobian matrix of $F$ in $z$. Since both $e^{\sqrt{-1}\theta(x)}$ and $b(x)(\not=0)$ are real analytic for $x\approx 0$, we conclude that $U(r)$ is real analytic in $x$. Hence, $U(w)$ is also holomorphic in $w$. Still write $U(x)$ for $U(x)e^{i\theta(x)}$. We see the proof of Proposition 3.1 in the case of (3.2). Suppose that $a\neq 0$. Still write $G(w)=wb(w)\overline{b}(w)$ with $b(0)\not=0$. We have $F(z,r^{2})=b(r^{2})\frac{ra(r)-\frac{\langle z,\bar{a}(r)\rangle}{\langle a(r),\bar{a}(r)\rangle}a(r)+v\left(z-\frac{\langle z,\bar{a}(w)\rangle}{\langle a(r),\bar{a}(r)\rangle}a(r)\right)}{1-\langle z,{\bar{a}(r)\over r}\rangle}e^{i\theta}U(r)\\\ $ Since $f(z,w)$ is holomorphic in $(z,w)$ and $f(0,w)=b(w)\sqrt{w}a(\sqrt{w})U^{*}(\sqrt{w})$ with $U^{*}=e^{i\theta}U$, we see that $\sqrt{w}a(\sqrt{w})U^{*}(\sqrt{w})$ is holomorphic in $w$ . In particular, $|a(\sqrt{w})|^{2}$ is real analytic in $w$. Moreover, $\frac{\partial F}{\partial z_{i}}(0,w)=b(w)\left(\frac{|a|^{2}-v-1}{|a|^{2}}\bar{a_{i}}a+ve_{i}\right)U^{*}(\sqrt{w})$ is analytic. Since $\sqrt{w}a(\sqrt{w})U^{*}(\sqrt{w})$ is real analytic, we see that $\left(\frac{|a|^{2}-v-1}{|a|^{2}}\bar{a_{i}}a+ve_{i}\right)U^{*}(\sqrt{w})\overline{U^{*}(\sqrt{w})}^{t}\overline{a(\sqrt{w})}^{t}r=\left((|a|^{2}-v-1)+v\right)r\bar{a_{i}}$ is real analytic, too. Here $(\cdot)^{t}$ denotes the matrix transpose. Since $(|a|^{2}-v-1)+v=|a|^{2}-1$ is real analytic, we conclude that both $ra_{i}$ and $a_{i}/r$ are real analytic in $w$. Since both $\sqrt{w}a(\sqrt{w})U^{*}(\sqrt{w})$ and $ra_{i}$ are real analytic, we see that $U^{*}(\sqrt{w})$ is real analytic. Still denote $a$ for $a/r$, we further obtain the following with the given properties stated in the Proposition: $\left\\{\begin{array}[]{l}F(z,w)=b(w)\frac{wa(w)-\frac{\langle z,\bar{a}(w)\rangle}{\langle a(w),\bar{a}(w)\rangle}a(w)+\sqrt{1-wa(w)\bar{a}(w)}\left(z-\frac{\langle z,\bar{a}(w)\rangle}{\langle a(w),\bar{a}(w)\rangle}a(w)\right)}{1-\langle z,\bar{a}(w)\rangle}U^{*}(w)\\\ G(w)=b(w)\bar{b}(w)w.\end{array}\right.$ This completes the proof of Proposition 3.1. Remark 3.2: In Proposition 3.1, if we let $a(w),b(w),U(w)$ be formal power series in $w$ with $a(0),\ b(0)\not=0$ and $\langle a(0),\overline{a}(0)\rangle<1$, $U(x)\cdot U(x)^{t}=I$, then (3.1) and (3.2) give formal automorphisms of $M_{\infty}$, which are not convergent. Write the set of automorphisms obtained in this way as $aut_{0}(M_{\infty})$. One may prove that $aut_{0}(M_{\infty})$ consists of all the formal automorphisms of $(M_{\infty},0)$ . We now suppose that $H=(F,G)$ is a formal equivalence self-map of $({{\mathbb{C}}}^{n+1},0)$, mapping a formal submanifold of the form $w=|z|^{2}+O(|z|^{3})$ to a submanifold of the form $w=|z|^{2}+O(|z|^{3})$. The following lemma shows that we can always normalize $H$ by composing it from the left with an element from $aut_{0}(M_{\infty})$ to get a normalized mapping. This fact will be used in the proof of Theorem 1. In what follows, we set $v(g,a)=\sqrt{1-g\cdot a(g)\cdot\bar{a}(g)}$. Lemma 3.3: There exists a unique automorphism $T\in aut_{0}(M_{\infty})$ such that $T\circ H$ satisfies the normalized condition in (2.7). When $H$ is biholomorphic, $T\in Aut_{0}(M_{\infty})$ Proof of Lemma 3.3: First, it is easy to see that by composing an automorphism of the form $w^{\prime}=|c|^{2}w,\ z^{\prime}=czU$, we can assume that $F=z+O_{wt}(2)$ and $G=w+O_{wt}(3)$ (see [Hu1]). Here $c$ is a non-zero constant and $U$ is a certain $n\times n$-unitary matrix. Let $b(w)=1,a_{j}=\alpha_{j}(w),a_{1}=\cdots=a_{j-1}=a_{j+1}=\cdots=a_{n}=0$, and $U=I$ in (3.1). We get the following automorphism of $M_{\infty}$ $T_{j}=\left(\frac{v(w,\alpha_{j})z_{1}}{1-\bar{\alpha_{j}}z_{j}},\cdots,\frac{v(w,\alpha_{j})z_{j-1}}{1-\bar{\alpha_{j}}z_{j}},\frac{z_{j}-w\alpha_{j}}{1-\bar{\alpha_{j}}z_{j}},\frac{v(w,\alpha_{j})z_{j+1}}{1-\bar{\alpha_{j}}z_{j}},\cdots,\frac{v(w,\alpha_{j})z_{n}}{1-\bar{\alpha_{j}}z_{j}},w\right).$ Write $\left\\{\begin{array}[]{l}H_{j}=(_{(j)}F,_{(j)}G)=T_{j}\circ T_{j-1}\circ\cdots\circ T_{1}\circ H,\ H_{0}=H;\\\ \alpha_{j}=\frac{{}_{(j-1)}F_{j,(0)}(u)}{{}_{(j-1)}G_{(0)}(u)}\circ\left({}_{(j-1)}G_{(0)}(u)\right)^{-1}.\end{array}\right.$ (3.5) Then a direct computation shows that $(_{(j)}F)_{i,(0)}(u)=0$ for $1\leq i\leq j$. In particular, we have $(_{(j)}F)_{i,(0)}(u)=0$ for all $1\leq i\leq n$. Still write $H$ for $H_{n}$. Next, for $i<j$, let $b(w)=1,a=0$, and let $U_{j}^{i}=\left(\begin{array}[]{ccccc}I&0&0&0&0\\\ 0&\cos(\theta_{j}^{i})&0&-\sin(\theta_{j}^{i})&0\\\ 0&0&I&0&0\\\ 0&\sin(\theta_{j}^{i})&0&\cos(\theta_{j}^{i})&0\\\ 0&0&0&0&I\\\ \end{array}\right)$ in (3.1), where $\cos(\theta_{j}^{i})$ is at the $i^{th}$ row and the $j^{th}$ column. Then we get an automorphism $T_{j}^{i}$. Set $\begin{array}[]{l}H_{j}^{i}=(_{j}^{i}F,_{j}^{i}G)=T_{j}^{i}\circ\cdots\circ T_{i+1}^{i}\circ T_{n}^{i-1}\circ\cdots T_{i}^{i-1}\circ\cdots\circ T_{n}^{1}\circ\cdots\circ T_{2}^{1}\circ H,\\\ \theta_{j}^{i}=\left\\{\begin{array}[]{ll}\tan^{-1}\left(\frac{(_{n}^{i-1}F)_{j,(e_{i})}}{(_{n}^{i-1}F)_{i,(e_{i})}}\right)\circ\left({}_{n}^{i-1}G_{(0)}(w)\right)^{-1},\hskip 14.22636pt&j=i+1;\\\ \tan^{-1}\left(\frac{(_{j-1}^{i}F)_{j,(e_{i})}}{(_{j-1}^{i}F)_{i,(e_{i})}}\right)\circ\left({}_{j-1}^{i}G_{(0)}(w)\right)^{-1},&j\neq i+1.\\\ \end{array}\right.\end{array}$ (3.6) Then we can inductively prove that $H_{j}^{i}$ satisfies $(_{j}^{i}F)_{(0)}=0\ ,\ (_{j}^{i}F)_{k,(e_{l})}=0\ \mbox{for}\ l=i,i+1\leq k\leq j\ \mbox{or}\ l<i,l+1\leq k\leq n.$ In particular, we see that $H_{n}^{n-1}$ satisfies $(_{n}^{n-1}F)_{(0)}=0\ ,\ (_{n}^{n-1}F)_{i,(e_{j})}=0\ \mbox{for}\ 1\leq j<i\leq n$. Still write $H$ for $H_{n}^{n-1}$ and set $H^{\prime}=T\circ H=(F^{\prime},G^{\prime})$ with $T=\left(d(w)z,d(w)\bar{d}(w)w\right)\ ,\ d=\frac{1}{F_{1,(e_{1})}\left(w\right)}\circ\left(G_{(0)}(w)\right)^{-1}.$ Then $H^{\prime}$ satisfies $(F^{\prime})_{(0)}=0\ ,\ (F^{\prime})_{1,(e_{1})}=1\ ,\ (F^{\prime})_{i,(e_{j})}=0\ \mbox{for}\ 1\leq j<i\leq n.$ At last, a composition from the left with the rotation map as follows: $\hat{T}=(z_{1},\beta_{2}z_{2},\cdots,\beta_{n}z_{n},w),\ \beta_{i}=\frac{(\bar{F^{\prime}})_{i,(e_{i})}(w)}{\sqrt{(F^{\prime})_{i,(e_{i})}(w)\cdot(\bar{F^{\prime}})_{i,(e_{i})}(w)}}\circ\left({G^{\prime}}_{(0)}(w)\right)^{-1}$ makes $H^{\prime}$ satisfy the normalization condition (2.7). This proves the existence part of the lemma. Next, suppose both $H=(F,G)=(z+O_{wt}(2),w+O_{wt}(3))$ and $\hat{H}=(\hat{F},\hat{G})=T\circ H=(z+O_{wt}(2),w+O_{wt}(3))$ satisfy the normalization condition (2.7). Here T is an automorphism of $M_{\infty}$. Then $T$ must be of the form in (3.2), for $T(0,w)=0$. Hence, $T=(b(w)zU(w),b(w)\bar{b}(w)w).$ By the normalization condition (2.7) on $H,\ \hat{H}$, we have $\left(\begin{array}[]{cc}\begin{array}[]{ll}1&\\\ &\hat{F}_{2,(e_{2})}\end{array}&0\\\ \ast&\begin{array}[]{ll}\ddots&\\\ &\hat{F}_{n,(e_{n})}\end{array}\end{array}\right)=b\left(G_{(0)}(w)\right)\left(\begin{array}[]{cc}\begin{array}[]{ll}1&\\\ &F_{2,(e_{2})}\end{array}&0\\\ \ast&\begin{array}[]{ll}\ddots&\\\ &F_{n,(e_{n})}\end{array}\end{array}\right)U(G_{0}(w)).$ (3.7) with $U(x)$ unitary and $Im(\hat{F}_{i,(e_{i})}(0,u))=Im(F_{i,(e_{i})}(0,u))=0$. Considering the norm of the first row of the right hand side, we get $b(G_{(0)}(w))\cdot\overline{b}(G_{(0)}(w))=1$ in case $G_{(0)}(w)=\overline{G_{(0)}(w)}$. Since $G_{0}(w)=w+o(w)$, this implies that $b(w)\overline{b}(w)\equiv 1$ and thus $T=(b(w)zU(w),w)$. Write $b(w)U(w)=\widetilde{U}(w)=\left(\begin{array}[]{ccc}u_{11}&\cdots&u_{nn}\\\ \vdots&\ddots&\vdots\\\ u_{n1}&\cdots&u_{nn}.\end{array}\right).$ We notice that $\widetilde{U}$ is a lower triangular matrix and is unitary when $w=x$. Thus we have $u_{ii}(w)\overline{u_{ii}}(w)=1$ and $u_{ij}=0$ for $i\not=j$. Notice that $u_{11}\equiv 1,\ \hat{F}_{i,(e_{i})}(w)=u_{ii}(w)\cdot F_{i,(e_{i})}(w)\ \rm{for}\ 2\leq i\leq n.$ Since $\hat{F}_{i,(e_{i})}(x),F_{i,(e_{i})}(x)=1+o(x)$ are real, we get $u_{ii}(x)=1$. This proves the uniqueness part of the lemma. Lemma 3.4: Suppose that $H$ with $H(0)=0$ is an equivalence map from $w=|z|^{2}+\varphi(z,\bar{z})$ to $w^{\prime}=|z^{\prime}|^{2}+\varphi^{\prime}(z^{\prime},\bar{z^{\prime}})$. Here $\varphi$ and $\varphi^{\prime}$ are normalized as in (2.10). Let $s,s^{\prime}$ be the lowest order of vanishing in $\varphi$ and $\varphi^{\prime}$, respectively, then $s=s^{\prime}$. Proof of Lemma 3.4: We seek for a contradiction if $s\not=s^{\prime}$. Assume, for instance, that $s<s^{\prime}$. Let $T$ be an automorphism of $M_{\infty}$ with $T\circ H$ being normalized as in (2.7). Suppose that $T$ transforms $w^{\prime}=|z^{\prime}|^{2}+\varphi^{\prime}(z^{\prime},\bar{z^{\prime}})$ to $w^{\prime\prime}=|z^{\prime\prime}|^{2}+\varphi^{\prime\prime}(z^{\prime\prime},\overline{z^{\prime\prime}})$ with $s^{\prime\prime}$ the lowest vanishing order for $\varphi^{\prime\prime}$. We claim that $s^{\prime}=s^{\prime\prime}$. Suppose not. We assume, without loss of generality, that $s^{\prime}<s^{\prime\prime}$. Write the linear part of $T$ (in $(z^{\prime},w^{\prime})$) as $(z^{\prime\prime}=z^{\prime}B+Dw^{\prime},w^{\prime\prime}=dw^{\prime})$ with $B\in GL(n,{{\mathbb{C}}}),\ d\not=0$. Then a direct computation shows that $\varphi^{\prime\prime(s^{\prime})}(z^{\prime}B,\overline{z^{\prime}B})=d\cdot\varphi^{\prime(s^{\prime})}(z^{\prime},z^{\prime}).$ This is a contradiction. Now, $T\circ H$ transforms $w=|z|^{2}+\varphi$ to $w^{\prime\prime}=|z^{\prime\prime}|^{2}+\varphi^{\prime\prime}$ with $T\circ H,\ \varphi$ being normalized as in (2.7) and (2.10), respectively. Also $s<s^{\prime\prime}$. we see that $T\circ H$ transforms $w=|z|^{2}+\varphi^{(s)}$ to $w=|z|^{2}$, moduling $O(|(z_{1},\cdots,z_{n})|^{s+1})$. This contradicts the uniqueness part of Theorem 2.3. The proof of Lemma 3.4 is complete. We say that a formal submanifold $(M,0)$ of real dimension $2n$ defined by (2.5) can be formally flattened if there is a formal change of coordinates $(z^{\prime},w^{\prime})=H(z,w)$ with $H(0)=0$ such that in the new coordinates $(M,0)$ is defined by a formal function of the form $w^{\prime}=E^{*}(z^{\prime},\overline{z^{\prime}})$ with $E^{*}(z^{\prime},\overline{z^{\prime}})=\overline{{E^{*}(z^{\prime},\overline{z^{\prime}})}}$. We also say a pseudo-normal form of $(M,0)$ given by $w=|z|^{2}+\varphi(z,\overline{z})$ with $\varphi$ satisfying the normalizations in (2.10) is a flat pseudo-normal form if $\varphi$ is formally real-valued. An immediate application of Lemma 3.3 and Remark 2.4 (b) is that if $(M,0)$ has a flat pseudo-normal form, then all of its other pseudo-normal forms are flat. Indeed, for a given pseudo-normal form of $(M,0)$, there is a formal equivalence map $H$ mapping it into $Imw=0$. Now, by Lemma 3.3, we can compose $H$ with an element $T$ of $aut_{0}(M_{\infty})$ to normalize $H$. Next, since $T$ maps any flattened submanifold to a flattened submanifold, there is a formal transformation $H^{*}$ such that $H^{*}\circ T\circ H$ maps the pseudo-normal form given at the beginning to a flat one. On the other hand, since $H^{*}\circ T\circ H$ satisfies the normalizations in (2.7), by Theorem 2.3, we see that $H^{*}\circ T\circ H=id$ and two pseudo-normal forms are the same. Summarizing the above, we proved the following: Theorem 3.5: Let $(M,0)$ be a formal submanifold defined by an equation of the form: $w=|z|^{2}+E(z,\overline{z})$ with $E=O(|z|^{3})$. Then the following statements are equivalent: (I). $(M,0)$ can be flattened (II). $(M,0)$ has a flat pseudo-normal form. Namely, $M$ has a pseudo-normal form given by an equation of the form: $w^{\prime}=|z^{\prime}|^{2}+\varphi(z^{\prime},\overline{z^{\prime}})$ with $\varphi$ satisfying the normalizations in (2.10) and the reality condition $\varphi(z^{\prime},\overline{z^{\prime}})=\overline{\varphi(z^{\prime},\overline{z^{\prime}})}.$ (III). Any pseudo-normal form of $(M,0)$ is flat. Remark 3.6: By Theorem 3.5, we see that $M$ defined in (2.49) can be formally flattened if and only if $b_{i\overline{j}}=\overline{b_{j\overline{i}}}$ for all $i,j.$ ## 4 Proof of Theorem 1 We now give a proof of Theorem 1 by using the rapidly convergent power series method. We let $M$ be defined by $w=\Phi(z,\bar{z})=|z|^{2}+E(z,\bar{z})$ (4.1) where $E(z,\xi)$ is holomorphic near $z=\xi=0$ with vanishing order $\geq 3$. Assume that $H=(F,G)=(z+f,w+g)$ is a formal map satisfying the normalization condition in (2.6). We define $R=(r_{1},r_{2},\cdots,r_{n})=(2^{-\frac{n-2}{2}}r,2^{-\frac{n-2}{2}}r,2^{-\frac{n-3}{2}}r\cdots,2^{-\frac{1}{2}}r,r).$ (4.2) Then $|R|^{2}=2^{-(n-2)}r^{2}+\sum_{h=2}^{n}(2^{-\frac{n-i}{2}}r)^{2}=2r^{2}.$ Define the domains: $\begin{array}[]{l}\Delta_{r}=\\{(z,w):|z_{i}|<r_{i},|w|<2r^{2}\\},\\\ D_{r}=\\{(z,\xi):|z_{i}|<r_{i},|\xi_{i}|<r_{i}\ \mbox{for}\ 1\leq i\leq n\\}.\\\ \end{array}$ (4.3) When $E(z,\xi)$ is defined over $\overline{D_{r}}$, we set the norm of $E(z,\bar{z})$ on $D_{r}$ by $\|E\|_{r}=\sup\limits_{(z,\xi)\in D_{r}}|E(z,\xi)|.$ (4.4) Also for a holomorphic map $h(z,w)$ defined on $\overline{\Delta_{r}}$, we define $|h|_{r}=\sup\limits_{(z,w)\in\Delta_{r}}|h(z,w)|.$ (4.5) After a scaling transformation $(z,\xi,w)\longrightarrow(az,a\xi,a^{2}w)$, we may assume that $E$ is holomorphic on $\overline{D_{1}}$ with $|E|_{1}\leq\eta$ for a given small $\eta>0$. Suppose that $H$ maps $M$ to the quadric $w^{\prime}=|z^{\prime}|^{2}$. Then we have the following equation: $\begin{array}[]{l}E(z,\bar{z})+g(z,\Phi)=2Re\Big{(}\sum\limits_{i=1}^{n}\bar{z_{i}}f_{i}(z,\Phi)\Big{)}+|f(z,\Phi)|^{2}.\end{array}$ (4.6) We consider the following linearized equation of (4.6) with $(f,g,\varphi)$ as its unknowns: $\begin{array}[]{l}E(z,\bar{z})=-g(z,u)+2Re\Big{(}\sum\limits_{i=1}^{n}\bar{z_{i}}f_{i}(z,u)\Big{)}+\varphi(z,\bar{z}),\end{array}$ (4.7) where $\varphi$ satisfies (2.10). The unique solution of (4.7) is given in the formula (2.47). However, we will make a certain truncation to $(f,g,\varphi)$ to faciliate the estimates. Suppose that $Ord(E)\geq d\geq 3$. Set $\left\\{\begin{array}[]{l}f=\hat{f}+O_{wt}(2d-3),\ deg_{wt}(\hat{f})\leq 2d-4,\\\ g=\hat{g}+O_{wt}(2d-2),\ deg_{wt}(\hat{g})\leq 2d-3.\end{array}\right.$ (4.8) Define $\hat{F}=z+\hat{f}\ ,\ \hat{G}=w+\hat{g},\ \hat{H}=(\hat{F},\hat{G}).$ Write $\Theta=(\hat{F},\hat{G})$ and write $\begin{array}[]{l}\hat{\varphi}(z,\bar{z})=E(z,\bar{z})+\hat{g}(z,u)-2Re\Big{(}\sum\limits_{i=1}^{n}\bar{z_{i}}\hat{f}_{i}(z,u)\Big{)}.\end{array}$ (4.9) Then $\hat{\varphi}(z,\bar{z})-\varphi(z,\bar{z})=O(|z|^{2d-2}).$ Assume that $M^{\prime}=\Theta(M)$ is defined by $w^{\prime}=|z^{\prime}|^{2}+E^{\prime}(z^{\prime},\bar{z^{\prime}})$. Choose $r^{\prime},\sigma,\varrho,r$ to be such that $\frac{1}{2}<r^{\prime}<\sigma<\varrho<r\leq 1,\ \varrho=\frac{1}{3}(2r^{\prime}+r),\ \sigma=\frac{1}{3}(2r^{\prime}+\varrho).$ As in the paper of Moser [Mos], the following lemma will be fundamental for applying the rapid iteration procedure of Moser to prove Theorem 1. Lemma 4.1: Let $M:w=|z|^{2}+E(z,\bar{z})$ be as in Theorem 1. Suppose that $Ord(E)\geq d$. Let $\hat{H}$ and $E^{\prime}$ be defined above. Then $Ord(E^{\prime})\geq 2d-2.$ Proof of Lemma 4.1: Making use of (4.9), we have $\begin{array}[]{l}E^{\prime}(z^{\prime},\overline{z^{\prime}})=\Big{(}\hat{g}(z,\Phi)-\hat{g}(z,u)\Big{)}-2Re\Big{(}\sum\limits_{i=1}^{n}\overline{z_{i}}(\hat{f}_{i}(z,\Phi)-\hat{f}_{i}(z,u))\Big{)}-|\hat{f}(z,\Phi)|^{2}+\hat{\varphi}(z,\bar{z}).\end{array}$ (4.10) Since $Ord(E)\geq d$, by (2.47) and (2.48), we see that $Ord(\hat{f})\geq d-1$ and $Ord(\hat{g})\geq d$. Hence, we have $\begin{array}[]{l}Ord\Big{(}\hat{g}(z,\Phi)-\hat{g}(z,u)\Big{)}\geq\min\\{(d-1)+d,2d-2\\}=2d-2,\\\ Ord\Big{(}\hat{f}_{i}(z,\Phi)-\hat{f}_{i}(z,u)\Big{)}\geq\min\\{(d-2)+d,2d-3\\}=2d-3,\\\ Ord\Big{(}\left|\hat{f}(z,\Phi)\right|^{2}\Big{)}\geq 2(d-1)=2d-2.\end{array}$ Thus $Ord(E^{\prime}-\hat{\varphi})\geq 2d-2$. By Lemma 3.3 and the assumption that $w=|z|^{2}+E$ is formally equivalent to $w=|z|^{2}$, we have $s=\infty$. Hence we have $Ord(\varphi)\geq 2d-2$. The lemma follows. Before proceeding to the estimates of the solution given in (2.47), we need the following lemma: Lemma 4.2: If $E$ is holomorphic in $\overline{D_{r}}$, then we have $\begin{array}[]{l}|E^{(ke_{1})}_{(I,T)}|\leq\frac{(k+2)^{n}\|E\|_{r}}{R^{I+T}\cdot(2r^{2})^{k}}\ ,\ |E^{(ke_{1}+e_{j})}_{(I,T)}|\leq\frac{2^{n}(k+2)^{n}\|E\|_{r}}{R^{I+T}(2r^{2})^{k+1}}.\end{array}$ Proof of Lemma 4.2: We here give the estimates for $|E^{(ke_{1})}_{(0,I)}|$, $|E^{(ke_{1}+e_{j})}_{(0,I)}|$. The others can be done similarly. Suppose that $E=\sum a_{i_{1}\cdots i_{n}j_{1}\cdots j_{n}}z_{1}^{i_{1}}\cdots z_{n}^{i_{n}}\overline{z_{1}}^{j_{l}}\cdots\overline{z_{1}}^{j_{l}}$. Then by (2.3), we have $\begin{array}[]{lll}E_{(0,I)}&=&\sum\limits_{J}a_{j_{1}\cdots j_{n}(i_{1}+j_{1})\cdots(i_{n}+j_{n})}|z_{1}|^{2j_{1}}\cdots|z_{n}|^{2j_{n}}\\\ &=&\sum\limits_{J}a_{J(I+J)}\left(2^{1-n}(u+\sum\limits_{i=2}^{n}2^{n-i}v_{i})\right)^{j_{1}}\Pi_{h=2}^{n}\left(2^{h-n-1}(u+\sum\limits_{i=h+1}^{n}2^{n-i}v_{i}-v_{h})\right)^{j_{h}}\\\ &=&\sum\limits_{J}a_{J(I+J)}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}\Big{(}u^{|J|}+\sum\limits_{k=2}^{n}2^{n-k}\left(\sum\limits_{h=1}^{k-1}j_{h}-j_{k}\right)u^{|J|-1}v_{k}\\\ &&\hskip 10.0pt+O\left(\left|(v_{2},\cdots,v_{n})\right|^{2}\right)\Big{)}.\\\ \end{array}$ Thus we obtain $\begin{array}[]{l}E_{(0,I)}^{(ke_{1})}=\sum\limits_{|J|=k}a_{J(I+J)}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)},\\\ E_{(0,I)}^{(ke_{1}+e_{l})}=\sum\limits_{|J|=k+1}a_{J(I+J)}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}2^{n-l}\left(\sum\limits_{h=1}^{l-1}j_{h}-j_{l}\right).\end{array}$ (4.11) By the Cauchy estimates, we get $\begin{array}[]{lll}|E_{(0,I)}^{ke_{1}}|&=&|\sum\limits_{|J|=k}a_{J(I+J)}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}|\\\ &\leq&\sum\limits_{|J|=k}\frac{\|E\|_{r}}{R^{I+2J}}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}\\\ &=&\sum\limits_{|J|=k}\frac{\|E\|_{r}}{R^{I}}\frac{2^{-\left((n-1)j_{1}+(n-1)j_{2}+\cdots+j_{n}\right)}}{(2^{2-n}r^{2})^{j_{1}}\cdot(2^{2-n}r^{2})^{j_{2}}\cdots(r^{2})^{j_{n}}}\\\ &\leq&\frac{(k+1)^{n}\|E\|_{r}}{R^{I}\cdot(2r^{2})^{k}},\\\ |E_{(0,I)}^{ke_{1}+e_{l}}|&=&|\sum\limits_{|J|=k+1}a_{J(I+J)}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}2^{n-l}\left(\sum\limits_{h=1}^{l-1}j_{h}-j_{l}\right)|\\\ &\leq&\sum\limits_{|J|=k+1}\frac{\|E\|_{r}}{R^{I+2J}}2^{-\left((n-1)j_{1}+\sum\limits_{h=2}^{n}(n-h+1)j_{h}\right)}2^{n-l}|\sum\limits_{h=1}^{l-1}j_{h}-j_{l}|\\\ &\leq&\sum\limits_{|J|=k+1}\frac{\|E\|_{r}}{R^{I}\cdot(2r^{2})^{k+1}}2^{n}(k+1)\\\ &=&\frac{2^{n}(k+2)^{n}\|E\|_{r}}{R^{I}\cdot(2r^{2})^{k+1}}.\end{array}$ Here we have used the fact that $\sharp\left\\{(j_{1},j_{2},\cdots,j_{n})\in\mathbb{Z}^{n}:j_{h}\geq 0\ \mbox{for}\ 1\leq h\leq n,\ j_{1}+j_{2}+\cdots+j_{n}=k\right\\}\leq(k+1)^{n-1}.$ This completes the proof of Lemma 4.2. To carry out the rapid iteration procedure, we need the following estimates of the solution given by (2.47) for the equation (4.7). Proposition 4.3: Suppose that $w=|z|^{2}+E(z,\bar{z})$ is formally equivalent to $M_{\infty}$ with $E$ holomorphic over $\overline{D_{r}}$ and $Ord(E)\geq d$. Then the solution given in (2.47) satisfies the following estimates: $\begin{array}[]{cll}|\hat{f}_{h}|_{\varrho},|\hat{g}|_{\varrho}&\leq&\frac{C(n)(2d)^{2n}\|E\|_{r}}{r-\varrho}(\frac{\varrho}{r})^{d-1},\\\ |\nabla\hat{f}_{h}|_{\varrho},|\nabla\hat{g}|_{\varrho}&\leq&\frac{C(n)(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}},\\\ |\hat{\varphi}|_{\varrho}&\leq&\frac{(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{2n}}(\frac{\varrho}{r})^{2d-2},\end{array}$ (4.12) where $C(n)=3^{3}n(n+1)2^{n+3}$. Proof of Proposition 4.3: Notice that by the definition of $\hat{f}$ given in (4.8), we have $Ord(\hat{f})\geq d-1$ and $deg_{wt}\leq 2d-4$. In terms of (2.47), we can write, for $2\leq h\leq n$, $\hat{f}_{h}=A_{1}+A_{2}+A_{3}+A_{4},$ where $\begin{array}[]{l}A_{1}=\sum\limits_{d\leq 2k+2\leq 2d-3}\frac{1}{2}z_{h}\left(-\sum_{j=2}^{h-1}Re(E_{(0,0)}^{(ke_{1}+e_{j})}u^{k})-2Re(E_{(0,0)}^{(ke_{1}+e_{h})}u^{k})\right),\\\ A_{2}=\sum\limits_{i>h,d\leq 2k+2\leq 2d-3}z_{i}E_{(e_{i},e_{h})}^{(ke_{1})}u^{k},\ A_{3}=\sum\limits_{|J|\geq 1,d\leq|J|+2k+1\leq 2d-3}z^{J}E_{(J,e_{h})}^{(ke_{1})}u^{k},\\\ A_{4}=\sum\limits_{|I|\geq 1,d\leq|I|+2k\leq 2d-3}z^{I+e_{h}}u^{k-1}\overline{E^{(ke_{1})}_{(0,I)}}+\sum\limits_{|I|\geq 1,d\leq|I|+2k+2\leq 2d-3}\sum\limits_{i=0}^{n-h-1}z^{I+e_{h}}u^{k}\overline{E^{(ke_{1}+e_{n-i})}_{(0,I)}}\\\ \hskip 28.45274pt-\sum\limits_{|I|\geq 1,d\leq|I|+2k+2\leq 2d-3}z^{I+e_{h}}u^{k}\overline{E^{(ke_{1}+e_{h})}_{(0,I)}}\\\ \hskip 11.38092pt:=B_{1}+B_{2}+B_{3}.\end{array}$ By Lemma 4.2, we have, for $B_{1}$, the following $\begin{array}[]{lll}\|B_{1}\|_{\varrho}&=&\|\sum\limits_{|I|\geq 1,d\leq|I|+2k\leq 2d-3}z^{I+e_{h}}u^{k-1}\overline{E^{(ke_{1})}_{(0,I)}}\|_{\varrho}\\\ &\leq&\sum\limits_{|I|\geq 1,d\leq|I|+2k\leq 2d-3}(R^{\prime})^{I+e_{h}}(2{\varrho}^{2})^{k-1}\frac{(k+2)^{n}\|E\|_{r}}{R^{I}\cdot(2r^{2})^{k}}\\\ &\leq&\sum\limits_{|I|\geq 1,d\leq|I|+2k\leq 2d-3}\left(\frac{\varrho}{r}\right)^{|I|+2k-1}\frac{(k+2)^{n}\|E\|_{r}}{2r}\\\ &\leq&\sum\limits_{|I|\geq 1,d\leq|I|+2k\leq 2d-3}\left(\frac{\varrho}{r}\right)^{|I|+2k-1}(2d)^{n}\|E\|_{r}\\\ &\leq&\frac{(2d)^{2n}\|E\|_{r}}{r-\varrho}\left(\frac{\varrho}{r}\right)^{d-1}.\end{array}$ Here and in what follows, we write $R^{\prime}=(2^{-\frac{n-2}{2}}\varrho,2^{-\frac{n-2}{2}}\varrho,2^{-\frac{n-3}{2}}\varrho\cdots,2^{-\frac{1}{2}}\varrho,\varrho)$. Wee have also used the fact that $\sharp\\{(i_{1},i_{2},\cdots,i_{n},k)\in\mathbb{Z}^{n+1}:i_{h},k\geq 0\ \mbox{for}\ 1\leq h\leq n,\sum_{h=1}^{n}i_{h}+2k=2d-1\\}\leq(2d)^{n}.$ For $B_{2}$, we have $\begin{array}[]{lll}\|B_{2}\|_{\varrho}&=&\|\sum\limits_{|I|\geq 1,d\leq|I|+2k+2\leq 2d-3}\sum\limits_{i=0}^{n-h-1}z^{I+e_{h}}u^{k}\overline{E^{(ke_{1}+e_{n-i})}_{(0,I)}}\|_{\varrho}\\\ &\leq&\sum\limits_{|I|\geq 1,d\leq|I|+2k+2\leq 2d-3}(R^{\prime})^{I+e_{h}}(2{r^{\prime}}^{2})^{k}\cdot n\frac{2^{n}(k+2)^{n}\|E\|_{r}}{R^{I}\cdot(2r^{2})^{k+1}}\\\ &\leq&\sum\limits_{|I|\geq 1,d\leq|I|+2k+2\leq 2d-3}\left(\frac{\varrho}{r}\right)^{|I|+2k+1}\frac{n2^{n}(k+2)^{n}\|E\|_{r}}{2r}\\\ &\leq&\frac{n2^{n}(2d)^{2n}\|E\|_{r}}{r-\varrho}\left(\frac{\varrho}{r}\right)^{d-1}.\end{array}$ Similarly, we have $\|B_{3}\|_{\varrho}\leq\frac{2^{n}(2d)^{2n}\|E\|_{r}}{r-\varrho}\left(\frac{\varrho}{r}\right)^{d-1}$. Thus we obtain: $\|A_{4}\|_{\varrho}\leq\frac{n2^{n+1}(2d)^{2n}\|E\|_{r}}{r-\varrho}\left(\frac{\varrho}{r}\right)^{d-1}.$ In the same manner, we have $\left\|A_{1}\right\|_{\varrho},\left\|A_{2}\right\|_{\varrho},\left\|A_{3}\right\|_{\varrho}\leq\frac{n\cdot 2^{n+1}(2d)^{2n}\|E\|_{r}}{r-\varrho}(\frac{\varrho}{r})^{d-1},$ Hence we get $|\hat{f}_{h}|_{\varrho}\leq\frac{n\cdot 2^{n+3}(2d)^{2n}\|E\|_{r}}{r-\varrho}(\frac{\varrho}{r})^{d-1}.$ Now letting $\tau=\frac{r+2\varrho}{3}$ and using the Cauchy estimates, we have the following estimate of the derivatives of $F$: $\begin{array}[]{l}|(\hat{f}_{h})^{\prime}_{z_{i}}|_{\varrho}\leq\frac{\tau|\hat{f}_{h}|_{\tau}}{(\tau-\varrho)^{2}}\leq\frac{3^{3}n\cdot 2^{n+3}(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}},\\\ |(\hat{f}_{h})^{\prime}_{w}|_{\varrho}\leq\frac{2\tau^{2}|\hat{f}_{h}|_{\tau}}{(2\tau^{2}-2(\varrho)^{2})^{2}}\leq\frac{3^{3}n\cdot 2^{n+3}(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}}.\end{array}$ (4.13) Here we have used the fact that $(\frac{\tau}{r})^{2}\leq\frac{\varrho}{r}\ \mbox{for}\ \frac{1}{2}<\varrho<\tau<r\leq 1\ ,\ \tau=\frac{r+2\varrho}{3}.$ (4.14) The inequality (4.13) shows that $|\nabla\hat{f}_{h}|_{\varrho}\leq\frac{3^{3}n(n+1)\cdot 2^{n+3}(2d)^{2n}\|E\|_{r}}{(r-r^{\prime})^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}}$. The corresponding estimates on $\hat{f}_{1}$ and $\hat{g}$ can be achieved similarly. We next estimate $\hat{\varphi}$. Notice that $-\hat{g}(z,u)+2Re\left(\sum_{h=1}^{n}\bar{z_{i}}\hat{f}_{i}(z,u)\right)$ is only used to cancel terms of with weight $<2d-2$ in $E$. By (4.9), we have the following: $\begin{array}[]{l}\|\hat{\varphi}\|_{\varrho}=\|\sum\limits_{t\geq 2d-2}E^{(t)}\|_{\varrho}\\\ =\|\sum\limits_{|I|+|J|\geq 2d-2}a_{i_{1}\cdots i_{n}j_{1}\cdots j_{n}}z_{1}^{i_{1}}\cdots z_{n}^{i_{n}}{\overline{z_{1}}^{j_{1}}}\cdots{\overline{z_{n}}^{j_{n}}}\|_{\varrho}\\\ \leq\sum\limits_{|I|+|J|\geq 2d-2}\|E\|_{r}(\frac{R^{\prime}}{R})^{I+J}\\\ \leq\sum\limits_{|I|+|J|=2d-2,|K|,|L|\geq 0}\|E\|_{r}(\frac{\varrho}{r})^{|I|+|J|}\cdot(\frac{\varrho}{r})^{k_{1}}\cdots(\frac{\varrho}{r})^{k_{n}}\cdot(\frac{\varrho}{r})^{l_{1}}\cdots(\frac{\varrho}{r})^{l_{n}}\\\ \leq\sum\limits_{|I|+|J|=2d-2}\|E\|_{r}(\frac{\varrho}{r})^{2d-2}\cdot(\frac{1}{1-\frac{\varrho}{r}})^{2n}\\\ \leq\frac{(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{2n}}(\frac{\varrho}{r})^{2d-2}.\end{array}$ Here we have used the fact that $\sharp\\{(i_{1},\cdots,i_{n},j_{1},\cdots,j_{n})\in\mathbb{Z}^{2n}:i_{h},j_{h}\geq 0\ \mbox{for}\ 1\leq h\leq n,\sum_{h=1}^{n}(i_{h}+j_{h})=k\\}\leq(k+1)^{2n}.$ This finishes the proof of Proposition 4.3. Proposition 4.4: Let $E,r,\varrho,C(n)$ be as in Proposition 4.3. Then there exists a constant $\delta>0$ such that for $\frac{C(n)(2d)^{2n}\|E\|_{r}}{(r-\varrho)^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}}<\delta,$ (4.15) $\Psi(z^{\prime},w^{\prime}):=\Theta^{-1}(z^{\prime},w^{\prime})$ is well defined in $\overline{\triangle_{\sigma}}$. Moreover, it holds that $\Psi(\triangle_{r^{\prime}})\subset\triangle_{\sigma}$, $\Psi(\triangle_{\sigma})\subset\triangle_{\varrho}$, $E^{\prime}(z,\xi)$ is holomorphic in $\overline{\triangle_{\sigma}}$ and $\|E^{\prime}\|_{r^{\prime}}\leq C_{d}\|E\|_{r}^{2}+\widetilde{C_{d}}\|E\|_{r}.$ (4.16) Here $\begin{array}[]{l}C_{d}=\frac{(2n+1)\cdot 3^{3}C(n)(2d)^{2n}}{(r-r^{\prime})^{3}}(\frac{r^{\prime}}{r})^{\frac{d-1}{4}}+(\frac{r^{\prime}}{r})^{d-1}n\cdot\left(\frac{3C(n)(2d)^{2n}}{r-r^{\prime}}\right)^{2},\ \ \widetilde{C_{d}}=\frac{3^{2n}\cdot(2d)^{2n}}{(r-r^{\prime})^{2n}}(\frac{r^{\prime}}{r})^{d-1}.\end{array}$ Proof of Proposition 4.4: We need to show that for each $(z^{\prime},w^{\prime})\in\overline{\triangle_{\sigma}}$, we can uniquely solve the system: $\left\\{\begin{array}[]{l}z^{\prime}=z+\hat{f}(z,w)\\\ w^{\prime}=w+\hat{g}(z,w)\end{array}\right.$ with $(z,w)\in\triangle_{\varrho}$. By (4.12), choosing $\delta$ sufficiently small such that $|\nabla\hat{f}|_{\varrho}+|\nabla\hat{g}|_{\varrho}<\frac{1}{2n+4}$ and $|\hat{f}|_{\varrho}+|\hat{g}|_{\varrho}<\frac{1}{2n+4}.(r-\varrho)$ . Define $(z^{[1]},w^{[1]})=(z^{\prime},w^{\prime})\in\triangle_{\sigma}$ and $(z^{[j]},w^{[j]})$ inductively by $\left\\{\begin{array}[]{l}z^{[j+1]}=z^{\prime}-\hat{f}(z^{[j]},w^{[j]})\\\ w^{[j+1]}=w^{\prime}-\hat{g}(z^{[j]},w^{[j]}).\end{array}\right.$ By a standard argument on the Picard iteration procedure, we can get a unique $(z,w)\in\triangle_{\varrho}$ satisfying $\Psi^{-1}(z,w)=(z^{\prime},w^{\prime})$, which gives that $\Psi(\triangle_{\sigma})\subset\triangle_{\varrho}$. Similarly, we have $\Psi(\triangle_{r^{\prime}})\subset\triangle_{\sigma}$. Hence we conclude that $E^{\prime}$ is holomorphic in $\triangle_{\sigma}$. Moreover, $\|E^{\prime}\|_{r^{\prime}}\leq\|Q\|_{\sigma}$ (4.17) where $\begin{array}[]{l}Q=\left(\hat{g}(z,\Phi)-\hat{g}(z,u)\right)-2Re\left(\sum\limits_{i=1}^{n}\overline{z_{i}}\Large(\hat{f}_{i}(z,\Phi)-\hat{f}_{i}(z,u)\Large)\right)-|\hat{f}(z,\Phi)|^{2}+\hat{\varphi}(z,\bar{z}).\end{array}$ (4.18) Notice that $\begin{array}[]{lll}|(\hat{g}(z,\Phi)-\hat{g}(z,u)|_{\sigma}&\leq&|\nabla\hat{g}|_{\varrho}\cdot\|E\|_{r}\leq\frac{C(n)(2d)^{2n}\|E\|_{r}^{2}}{(r-\varrho)^{3}}(\frac{\varrho}{r})^{\frac{d-1}{2}}\\\ &\leq&\frac{3^{3}C(n)(2d)^{2n}\|E\|_{r}^{2}}{(r-r^{\prime})^{3}}(\frac{r^{\prime}}{r})^{\frac{d-1}{4}}.\end{array}$ (4.19) Here we have used the fact that $(\frac{\varrho}{r})^{2}<\frac{r^{\prime}}{r}$. (This can be achieved by the same token as for (4.14).) We also have $\begin{array}[]{l}|(\hat{f}_{i}(z,\Phi)-\hat{f}_{i}(z,u)|_{\sigma}\leq\frac{3^{3}C(n)(2d)^{2n}\|E\|_{r}^{2}}{(r-r^{\prime})^{3}}(\frac{r^{\prime}}{r})^{\frac{d-1}{4}}\ \rm{for}\ 1\leq i\leq n.\\\ |\hat{f}(z,\Phi)|^{2}_{\sigma}\leq n\cdot\left(\frac{C(n){(2d)}^{2n}\|E\|_{r}}{r-\sigma}(\frac{\sigma}{r})^{d-1}\right)^{2}\leq n\cdot\left(\frac{3C(n){(2d)}^{2n}\|E\|_{r}}{r-r^{\prime}}\right)^{2}(\frac{r^{\prime}}{r})^{d-1}\\\ \|\hat{\varphi}\|_{\sigma}\leq\frac{(2d)^{2n}\|E\|_{r}}{(r-\sigma)^{2n}}(\frac{\sigma}{r})^{2d-2}\leq\frac{3^{2n}(2d)^{2n}\|E\|_{r}}{(r-r^{\prime})^{2n}}(\frac{r^{\prime}}{r})^{d-1}.\end{array}$ (4.20) By (4.18)-(4.20), we obtain: $\begin{array}[]{l}\|E^{\prime}\|_{r^{\prime}}\leq\left\\{\frac{(2n+1)\cdot 3^{3}C(n)(2d)^{2n}}{(r-r^{\prime})^{3}}(\frac{r^{\prime}}{r})^{\frac{d-1}{4}}+(\frac{r^{\prime}}{r})^{d-1}n\cdot\left(\frac{3C(n)(2d)^{2n}}{r-r^{\prime}}\right)^{2}\right\\}\|E\|_{r}^{2}+\frac{3^{2n}\cdot(2d)^{2n}}{(r-r^{\prime})^{2n}}(\frac{r^{\prime}}{r})^{d-1}\|E\|_{r}\end{array}$ This completes the proof of Proposition 4.4. Now we turn to the proof of Theorem 1. Set $r_{\upsilon},\varrho_{\upsilon},\sigma_{\upsilon}$ as follows: $r_{\upsilon}=\frac{1}{2}\left(1+\frac{1}{\upsilon+1}\right)\ ,\ \varrho_{\upsilon}=\frac{1}{3}(2r_{\upsilon}+r_{\upsilon+1})\ ,\ \sigma_{\upsilon}=\frac{1}{3}(2r_{\upsilon}+\varrho_{\upsilon}).$ We will apply the previous estimates with $r=r_{\upsilon},\varrho=\varrho_{\upsilon},\sigma=\sigma_{\upsilon},r^{\prime}=r_{\upsilon+1},\Psi=\Psi_{v},\cdots,$ with $v=0,1,\cdots,.$ Then we have the following (see [(4.5), Moser]): $(r_{\upsilon}-r_{\upsilon+1})^{-1}=2(\upsilon+1)(\upsilon+2),\ \frac{r_{\upsilon+1}}{r_{\upsilon}}=1-\frac{1}{(\upsilon+2)^{2}}$ (4.21) Define a sequence of real analytic submanifolds $M_{k}\ :\ w=|z|^{2}+E_{k}(z,\bar{z})$ by $M_{0}=M$, $M_{\upsilon+1}=\Psi^{-1}_{\upsilon}(M_{\upsilon})$ for all $\upsilon=0,1,2,\cdots$, where $\Psi_{\upsilon}$ is the biholomorphic mapping taking $\triangle_{\sigma_{\upsilon}}$ into $\triangle_{\varrho_{\upsilon}}$. And let $d_{\upsilon}=Ord(E_{\upsilon})\ ,\ \Phi_{\upsilon}=\Psi_{0}\circ\Psi_{1}\circ\cdots\circ\Psi_{\upsilon}.$ Since $s=\infty$, we find that $Ord(E_{\upsilon})=d_{\upsilon}\geq 2^{\upsilon}+2\ \mbox{for}\ \upsilon\geq 0.$ We next state the following elementary fact: Lemma 4.5: Suppose that there is a constant C and number $a>1$ such that $d_{v}\geq Ca^{v}$. Then for any integer $m_{1},m_{2},m_{3}>0$, $\lim_{v\rightarrow\infty}v^{m_{3}}d_{v}^{m_{1}}(1-\frac{1}{v^{m_{2}}})^{d_{v}}=0.$ Then one can prove, by using (4.21) and Lemma 4.5, that $\lim_{\upsilon\rightarrow\infty}C_{d_{\upsilon}}=0\ ,\ \lim_{\upsilon\rightarrow\infty}\widetilde{C_{d_{\upsilon}}}=0.$ Hence $C_{d_{\upsilon}}$ and $\widetilde{C_{d_{\upsilon}}}$ are bounded. Set $C_{d_{\upsilon}},\widetilde{C_{d_{\upsilon}}}<C$, where $C$ is a fixed positive constant. Also, one can verify that the hypothesis in (4.15) holds for all $\upsilon\geq 0$, by choosing $\eta^{*}_{0}=\|E_{0}\|_{r_{0}}$ sufficiently small. Indeed, we can even have $\|E_{\upsilon}\|_{r_{\upsilon}}\leq\epsilon 2^{-\upsilon}$ for all $\upsilon\geq 0$ and any given $1>\epsilon>0$. Choose $N$ large enough such that $C_{d_{\upsilon}},\widetilde{C_{d_{\upsilon}}}\leq\frac{1}{4}$ when $\upsilon\geq N$. Suppose $C>1$ and choose $E_{0}$ such that $\eta^{*}_{0}=\epsilon(2C)^{-2N}<1$. Then we have the following (I) When $\upsilon\leq N$, we have $\begin{array}[]{l}\|E_{\upsilon}\|_{r_{\upsilon}}\leq C(\|E_{\upsilon-1}\|_{r_{\upsilon-1}}+1)\|E_{\upsilon-1}\|_{r_{\upsilon-1}}\leq 2C\cdot\|E_{\upsilon-1}\|_{r_{\upsilon-1}}\leq(2C)^{\upsilon}\|E_{0}\|_{r_{0}}\leq\epsilon(2C)^{\upsilon-2N}\leq\epsilon 2^{-N}.\end{array}$ (II) When $\upsilon>N$, we have $\begin{array}[]{l}\|E_{\upsilon}\|_{r_{\upsilon}}\leq\frac{1}{4}\cdot 2\cdot\|E_{\upsilon-1}\|_{r_{\upsilon-1}}\leq(\frac{1}{2})^{\upsilon-N}\|E_{N}\|_{r_{N}}\leq\epsilon 2^{-\upsilon}.\end{array}$ Now, choose $\epsilon$ sufficiently small. Then it follows from (4.12) and Proposition 4.4 that $\|d\Psi_{\upsilon}^{-1}\|_{\triangle_{\varrho_{\upsilon}}}\leq 1+C_{0}\|E_{\upsilon}\|_{r_{\upsilon}}\leq 1+C_{0}\epsilon 2^{-\upsilon}$ for some constant $C_{0}$. Notice that $\Psi_{\upsilon}$ maps $\triangle_{\sigma_{\upsilon}}$ into $\triangle_{\varrho_{\upsilon}}$. By Cramer’s rule, we have $\|d\Psi_{\upsilon}\|_{\triangle_{\sigma_{\upsilon}}}\leq 1+\epsilon C_{1}2^{-\upsilon}$ for some constant $C_{1}$. Now the convergence of $\Phi_{\upsilon}$ in $\triangle_{\frac{1}{2}}$ follows from the fact that $\Pi_{\upsilon=0}^{\infty}\|d\Psi_{\upsilon}\|_{\triangle_{\sigma_{\upsilon}}}\leq\Pi_{\upsilon=0}^{\infty}(1+\epsilon C_{1}2^{-\upsilon})<\infty,$ which completes the proof of Theorem 1. Remark 4.6: We notice that the formal map in Theorem 1 sending $(M,0)$ to its quadric $(M_{\infty},0)$ may not be convergent as $aut_{0}(M_{\infty})$ contains many non-convergent elements. This is quite different from the setting for CR manifolds, where formal maps are always convergent under certain not too degenerate assumptions. We refer the reader to the survey article [BER1] for discussions and references on this matter. ## References * [BER1] S. Baouendi, P. Ebenfelt and L. Rothschild, Local geometric properties of real submanifolds in complex space, Bull. Amer. Math. Soc. (N.S.) 37 (2000), no. 3, 309–33. * [BER2] S. Baouendi, P. Ebenfelt and L. Rothschild, Real Submanifolds in Complex Space and Their Mappings, Princeton Mathematical Series, 47, Princeton University Press, Princeton, NJ, 1999\. * [BG] E. Bedford and B. Gaveau, Envelopes of holomorphy of certain 2-spheres in ${\mathbf{C}}^{2}$, Amer. J. Math. (105), 975-1009, 1983\. * [Bis] E. Bishop, Differentiable manifolds in complex Euclidean space, Duke Math. J. (32), 1-21, 1965. * [CM] S. S. Chern and J. K. Moser, Real hypersurfaces in complex manifolds, Acta Math. 133, 219-271(1974). * [Cof1] A. Coffman, Unfolding CR singularities, 2006, preprint. (to appear in Memoirs of the AMS.) * [Cof2] A. Coffman, Analytic stability of the CR cross-cap, Pacif. Jour. of Math. (2) 226 (2006), 221-258. * [DTZ] P. Dolbeault, G. Tomassini and D. Zaitsev, On Levi-flat hypersyrfaces with prescribed boundary, preprint. (Announcement of the paper appeared at C.R. Acd. Sci. Paris, Ser. I 341 (2005), 343-348.) * [Gon1] X. Gong, On the convergence of normalilations of real analytic surfaces near hyperbolic complex tangents, Comment. Math. Helv. 69 (1994), no. 4, 549–574. * [Gon2] X. Gong, Normal forms of real surfaces under unimodular transformations near elliptic complex tangents, Duke Math. J. 74 (1994), no. 1, 145–157. * [Hu1] X. Huang, Local Equivalence Problems for Real Submanifolds in Complex Spaces, Lecture Notes in Mathematics 1848 (C.I.M.E. series), Springer-Verlag, pp 109-161, Berlin-Heidelberg-New York, 2004\. * [HK] X. Huang and S. Krantz, On a problem of Moser, Duke Math. J. (78), 213-228, 1995. * [HY] X. Huang and W. Yin, A Bishop surface with a vanishing Bishop invariant, preprint, March 2007. (arXiv:0704.2040 ) * [Mir] N. Mir, Convergence of formal embeddings between real-analytic hypersurfaces in codimension one, J. Differential Geom. 62 (2002), 163–173. * [Mos] J. Moser, Analytic surfaces in ${{\mathbb{C}}}^{2}$ and their local hull of holomorphy, Annales Aca -demiæFennicae Series A.I. Mathematica (10), 397-410, 1985. * [MW] J. Moser and S. Webster, Normal forms for real surfaces in ${{\mathbb{C}}}^{2}$ near complex tangents and hyperbolic surface transformations, Acta Math. (150), 255-296, 1983. * [Rud] W. Rudin, Function theory in the unit ball of ${{\mathbb{C}}}^{n}$, New York, Springer-Verlag, 1980. * [Sto] L. Stolovitch, Family of intersecting totally real manifolds of $({{\mathbb{C}}}^{n},0)$ and CR-singularities, preprint. (arXiv:math/0506052). Xiaojun Huang (huangx@math.rutgers.edu), Department of Mathematics, Rutgers University at New Brunswick, NJ 08903, USA; Wanke Yin, School of Mathematical Sciences, Wuhan University, Wuhan 430072, P. R. China.

arxiv-papers

2024-09-04T02:48:54.071491

0803.0102

# Generalization of distance to higher dimensional objects Steven S. Plotkin†***e-mail: steve@physics.ubc.ca † Department of Physics and Astronomy, University of British Columbia, 6224 Agricultural Road, Vancouver, BC V6T1Z1, Canada ###### Abstract The measurement of distance between two objects is generalized to the case where the objects are no longer points but are one-dimensional. Additional concepts such as non-extensibility, curvature constraints, and non-crossing become central to the notion of distance. Analytical and numerical results are given for some specific examples, and applications to biopolymers are discussed. ## I Introduction The distance, as conventionally defined between two zero-dimensional objects (points) $A$ and $B$ at positions ${\bf r}_{\mbox{\tiny{A}}}$ and ${\bf r}_{\mbox{\tiny{B}}}$, is the minimal arclength travelled in the transformation from $A$ to $B$. A transformation ${\bf r}(t)$ between $A$ and $B$ is a vector function which may be parametrized by a scalar variable $t$: $0\leq t\leq T$, ${\bf r}(0)={\bf r}_{\mbox{\tiny{A}}}$, ${\bf r}(T)={\bf r}_{\mbox{\tiny{B}}}$, and the distance travelled is a functional of ${\bf r}(t)$. The (minimal) transformation ${\bf r}^{\ast}(t)$ is an object of dimension one higher than $A$ or $B$, i.e. it yields a distance that is one- dimensional. The distance $\mathcal{D}^{\ast}$ is found through the variation of the functional GelfandIM00 : $\displaystyle\mathcal{D}^{\ast}$ $\displaystyle=$ $\displaystyle\mathcal{D}\left[{\bf r}^{\ast}(t)\right]\mbox{where ${\bf r}^{\ast}(t)$ satisfies}$ (1b) $\displaystyle\delta\int_{0}^{T}\\!\\!\\!dt\>\left(g_{\mu\nu}\dot{x}^{\mu}(t)\dot{x}^{\nu}(t)\right)^{1/2}=0\>.$ or $\displaystyle\delta\int_{0}^{T}\\!\\!\\!dt\>\sqrt{\dot{{\bf r}}^{2}}=0\;\;\;\;\mbox{(Euclidean metric)}$ (1c) Here $\dot{x}=dx/dt$, and $\dot{{\bf r}}=d{\bf r}/dt$. The boundary conditions mentioned above are present at the end points of the integral. The Einstein summation convention will be used where convenient, e.g. eq. (1b), however all the analysis here deals with spatial coordinates, $\nu=1,2,3$ on a Euclidean metric. Generalizations to dimension higher than $3$, as well as non-Euclidean metrics, are straightforward to incorporate into the formalism. On a Euclidean metric, $g_{\mu\nu}=\delta_{\mu\nu}$ and the minimal distance becomes the diagonal of a hypercube. However, formulated as above, the solutions minimizing $\mathcal{D}$ are infinitely degenerate, because particles moving at various speeds but tracing the same trajectory over the total time $T$ all give the same distance. To circumvent this problem what is typically done is to let one of the space variables (e.g $x$) become the independent variable. However for higher dimensional objects, or zero dimensional objects on a manifold with nontrivial topology, there is no guarantee that the dependent variables ($y$, $z$) constitute single valued functions of $x$. Alternatively, one can study the ’time’ trajectory of the parametric curve defined above, but under a gauge that fixes the speed to a constant $v_{o}$, for example. One can either fix the gauge from the outset with Lagrange multipliers, or choose a gauge that may simplify the problem after finding the extremum equations. The latter is often simpler in practice. To be specific, the effective Lagrangian $\mathcal{L}$ appearing in the above problem is $\sqrt{\dot{{\bf r}}^{2}}$, and the Euler-Lagrange (EL) equations are $\frac{d}{dt}\left(\frac{\partial\mathcal{L}}{\partial\dot{{\bf r}}}\right)=0\;\;\;\;\;\mbox{or}\;\;\;\;\;\frac{d}{dt}\left(\frac{\dot{{\bf r}}}{\left|\dot{{\bf r}}\right|}\right)=\dot{\hat{{\bf v}}}=0$ (2) with $\hat{{\bf v}}$ the unit vector in the direction of the velocity. The boundary conditions are ${\bf r}^{\ast}(0)={\bf r}_{\mbox{\tiny{A}}}\;\;\;\mbox{and}\;\;\;{\bf r}^{\ast}(T)={\bf r}_{\mbox{\tiny{B}}}\;.$ (3) Since the derivative of a unit vector is always orthogonal to that vector, equation (2) says that the direction of the velocity cannot change, and therefore straight line motion results. Applying the boundary conditions gives $\hat{{\bf v}}=({\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}})/\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$. However, any function ${\bf v}(t)=\left|v_{o}(t)\right|\hat{{\bf v}}$ satisfying the boundary conditions is a solution, so long as $\int_{0}^{T}\\!\\!dt\>\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$. This is the infinite degeneracy of solutions mentioned above. Then ${\bf r}^{\ast}(t)={\bf r}_{\mbox{\tiny{A}}}+\frac{{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}}{\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|}\int_{0}^{t}\\!\\!dt\>\left|v_{o}(t)\right|$, and ${\cal D}^{\ast}=\int_{0}^{T}\\!\\!dt\>\sqrt{\dot{{\bf r}}^{\ast^{2}}}=\int_{0}^{T}\\!\\!dt\>\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|$. At this point we could fix the parameterization by choosing $\left|v_{o}(t)\right|=\left|{\bf r}_{\mbox{\tiny{B}}}-{\bf r}_{\mbox{\tiny{A}}}\right|/T$ (constant speed), for example. The extremum is a minimum, as can be shown by analyzing the eigenvalues of the matrix $\partial^{2}\mathcal{D}/\partial x_{\nu}(t)\partial x_{\mu}(t^{\prime})=-\delta_{\mu\nu}\delta^{\prime\prime}(t-t^{\prime})$. Diagonalizing by Fourier transform gives positive elements $+\omega_{n}^{2}\,\delta_{\mu\nu}\delta(\omega_{n}-\omega_{n}^{\prime})$ for the stability matrix and thus positive eigenvalues. In what follows we generalize the notion of distance to higher dimensional objects, specifically space-curves. We will see many of the above themes reiterated, as well as some fundamentally new features that emerge when one treats the space curves as non-extensible, having some persistence length or curvature constraint, and non-crossing or unable to pass through themselves. We provide analytical and numerical results for some prototypical examples for non-extensible chains, and we lay the foundations for treating curvature and non-crossing constraints. ## II Distance metric for one dimensional objects The distance ${\cal D}^{\ast}$ between two one-dimensional objects (which we refer to as space curves or strings) $A$ and $B$ having configurations ${\bf r}_{\mbox{\tiny A}}(s)$ and ${\bf r}_{\mbox{\tiny B}}(s)$, $0\leq s\leq L$, is obtained from the transformation from $A$ to $B$ that minimizes the integrated distance travelled. By integrated distance we mean the cumulative arclength all elements of the string had to move in the transformation from $A$ to $B$. For the transformation to exist, strings $A$ and $B$ must have the same length (although this condition may be relaxed by allowing specific extensions or contractions). For the distance to be finite, open space curves must be finite in length. For closed non-crossing space curves, $A$ and $B$ must be in the same topological class for the transformation to exist. Describing the transformation ${\bf r}(t,s)$ requires two scalar parameters, one for arc length $s$ along the string and another measuring progress as in the zero- dimensional case, say $t$: $0\leq t\leq T$, so that ${\bf r}(s,0)={\bf r}_{\mbox{\tiny A}}(s)$ and ${\bf r}(s,T)={\bf r}_{\mbox{\tiny B}}(s)$. The distance travelled is a functional of the vector function ${\bf r}(s,t)$. The minimal transformation ${\bf r}^{\ast}(t,s)$ is an object of dimension one higher than $A$ or $B$, i.e. it yields a distance that is two-dimensional. The problem does not map to a simple soap film, since there are many configuration pairs that have zero area between them but nonzero distance travelled, e.g. a straight line displaced along its own axis, or that in figure 1C. The analogue to a higher-dimensional surface of minimal area when the ’time’ $t$ is included is closer but inexact (see footnote below). We can construct the effective Lagrangian along the same lines as the zero- dimensional case. Using the shorthand ${\bf r}\equiv{\bf r}(s,t)$, $\dot{{\bf r}}\equiv\partial{\bf r}/\partial t$, ${\bf r}^{\prime}\equiv\partial{\bf r}/\partial s$, the distance travelled is†††The distance-metric action in eq. (4) bears a strong resemblance to the Nambu-Goto action for a classical relativistic string ZwiebachB04 : $S_{\mbox{{\tiny NG}}}[{\bf r}(s,t)]=\int\\!\\!d\sigma\,d\tau\>\sqrt{(\dot{{\bf r}}\cdot{\bf r}^{\prime})^{2}-(\dot{{\bf r}})^{2}({\bf r}^{\prime})^{2}}$, where ${\bf r}$ in $S_{\mbox{{\tiny NG}}}$ is now a four-vector and the dot product is the relativistic dot product. This action is physically interpreted as the (Lorentz Invariant) world-sheet area of the string. If eq. (4) could be mapped by suitable choice of gauge to the minimization of the Nambu-Goto action, one could exploit here the same reparameterization invariance that results in wave equation solutions to the equations of motion for the classical relativistic string, by choosing a parameterization such that $\dot{{\bf r}}\cdot{\bf r}^{\prime}=0$ (for the purely geometrical problem, the discriminant under the square root in the action has opposite sign). Unfortunately however, because the velocity in the distance-metric action is a $3$-velocity rather than a $4$-velocity, our action only accumulates area when parts of the string move in $3$-space, in contrast to the Nambu-Goto action which accumulates area even for a static string. The distance-metric action eq. (4) has a lower symmetry than that for the classical relativistic string. ${\cal D}^{\ast}$ cannot depend on the time the transformation took, while the world sheet area does. Conversely, if we take e.g. configuration $A$ at $t=0$ to be a straight line of length $L$, and configurations $B$ at $t=T$ to be the same straight line but displaced along its own axis by varying amounts $d$, the geometrical area for all transformations would be $LT$, while the distances ${\cal D}^{\ast}_{\mbox{{\tiny AB}}}$ for each transformation would be $Ld$. $\mathcal{D}=\int_{0}^{L}\\!\\!\\!ds\\!\\!\int_{0}^{T}\\!\\!\\!dt\>\sqrt{\dot{x}^{\nu}\dot{x}_{\nu}}=\int_{0}^{L}\\!\\!\\!ds\\!\\!\int_{0}^{T}\\!\\!\\!dt\>\sqrt{\dot{{\bf r}}^{2}}\>.$ (4) However to meaningfully represent the distance a string must move to reconfigure itself from conformation $A$ to $B$, the transformation must be subject to several auxiliary conditions. The first of these is non-extensibility. Points along the space curve cannot move independently of one another but are constrained to integrate to fixed length, so the curve cannot stretch or contract. Thus there is a Lagrange multiplier $\lambda(s,t)$ weighting the (non-holonomic) constraint: $\sqrt{{\bf r}^{\prime 2}}=1\>.$ (5) This constraint ensures a parameterization of the string with unit tangent vector $\hat{{\bf t}}={\bf r}^{\prime}$, so that the total length of the string is $L=\int_{0}^{L}\\!\\!ds\>\sqrt{{\bf r}^{\prime 2}}=\int_{0}^{L}\\!\\!ds$. In the language of differential geometry, the space curve is a unit-speed curve. If the constraint (5) were not present in eq. (4), each point along the space- curve could follow a straight line path from $A$ to $B$ and the problem of minimizing the distance would be trivial. Equivalently, setting $\lambda=0$ should reduce the problem to a sum of straight lines analogously to the zero- dimensional case above. As in the case of distance between points, one can fix the $t$-parameterization from the outset by introducing a Lagrange multiplier $\alpha(t)$ that fixes the total distance covered per time $\int_{0}^{L}\\!\\!ds\sqrt{\dot{{\bf r}}^{2}}$ to a known function $f(t)$. While this approach removes the infinite degeneracy mentioned above, as a global isoperimetric condition it reduces the symmetry of the problem. For example there would then be no conservation law that could be written to capture the invariance of the effective Lagrangian with respect to the independent variable $t$. For these reasons we choose to leave the answer as unparamaterized with respect to $t$, analogous to the point-distance case above. ### II.1 Ideal chains There are many examples of nontrivial transformations between two strings $A$ and $B$ where chain non-crossing is unimportant (c.f. figures 1A and 1B). Here we derive the Euler-Lagrange equations for this case. From equations (4)-(5), the extrema of the distance ${\cal D}$ are found from $\displaystyle\delta{\cal D}=$ $\displaystyle\delta$ $\displaystyle\\!\\!\\!\int_{0}^{L}\\!\\!\int_{0}^{T}\\!\\!\\!ds\,dt\>\mathcal{L}\left(\dot{{\bf r}},{\bf r}^{\prime}\right)=0$ or $\displaystyle\delta$ $\displaystyle\\!\\!\\!\int_{0}^{L}\\!\\!\int_{0}^{T}\\!\\!\\!ds\,dt\>\left(\sqrt{\dot{{\bf r}}^{2}}-\lambda\sqrt{{\bf r}^{\prime 2}}\right)=0$ (6) Performing the variation gives $\displaystyle\delta{\cal D}$ $\displaystyle=$ $\displaystyle\int_{0}^{L}\\!\\!\\!ds\>\left[{\bf p}_{t}\cdot\delta{\bf r}\right]_{0}^{T}+\int_{0}^{T}\\!\\!\\!dt\>\left[{\bf p}_{s}\cdot\delta{\bf r}\right]_{0}^{L}$ (7) $\displaystyle-$ $\displaystyle\int_{0}^{L}\\!\\!\int_{0}^{T}\\!\\!\\!ds\,dt\>\delta{\bf r}\cdot\left[\frac{d{\bf p}_{s}}{ds}+\frac{d{\bf p}_{t}}{dt}\right]=0$ where the generalized momenta ${\bf p}_{t}$ and ${\bf p}_{s}$ are given by: ${\bf p}_{t}=\frac{\partial\mathcal{L}}{\partial\dot{{\bf r}}}=\hat{{\bf v}}\;\;\;\mbox{and}\;\;\;{\bf p}_{s}=\frac{\partial\mathcal{L}}{\partial{\bf r}^{\prime}}=-\lambda\hat{{\bf t}}$ (8) where $\hat{{\bf v}}$ is again the unit velocity vector, and $\hat{{\bf t}}$ is the unit tangent to the curve. The EL equation follows from the last term in (7), and yields a partial differential equation for the minimal transformation ${\bf r}^{\ast}(s,t)$: $\left(\dot{{\bf r}}^{2}\right)\ddot{{\bf r}}-\left(\dot{{\bf r}}\cdot\ddot{{\bf r}}\right)\dot{{\bf r}}=\left|\dot{{\bf r}}\right|^{3}\left(\lambda{\bf r}^{\prime\prime}+\lambda^{\prime}{\bf r}^{\prime}\right)$ (9) where we have used the facts that $\left|{\bf r}^{\prime}\right|=1$ and ${\bf r}^{\prime}\cdot{\bf r}^{\prime\prime}\equiv\hat{{\bf t}}\cdot\boldsymbol{\kappa}=0$, since the tangent is always orthogonal to the curvature at any given point along a space curve. Equation (9) can be written in terms easier to understand intuitively by using the unit velocity vector $\hat{{\bf v}}$, tangent $\hat{{\bf t}}$, and curvature $\boldsymbol{\kappa}$:‡‡‡The invariance of the Lagrangian to $(s,t)$ leads to conservation laws by Noether’s theorem GelfandIM00 , which here take the form of divergence conditions. However these generally contain no new information beyond the EL equations, and can be obtained by dotting eq. (10) with either ${\bf r}^{\prime}$ to give $\lambda^{\prime}=\dot{\hat{{\bf v}}}\cdot\hat{{\bf t}}$, or $\dot{{\bf r}}$ to give ${\bf v}\cdot(\lambda\hat{{\bf t}})^{\prime}=0$. $\dot{\hat{{\bf v}}}=\lambda\boldsymbol{\kappa}+\lambda^{\prime}\hat{{\bf t}}\>.$ (10) Comparison of equations (10) and (2) illustrates the point made earlier that setting the Lagrange multiplier $\lambda$ corresponding to the non- extensibility condition to zero results in straight line solutions for all points along the space curve. Conversely the condition that the space curve form a contiguous object results generally in nonzero deviation from straight line motion. So in comparing various extremal solutions to eq. (10), the minimal solution will minimize $\left|\lambda\right|$ everywhere. The boundary conditions are obtained from the first two terms in (7). Since the initial and final configurations are specified, the variation $\delta{\bf r}$ vanishes at $t=0,T$, and the corresponding boundary conditions, or initial and final conditions, are: ${\bf r}^{\ast}(s,0)={\bf r}_{\mbox{\tiny{A}}}(s)\;\;\;\;\mbox{and}\;\;\;\;{\bf r}^{\ast}(s,T)={\bf r}_{\mbox{\tiny{B}}}(s)\>.$ (11) Since the end points of the string are free during the transformation, $\delta{\bf r}\neq 0$ at $s=0,L$, and so the conjugate momenta must vanish: ${\bf p}_{s}(0,t)={\bf p}_{s}(L,t)=0$. This means that $\lambda\hat{{\bf t}}=0$ at the end points. However since $\hat{{\bf t}}$ cannot be zero, the only way this can occur is for $\lambda(0,t)=\lambda(L,t)=0$. The Lagrange multiplier, which represents the conjugate force or tension to ensure an inextensible chain, must vanish at the end points of the string. If $\lambda=0$, the EL equation (10) gives $\dot{\hat{{\bf v}}}=\lambda^{\prime}\hat{{\bf t}}$ at the end points. However since $\hat{{\bf v}}$ is a unit vector, $\dot{\hat{{\bf v}}}$ is orthogonal to $\hat{{\bf v}}$ (or ${\bf v}$), and we have finally the boundary conditions at the end points of the string: $\lambda^{\prime}{\bf v}\cdot\hat{{\bf t}}=0\;\;\;\;\;\mbox{(at the end points).}$ (12) Equation (12) has three possible solutions. One is that ${\bf v}\cdot\hat{{\bf t}}=0$ or equivalently $\dot{{\bf r}}\cdot{\bf r}^{\prime}=0$, which corresponds to pure rotation of the end points. It is worth mentioning that the end points of the classical relativistic string also move transversely to the string. Moreover because of the Minkowski metric the end points must also move at the speed of light. Here however because Lorentz invariance is not at issue, additional solutions are possible. The end-points of our string can be at rest, ${\bf v}=0$, and satisfy the boundary condition (12). The last solution of eq. (12) is for $\lambda^{\prime}=0$. Because $\lambda$ also vanishes at the end points, eq. (10) gives $\dot{\hat{{\bf v}}}=0$, or straight line motion. In summary the three possible boundary conditions for the string end points are: $\displaystyle{\bf v}\cdot\hat{{\bf t}}$ $\displaystyle=$ $\displaystyle 0\;\;\;\mbox{(pure rotation)}$ (13a) $\displaystyle{\bf v}$ $\displaystyle=$ $\displaystyle 0\;\;\;\mbox{(at rest)}$ (13b) $\displaystyle\dot{\hat{{\bf v}}}$ $\displaystyle=$ $\displaystyle 0\;\;\;\mbox{(straight line motion)}$ (13c) Whether an extremal transformation is a minimum can be determined by examining the second variation of the functional (6): $\delta^{2}{\cal D}=\frac{1}{2}\int_{0}^{L}\\!\\!\int_{0}^{T}\\!\\!\\!ds\,dt\>\left[\delta\dot{{\bf r}}\cdot\mathbf{I}\cdot\delta\dot{{\bf r}}+\delta{\bf r}^{\prime}\cdot\boldsymbol{\Lambda}\cdot\delta{\bf r}^{\prime}\right]\>,$ (14) where $\mathbf{I}_{ij}=(\dot{{\bf r}}^{2}\delta_{ij}-\dot{x}_{i}\dot{x}_{j})/|\dot{{\bf r}}|^{3}$ and $\boldsymbol{\Lambda}_{ij}=-\lambda(s,t)\,\delta_{ij}$, and $\delta{\bf r}^{\prime}$ and $\delta\dot{{\bf r}}$ are the $s$ and $t$ derivatives of the variation $\delta{\bf r}$ from the extremal path. We now apply these concepts to some specific examples. ### II.2 Examples Translations. If two space curves differ by a translation, ${\bf r}_{\mbox{\tiny{B}}}(s)={\bf r}_{\mbox{\tiny{A}}}(s)+{\bf d}$ with ${\bf d}$ a constant vector. The appropriate boundary condition for the end points is (13c). The points along the string can all satisfy (10) with $\dot{\hat{{\bf v}}}=0$ and $\lambda=0$ everywhere (since $\hat{{\bf t}}$, $\boldsymbol{\kappa}\neq 0$), and straight line motion results: ${\bf r}^{\ast}(s,t)={\bf r}_{\mbox{\tiny{A}}}(s)+({\bf r}_{\mbox{\tiny{B}}}(s)-{\bf r}_{\mbox{\tiny{A}}}(s))t/T$. The distance ${\cal D}^{\ast}=L\left|{\bf d}\right|$. This is the 1-dimensional analogue to eq.s (2), (3). Piece-wise linear space curves. Suppose initially the curvature of some section of the string is zero. Then, taking the dot product of ${\bf v}$ with eq. (10), we see that eq. (12) holds for all points along the string. So the string either rotates or translates (or remains at rest if that segment has completed the transformation). Generally if one string partner has curvature (e.g. ${\bf r}_{\mbox{\tiny{A}}}$ in fig. 1B) the transformation is more complicated, but if both ${\bf r}_{\mbox{\tiny{A}}}$ and ${\bf r}_{\mbox{\tiny{B}}}$ are straight lines as in figure 1A, equation (12) holds for both. It is then reasonable to seek solutions ${\bf r}^{\ast}$ of the EL equation such that equation (12) holds for all (s,t). Consider the two space curves shown in figure 1A with ${\bf r}_{\mbox{\tiny A}}(s)=s\,\hat{{\bf x}}$ and ${\bf r}_{\mbox{\tiny B}}(s)=s\,\hat{{\bf y}}$, both with curvature $\boldsymbol{\kappa}=0$. We first investigate rotation from $A$ to $B$. This transformation satisfies the EL equation so appears to be extremal: ${\bf r}=s\hat{{\bf r}}=s(\cos\omega t\hat{{\bf x}}+\sin\omega t\hat{{\bf y}})$. The velocity $\dot{{\bf r}}=s\omega\hat{\boldsymbol{\theta}}$, so the Distance ${\cal D}[{\bf r}_{\mbox{\tiny{ROT}}}(s,t)]=\pi L^{2}/4$. Taking the dot product of $\hat{{\bf t}}$ with eq. 10 gives $\lambda^{\prime}=\hat{{\bf t}}\cdot\dot{\hat{{\bf v}}}=-\omega$, or $\lambda(s,t)=\lambda_{o}-\omega s$. For the transformation to be extremal, the conjugate momenta must also vanish at the string end points, or $\lambda(0,t)=\lambda(L,t)=0$. This is impossible to achieve with this functional form, so the transformation is not extremal. We may however include the subsidiary condition here that ${\bf r}_{\mbox{\tiny{A}}}(0,t)={\bf r}_{\mbox{\tiny{B}}}(0,t)$. Then the end point of the string at $s=0$ is determined, and the variations $\delta{\bf r}(0,t)$ must vanish. Now only $\lambda(L,t)=0$, and so $\lambda(s,t)=\omega(L-s)$. The transformation is extremal. Whether it is a minimum can be determined by examining the second variation (14). For the transformation ${\bf r}_{\mbox{\tiny{ROT}}}(s,t)$, the matrix $\mathbf{I}$ in (14) is non-negative definite, a necessary condition for a local minimum GelfandIM00 , however $\boldsymbol{\Lambda}$ is negative definite, so the character of the extremum is determined by the interplay of the two terms in (14). Variations $\delta{\bf r}$ that preserve ${\bf r}^{\prime 2}=1$ or $2\hat{{\bf t}}\cdot\delta{\bf r}^{\prime}=0$ are satisfied in this example by $\delta{\bf r}=f(s,t)\hat{\boldsymbol{\theta}}$, where $f(s,t)$ must satisfy the boundary conditions $\delta{\bf r}(0,t)=\delta{\bf r}(s,0)=\delta{\bf r}(s,T)=0$. We thus let the variations have the functional form: $\delta{\bf r}=\epsilon\sin(ks)\sin(n\pi t/T)\hat{\boldsymbol{\theta}}$, where $\hat{\boldsymbol{\theta}}=-\sin\omega t\hat{{\bf x}}+\cos\omega t\hat{{\bf y}}$, $n$ is a positive integer, and $k$ is unrestricted. Inserting this functional form for the variations into eq. (14) gives $\delta^{2}{\cal D}=(\epsilon^{2}\pi/8)\mathcal{F}(kL)$, where $\mathcal{F}(x)$ is a non-positive, monotonically decreasing function, with a maximum of zero at $kL=0$. In fact to lowest order $\mathcal{F}(kL)\approx-(\pi\epsilon^{2}/2160)\,(kL)^{6}$. The extremum corresponding to pure rotation of curve ${\bf r}_{\mbox{\tiny{A}}}$ into ${\bf r}_{\mbox{\tiny{B}}}$ is a maximum! The only other solution to equations (10) and (12) for all $(s,t)$ is for each point $s$ on ${\bf r}_{\mbox{\tiny A}}(s)$ to be connected to a corresponding point on ${\bf r}_{\mbox{\tiny B}}(s)$ by a straight line, corresponding to equation (13c). Equation (12) holds everywhere because $\lambda^{\prime}(s,t)=0$. Because $\lambda$ is zero at the boundaries it is thus zero everywhere. An intermediate configuration then has the shape of a piecewise linear curve with a right angle ’kink’ at $s^{\ast}(t)$ (see fig 2). As $t$ progresses, the kink propagates along curve ${\bf r}_{\mbox{\tiny B}}$, and the horizontal part of the chain follows straight line diagonal motion, shrinking as its left end is overlaid onto curve ${\bf r}_{\mbox{\tiny B}}$. The solution for the velocity at all $(s,t)$ is given by ${\bf v}(s,t)=v_{o}(t)\Theta\left(s-s^{\ast}(t)\right)\hat{\mbox{{\bf e}}}_{v}$ where $s^{\ast}(t)$ is the position of the tangent discontinuity in figure 2, which goes from $s^{\ast}(0)=0$ to $s^{\ast}(T)=L$ as $t$ goes from $0$ to $T$. $\hat{\mbox{{\bf e}}}_{v}$ is a unit vector along the direction of the velocity, $\hat{\mbox{{\bf e}}}_{v}=(-\hat{{\bf x}}+\hat{{\bf y}})/\sqrt{2}$, and $v_{o}(t)$ is a speed which can be taken constant. By simple geometry, $v_{o}=\sqrt{2}\,\dot{s}^{\,\ast}$. Because $s^{\,\ast}(T)=L$, $v_{o}=\sqrt{2}L/T$ and $s^{\,\ast}(t)=L\,t/T$. The total distance travelled from equation (4) is then ${\cal D}^{\ast}=L^{2}/\sqrt{2}$. Because the transformation involves straight line motion, it is minimal. This can be seen from the second variation eq. (14). The shape of the curve at all times is given by $\displaystyle{\bf r}^{\ast}(s,t)$ $\displaystyle=$ $\displaystyle s\,\,\Theta(Lt/T-s)\,\hat{{\bf y}}+(Lt/T)\Theta(s-Lt/T)\,\hat{{\bf y}}$ (15) $\displaystyle+$ $\displaystyle(s-Lt/T)\Theta(s-Lt/T)\,\hat{{\bf x}}$ Taking variations from the extremal path as before, let $\delta{\bf r}=\epsilon\sin k(s-Lt/T)\sin(n\pi t/T)\Theta(s-Lt/T)\hat{{\bf y}}$. These variations only act on the “free” part of the string and preserve a unit tangent to first order. The matrix $\boldsymbol{\Lambda}$ in (14) is zero for straight line transformations where $\lambda=0$. The quadratic form $\delta\dot{{\bf r}}\cdot\mathbf{I}\cdot\delta\dot{{\bf r}}$ is non-negative, and results in a 2nd variation $\delta^{2}{\cal D}=\epsilon^{2}(32\sqrt{2})^{-1}[(kL)^{2}+(n\pi)^{2}(1-\mbox{sinc}^{2}(kL))]$, which is non-negative, monotonically increasing in $kL$, and quadratic to lowest order, with a minimum of zero at $kL=0$. The transformation is indeed minimal. Likewise, the minimal distance to fold a string of total length $L$ upon itself starting from a straight line (to form a hairpin) is ${\cal D}^{\ast}=L^{2}/4$. Solution Degeneracy. The above example illustrates that there are essentially an infinite number of extremal transformations: one can piece together various rotations and translations for parts or all of the chain while still satisfying the EL equations. This infinity of extrema is likely to lead to nearly insurmountable difficulties for the solution of eq. (9) by direct numerical integration. For these reasons we apply a method based on analytic geometry to obtain numerical solutions. This described in more detail below. There is also an infinite degeneracy of solutions having the minimal distance in the above example. To see a second minimal transformation, imagine running the above solution backwards in time, so the kink propagates from $s=L$ to $s=0$ along ${\bf r}_{\mbox{\tiny B}}$. But this solution should hold forwards in time for the original problem if we permute ${\bf r}_{\mbox{\tiny B}}$ and ${\bf r}_{\mbox{\tiny A}}$. Now intermediate states ${\bf r}^{\ast}$ first run along $\hat{{\bf x}}$, then $\hat{{\bf y}}$. But then we can introduce multiple right angle kinks in various places, without causing the trajectories in the transformation to deviate from straight lines, so that intermediate states look like staircases. As there are an infinite number of possible staircases in the continuum limit, there is an infinite degeneracy. This can lead to a tangent vector ${\bf r}^{\prime}$ whose magnitude is length-scale dependent, and less than unity until $s\rightarrow 0$. For example an intermediate configuration can be drawn in figure 2 which appears as a straight diagonal line from ${\bf r}^{\ast}(0,t)$ to ${\bf r}^{\ast}(L,t)$, until $s\rightarrow 0$ when an infinite number of step discontinuities are revealed. This problem is resolved in practice through finite-size effects involving different critical angles of rotation described below. In the continuum limit it is resolved by introducing curvature constraints. Curvature constraints. In applications to polymer physics, chains have a stiffness characterized by bending potential in the analysis that is proportional to the square of the local curvature. Here we may choose to characterize stiffness by introducing a constraint on the configurations of the space curve, so that the curvature simply cannot exceed a given number: $V_{\kappa}\left({\bf r}^{\prime\prime}\right)=\Theta\left(\left|{\bf r}^{\prime\prime}\right|<\kappa_{\mbox{\tiny C}}\right)\>.$ (16) This term lifts the infinite degeneracy mentioned above, as each near-kink (with putative $\kappa>\kappa_{\mbox{\tiny C}}$) would result in slight deviations from linear motion in the above example, and thus an additional cost in the effective action. Other functional forms for $V_{\kappa}$ are also possible. For some applications a more conventional stiffness potential of the form $V_{\kappa}\left({\bf r}^{\prime\prime}\right)=\frac{1}{2}A_{\kappa}{\bf r}^{\prime\prime 2}$ may be more appropriate. However then the action would no longer consist of a true distance functional, and its minimization would involve the detailed interplay of the parameter $A_{\kappa}$ favouring globally minimal curvature with other factors affecting distance in the problem. Discrete Chains. Strings with a finite number of elements (chains) provide a more accurate representation of real-world systems such as biopolymers. Discretization is also essential for numerical solutions in these more realistic cases. Monomers on a discretized chain travel along a curved metric GrosbergAY04 , and Lagrange multipliers explicitly account for this fact here. We start by discretizing the string into a chain of $N$ links each with length $ds=L/N$, so that equation (4) becomes $(ds)\int\\!dt\,\sum_{i=1}^{N+1}\sqrt{\dot{{\bf r}}_{i}^{2}}$, with each ${\bf r}_{i}(t)$ a function of $t$ only. The total distance is the accumulated distance of all the points joining the links, plus that of the end points, all times $ds$. This approach is essentially the method of lines for solving equation 10: the PDE becomes a set of $N+1$ coupled ODEs. Equation (5) becomes $N$ constraint equations added to the effective Lagrangian: $\sum_{i=1}^{N}\hat{\lambda}_{i,i+1}\sqrt{({\bf r}_{i+1}-{\bf r}_{i})^{2}}$. We rewrite this strictly for convenience as $\sum\frac{\lambda_{i,i+1}}{2}\>{\bf r}_{i+1/i}^{2}$, where ${\bf r}_{i+1/i}\equiv{\bf r}_{i+1}-{\bf r}_{i}$, and $|{\bf r}_{i+1/i}|=L/N$. The PDE in (10) then becomes $N+1$ coupled (vector) ODEs, each of the form $\dot{\hat{{\bf v}}}_{i}+\lambda_{i-1,i}\,{\bf r}_{i/i-1}-\lambda_{i,i+1}\,{\bf r}_{i+1/i}=0$ (17) with $\lambda_{0,1}=\lambda_{N+1,N+2}=0$. Equation (17) is consistent with (10) after suitable definitions, for example the curvature at point $i$ after discretization is given by $({\bf r}_{i+1/i}-{\bf r}_{i/i-1})/ds^{2}$. One link. We turn to the simplest problem of one link with end points $A$ and $B$ (see fig. 3), for which the action reads $L\int_{0}^{T}\\!dt\,(\sqrt{\dot{{\bf r}}_{\mbox{{\tiny A}}}^{2}}+\sqrt{\dot{{\bf r}}_{\mbox{{\tiny B}}}^{2}}-\frac{\lambda(t)}{2}{\bf r}_{\mbox{\tiny{B/A}}}^{2})$. Points $A$ and $B$ have boundary conditions ${\bf r}_{\mbox{\tiny{A}}}(0)=\bf{A}$, ${\bf r}_{\mbox{\tiny{B}}}(0)=\bf{B}$, ${\bf r}_{\mbox{\tiny{A}}}(T)=\bf{A^{\prime}}$, ${\bf r}_{\mbox{\tiny{B}}}(T)=\bf{B^{\prime}}$. The link in our problem is taken to have a direction, so point $A$ cannot transform to point $B$. The Euler- Lagrange equations become: $\displaystyle\begin{matrix}\dot{\hat{{\bf v}}}_{\mbox{{\tiny A}}}-\lambda\,{\bf r}_{\mbox{\tiny{B/A}}}=0\\\ \dot{\hat{{\bf v}}}_{\mbox{{\tiny B}}}+\lambda\,{\bf r}_{\mbox{\tiny{B/A}}}=0\end{matrix}\quad\text{or}\quad\begin{matrix}\lambda\,{\bf v}_{\mbox{\tiny{A}}}\cdot{\bf r}_{\mbox{\tiny{B/A}}}=0\\\ \lambda\,{\bf v}_{\mbox{\tiny{B}}}\cdot{\bf r}_{\mbox{\tiny{B/A}}}=0\end{matrix}$ (18) where the orthogonality of $\bf{v}$ and $\dot{\hat{{\bf v}}}$ has been used. Reminiscent of eq. (12), equations (18) each have $3$ solutions. For point $A$ these are: (1) ${\bf v}_{\mbox{\tiny{A}}}\cdot{\bf r}_{\mbox{\tiny{B/A}}}=0$, or pure rotation of $A$ about $B$, (2) ${\bf v}_{\mbox{\tiny{A}}}=0$ or point $A$ is stationary, or (3) $\lambda=0$ and thus $\dot{\hat{{\bf v}}}_{\mbox{{\tiny A}}}=0$ from the EL equations, indicating straight-line motion. Moreover, (1) implies ${\bf v}_{\mbox{\tiny{B}}}=0$, or both points rotate about a common center, (2) implies ${\bf v}_{\mbox{\tiny{B}}}\cdot{\bf r}_{\mbox{\tiny{B/A}}}=0$ or $B$ rotates, and (3) implies $\dot{\hat{{\bf v}}}_{\mbox{{\tiny B}}}=0$ as well, so that both points move in straight lines. An extremal transformation thus involves either straight line motion, or rotations of one point about the other at rest (or common center). Once again, there are an infinite number of solutions: any combination of translations and rotations satisfies the EL equations, such as those shown in figure 3B-F. The Lagrange multiplier may be found from the first integral: taking the dot product of the EL equation for $B$ with ${\bf r}_{\mbox{\tiny{B/A}}}$ gives $-ds^{2}\lambda={\bf r}_{\mbox{\tiny{B/A}}}\cdot\dot{\hat{{\bf v}}}_{\mbox{{\tiny B}}}$. Thus when $B$ moves in a straight line $\lambda=0$. When $B$ rotates about $A$, its acceleration ${\bf a}_{\mbox{{\tiny B}}}$ follows from rigid body kinematics as ${\bf a}_{\mbox{{\tiny A}}}+\boldsymbol{\alpha}\times{\bf r}_{\mbox{\tiny{B/A}}}-\omega^{2}{\bf r}_{\mbox{\tiny{B/A}}}$, where $\boldsymbol{\omega}$ and $\boldsymbol{\alpha}$ are the angular velocity and acceleration respectively, and ${\bf a}_{\mbox{{\tiny A}}}=0$. Thus $\lambda=1/L$. The minimal solution is the one that involves the minimal amount of rotation (and monotonic approach to $A^{\prime}B^{\prime}$). This may be obtained from analytic geometry: for the example configurations in fig. 3F, point $B$ rotates about point $A$ until $B^{\prime\prime}$, where the straight line $\overline{B^{\prime\prime}B^{\prime}}$ is tangent to the circle of radius $ds=L$ about $A$. The distance (over $ds$) is $AA^{\prime}+L\theta_{c}+B^{\prime\prime}B^{\prime}$, where $\sin\theta_{c}=L/(L+AA^{\prime})$ and $B^{\prime\prime}B^{\prime}=\sqrt{(AA^{\prime})^{2}+2L(AA^{\prime})}$, so for example if $AA^{\prime}=2L$, ${\cal D}\approx 5.168\,L^{2}$. Chains with curvature. We can now investigate the transformation shown in figure 1B with the above methods. This is the canonical example when at least one of the space curves has non-zero curvature $\boldsymbol{\kappa}$. Let ${\bf r}_{\mbox{\tiny A}}=R\sin(\pi s/2L)\hat{{\bf x}}+R\cos(\pi s/2L)\hat{{\bf y}}$ and ${\bf r}_{\mbox{\tiny B}}=s\hat{{\bf x}}+R\hat{{\bf y}}$, with $0\leq s\leq L$ and $R=2L/\pi$. We then discretize the chain into $N$ segments. According to eq. (17), the end point velocities $\dot{\hat{{\bf v}}}_{1}$, $\dot{\hat{{\bf v}}}_{N+1}$ obey EL equations of the same form as equations (18), and thus either rotate or translate. The situation for these links is analogous to figures 3B and 3F, in that the angle the link must rotate depends on the order of translation and rotation. The geometry in figure 1B is analogous to transformations $A^{\prime}B^{\prime}\rightarrow AB$ in figures 3B, 3F, in that the critical angle $\theta_{c}$ the link must rotate before translating is smaller if translation occurs first. Figure 4 shows the two minimal solutions thus obtained. The transformation in fig. 4A undergoes translation away from curve ${\bf r}_{\mbox{\tiny{A}}}$, and rotation at ${\bf r}_{\mbox{\tiny{B}}}$. It is the global minimum. The transformation in 4B rotates from ${\bf r}_{\mbox{\tiny{A}}}$ through a larger critical angle (see 4B inset), and then translates to ${\bf r}_{\mbox{\tiny{B}}}$. Both solutions have a soliton-like kink that propagates across either space-curve ${\bf r}_{\mbox{\tiny{B}}}$ or ${\bf r}_{\mbox{\tiny{A}}}$. The minimal transformation follows these steps: (1) Link ${\bf r}_{2/1}$ rotates about ${\bf r}_{1}$, ${\bf v}_{1}=0$, ${\bf v}_{2}\cdot{\bf r}_{2/1}=0$, and the Lagrange multiplier representing the conjugate ’force’ $\lambda_{12}\neq 0$. During this rotation, nodes $3,4,\ldots$ move in straight lines formed by their initial values ${\bf r}_{\mbox{{\tiny A}}3},{\bf r}_{\mbox{{\tiny A}}4},\ldots$ and the tangent points to circles of radius $ds$ centered at ${\bf r}_{\mbox{{\tiny B}}2},{\bf r}_{\mbox{{\tiny B}}3},\ldots$. The corresponding Lagrange constraint forces $\lambda_{23},\lambda_{34},\ldots$ are all zero. Links ${\bf r}_{3/2},{\bf r}_{4/3},\ldots$ all adjust their orientation to ensure straight-line motion of their end points (dashed lines in fig. 4A), except for ${\bf r}_{2}$ which follows a curved path. (2) When link ${\bf r}_{2/1}$ completes its rotation, it coincides with curve ${\bf r}_{\mbox{\tiny{B}}}$, and the process starts again with link ${\bf r}_{3/2}$ which begins its rotation about ${\bf r}_{2}$, while nodes $4,5,\ldots$ move in straight lines. This process continues until the final link ${\bf r}_{N+1/N}$ rotates into place on ${\bf r}_{\mbox{\tiny{B}}}$. The transformation in 4B is essentially the time- reverse of the above, but starting at curve ${\bf r}_{\mbox{\tiny{B}}}$ and ending on ${\bf r}_{\mbox{\tiny{A}}}$. For ideal chains without curvature constraints, the distances obtained from the two transformations in 4A,B differ non-extensively as the number of links $N\rightarrow\infty$. Moreover, the distance for each transformation itself differs non-extensively from the Mean Root Square distance $MRSD=N^{-1}\sum_{i=1}^{N}\sqrt{({\bf r}_{\mbox{{\tiny A}}i}-{\bf r}_{\mbox{{\tiny B}}i})^{2}}$ as $N\rightarrow\infty$. §§§The MRSD is always less than or equal to the Root Mean Square Deviation or RMSD between structures, as can be shown by applying Hölder’s inequality. Specifically, the distance travelled by straight line motion scales as $ds\,NL\sim L^{2}$, while the distance travelled by rotational motion scales as $ds\,(N\overline{\theta}_{c}ds)\sim L^{2}/N$. On the other hand, curvature constraints as in eq. (16) become more severe on consecutive links as $N\rightarrow\infty$, and can yield extensive corrections to the distance. Specifically, the increase in distance $\Delta{\cal D}$ due to curvature constraints scales like the radius of curvature $R$ times $N$, since every node is affected by the rounded kink as it propagates. So $\Delta{\cal D}\sim ds\,NR\sim LR$. The importance of this effect then depends on how $R$ compares to $L$ (the ratio of the persistence length to the total length). It does not vanish as $N\rightarrow\infty$. Non-crossing constraints described below also yield extensive corrections to the distance travelled. ### II.3 Non-crossing space curves The minimal transformation may be qualitatively different when chain crossing is explicitly disallowed. Figure 1C illustrates a pair of curves that differ only by the order of chain crossing. They are displaced in the figure for easier visualization but should be imagined to overlap so the quantity $\int_{0}^{L}\left|{\bf r}_{\mbox{\tiny A}}-{\bf r}_{\mbox{\tiny B}}\right|\approx 0$, i.e. if they were ghost chains their distance would be nearly zero, and most existing metrics give zero distance between these curve pairs (see Table I). Analogous to the construction of Alexander polynomials for knots, if we form the orthogonal projection of these space curves onto a plane there will be double points indicating one part of the curve crossing over or under another. To transform from configuration ${\bf r}_{\mbox{\tiny A}}$ to ${\bf r}_{\mbox{\tiny B}}$ without crossing, the curves must always go through configurations having zero double points. If we trace the curve in an arbitrary but fixed direction, each double point occurs twice, once as underpass and once as an overpass. We may call the part of the curve between two consecutive passes a bridge. If the bridge ends in an overpass we assign it +1, if the bridge ends in an underpass we assign it -1, so traversing from the left in figure 1C, curve ${\bf r}_{\mbox{\tiny B}}$ has (+1) sense, and curve ${\bf r}_{\mbox{\tiny A}}$ (-1). The change in sense during any transformation obeying non-crossing is always $\pm 1$, while ghost chains can have changes of $\pm 2$. The non-crossing condition means that the Lagrangian for the minimal transformation now depends on the position ${\bf r}(s,t)$ of the space curve, which may be accounted for using an Edwards potential: $V_{\mbox{\tiny NC}}([{\bf r}(s,t)])=\int_{0}^{L}\\!\\!\\!ds_{1}\\!\\!\\!\int_{0}^{L}\\!\\!ds_{2}\>\,\delta({\bf r}(s_{1},t)-{\bf r}(s_{2},t))$ In practice a Gaussian may be used to approximate the delta function, with a variance that may be adjusted to account for the thickness or volume of the chain. The Euler-Lagrange equation now becomes $(V_{\mbox{\tiny NC}})_{{\bf r}}=(\mathcal{L}_{{\bf r}^{\prime}})_{s}+(\mathcal{L}_{\dot{{\bf r}}})_{t}-[(V_{\kappa})_{{\bf r}^{\prime\prime}}]_{ss}$ (19) where the curvature potential in eq. (16) has been included, and the notation $(\mathcal{L}_{{\bf r}^{\prime}})_{s}\equiv(d/ds)(\partial\mathcal{L}/\partial{\bf r}^{\prime})$ has been used. Equation (10) is now modified to $\hat{{\bf v}}_{t}=\left(\lambda\hat{{\bf t}}\right)_{s}+\nabla V_{\mbox{\tiny NC}}+[(V_{\kappa})_{{\bf r}^{\prime\prime}}]_{ss}$ (20) To access various conformations, the minimal transformation must now abide by the non-trivial geometrical constraints that are induced by non-crossing. In general this renders the problem difficult, however the example in figure 1C is simple enough to propose a mechanism for the minimal transformation consistent with the developments above, without explicitly solving the EL equations in this case. In analogy with the hairpin transformation described below eq. (15), the transformation here involves essentially forming and then unforming a hairpin. ${\bf r}_{\mbox{\tiny{A}}}(N)$ (the blue end of curve ${\bf r}_{\mbox{\tiny{A}}}$ in fig 1C) propagates back along its own length until it reaches the junction, where it then rotates over it to become the overpass (this takes essentially zero distance in the continuum limit). The curve then doubles back following its path in reverse to its starting point. This transformation is fully consistent with the allowed extremal rotations and translations of the discretized chain. The distance in the continuum limit is ${\cal D}=\int_{0}^{\ell}\\!\\!ds\,(2s)=\ell^{2}$, where $\ell$ is the length of the shorter arm extending from the junction in fig 1C. ## III Discussion The distance between finite objects of any dimension $d$ is a variational problem, and may be calculated by minimizing a vector functional of $d+1$ independent variables. Here we formulated the problem for space curves, where the function ${\bf r}^{\ast}(s,t)$ defining the transformation from curve ${\bf r}_{\mbox{\tiny A}}$ to curve ${\bf r}_{\mbox{\tiny B}}$ gives the minimal distance $\mathcal{D}$. We provided a general recipe for the solution to the problem through the calculus of variations. For simple cases the solution is analytically tractable. Generally there are an infinity of extrema, and direct numerical methods are unlikely to be fruitful. We employed a method that interpreted the discretized EL equations geometrically to obtain minimal solutions. The various solutions obtained here are summarized in Table I, and compared with other similarity measures currently used. The distance metric may be generalized to higher dimensional manifolds, for example a two dimensional surface needs three independent parameters to describe the transformation. The distance becomes $\mathcal{D}=\int\\!du\int\\!dv\int\\!dt\,|\dot{{\bf r}}|$ and the constant unit area condition becomes $\left|\frac{\partial{\bf r}}{\partial u}\times\frac{\partial{\bf r}}{\partial v}\right|=1$. The question of a distance metric between configurations of a biopolymer has occupied the minds of many in the protein folding community for some time (c.f. for example Leopold92 ; Chan94 ; FalicovA96 ; DuR98:jcp ; ChoSS06 ). Such a metric is of interest for comparison between folded structures, as well as to quantify how close an unfolded or partly folded structure is to the native. Chan and Dill Chan94 investigated the minimum number of moves necessary to transform one lattice structure to another, in particular while breaking the smallest number of hydrogen bonds. Leopold et al Leopold92 investigated the minimum number of monomers that had to be moved to transform one compact conformation to another. Falicov and Cohen investigated structural comparison by rotation and translation until the minimal area surface by triangulation was obtained between two potentially dissimilar protein structures FalicovA96 . The present theoretical framework allows computation of a minimal distance between proteins of the same length by rotating and translating until $\mathcal{D}$ is minimized, as done in the calculation of RMSD. Comparison between different length proteins would involve the further optimization with respect to insertion or deletion of protein chain segments. It is interesting to ask which folded structures have the largest, or smallest average distance $\left<\mathcal{D}\right>$ from an ensemble of random coil structures, and also whether the accessibility of these structures in terms of $\mathcal{D}$ translates to their folding rates. It can also be determined whether the distance to a structure correlates with kinetic proximity in terms of its probability $p_{\mbox{\tiny{F}}}$ to fold before unfolding DuR98:jcp , by calculating $\left<\mathcal{D}p_{\mbox{\tiny{F}}}\right>$. The question of the most accessible or least accessible structure may be formulated variationally as a free-boundary or variable end-point problem. It is an important future question to address whether the entropy of paths to a particular structure is as important as the minimal distance. In this sense it may be the finite ”temperature” ($\beta<\infty$) partition function $Z(\beta)=\int d[{\bf r}(s,t)]\exp\left(-\beta\mathcal{D}[{\bf r}(s,t)]\right)$, i.e. the sum over paths weighted by their ’actions’, which is the most important quantity in determining the accessibility between structures. This has an analogue to the quantum string: we investigated only $Z(\infty)$ here. We hope that this work proves useful in laying the foundations for unambiguously defining distance between biomolecular structures in particular and high-dimensional objects in general. ## IV Acknowledgements We are grateful to Ali Mohazab, Moshe Schecter, Matt Choptuik, and Bill Unruh for insightful discussions. Support from the Natural Sciences and Engineering Research Council and the A. P. Sloan Foundation is gratefully acknowledged. ## References * (1) Gelfand, I. M & Fomin, S. V. (2000) Calculus of Variations. (Dover). * (2) Zwiebach, B. (2004) A first course in string theory. (Cambridge University Press, New York). * (3) Grosberg, A. Y. (2004) in Computational Soft Matter: From Synthetic Polymers to Proteins, eds. Attig, N, Binder, K, Grubmüller, H, & Kremer, K. (John von Neumann Institut für Computing, Bonn) Vol. NIC series vol. 23, pp. 375–399. * (4) Leopold, P. E, Montal, M, & Onuchic, J. N. (1992). Protein folding funnels: Kinetic pathways through compact conformational space. Proc. Natl Acad. Sci. USA 89, 8721–8725. * (5) Chan, H. S & Dill, K. A. (1994). Transition States and Folding Dynamics of Proteins and Heteropolymers. J. Chem. Phys. 100, 9238–9257. * (6) Falicov, A & Cohen, F. E. (1996). A surface of minimum area metric for the structural comparison of proteins. J Mol Biol 258, 871–892. * (7) Du, R, Pande, V. S, Grosberg, A. Y, Tanaka, T, & Shakhnovich, E. S. (1998). On the transition coordinate for protein folding. J Chem Phys 108, 334–350. * (8) Cho, S. S, Levy, Y, & Wolynes, P. G. (2006). P versus Q: Structural reaction coordinates capture protein folding on smooth landscapes. Proc. Natl Acad. Sci. USA 103, 586–591. * (9) Veitshans, T, Klimov, D, & Thirumalai, D. (1996). Protein folding kinetics: Timescales, pathways and energy landscapes in terms of sequence-dependent properties. Folding and Design 2, 1–22. FIGURE CAPTIONS FIGURE 1: Three representative pairs of curves. A Straight line curve rotated by $\pi/2$. B One string has a finite radius of curvature, the other is straight. C A canonical example where non-crossing is important- the curves are displaced for easy visualization but should be imagined to be superimposed. FIGURE 2: The minimal transformation from A to B in figure 1A involves the propagation of a kink along curve B. The end point of the curve at intermediate states satisfies $x+y=L$, the equation for a straight line. A similar linear equation holds for any point on the curve, thus no solution with shorter distance can exist. An intermediate configuration is shown in red. Alternative transformations are possible with kinks along A, as well as multiple kinks (see text). FIGURE 3: Transformations between two rigid rods. (A) undergoes simultaneous translation and rotation and so is not extremal. (B) is extremal and minimal. The rod cannot rotate any less given that it translates first. However this transformation is a weak or local minimum. (C), (D), and (E) are extremal but not minimal. (F) Is the global minimum. It rotates the minimal amount, and both $A$ and $B$ move monotonically towards $A^{\prime}$, $B^{\prime}$. A purely straight-line transformation exists but involves moving point $A$ away from $A^{\prime}$ before moving towards it (similar to (D)), thus covering a larger distance than the minimal transformation. FIGURE 4: Two minimal transformations between the curves shown in fig. 1B, for $N=10$ links. Fig (A) is the global minimal transformation ${\bf r}^{\ast}(s,t)$, with ${\cal D}^{\ast}\approx 0.330\,L^{2}$, figure (B) is a local minimum with ${\cal D}\approx 0.335\,L^{2}$. In (A), links with one end touching curve ${\bf r}_{\mbox{\tiny{B}}}$ rotate, the others translate first from ${\bf r}_{\mbox{\tiny{A}}}$, rotating only when one end of a link has touched ${\bf r}_{\mbox{\tiny{B}}}$. In (B) they rotate first from ${\bf r}_{\mbox{\tiny{A}}}$, then translate into ${\bf r}_{\mbox{\tiny{B}}}$. Dashed lines in (A) show the paths travelled for each bead. The inset of (A) plots the total distance travelled as a function of the number of links $N$, with various $N$ plotted as filled circles to indicate the rapid decrease and asymptotic limit to ${\cal D}_{\infty}\approx 0.251\,L^{2}$ The inset in (B) shows the minimal angle each link must rotate during the transformation- it is less for the transformation in (A). Movie animations of these transformations are provided as Supporting Information. TABLES AND TABLE CAPTIONS Table 1: Values of the distance for various examples considered here, compared to other metrics. Curve Pair | ${\cal D}^{\ast}\,(L^{2})$ | RMSD${}^{\star}\,$ (L) | (1-Q)† | $\chi^{\sharp}$ ---|---|---|---|--- Trivial translation | $|{\bf d}|/L$ | $|{\bf d}|/L$ | 0 | 0 “L-curves”, fig 1A | $1/\sqrt{2}$ | $\sqrt{2/3}$ | –‡ | 0 Straight line to Hairpin | $1/4$ | $1/\sqrt{6}$ | 1 | 1/2 “C-curve”- st. line, fig 4A | $0.330$ | 0.371 | –‡ | 0.417 “C-curve”- st. line, fig 1A♮ | $0.251$ | $0.334$ | –‡ | 1 “Over/under” curves, fig 1C | $(\ell/L)^{2}$ | $\approx 0$ | $0^{\wr}$ | 0 Single link, fig 3F♭ | $5.168$ | $\sqrt{7}^{\lambda}$ | –δ | –δ ⋆ $RMSD\equiv\surd{N^{-1}\sum_{i}({\bf r}_{\mbox{{\tiny A}}i}-{\bf r}_{\mbox{{\tiny B}}i})^{2}}$ † Fraction of shared contacts $A$ has with $B$, --- see DuR98:jcp ; ChoSS06 for definitions. ♯ Structural overlap function equal to $1$ minus the fraction of residue pairs with similar distances in structures $A$ and $B$. The formula in ref. VeitshansT96 is used. ♮ i.e. In the continuum limit. ♭ For $AA^{\prime}=2\times\mbox{link length}$. ‡ $0/0$ or undefined ≀ Assuming a contact is made at the junction. δ Undefined for a single link λ${\cal D}$ is larger than the RMSD here because RMSD contains a factor of $2$ while ${\cal D}$ did not. We could have computed the “effective distance” for the rod by dividing by $2$. Figure 1: Figure 2: Figure 3: Figure 4:

arxiv-papers

2024-09-04T02:48:54.078970

0803.0130

# Raman scattering in a Heisenberg $S=1/2$ antiferromagnet on the triangular lattice Natalia Perkins Department of Physics, University of Wisconsin-Madison, Madison, WI 53706, USA Institute for Theoretical Physics, Technical University Braunschweig, Mendelssohnstr. 3, 38106 Braunschweig, Germany Wolfram Brenig Institute for Theoretical Physics, Technical University Braunschweig, Mendelssohnstr. 3, 38106 Braunschweig, Germany ###### Abstract We investigate two-magnon Raman scattering from the $S=1/2$ Heisenberg antiferromagnet on the triangular lattice, considering both the effect of renormalization of the one-magnon spectrum by $1/S$ corrections and final- state magnon-magnon interactions. The bare Raman intensity displays two peaks related to one-magnon van-Hove singularities. We find that $1/S$ self-energy corrections to the one-magnon spectrum strongly modify this intensity profile. The central Raman-peak is significantly enhanced due to plateaus in the magnon dispersion, the high frequency peak is suppressed due to magnon damping, and the overall spectral support narrows considerably. Additionally we investigate final-state interactions by solving the Bethe-Salpeter equation to $O(1/S)$. In contrast to collinear antiferromagnets, the non-collinear nature of the magnetic ground state leads to an irreducible magnon scattering which is retarded and non-separable already to lowest order. We show that final-state interactions lead to a rather broad Raman-continuum centered around approximately twice the ’roton’-energy. We also discuss the dependence on the scattering geometry. ## I Introduction Raman scattering is an effective tool to study the excitation spectrum of magnetic systems since the intensity of the inelastically scattered light is directly related to the density of singlet states at zero momentum. In local- moment magnets with well defined magnon excitations this quantity is linked to the two-magnon density of states. Therefore, magnetic Raman scattering plays an important role in understanding the dynamics and interactions of magnons in conventional spin systems Elliot1963 ; Fleury1966 ; Parkinson1969 ; Fleury1970 . This is particularly true for the spin-$1/2$ square-lattice Heisenberg antiferromagnet (HAF) of the high-Tc superconductor parent compounds, where experimental Lyons1988 ; Sugai1988 ; Sulewski1991 and theoretical Singh1989 ; girvin ; Chubukov1995 ; Nori1995 studies of the magnetic correlations by Raman scattering may provide important insight into the energy scales relevant to the pairing mechanism (for reviews see Refs. Blumberg1997 ; Devereaux2007 ). Raman scattering from HAFs can be understood in terms of the Loudon-Fleury (LF) processes fleury , in which two magnons are simultaneously created by light absorption and emission. In the limit of large on-site Coulomb correlations $U$, the Hamiltonian describing these processes can be obtained as a leading term of the expansion in $t/(U-\omega)$, where $t$ is the nearest-neighbor (NN) hopping, and $\omega$ is of the order of photon frequencies Shastry1990 . The Raman intensity of HAFs on hypercubic lattices with unfrustrated NN exchange and collinear type of antiferromagnetic (AFM) order allows for a straightforward semi-quantitative interpretation in terms of the LF processes. In fact, in real space, exchanging two NN spins of $S=1/2$ leads to an excitation with energy $\Omega\sim(z-1)J$, where $z$ is the coordination number and $J$ is the AFM exchange energy. The reduction of $\Omega/J$ by $-1$ is a consequence of the exchange link between the NN sites and can be interpreted in terms of a two-magnon interactions in the final state. In momentum space, the linear spin-wave theory yields non-dispersive magnons along the the magnetic Brillouin zone (BZ) boundary, leading to a square-root divergence of the bare two-magnon density of states at $\Omega=zJ$. Inclusion of the final state magnon-magnon interactions broadens the singularity and shifts it down to $\Omega\sim 2.9J$ Singh1989 ; girvin ; Chubukov1995 ; Nori1995 in two dimensions, which is consistent, both with the real-space result $\Omega=J(4-1)=3J$, and with the experimentally observed Raman profile. In contrast to conventional collinear HAFs, very little is known theoretically about Raman scattering from frustrated HAFs. This is intriguing, since the singlet spectrum is believed to be an essential fingerprint of such magnets. The spin $S=1/2$ HAF on the triangular lattice (THAF) with NN exchange interactions is a prominent example of strongly frustrated spin systems. It has a ground state with non-collinear $120^{\circ}$ degree ordering of the spins. Due to this non-collinearity of the classical ground state, nontrivial corrections to the spin-wave spectrum appear already to first order in $1/S$. It has been shown in Refs.chubukov ; chubukov06 ; mike that $1/S$ corrections strongly modify the form of the magnon dispersion of the triangular HAF. The resulting dispersion turns out to be almost flat in a wide range of momenta in which it possesses shallow local minima, ”rotons”, at the midpoint of the faces of the hexagonal BZ. This differs strongly from the classical spin-wave spectrum, which lacks such minima and flat zones. Similar results have been obtained in series expansion studies Zheng2006 . Motivated by these recent findings, in this paper, we analyze the Raman scattering from the THAF by $1/S$ expansion. This is complementary to the recent analysis of Raman scattering on finite, 16 sites THAFs by means of exact diagonalization Vernay2007 . First, our results show that the Raman intensity is very sensitive to both $1/S$ corrections of the magnon spectrum and the magnon-magnon interactions in the final state. Moreover, we find that the Loudon-Fleury process on the THAF leads to a Raman profile, which is independent at $O(1/S)$ of the scattering geometry. The manuscript is organized as follows. In section II, we review results from existing calculations chubukov ; chubukov06 of the one magnon excitations in the THAF to first order in $1/S$ needed for our study of the Raman spectra. In section III we consider the LFP to leading order in $1/S$. In section IV we calculate the Raman spectrum on various levels of approximation in $1/S$, i.e. bare, one-magnon renormalized, and including final state interactions, and show that Raman profile is very sensitive to the magnon-magnon interactions. We discuss our results in section V. ## II Model The Hamiltonian of the THAF reads $H=J\sum_{\langle ij\rangle}{\bf S}_{i}{\bf S}_{j},$ (1) where ${\bf S}_{i}$ are spin$-1/2$ operators, $i$ refers to sites on the triangular lattice, $\langle\rangle$ denotes NN summation, and $J$ is the exchange interaction. The classical ground state of the THAF Jolicoeur1989 is a non-collinear $120^{\circ}$ degree ordering of spins which is shown in Fig.1a). To avoid the complexity of a three-sublattice notation it is convenient to work within a locally rotated frame of reference in which the magnetic order is ferromagnetic. To achieve this we assume a gauge in which the $(x,z)$-coordinates label the lattice plane and a uniform twist with a pitch vector ${\bf Q}=(4\pi/3,0)$ is applied to the $y$-axis. The laboratory frame-of-reference spin ${\bf S}_{i}$ is related to the spin $\tilde{{\bf S}}_{i}$ in the rotated frame through ${\bf S}_{i}=\left[\begin{array}[]{ccc}\sin(q_{i})&-\cos(q_{i})&0\\\ 0&0&-1\\\ \cos(q_{i})&\sin(q_{i})&0\end{array}\right]^{-1}\tilde{{\bf S}}_{i},$ (2) where $q_{i}=2\pi(2l_{i}+m_{i})/3$ and $(l_{i},m_{i})$ are integers labeling the points on the triangular lattice, which is depicted in Fig. 1a). In contrast to ${\bf S}_{i}$, the spin $\tilde{{\bf S}}_{i}$ is amenable to a representation in terms of a single Holstein-Primakoff boson field on all sites $\displaystyle\tilde{S}^{z}_{i}$ $\displaystyle=$ $\displaystyle S-a^{+}_{i}a^{\phantom{+}}_{i}$ (3) $\displaystyle\tilde{S}^{+}_{i}$ $\displaystyle=$ $\displaystyle(2S-a^{+}_{i}a^{\phantom{+}}_{i})^{1/2}a^{\phantom{+}}_{i}$ $\displaystyle\tilde{S}^{-}_{i}$ $\displaystyle=$ $\displaystyle a^{+}_{i}(2S-a^{+}_{i}a^{\phantom{+}}_{i})^{1/2}~{}.$ Figure 1: a) Classical $120^{\circ}$ degree non-collinear spin order on the triangular lattice. Basic vectors of triangular lattice: ${\vec{\delta}}_{1}=(\frac{1}{2},\frac{\sqrt{3}}{2})$, ${\vec{\delta}}_{2}=(\frac{1}{2},-\frac{\sqrt{3}}{2})$ and ${\vec{\delta}}_{3}=(1,0)$. b) Definition of scattering angles for LF vertex. Because we intend to study magnon interactions to first order in $1/S$, we need to expand the Hamiltonian in Eqn. (1) up to quartic order in the boson fields. We have $\displaystyle H-E_{0}=3JS(H_{2}+H_{3}+H_{4})~{},$ (4) where $E_{0}=3JS^{2}/2$ is the classical ground state energy and $\displaystyle H_{2}=\sum_{\bf k}A_{\bf k}a_{\bf k}^{\dagger}a^{\phantom{\dagger}}_{\bf k}+\frac{B_{\bf k}}{2}(a_{\bf k}^{\dagger}a_{-{\bf k}}^{\dagger}+a^{\phantom{\dagger}}_{\bf k}a^{\phantom{\dagger}}_{-{\bf k}})\phantom{aaaaaaaaa}$ (5) $\displaystyle H_{3}=-i\sqrt{\frac{3}{8S}}\sum_{{\bf k}_{1},{\bf k}_{2},{\bf k}_{3}}(a_{{\bf k}_{1}}^{\dagger}a_{{\bf k}_{2}}^{\dagger}a^{\phantom{\dagger}}_{{\bf k}_{3}}-a_{{\bf k}_{3}}^{\dagger}a^{\phantom{\dagger}}_{{\bf k}_{2}}a^{\phantom{\dagger}}_{{\bf k}_{1}})\times\phantom{aa}$ (6) $\displaystyle(\bar{\nu}_{{\bf k}_{1}}+\bar{\nu}_{{\bf k}_{2}})\delta_{{\bf k}_{3},{\bf k}_{1}+{\bf k}_{2}}$ $\displaystyle H_{4}=-\frac{1}{16S}\sum_{{\bf k}_{1},{\bf k}_{2},{\bf k}_{3},{\bf k}_{4}}\delta_{{\bf k}_{3}+{\bf k}_{4},{\bf k}_{1}+{\bf k}_{2}}\,a_{{\bf k}_{1}}^{\dagger}a_{{\bf k}_{2}}^{\dagger}a^{\phantom{\dagger}}_{{\bf k}_{3}}a^{\phantom{\dagger}}_{{\bf k}_{4}}\times\phantom{a}$ (7) $\displaystyle(4(\nu_{{\bf k}_{1}-{\bf k}_{3}}+\nu_{{\bf k}_{2}-{\bf k}_{3}})+\nu_{{\bf k}_{1}}+\nu_{{\bf k}_{2}}+\nu_{{\bf k}_{3}}+\nu_{{\bf k}_{4}}))-$ $\displaystyle 2\,\delta_{{\bf k}_{1}+{\bf k}_{2}+{\bf k}_{3},{\bf k}_{4}}\,(a_{{\bf k}_{1}}^{\dagger}a_{{\bf k}_{2}}^{\dagger}a_{{\bf k}_{3}}^{\dagger}a^{\phantom{\dagger}}_{{\bf k}_{4}}+a_{{\bf k}_{4}}^{\dagger}a^{\phantom{\dagger}}_{{\bf k}_{3}}a^{\phantom{\dagger}}_{{\bf k}_{2}}a^{\phantom{\dagger}}_{{\bf k}_{1}})$ $\displaystyle(\nu_{{\bf k}_{1}}+\nu_{{\bf k}_{2}}+\nu_{{\bf k}_{3}})~{},$ where momentum ${\bf k}$ is defined in the first magnetic BZ. We use the following notations: $\displaystyle A_{\bf k}=1+\nu_{\bf k}/2,~{}~{}B_{\bf k}=-3\nu_{\bf k}/2~{},$ (8) and the momentum dependent functions are $\displaystyle\nu_{\bf k}$ $\displaystyle=$ $\displaystyle\frac{1}{3}(\cos k_{x}+2\cos\frac{k_{x}}{2}\cos\frac{k_{y}\sqrt{3}}{2})~{},$ (9) $\displaystyle\bar{\nu}_{\bf k}$ $\displaystyle=$ $\displaystyle\frac{2}{3}\sin\frac{k_{x}}{2}(\cos k_{x}-\cos\frac{k_{y}\sqrt{3}}{2})~{}.$ (10) The expressions for $H_{3}$ and $H_{4}$ have been obtain first in Ref. chubukov . The essential difference between Eqns. (4) \- (7) and a corresponding expansion around a Neél state on a hypercubic lattice is the occurrence of the term $H_{3}$, which is present due to the non-collinearity of the classical ground state configuration of the THAF. In the remainder of this paper we set the scale of energy to $3J/2=1$, i.e. for $S=1/2$ the prefactor in Eqn. (4) is unity. Figure 2: One magnon dispersion. Top: linear spin-wave dispersion $E_{\bf k}$ of from Eqn. (14). Middle and bottom: real and imaginary part $Re(Im)E^{r}_{\bf k}$ of one magnon dispersion to $O(1/S)$ from a solution of Eqns. (10) - (12) of Ref. chubukov06 on a lattice of $252\times 252$ ${\bf k}$-points with artificial line broadening of $\eta=0.05$. To proceed, we diagonalize the quadratic part of the Hamiltonian $H_{2}$ by a Bogoliubov transformation to a set of magnon quasiparticles $\displaystyle a^{\phantom{\dagger}}_{\bf k}$ $\displaystyle=$ $\displaystyle u_{\bf k}c^{\phantom{\dagger}}_{\bf k}+v_{\bf k}c_{-{\bf k}}^{\dagger}$ (11) $\displaystyle a_{\bf k}^{\dagger}$ $\displaystyle=$ $\displaystyle u_{\bf k}c_{\bf k}^{\dagger}+v_{\bf k}c^{\phantom{\dagger}}_{-{\bf k}}~{},$ where $c^{(\dagger)}_{\bf k}$ are bosons, and the coherence coefficients $\displaystyle u_{\bf k}$ $\displaystyle=$ $\displaystyle\sqrt{\frac{A_{\bf k}+E_{\bf k}}{2E_{\bf k}}}$ (12) $\displaystyle v_{\bf k}$ $\displaystyle=$ $\displaystyle-\frac{B_{\bf k}}{|B_{\bf k}|}\sqrt{\frac{A_{\bf k}-E_{\bf k}}{2E_{\bf k}}}~{}.$ satisfy $u_{\bf k}^{2}-v_{\bf k}^{2}=1$. The Hamitonian $H_{2}$ in terms of the Bogoliubov quasiparticles reads $\displaystyle H_{2}=\sum_{\bf k}E_{\bf k}c_{\bf k}^{\dagger}c_{\bf k}~{},$ (13) and the dispersion is given by $\displaystyle E_{\bf k}=\sqrt{A_{\bf k}^{2}-B_{\bf k}^{2}}=\sqrt{(1-\nu_{\bf k})(1+2\nu_{\bf k})}~{}.$ (14) The magnon dispersion $E_{\bf k}$ is depicted in Fig. 2. It vanishes at the center of the zone, $k_{x}=0,k_{y}=0$, where $\nu_{\bf k}=1$ and at the corners of the BZ, where $\nu_{\bf k}=-1/2$. There are two van-Hove singularities, i.e. at $E=3/(2\sqrt{2})\simeq 1.061$ from the maximum energy and at $E=2/3\simeq 0.6667J$ from the zone-boundary. To treat the interaction between magnons we need to express the triplic and quartic part of the Hamiltonian, $H_{3}$ and $H_{4}$, in terms of the quasiparticles $c^{(\dagger)}_{\bf k}$ using the transformation of Eqn. (11). For the triplic part we obtain $\displaystyle H_{3}=\sum_{{\bf k},{\bf p}}\,\left(c_{{\bf k}}c_{{\bf p}}^{\dagger}c_{{\bf k}-{\bf p}}^{\dagger}\,f\left({\bf k},{\bf p}\right)+\right.\phantom{aaaaaa}$ $\displaystyle\left.c_{{\bf k}}c_{{\bf p}}^{\dagger}c^{\phantom{\dagger}}_{-{\bf k}+{\bf p}}\,g\left({\bf k},{\bf p}\right)+O(c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}+h.c.)\right)~{}.$ (15) Terms with three creation(destruction) operators are present in principle, but are not expressed explicitly for notational simplicity. As will be come clear in section IV they play no role in evaluating the magnon interactions within the Raman response. Moreover $\displaystyle f({\bf k},{\bf p})$ $\displaystyle=$ $\displaystyle i\sqrt{3}(\bar{\nu}_{{\bf p}}(u_{{\bf k}}u_{{\bf{\bf k}}-{\bf p}}+v_{{\bf k}}v_{{\bf k}-{\bf p}})(u_{{\bf p}}+v_{{\bf p}})-$ (16) $\displaystyle\bar{\nu}_{{\bf k}}(u_{{\bf k}}+v_{{\bf k}})(u_{{\bf p}}v_{{\bf{\bf k}}-{\bf p}}+u_{{\bf k}-{\bf p}}v_{{\bf p}})+$ $\displaystyle\bar{\nu}_{{\bf k}-{\bf p}}(u_{{\bf k}-{\bf p}}+v_{{\bf k}-{\bf p}})(u_{{\bf k}}u_{{\bf p}}+v_{{\bf k}}v_{{\bf p}}))~{},$ $\displaystyle g({\bf k},{\bf p})$ $\displaystyle=$ $\displaystyle i\sqrt{3}(\bar{\nu}_{{\bf p}}(u_{{\bf{\bf k}}-{\bf p}}v_{{\bf k}}+u_{{\bf k}}v_{{\bf k}-{\bf p}})(u_{{\bf{\bf p}}}+v_{{\bf p}})+$ (17) $\displaystyle\bar{\nu}_{{\bf k}-{\bf p}}(u_{{\bf k}-{\bf p}}+v_{{\bf{\bf k}}-{\bf p}})(u_{{\bf k}}u_{{\bf p}}+v_{{\bf k}}v_{{\bf p}})-$ $\displaystyle\bar{\nu}_{{\bf{\bf k}}}(u_{{\bf k}}+v_{{\bf{\bf k}}})(u_{{\bf k}-{\bf p}}u_{{\bf p}}+v_{{\bf k}-{\bf p}}v_{{\bf p}}))~{}.$ For the quartic part we obtain $\displaystyle H_{4}=-\frac{1}{16S}\sum_{{\bf k},{\bf p}}\,h({\bf k},{\bf p})\,c_{{\bf k}}^{\phantom{\dagger}}c_{-{\bf k}}^{\phantom{\dagger}}c_{{\bf p}}^{\dagger}c_{-{\bf p}}^{\dagger}$ $\displaystyle+O(c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}+c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}c^{\dagger}_{\phantom{\bf k}}c^{\phantom{\dagger}}_{\phantom{\bf k}}+h.c.)$ $\displaystyle~{},$ (18) where again terms irrelevant for the Raman scattering are not displayed explicitly and $\displaystyle h({\bf k},{\bf p})=2((u_{{\bf k}}^{2}u_{{\bf p}}^{2}+v_{{\bf k}}^{2}v_{{\bf p}}^{2})(\nu_{{\bf k}}+4\nu_{{\bf k}-{\bf p}}+\nu_{{\bf p}})-3(u_{{\bf k}}^{2}+$ $\displaystyle v_{{\bf k}}^{2})u_{{\bf p}}v_{{\bf p}}(2\nu_{{\bf k}}+\nu_{{\bf p}})-3(u_{{\bf p}}^{2}+v_{{\bf p}}^{2})u_{{\bf k}}v_{{\bf k}}(\nu_{{\bf k}}+2\nu_{{\bf p}})\phantom{aa}$ $\displaystyle+4u_{{\bf k}}v_{{\bf k}}u_{{\bf p}}v_{{\bf p}}(2+\nu_{{\bf k}}+\nu_{{\bf p}}+2\nu_{{\bf k}+{\bf p}}))\phantom{aaaa}$ $\displaystyle~{}.$ (19) Eqns. (II) – (II) allow to construct all vertices relevant to the final state two-magnon interactions in the Raman scattering. Apart from that Eqns. (4) – (10) can be used to derive the one-magnon selfenergy to $O(1/S)$. This has been done in Ref. chubukov, , to which we refer the reader for details. For the purpose of the present work it is sufficient to employ Eqns. (10) - (12) from Ref.chubukov, to calculate the renormalized magnon dispersion $E^{r}_{\bf k}$ to $O(1/S)$. Fig. 2 (middle and bottom panel) shows the result of such calculations. It is evident that in the real part of the magnon energy, the interactions lead to extended and almost flat regions with a shallow ’roton’-like minimum along the BZ faces. Moreover, as the $ImE^{r}_{\bf k}$ almost vanishes at these regions, the life time of a quasi- particle with corresponding momenta is very large. On the other hand quasi- particles with near-maximum energies are located in the momentum regions of rather large damping. ## III Loudon-Fleury Vertex We use the framework of the Loudon-Fleury (LF) model for the interaction of light with spin degrees of freedom for the calculation of the two-magnon Raman scattering. The LF vertex has the form $\displaystyle\begin{array}[]{l}R=\sum_{i\delta}(\hat{\varepsilon}_{\rm in}\cdot{\bf\delta})(\hat{\varepsilon}_{\rm out}\cdot{\bf\delta})\tilde{{\bf S}}_{i}\tilde{{\bf S}}_{i+\delta}~{},\end{array}$ (21) where the polarizations $\hat{\varepsilon}_{\rm in}=\cos\theta\hat{x}+\sin\theta\hat{y}$ and $\hat{\varepsilon}_{\rm out}=\cos\phi\hat{x}+\sin\phi\hat{y}$ of the incoming and the outgoing light are determined by angles $\theta$ and $\phi$, defined with respect to the $x$-axis. To derive the final form of the scattering LF vertex, we first write spin operators in terms of Holstein-Primakoff bosonic $a$-operators (3), and then express the latter in terms of the boson quasi-particle operators $c$. We get the following expression $\displaystyle\begin{array}[]{l}R=\sum_{k}M_{\bf k}(c_{\bf k}c_{-{\bf k}}+c_{\bf k}^{\dagger}c_{-{\bf k}}^{\dagger})\equiv r^{-}+r^{+}~{},\end{array}$ (23) where $M_{\bf k}$ is given by $\displaystyle\begin{array}[]{ll}M_{\bf k}=&(F_{1}(\theta,\phi)+F_{2}({\bf k},\theta,\phi))u_{\bf k}v_{\bf k}-\\\ &\frac{3}{4}F_{2}({\bf k},\theta,\phi)(u_{\bf k}^{2}+v_{\bf k}^{2})~{},\end{array}$ (26) and we have introduced the following notations: $\displaystyle\begin{array}[]{ll}&F_{1}(\theta,\phi)=2S\sum_{\mu=1}^{3}f_{\mu}(\theta,\phi)~{},\\\\[5.69046pt] &F_{2}({\bf k},\theta,\phi)=2S(f_{3}(\theta,\phi)\cos k_{x}+\\\ &f_{1}(\theta,\phi)\cos(\frac{k_{x}}{2}+\frac{\sqrt{3}}{2}k_{y})+f_{2}(\theta,\phi)\cos(\frac{k_{x}}{2}-\frac{\sqrt{3}}{2}k_{y}))~{},\\\\[5.69046pt] &f_{\mu}(\theta,\phi)=\hat{\varepsilon}_{\rm in}\cdot\vec{\delta_{\mu}}~{}.\end{array}$ (31) In principle the Raman vertex contains also $c^{\dagger}_{\bf k}c_{\bf k}$ terms. However, at zero momentum and to lowest order in $1/S$ these terms do not contribute to the Raman response at finite frequency, and we dropped them. Note that $R$ is explicitly Hermitian. Figure 3: Diagrams for the Raman intensity: a) Bare Raman vertex $R$ from Eqn. (23); b) Raman susceptibility (both bare, $G^{0}$, and dressed, $G$, magnon propagators will be considered (see text). c) The integral equation for the dressed Raman vertex $\Gamma$ in terms of the irreducible magnon particle- particle (IPP) vertex $\gamma$. d) Leading order $1/S$ contributions to the IPP vertex. ## IV Raman Intensity We now calculate the Raman intensity including one- and two-magnon renormalizations up to $O(1/S)$. The Raman intensity $I(\Omega)$ is obtained via Fermi’s golden-rule from $I(\omega_{n})=const\times Im[\int_{0}^{\beta}d\tau\,e^{i\omega_{m}\tau}\left\langle T_{\tau}(R(\tau)R)\right\rangle]$ (32) by analytic continuation of the bosonic Matsubara frequencies $\omega_{m}=2\pi mT$ onto the real axis as $i\omega_{m}\rightarrow\Omega+i\eta$, where $\Omega=\omega_{\rm in}-\omega_{\rm out}$ refers to the inelastic energy transfer by the photon, and for the remainder of this paper we assume the temperature $T=1/\beta$ to be zero. The prefactor ’$const$’ refers to some arbitrary units by which the observed intensities are scaled. The role of interactions is summarized in Fig. 3. Two effects have to be distinguished, namely renormalizations of the one-magnon propagators, i.e. $G^{0}\rightarrow G$, and vertex corrections to the Raman intensity (final state interactions), i.e $R\rightarrow\Gamma$. All propagators are expressed in terms $c_{k}^{(\dagger)}$-type of Bogoliubov particles. To orders higher than $O((1/S)^{0})$ the propagators of these particles are not diagonal i.e. both normal $G_{cc}({\bf k},\tau)=-\langle T_{\tau}(c_{{\bf k}}(\tau)c_{{\bf k}}^{\dagger})\rangle$ and anomalous $D_{cc}({\bf k},\tau)=-\langle T_{\tau}(c_{{\bf k}}(\tau)c_{-{\bf k}})\rangle$ propagators do occur. However, anomalous propagators are smaller by one factor of $1/S$ as compared to the normal propagators and, therefore, can be neglected. To first order in $1/S$, the normal propagators read $G({\bf k},i\omega_{n})=1/(i\omega_{n}-E^{r}_{\bf k})$, i.e. the quasi-particle residue remains unity, and $E^{r}_{\bf k}$ is taken from Eqns. (10) - (12) of Ref. chubukov06, . In order to evaluate Raman intensity (Eqn. (32)) we have to calculate the Raman susceptibility $\left\langle T_{\tau}(R(\tau)R)\right\rangle$. In principle, the latter can contain terms of type $\left\langle T_{\tau}(r^{-}(\tau)r^{-})\right\rangle$, with $r^{\pm}$ specified in Eqn. (23) and Fig. 3 a). However, these terms need at least one $O(1/S)$ interaction-event to occur or require anomalous propagators, i.e., they are smaller by one order of $1/S$ and will be dropped. In the following we consider only $\left\langle T_{\tau}(r^{+}(\tau)r^{-})\right\rangle+\left\langle T_{\tau}(r^{-}(\tau)r^{+})\right\rangle$. By Hermitian conjugation, it is sufficient to calculate $J(\tau)=\left\langle T_{\tau}(r^{-}(\tau)r^{+})\right\rangle$, which is depicted in Fig. 3 b). Fig. 3 b) shows the two-particle reducible Raman vertex $\Gamma({\bf k},\omega_{n},\omega_{m})$, which includes a series of magnon-magnon interaction events. It satisfies the Bethe-Salpeter equation expressed in terms of the two-particle irreducible vertex $\gamma$, depicted in Fig. 3 c) $\displaystyle\Gamma({\bf k},\omega_{n},\omega_{m})=r^{-}({\bf k})+\sum_{{\bf p},\omega_{o}}\gamma({\bf k},{\bf p},\omega_{n},\omega_{o})$ $\displaystyle G({\bf p},\omega_{o}+\omega_{m})G(-{\bf p},-\omega_{o})\Gamma({\bf p},\omega_{o},\omega_{m})$ (33) In this work we consider only the leading order contributions in $1/S$ to $\gamma$. They are shown in Fig. 3 d). The quartic vertex $\gamma_{4}({\bf k},{\bf p})$ is identical to the two-particle-two-hole contribution from $H_{4}$ of Eqns.(II)-(II) and reads $\displaystyle\begin{array}[]{l}\gamma_{4}({\bf k},{\bf p})=-\frac{1}{2S}h({\bf k},{\bf p})\end{array}$ (35) The two addends forming the irreducible vertex $\gamma_{3}({\bf k},{\bf p},\omega_{n},\omega_{o})$ are assembled from $H_{3}$ and one intermediate propagator, and can be written as $\displaystyle\gamma_{3}({\bf k},{\bf p},\omega_{n},\omega_{o})=\frac{1}{2S}\sum_{{\bf k},{\bf p}}\,(f({\bf k},{\bf p})g(-{\bf k},-{\bf p})\times$ $\displaystyle G^{0}({\bf k}-{\bf p},i\omega_{o}-i\omega_{n})c_{\bf k}^{\phantom{\dagger}}c_{-{\bf k}}^{\phantom{\dagger}}c_{\bf p}^{\dagger}c_{-{\bf p}}^{\dagger}+\phantom{aaa}$ (36) $\displaystyle g({\bf k},{\bf p})f(-{\bf k},-{\bf p})G^{0}({\bf p}-{\bf k},i\omega_{n}-i\omega_{o}))c_{\bf k}^{\phantom{\dagger}}c_{-{\bf k}}^{\phantom{\dagger}}c_{\bf p}^{\dagger}c_{-{\bf p}}^{\dagger}$ $\displaystyle~{},$ where the functions $f({\bf k},{\bf p})$ and $g({\bf k},{\bf p})$ obey the symmetry relation $f[g](-{\bf k},-{\bf q})=-f[g]({\bf k},{\bf q})$. To keep $\gamma_{3}$ to leading order in $1/S$ we retain only the zeroth order propagators $G^{0}$ for each intermediate line. In principle, $H_{3}$ allows for an additional two-particle irreducible graph, with the incoming(outgoing) legs placed into the particle-particle(hole-hole) channel and one intermediate line at zero momentum and frequency. However, we verified that these contributions vanish exactly. Due to $\gamma_{3}$, Eqn. (IV) is an integral equation with respect to both, momentum and frequency. This is the first major difference to Raman scattering from collinear HAFs, where only $\gamma_{4}$ exists at $O(1/S)$. To proceed, further approximations have to be made. Here we simplify $\gamma_{3}$ by assuming the dominant contribution from the frequency summations to result from the mass-shell of the propagators in the intermediate particle-particle reducible sections of Fig. 3c) $\displaystyle-i\omega_{n}$ $\displaystyle\approx$ $\displaystyle E_{\bf k}$ $\displaystyle-i\omega_{o}$ $\displaystyle\approx$ $\displaystyle E_{\bf p}~{}.$ (37) This approximation will be best for sharp magnon lines and the transferred frequencies $i\omega_{m}$ close to the van-Hove singularities of $2E_{\bf k}$. In this approximation for $\gamma_{3}$, the two-particle irreducible vertex $\gamma$ simplifies to $\displaystyle\gamma({\bf k},{\bf p},\omega_{n},\omega_{o})\approx\gamma({\bf k},{\bf p})=\gamma_{3}({\bf k},{\bf p})+\gamma_{4}({\bf k},{\bf p})$ (38) $\displaystyle=-\frac{1}{2S}\left(\frac{2E_{{\bf k}-{\bf p}}f({\bf k},{\bf p})g({\bf k},{\bf p})}{(E_{\bf k}-E_{\bf p})^{2}-E^{2}_{{\bf k}-{\bf p}}}+h({\bf k},{\bf p})\right)~{}.$ Now we can perform the frequency summation over $\omega_{o}$ on the right hand side of Eqn. (IV) as well as the analytic continuation $i\omega_{m}\rightarrow\Omega+i\eta\equiv z$. With this $\Gamma$ in the latter equation turns into a function of ${\bf p}$ and $z$ only, leading to $\displaystyle\sum_{\bf p}L_{{\bf k},{\bf p}}(z)\Gamma_{\bf p}(z)=r^{-}({\bf k})$ (39) $\displaystyle L_{{\bf k},{\bf p}}(z)=\delta_{{\bf k},{\bf p}}-\frac{\gamma({\bf k},{\bf p})}{z-2E^{(r)}_{\bf p}}~{},$ (40) which is an integral equation with respect to momentum only. In the rest of the paper the superscript ’$r$’ refers to the case when renormalized propagators with $E^{r}_{\bf p}$ are taken in the two-particle reducible part of the Raman intensity, while $E_{\bf p}$ corresponds to the usage of bare propagators. Close inspection of the vertex $\gamma({\bf k},{\bf p})$ shows, that it does not separate into a finite sum of products of lattice harmonics of the triangular lattice. Therefore, Eqns. (39)-(40) cannot be solved algebraically in terms of a finite number of scattering channels, but require a numerical solution. On finite lattices this can be done by treating Eqn. (39) as a linear equation for $\Gamma_{\bf p}(z)$ at fixed $z$. This marks another significant difference between Raman scattering from collinear and non- collinear antiferromagnets. Figure 4: Raman intensity neglecting final state interactions, i.e. replacing $\Gamma$ by $M$ in eqn. (42). Scattering geometry: $\phi=\theta=0$. Number of k-points: $N\times N$. Dashed (BB): bubble using bare magnon energies $E_{\bf k}$ from eqn. (14), shown in top panel of fig. 2. Imaginary broadening $\eta=0.003$ choosen such as to retain visible but small finite-size oscillations. Solid (RB): bubble using renormalized magnon energies $E^{r}_{\bf k}$ to $O(1/S)$ obtained from eqns. (10) - (12) of ref. chubukov06 and shown in the middle and lower panel of fig. 2. Finite-size oscillations are suppressed by $Im[E^{r}_{\bf k}]$. The absolute scale of $I(\Omega)$ is set to unity, but the relative scale of BB and RB is kept. Finally, the expression for the Raman intensity from Eqn. (32) can be written as $\displaystyle I(\Omega)$ $\displaystyle=$ $\displaystyle const\times(J(\Omega)-J(-\Omega))$ (41) $\displaystyle J(\Omega)$ $\displaystyle=$ $\displaystyle Im\,[\sum_{\bf k}\frac{M_{\bf k}\,\Gamma_{\bf k}(\Omega+i\eta)}{\Omega+i\eta-2E^{(r)}_{\bf k}}]~{}.$ (42) We now discuss the Raman intensity for several levels of approximations. First, we neglect final state interactions and set $\Gamma_{\bf k}(z)\rightarrow M_{\bf k}$. Fig. 4 shows the Raman intensity as a function of the transferred photon frequencies $\Omega$ for this case. This figure contrasts the Raman bubble with bare propagators against that with renormalized ones. Such results can be obtained on fairly large lattices, since they do not involve a solution of the integral equation (39), but only a calculation of the one magnon self-energy chubukov06 . We keep the shift $i\eta$ off the real axis deliberately small in this figure, in order to discriminate its effect from that of the actual life-time broadening due to the imaginary part of $E^{r}_{\bf k}$. First we would like to note that we find the line shape to be insensitive to the scattering geometry. This is in a sharp contrast to Raman scattering from the square lattice HAF, where Raman amplitudes in $A_{1g}$, $B_{1g}$ and $B_{2g}$ symmetries are very different. In case of the bare Raman bubble, one can see two well-defined peaks, one at energy $\Omega=3/\sqrt{2}$ and one at $\Omega=4/3$ \- both in units of $3J/2$. These energies correspond to $2$ times that of the maximum and of the BZ- boundary saddle-point of the classical spin-wave spectrum $E_{\bf k}$. Clearly, the dominant spectral weight does not stem from the k-points at the upper cut-off of the linear spin-wave energy but from the BZ-boundary. This does not reflect the bare two-magnon density of states but is an effect of the Raman matrix element $M_{\bf k}$, which samples the BZ regions preferentially. Switching on $1/S$ corrections, two modifications of the intensity occur. First, both maxima are shifted downwards by a factor of $\sim 0.7$ due to the corresponding renormalizations of the one magnon energies. Second, as the BZ- boundary saddle-point of $E_{\bf k}$ has turned into a flat region, occupying substantial parts of the BZ for $E^{r}_{\bf k}$, the intensity of the lower energy peak is strongly enhanced due to the very large density of one-magnon states. Equally important, the imaginary part of $E^{r}_{\bf k}$ is finite in the BZ-region which corresponds to the maximal one-magnon energies. This smears the peak at the upper frequency cut-off in $I(\Omega)$ almost completely - as can seen from the solid line in Fig. 4. In contrast to that, $Im[E^{r}_{\bf k}]$ almost vanishes in the flat regions on the BZ-boundary due to phase-space constraints mike , leading to a further relative enhancement of the intensity from there. Figure 5: Effect of final state interactions on Raman intensity. Scattering geometry: $\phi=\theta=0$. Number of k-points: $N\times N$. Imaginary shift off real axis is $\eta=0.03$. Dashed line (RB): replacing $\Gamma$ by $M$ in Eqn. (42) and using bubble with renormalized magnon energies $E^{r}_{\bf k}$ to $O(1/S)$ obtained from Eqns. (10) - (12) of Ref. chubukov06 . Solid line (RBV): using dressed vertex $\Gamma$ obtained from Eqn. (39) in Eqn. (42) and renormalized magnon energies $E^{r}_{\bf k}$. The absolute scale of $I(\Omega)$ is set to unity, but the relative scale of RB and RBV is kept. Next we turn to final-state interactions. In Fig.5, we compare $I(\Omega)$ from the Raman bubble obtained with propagators renormalized to $O(1/S)$ and only bare Raman vertices to the intensity obtained by including also the dressed Raman vertex $\Gamma_{\bf k}(z)$ from Eqn. (39). The numerical solution of the latter equation requires some comments. Since the kernel $L_{{\bf k},{\bf p}}(z)$ is not sparse and has rank $N^{2}\times N^{2}$, already moderate lattice sizes lead to rather large dimensions and storage requirements for the linear solver. We have chosen $N=69$, leading to a 4761$\times$4761 system which we have solved 200 times to account for 200 frequencies in the interval $\Omega\in[0,2.5]$. The kernel has points and lines of singular behavior in $({\bf k},{\bf p})$-space, which stem from the singularities of the Bogoliubov factors $u[v]({\bf k})$ in $f[g][h]({\bf k},{\bf p})$ and from the energy denominators in Eqn. (38). In principle, such regions are of measure zero with respect to the complete $({\bf k},{\bf p})$-space, yet we have no clear understanding of their impact on a solution of Eqn. (39) as $N\rightarrow\infty$. In our case, i.e. a finite lattice, we have chosen to regularize these points and lines by cutting off eventual singularities in $L_{{\bf k},{\bf p}}$. The comparatively small system sizes require a rather larger artificial broadering $i\eta$ in order to achieve acceptably smooth line shapes. This can be seen by contrasting the dashed curve in Fig. 5 and solid curve in Fig.4, which correspond to identical quantities, however for different finite systems, $252\times 252$ vs $69\times 69$. The main message put forward by Fig. 5 is that the final state interactions lead to a flattening of the peak from the two-magnon density of states, transforming it to a rather broad Raman continuum. This can be understood, at least qualitatively, from the RPA-like functional form of the Bethe-Salpeter equation. Discarding momentum dependencies and iterating the two-magnon bubble times the irreducible vertex $\gamma$, leads to a renormalization of the intensity by a factor, roughly of the form $\sim 1/(1-\gamma\cdot\rho(\Omega))$, where $\rho(\Omega)$ refers to the two-magnon bubble. Directly at the peak position of the Raman bubble this renormalization factor may get small, thereby suppressing the over-all intensity. While exactly the same mechanism is at work also for the square lattice HAF, its impact on the spectrum is complete different. In the latter case the peak- intensity without final state interactions is at the upper cut-off of the Raman intensity. Suppression of this peak-intensity simply shifts the maximum intensity to lower frequencies within the Raman spectrum. This shift is then interpreted in terms of a two-magnon binding energy. Such reasoning cannot be pursued in the present case. ## V Conclusion and Discussion To summarize, we have investigated magnetic Raman scattering from the two- dimensional triangular Heisenberg antiferromagnet considering various levels of approximation within a controlled $1/S$-expansion. Our study has revealed several key differences as compared to the well-known magnetic Raman scattering from the planar square lattice spin-$1/2$ antiferromagnet. First, we found that the intensity profile is insensitive to the in-plane scattering geometry of the incoming and outgoing light at $O(1/S)$. This has to be contrasted against the clear difference between A1g and B1g,2g symmetry for the square-lattice case. Second, on the level of linear spin-wave theory, we showed that the Raman intensity has two van-Hove singularities. The less intensive peak is located at the upper edge of the two-magnon density of states and stems from twice the maximum of the one-magnon energy. This is similar to the square lattice case. However, the dominant peak is located approximately in the center of the two- magnon density of states. This peak stems from the Loudon-Fleury Raman-vertex strongly selecting the Brillouin zone boundary regions where the one-magnon dispersion on the triangular lattice has an additional weak van-Hove singularity. This is absent on the square lattice. Next, we calculated the Raman intensity with the one-magnon spectrum, renormalized to $O(1/S)$, however neglecting final-state interactions within the Raman process. In this case we have obtained a sharp and almost $\delta$-functional Raman peak at energy $\sim 3J/2$. At this energy the real part of the renormalized one-magnon dispersion shows a large plateau-region at the Brillouin zone boundary with a roton-like shallow minimum. Moreover, due to phase-space constraints the one-magnon life-time is large in this region. Therefore, the two-magnon density of states in this region is strongly enhanced, as compared to the linear spin-wave result. In contrast to that, the intensity at the upper edge of the spectrum is suppressed further, since the $O(1/S)$ corrections lead to the significant one-magnon damping. Finally the overall width of the spectrum is reduced by a factor of approximately $0.7$. In a last step, we considered the impact of the final-state interactions to $O(1/S)$. Due to the non-collinear ordering on the triangular lattice, and in sharp contrast to the square-lattice case, we find, that even to lowest order the two-magnon scattering is neither instantaneous in time, nor separable in momentum space. Our solution of the corresponding Bethe-Salpeter equation reveals a broad continuum-like Raman profile which results from a smearing of the intensity of the two-roton peak by virtue of repeated two-magnon scattering. While, at this order in $1/S$ the over-all form of the Raman profile is reminiscent of that on the square-lattice, one has to keep in mind, that in the latter case the position of the maximum in the center of the Raman continuum has to interpreted rather differently, namely in terms of a two- magnon binding effect. In conclusion we hope that our theoretical investigation will stimulate further experimental analysis of triangular, and more generally frustrated magnetic systems by Raman scattering. Several novel materials with triangular structure have been investigated thoroughly over the last few years, among them the cobaltites, NaxCoO2 cob , and the spatially anisotropic triangular antiferromagents Cs2CuCl4 Cs and $\kappa$-(BEDT-TTF)2Cu2(CN)3 Shimizu2003 . To our knowledge however, magnetic Raman scattering on such systems remains a rather open issue. ## VI Acknowledgements We would like to thank A. Chubukov for useful discussions. One of us (W.B.) acknowledges partial support by the DFG through Grant No. BR 1084/4-1 and the hospitality of the KITP, where this research was supported in part by the NSF under Grant No. PHY05-51164. ## References * (1) Elliot, R. J., and R. Loudon, Phys. Lett. 3A, 189 (1963) * (2) P. A. Fleury, S. P. S. Porto, L. E. Cheesman, and H. J. Guggenheim, Phys. Rev. Lett. 17, 84 (1966) * (3) J. B. Parkinson, J. Phys. C 2, 2012 (1969) * (4) P. A. Fleury and H. J. Guggenheim, Phys. Rev. Lett. 24, 1346 (1970) * (5) K. B. Lyons, P. A. Fleury, J. P. Remeika, A. S. Cooper, and T. J. Negran, Phys. Rev. B 37, 2353 (1988); * (6) S. Sugai, S. I. Shamoto, and M. Sato, Phys. Rev. B 38, 6436 (1988); * (7) P. E. Sulewski, P. A. Fleury, K. B. Lyons, and S.-W. Cheong, Phys. Rev. Lett. 67, 3864 (1991) * (8) R. R. P. Singh, P. A. Fleury, K. B. Lyons, and P. E. Sulewski Phys. Rev. Lett. 62, 2736 (1989); * (9) C. M. Canali and S. M. Girvin, Phys. Rev. B 45, 7127 (1992) * (10) A. V. Chubukov, D. M. Frenkel, Phys. Rev. Lett. 74, 3057 (1995); * (11) F. Nori, R. Merlin, S. Haas, A. W. Sandvik, and E. Dagotto, Phys. Rev. Lett. 75, 553 (1995) * (12) G. Blumberg, M. Kang, M. V. Klein, K. Kadowaki, and C. Kendziora, Science 278, 1427 (1997); * (13) T. P. Devereaux, R. Hackl, Rev. Mod. Pphys. 79, 175 (2007) * (14) P. A. Fleury and R. Loudon, Phys. Rev. 166, 514 (1968) * (15) B. S. Shastry and B. I. Shraiman, Phys. Rev. Lett. 65, 1068 (1990). * (16) A. V. Chubukov, S.Sachdev and T. Senthil, J. Phys.: Condens. Matter 6, 8891 (1994) * (17) O.A. Starykh, A. V. Chubukov and A. G. Abanov, Phys. Rev. 74, 180403 (R) (2006) * (18) A. L. Chernyshev and M.E. Zhitomirsky, Phys. Rev. Lett 97, 207202 (2006) * (19) W. H. Zheng, J. O. Fjaerestad, R. R. P. Singh, R. H. McKenzie, R. Coldea, Phys. Rev. B 74, 224420 (2006) * (20) F. Vernay, T.P. Devereaux and M. J. P. Gingras, J. Phys.: Condens. Matter 19, 145243 (2007) * (21) T. Jolicoeur, J. C. Leguillou, Phys. Rev. B 40, 2727 (1989). * (22) K. Takada, H. Sakurai, E. Takayama-Muromachi, F. Izumi, R. A. Dilanian, and T. Sasaki, Nature 422, 53 (2003). * (23) R. Coldea, D. A. Tennant, K. Habicht, P. Smeibidl, C. Wolters, and Z. Tylczynski, Phys. Rev. Lett, 88, 137203, (2002). * (24) Y. Shimizu, K. Miyagawa, K. Kanoda, M. Maesato, and G. Saito, Phys. Rev. Lett. 91, 107001 (2003)

arxiv-papers

2024-09-04T02:48:54.085557

0803.0186

§ INTRODUCTION One of the most fascinating puzzles in the meson mass spectrum is $U(1)_A$ problem: why the mass of the flavor singlet pseudoscalar meson, $\eta'$, is so heavier, $\metap=957.78(14)$ MeV, than that of its flavor nonsinglet counterparts in nature, $m_{\pi^{0}}=134.9766(6)$ MeV, $m_{K^{0}}=497.648(22)$ MeV, and $m_\eta=547.51(18)$ MeV[1]. Nonsinglet mesons behave as Nambu-Goldstone (NG) bosons with the spontaneous breaking of $SU(3)_{A}$ symmetry in the quark massless limit $(m_{\rm quark}\to0)$, ignoring the small QED effects whereas $\eta'$ is not an NG boson because $U(1)_{A}$ symmetry is broken by the quantum effect, the $U(1)_{A}$ anomaly. The nonvanishing divergence of the flavor singlet axial current, ${\cal A}_\mu^0(x)$, in the axial Ward-Takahashi identity (AWTI) occurs for an operator ${\cal O}$ in the case of degenerate quarks up to the contact term, \begin{eqnarray} \partial_{\mu} \vev{{\cal{A}}_{\mu}^{b}(x) {\cal O}} = 2m_{\rm quark}\vev{P^{a}(x) {\cal{O}}} + \delta_{b,0} {2N_f} \vev{\rho_\text{top} (x) {\cal O}}, \label{eq:flavor-singlet-AWTI} \end{eqnarray} and expresses the anomalous breaking of chiral symmetry in the last term, which is proportional to the topological charge density, $\rho_\text{top}(x)$. For a sufficiently smooth gauge field, \begin{eqnarray} \rho_\text{top}(x)= {1\over 32\pi^2} \epsilon_{\mu\nu\rho\sigma} \text{tr} F_{\mu\nu}F_{\rho\sigma}(x). \end{eqnarray} The difference in the pseudoscalar meson masses between the flavor singlet sector, $m_{\eta'}$, and the non-singlet sector,$m_\pi$, was estimated by the Witten-Veneziano (WV) relation,[2, 3], \begin{eqnarray} m_{0}^{2}=m_{\eta'}^{2}-m_{\pi}^{2}={2N_{f}\over f_{\pi}^{2}} \chi_\text{top} \label{eq:witten-veneziano} \end{eqnarray} at the limit of $N_{c}\to\infty$. Here $\chi_\text{top}$ is the susceptibility of the topological charge ($Q_\text{top}$) : \begin{eqnarray} \chi_\text{top}={\vev{Q_\text{top}^2}\over VT},~~ Q_\text{top} = \int \rho_\text{top}(x)\, d^4x,~~ \label{eq:Qtop} \end{eqnarray} in pure Yang-Mills (YM) theory in a four dimensional volume $VT$. A recent result in $N_c=3$ YM theory with the overlap fermion[4] shows that $\chi = (191(5) {\rm MeV})^{4}$. ${\eta'}$ mass from this estimation for $N_f=3$ and in the chiral limit, $\mpi^{2}\to0$, is $\metap\approx970$ MeV, which is very close to experimental values. The direct numerical calculation of the $\eta'$ spectrum is important for checking theoretical scenarios such as the WV relation, and should result in its correction in finite $N_c$ and nonzero quark masses. Simulations of ${\eta'}$ physics in pure YM theory with quenched Wilson fermions were carried out in pioneer works[5, 6]. The relation between the topological charge and the mass of $\eta'$ was also explored[7]. Unquenched simulations[8, 9, 10, 11, 12] were performed for two-flavor and for 2+1-flavor[13] of Wilson fermions. Using staggered fermions, $m_{\eta'}$ has been calculated for $N_f=0,2$[14] and $N_f=2+1$[47, 48]. Recently there are other interesting investigations, such as using twisted-mass quarks [49] or a local imaginary $\theta$-term [50]. In this paper, we discuss the mass of $\eta'$ in $N_{f}=2$ QCD with domain wall fermions (DWF). DWF [18, 19, 20] is one of the lattice chiral fermions, which has both flavor and chiral symmetries even at finite lattice spacing $(a>0)$, and is thus suitable for investigation of nonperturbative physics of chiral anomalies. These features of DWF make their use preferable to the other alternative methods of discretization. Wilson fermions break chiral symmetry at $a>0$ and discretization errors start at ${\cal O}(a\Lambda_\text{\rm QCD})$. The singlet flavor meson in staggered fermions is a very important subject as it may be related to the potential issue about the locality of the formalisms in the continuum limit[16, 17]. Chiral and flavor symmetry are particularly important for $\eta'$ physics, and the DWF is the natural choice of lattice quark in investigations. Chiral symmetry in a DWF is not realized perfectly, it is broken due to its finite extent in the fifth direction, $L_s$. The amount of breaking can be measured by a shift in quark mass: $m_\text{quark}= m_f+m_\text{res}$, so that the nonsinglet axial current is conserved at $m_\text{quark}=0$. $m_\text{res}$ is called the residual quark mass and vanishes at large $L_s$ for a sufficiently smooth gauge configuration[21]. Although it is desirable to take $L_s\to\infty$ limit, to reduce the computational cost, we restrict ourselves to finite $L_s=12$ with the combination of DBW2 improved gauge action [22, 23], which smoothen gauge field at short distance and reduces $\mres$ significantly [24]. The RBC collaboration examined the first large scale dynamical DWF simulation Pseudoscalar meson masses and decay constants were computed and fit to the chiral perturbation theory (ChPT) formula. $m_{\pi}$, $m_{K}$, $m_{\rho}$, $f_{\pi}$, and $f_{K}$ calculated in their work are reasonably consistent with values obtained in experiments. The $J$ parameter is closer to the phenomenological value than the value obtained in the quenched simulation. The nonsinglet scalar meson, $a_{0}$, mass and the decay constant have also been examined both in dynamical QCD and partially quenched QCD using partially quenched ChPT [26]. We will mainly focus on the $\eta'$ meson in this paper, but we will also report on the results of other mesons belong to other Lorentz and flavor representations, and also investigate the signal of the mesons in their excited state and their decay constants. As the results are limited to the isospin symmetric case and the number of dynamical quarks is two, our focal interest in this paper is to provide a benchmark calculation for the study of the general meson spectrum on a dynamical DWF ensemble with various (smeared) meson field using a larger statistical sample than in the previous study. In 2, the theoretical expectations of $\eta'$ meson physics are summarized. We explain the details of the simulation including improvements in the signal-to-noise ratio and the fitting methods used to relate the simulation data to physical quantities in 3. The numerical results are presented in 4 with a list of their systematic uncertainties. We will summarize in 5. § THEORETICAL RESULTS ON PHYSICS OF FLAVOR SINGLET MESON In (continuum Euclidean) QCD with $N_{f}$ degenerated quarks, the operator of the flavor singlet pseudoscalar meson, $\eta'$: $I(J^{P})=0(0^{-})$, is defined by quark operators, $q_{f}$, as \begin{equation} \eta'(x)={1\over\sqrt {N_{f}}}\sum_{f=1}^{N_{f}} \bar q_{f}(x)i\gamma_{5}q_{f}(x), \label{eq:eta_operator} \end{equation} where $f=1, ..., N_{f}$ is the flavor index. The $\eta'$ propagator consists of two parts: \begin{eqnarray} &&\int d^3x \langle \eta'(\vec{x},t) \eta'^{\dag}(\vec{0},0)\rangle =C_{\gamma_5}(t) - N_{f}D_{\gamma_5}(t), \label{eq:eta_def}\\ &&C_{\gamma_5}(t)= - \int d^3x \left\langle{{1\over N_{f}} \sum_{f}^{N_{f}}\overbrace{\bar q_{f}(\vec{x},t)\gamma_{5}\underbrace{q_{f}(\vec{x},t) \bar q_{f}(\vec{0},0)}\gamma_{5}q_{f}(\vec{0},0)}}\right\rangle, \label{eq:eta_def2}\nonumber\\ &&D_{\gamma_5}(t)=\int d^3x \left\langle{{1\over N_{f}} \sum_{f}^{N_{f}}\overbrace{\bar q_{f}(\vec{x},t)\gamma_{5}q_{f}(\vec{x},t)}{1\over N_{f}} \sum_{g}^{N_{f}}\overbrace{\bar \end{eqnarray} The braces represent the contraction of the quark propagators, $S_q(0,t)$. Thus, for example, $C_{\gamma_5}(t)$ is $\langle S_q(t,0)\gamma_5 S_q(0,t)\gamma_5\rangle$, the same as the nonsinglet meson (pion) propagator, and $D_{\gamma_5}(t)$ is the correlation function between disconnected quark loops, which exists in the flavor singlet mesons. When $D_{\gamma_5}(t)$ is suppressed by the OZI rule, it propagates and acquires $U(1)_{A}$ anomaly. In dynamical QCD, in which the mass of quark polarizing the gluon, $\msea$, is equal to that of the valence quark consisting the meson operator, $\mval$, the $\eta'$ propagator is an exponential function of time with its damping factor being the mass of the meson, $m_\eta'$, \begin{eqnarray} \int d^3x {\langle \eta'(\vec{x},t) \eta'^{\dag}(\vec{0}, 0)\rangle} = C_{\gamma_5}(t) - N_{f}D_{\gamma_5}(t) = \label{eq:eta_prop} \end{eqnarray} at large $t$. Diagram of $\eta'$ propagator. A model of the $\eta'$ propagator is depicted in Fig. <ref>. The meson propagator is expressed as a series expansion in number of the quark loops with signs reflecting the Grassmannian feature of the quark, and the blobs at the ends are the meson operators (<ref>). The wavy lines connecting the quark loops represent the coupling between disconnected loops attached to the pseudoscalar density, which is related to the $U(1)_A$ anomaly. The meson propagator in momentum space can be calculated from the model. The first term in Fig. $\ref{fig:eta_expand}$ is the same as the nonsinglet pseudoscalar meson (pion), $1/(p^2+m^2_\pi)$, and the second term is given by two pion propagators coupled to each other by gluons, $1/(p^2+m^2_\pi)\times m_0^2/N_f \times 1/(p^2+m^2_\pi)$, whose coupling we parameterize as $m_0^2/N_f$. There are $N_f$ combinations of quark loops in the second term. Repeating the similar identification of connected pion propagators to the $n$th order, the momentum space representation of the $\eta'$ propagator can be written as a geometrical series: \begin{eqnarray} &&\langle \eta'(p) \eta'^{\dag}(-p) \rangle \nonumber\\ &&\propto {1\over p^{2}+m_{\pi}^{2}} - N_{f}{1\over p^{2}+m_{\pi}^{2}}{m_{0}^{2}\over N_{f}}{1\over p^{2}+m_{\pi}^{2}} +N_{f}^{2}{1\over p^{2}+m_{\pi}^{2}} {m_{0}^{2}\over N_{f}}{1\over p^{2}+m_{\pi}^{2}}{m_{0}^{2}\over N_{f}}{1\over p^{2}+m_{\pi}^{2}}-\cdots \nonumber\\ && = {1\over p^{2}+m_{\pi}^{2}}\sum_{n=0}^{\infty} \left({-m_{0}^{2}\over p^{2}+m_{\pi}^{2}}\right)^n= {1\over p^{2}+(m_{\pi}^{2}+m_{0}^{2})}, \label{eq:eta_expand} \end{eqnarray} The $m_\pi$ pole in the connected diagram, $C_{\gamma_5}$, is exactly canceled by part of the disconnected diagram, $D_{\gamma_5}$, and thus the square of the $\eta'$ meson mass, $m_{\eta'}^{2}$, is identified by $m_{\pi}^{2}+m_{0}^{2}$, which means $\eta'$ does not behave as an NG boson. In terms of this model, $\eta'$ spectroscopy calculated in lattice simulation should reveal the magnitude of $m^2_0$, and if it is consistent with the WV relation (<ref>). In a later section, we will also calculate the ratio between $D_{\gamma_5}(t)$ and $C_{\gamma_5}(t)\sim A_\pi e^{-m_\pi t}$. From (<ref>), the ratio at large $t$ should behave as \begin{eqnarray} &&{N_{f}D_{\gamma_5}(t)\over C_{\gamma_5}(t)} =1- B {e^{m_{\eta'} t}+e^{-m_{\eta'} (T-t)} \over e^{m_\pi t}+e^{-m_\pi (T-t)}} +\cdots \label{eq:eta_ratio_1} \\ &&\ \ \ \ \ \ \ \stackrel{ (T-t) \gg 1 }{\longrightarrow} 1-Be^{-\Delta mt}+\cdots, \label{eq:eta_ratio} \\ && \Delta m = m_{\eta'}-m_{\pi}, \ \ B={A_{\eta'}\over A_{\pi}}. \end{eqnarray} The ratio at large $t$ exponentially approaches unity with the exponent being the mass difference between $\eta'$ and pion, which is a signature of the dynamical sea quark. This is in contrast to the quenched QCD, in which $\msea$ is taken to be infinitely heavy while $\mval$ is kept finite. In this nonunitary theory, the third and higher terms in the quark loop expansion (<ref>) are missing due to the decoupling of the sea quark, and the resulting meson propagator has an unphysical double pole, \begin{eqnarray} &&\langle \eta'(p) \eta'^{\dag}(-p) \rangle_\text{quenched} \propto {1\over p^{2}+m_{\pi}^{2}} - N_{f}{1\over p^{2}+m_{\pi}^{2}}{m_{0}^{2}\over N_{f}}{1\over \end{eqnarray} The ratio $D_{\gamma_5}(t)/C_{\gamma_5}(t)$ in this case behaves as a linear function of time, \begin{eqnarray} &&{N_{f}D^\text{(quenched)}_{\gamma_5}(t)\over C_{\gamma_5}(t)} ={m_{0}^{2}\over 2m_{\pi}}t + {\rm const}+\cdots, \label{eq:eta_ratio_quenched} \end{eqnarray} which is clearly different from (<ref>). Thus, to obtain a physical $\eta'$ we must simulate the dynamical theory $(\msea=\mval)$. We will examine $m_{\eta'}$ in the domain wall QCD only at the dynamical points. § SIMULATION DETAILS §.§ Domain wall fermion (DWF) The DWF action is defined as \begin{eqnarray} && S_{F}=\sum_{x,y,s,s'}\bar\psi(x,s)D_{\rm DWF}(x,s;y,s')\psi(y,s'), \\ && D_{\rm DWF}(x,s;y,s')=\delta_{s,s'}D^\parallel_{x,y}+\delta_{x,y}D^\bot_{s,s'}, \label{eq:DWF-action}\\ && D^\parallel_{x,y}={1\over2}\sum_{\mu=1}^{4} \left[(1-\gamma_{\mu})U_{\mu}(x)\delta_{x+\hat\mu,y}+ (M_{5}-4)\delta_{x,y}, \\ && D^\bot_{s,s'}= \nonumber\\&&~~~~~~~+ \end{eqnarray} where $\psi(x,s)$ is a DWF that is located in five dimensional space, $(x,s)$, $L_s$ is the size of the fifth direction, and the parameter $M_5$ is the domain wall height. By setting $M_5$ in a region around $[0,2]$, from (<ref>), left-(right-)handed zero modes are localized around $s=0 (L_s-1)$ and the zero modes undergo exponential damping as $s,~(L_s-1-s)$ increases. When a four-dimensional fermion and antifermion, $q(x)$ and $\bar{q}(x)$, are defined as \begin{eqnarray} &&q(x)={1-\gamma_5 \over 2}\psi(x,0)+{1+\gamma_5 \over 2}\psi(x,L_s-1), \\ &&\bar{q}(x)=\bar{\psi}(x,0){1+\gamma_5 \over 2}+\bar{\psi}(x,L_s-1){1-\gamma_5 \over 2}, \end{eqnarray} chiral symmetry is fulfilled even with finite lattice spacing $(a>0)$ at the $L_s\to\infty$ limit. However, in the simulation, $L_s$ is restricted to be finite, and the AWTI is modified from its expression in the continuum theory to, \begin{eqnarray} \partial_{\mu} \vev{{\cal{A}}_{\mu}^{b}(x) {\cal O}} = 2(m_{f}+\mres)\vev{P^{b}(x) {\cal{O}}}, \end{eqnarray} i.e., the physical quark mass is shifted to $m_{\rm quark}= m_{f}+\mres$. $\mres$ is a small lattice artifact called the residual quark mass, defined as \begin{eqnarray} \mres=\lim_{t\to\infty}{\sum_{\vec{x}}\vev{J_{5q}^b(\vec{x},t) P^b(\vec{0},0)}\over \sum_{\vec{x}}\vev{P^b(\vec{x},t) P^b(\vec{0},0)}}, \label{eq:mres_def} \end{eqnarray} where $J_{5q}^b(\vec{x},t)$ is an operator similar to the pseudoscalar operator but made of fermions at the midpoint of the fifth direction[20], $s\sim L_s/2$, thus the numerator of (<ref>) includes the contractions between the surface fermions at $s=0$ or $s=L_s$ and the midpoint fermions at $s\sim L_s/2$. For the flavor non-singlet case, $b\ne0$, $\mres$ is an exponential function of $L_s$ as a consequence of the exponentially localized zero modes to the surface, and vanishes as $L_s\to\infty$. One could further argue[52] that the effective Lagrangian contains the diverging, ${\cal O}(1/a)$, discretization error, which can be corrected by the small shift of the quark mass, $m_\text{quark}= m_f+m_\text{res}$. The remaining error is ${\cal O}(a)$, similar to that of Wilson fermions, however, it is an exponentially small number, $e^{\alpha L_s}$, or ${\cal O}(m_\text{res})$. Although $m_\text{res}$ is small compared with the statistical errors we will have in most of observables, we will treat the shifted quark mass $m_\text{quark}=m_f+m_\text{res}$ as the physical quark mass so that our analysis is precise modulo ${\cal O}(m_\text{res} a, a2)$, which is a few percent in our simulation. On the other hand, for flavor singlet $(b=0)$ case, $J_{5q}^b(\vec{x},t)$ in (<ref>) can be attached to a quark loop that does not propagate in the entire $L_s$ in the fifth direction, and is free from suppression. the counterparts of $\mres$ in the flavor singlet case remains finite even as $L_s\to\infty$, and reproduces the following anomalous term[51] \begin{eqnarray} \sum_{\vec{x}}\vev{J_{5q}^b(\vec{x},t) \calO}\to\delta_{b,0} \sum_{\vec{x}}\vev{\rho_\text{top}(\vec{x},t)\calO}~. \end{eqnarray} In summary, DWF even for finite $L_s$ correctly reproduces the quantum anomaly of axial symmetry with small error due to lattice discretization. §.§ Ensemble: actions and parameters We employ the $N_{f}=2$ QCD ensemble[25] with DWF actions described in the previous subsection. Our gauge action contains an improvement in the sense of the renormalization group invariance, \begin{eqnarray} && S_{G}={\beta \over 3}\left[(1-8c_{1})\sum_{x,\mu>\nu} {\text{ReTr}}[1-R_{\mu\nu}(x)]\right], \\ && P_{\mu\nu}(x)=U_\mu(x)U_\nu(x+\hat\mu) U^{\dag}_\mu(x+\hat\nu)U^{\dag}_\nu(x), \\ && R_{\mu\nu}(x)=U_\mu(x)U_\mu(x+\hat\mu)U_\nu(x+2\hat\mu) \end{eqnarray} with $\beta=0.80$ and $c_{1}=-1.4069$. The parameters of the DWF action (<ref>) are set as $L_{s}=12$ and $M_{5}=1.8$. We measure observables on a 470-940 lattice configuration samples for three different masses, $m_{f}$=0.02, 0.03, and 0.04, which correspond to $m_{\pi}/m_{\rho}\approx$ 0.51-0.64. The lattice size is $16^{3}\times32$, the lattice scale is $a^{-1}\approx 1.5$ GeV ($a\approx 0.13$ fm), and the residual chiral breaking $m_{\rm res}=0.00137(4)$ which is about an order of magnitude smaller than the input quark masses. Throughout this paper we estimate the statistical error using the blocked jackknife method. The size of the block is determined to be 50 trajectories by monitoring the autocorrelation of the hadron propagators. A summary of lattice ensembles and parameters is given in Table <ref>. Other results on these ensembles can be found in [25, 26, 31, 32, 30]. Lattice ensembles and simulation parameters. $\beta$ $c_{1}$ $V\times T$ $a^{-1}$ [GeV] $a$ [fm] $Va^3$ [fm$^3$] $\mres$ 0.80 $-1.4069$ $16^{3}\times32$ 1.537(26) 0.1284(22) $(2.054)^3$ 0.00137(4) $m_{f}$ $m_{\pi}/m_{\rho}$ 3cbegin-end(step) traj. #config. $N_{\text{noise}}$ 0.02 0.5121(36) 3c656-5351(5) 940 1 0.03 0.5984(31) 3c615-6205(10) 560 3 0.04 0.6415(33) 3c625-1765(10), 2075-5615(10) $^{\rm a}$ 470 2 $^{\rm (a)}$ For the $m_{f}=0.04$ ensemble, we do not use trajectories 1775-2065 due to a hardware error on trajectory 1772 that was not detected until lattice generation was finished. §.§ Improvements: smearing and sources Before constructing the meson propagators, we describe an improvement for the quark propagators in this section. It is known to be difficult to reduce the statistical error of the flavor singlet meson spectrum. As we have seen in the previous section, the meson propagator includes the correlation function between disconnected loops, $D_\Gamma$ $(\Gamma=\gamma_5,\gamma_i,{\bf 1},\gamma_5\gamma_i,\gamma_i\gamma_j)$, whose statistical fluctuation is very large, particularly for large $t$ as we will see. We have implemented smearing for a quark operator in a gauge-covariant manner called Wuppertal smearing [27]. The smeared quark operator $q_S$ is a gauge-covariant superposition of the local quark operator $q_L$: \begin{eqnarray} && q^c_{L}(\vec{x},t)\to q^{c}_{S}(\vec{x},t)=\sum_{\vec{y},c'} F^{c,c'}(\vec{x},\vec{y}) q^{c'}_{L}(\vec{y},t), \\\ && F^{c,c'}(\vec{x},\vec{y})=\left[\left\{{\bf 1}+{\omega^2\over 4N}\sum_{i=1}^3 \left(\nabla_i+\nabla_{i}^{\dag}\right) \right\}^N\right]_{\vec{x},c;\vec{y},c'}, \label{eq:gauss-smear}\\ && [{\bf 1}]_{\vec{x},c;\vec{y},c'}=\delta^{c,c'}\delta_{\vec{x},\vec{y}}, \\ && [\nabla_{i}]_{\vec{x},c;\vec{y},c'} = U_{i}(\vec{x},t)^{c,c'}\delta_{\vec{x}+\hat i, \vec{y}}-\delta^{c,c'}\delta_{\vec{x},\vec{y}}, \\ && [\nabla^{\dag}_{i}]_{\vec{x},c;\vec{y},c'} = U_{i}^{\dag}(\vec{y},t)^{c,c'}\delta_{\vec{x}-\hat i, \vec{y}}-\delta^{c,c'}\delta_{\vec{x},\vec{y}}. \end{eqnarray} The shape of $q_S$ in terms of $q_L$ is Gaussian with width $\omega$ as $N\to\infty$. We set $\omega=4.35$ and $N=40$. The overlap between the ground state and the meson operator made of smeared quarks is expected to be larger the meson made of unsmeared quarks, and the excited state contamination is suppressed for small $t$, where the statistical error is smaller. Both the quark correlation functions, $C_\Gamma(t)$ and $D_\Gamma(t)$, are calculated for a complex ${\boldsymbol{Z}}_2$ noise source, $\xi$, defined by \begin{eqnarray} && \xi^{(n)}(\vec{x},t) = {1\over\sqrt2} [\xi_{1}^{(n)}(\vec{x},t)+i\xi_{2}^{(n)}(\vec{x},t)], \end{eqnarray} where $n=1, 2, \dots, N_{\rm noise}$ are random noise ensembles and $\xi_{1}$ and $\xi_2$ take values of $\pm 1$ randomly. $\xi(\vec{x},t)$ is statistically independent of space-time: thus, it \begin{eqnarray} && \lim_{N_{\rm noise}\to\infty}{1\over N_{\rm noise}} \sum_{n=1}^{N_{\rm noise}}\xi^{(n)}(\vec{x},t) \xi^{(n)}(\vec{y},t') = 0, \label{eq:noise_zero} \\ && \lim_{N_{\rm noise}\to\infty}{1\over N_{\rm noise}} \sum_{n=1}^{N_{\rm noise}}\xi^{(n)}(\vec{x},t) \xi^{(n)*}(\vec{y},t') = \delta_{\vec{x},\vec{y}}\delta_{t,t'}, \label{eq:noise_one} \end{eqnarray} which is useful for calculating the disconnected loops as we will see in the next subsection. We use the source restricted to a time slice (wall source) for $C_\Gamma(t)$ and a space-time volume source for $D_\Gamma(t)$, and $N_{\text{noise}}=1$, 3, and 2 for $m_f$=0.02, 0.03, and 0.04, respectively. §.§ Meson operators and correlation functions Our naming convention for meson fields is similar to that used by the particle data group [1], but our simulation is limited to having only up and down quarks ($N_f=2$) with degenerate masses and zero electric charges; thus, the meson spectra are inevitably different from those in the real world. The systematic error from these omission may be comparable or smaller to our target precision of $\sim 10$ %. This point certainly needs further investigation. The Hermitian interpolation fields for flavor nonsinglet meson in our simulation, $\pi$, $\rho$, $a_0$, $a_1$, and $b_1$, and singlet fields, $\eta'$, $\omega$, $f_0$, $f_1$, and $h_1$ are defined in terms of quark operators, $q_{I,f}$ and $\bar{q}_{J,f}$ as follows: \begin{eqnarray} &&\pi_I(\vec{x},t)={1\over\sqrt {2}}\sum_{f,g=1}^{2} \bar q_{I,f}(\vec{x},t)\tau^b_{f,g} i \gamma_{5}q_{I,g}(\vec{x},t), \\ &&\rho_I(\vec{x},t)={1\over\sqrt {6}} \sum_{i=1}^3 \sum_{f,g=1}^{2} \bar q_{I,f}(\vec{x},t)\tau^b_{f,g} i \gamma_{i}q_{I,g}(\vec{x},t) \\ &&a_{0I}(\vec{x},t)={1\over\sqrt {2}}\sum_{f,g=1}^{2} \bar q_{I,f}(\vec{x},t)\tau^b_{f,g}q_{I,g}(\vec{x},t), \\ &&a_{1I}(\vec{x},t)={1\over\sqrt {6}} \sum_{i=1}^3 \sum_{f,g=1}^{2} \bar q_{I,f}(\vec{x},t)\tau^b_{f,g}i\gamma_{5}\gamma_{i}q_{I,g}(\vec{x},t) \\ &&b_{1I}(\vec{x},t)={1\over\sqrt {6}} \sum_{\substack{1\le i\le 3\\i<j\le 3}} \sum_{f,g=1}^{2} \bar q_{I,f}(\vec{x},t)\tau^b_{f,g}i\gamma_{i}\gamma_{j}q_{I,g}(\vec{x},t) \\ &&\eta'_I(\vec{x},t)={1\over\sqrt {2}}\sum_{f=1}^{2} \bar q_{I,f}(\vec{x},t) i \gamma_{5}q_{I,f}(\vec{x},t), \\ &&\omega_I(\vec{x},t)={1\over\sqrt {6}} \sum_{i=1}^3\sum_{f=1}^{2} \bar q_{I,f}(\vec{x},t) i \gamma_{i}q_{I,f}(\vec{x},t)\\ &&f_{0I}(\vec{x},t)={1\over\sqrt {2}}\sum_{f=1}^{2} \bar q_{I,f}(\vec{x},t)q_{I,f}(\vec{x},t), \\ &&f_{1I}(\vec{x},t)={1\over\sqrt {6}} \sum_{i=1}^3\sum_{f=1}^{2} \bar q_{I,f}(\vec{x},t)i\gamma_{5}\gamma_{i}q_{I,f}(\vec{x},t)\\ &&h_{1I}(\vec{x},t)={1\over\sqrt {6}} \sum_{\substack{1\le i\le 3\\i<j\le 3}}\sum_{f=1}^{2} \bar q_{I,f}(\vec{x},t) i \gamma_{i}\gamma_{j}q_{I,f}(\vec{x},t) \end{eqnarray} where $\tau^b$ $(b=1, 2, 3)$ are the Pauli matrices for the flavor indices $f$ and $g$, and $I$ and $J$ denotes whether we use the local quark field ($L$) or the smeared field ($S$) to control the ground-state overlap. In Table <ref>, we summarize the quantum numbers of each meson field. Meson operators in the simulation and their quantum numbers. Meson type $J^{PC}$ $\Gamma$ nonsinglet singlet pseudoscalar $0^{-+}$ $i \gamma_5$ $\pi$ $\eta'$ vector $1^{--}$ $i {\gamma_i}$ $^{\rm a}$ $\rho$ $\omega$ scalar $0^{++}$ 1 $a_0$ $f_0$ pseudovector $1^{++}$ $i{\gamma_5\gamma_i}$ $^{\rm a}$ $a_1$ $f_1$ pseudovector $1^{+-}$ $i \gamma_i\gamma_j$ $^{\rm a}$ $b_1$ $h_1$ $^{\rm (a)}$ average over $i,j=1,2,3$ is taken. The two-point correlation functions between the interpolation fields are calculated as \begin{eqnarray} &&\sum_{\vec{x},\vec{y}}\langle \pi_I(\vec{x},t) \pi^{\dag}_J(\vec{y},0)\rangle =C_{IJ,\gamma_5}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle \rho_I(\vec{x},t) \rho^{\dag}_J(\vec{y},0)\rangle ={1 \over 3}\sum_{i=1}^3 C_{IJ,\gamma_i}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle a_{0I}(\vec{x},t) a_{0J}^{\dag}(\vec{y},0)\rangle =C_{IJ,\bf 1}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle a_{1I}(\vec{x},t) a_{1J}^{\dag}(\vec{y},0)\rangle ={1 \over 3}\sum_{i=1}^3 C_{IJ,\gamma_5\gamma_i}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle b_{I1}(\vec{x},t) b_{1J}^{\dag}(\vec{y},0)\rangle ={1 \over 3}\sum_{i<j} C_{IJ,\gamma_i\gamma_j}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle \eta'_I(\vec{x},t) \eta'^{\dag}_J(\vec{y},0)\rangle =C_{IJ,\gamma_5}(t) - 2 D_{IJ,\gamma_5}(t), \label{eq:eta_op_def}\\ &&\sum_{\vec{x},\vec{y}}\langle \omega_I(\vec{x},t) \omega^{\dag}_J(\vec{y},0)\rangle ={1 \over 3}\sum_{i=1}^3 \left[C_{IJ,\gamma_i}(t)-2 D_{IJ,\gamma_i}(t)\right], \\ &&\sum_{\vec{x},\vec{y}}\langle f_{0I}(\vec{x},t) f_{0J}^{\dag}(\vec{y},0)\rangle =C_{IJ,\bf 1}(t) - 2 D_{IJ,\bf 1}(t), \\ &&\sum_{\vec{x},\vec{y}}\langle f_{1I}(\vec{x},t) f_{1J}^{\dag}(\vec{y},0)\rangle ={1 \over 3}\sum_{i=1}^3 \left[C_{IJ,\gamma_5\gamma_i}(t)-2 D_{IJ,\gamma_5\gamma_i}(t)\right], \\ &&\sum_{\vec{x},\vec{y}}\langle h_{1I}(\vec{x},t) h_{1J}^{\dag}(\vec{y},0)\rangle ={1 \over 3}\sum_{i<j} \left[C_{IJ,\gamma_i\gamma_j}(t)-2 D_{IJ,\gamma_i\gamma_j}(t)\right], \end{eqnarray} in terms of the connected and disconnected quark loop contributions (${\rm Tr}$ is for the trace over color and spinor indices only): \begin{eqnarray} \overbrace{\bar q_I(\vec{x},t)\Gamma \underbrace{q_I(\vec{x},t) \bar q_J}(\vec{y},0)\Gamma q_J}(\vec{y},0)}\right\rangle \nonumber\\ &&\ \ \ \ \ \ \ \ =-\sum_{\vec{x},\vec{y}}\left\langle{\rm Tr} \left[ G_{IJ}(\vec{x},t;\vec{y},0)\Gamma G_{JI}(\vec{y},0;\vec{x},t)\Gamma \right]\right\rangle\ \ (\Gamma=i\gamma_5, i\gamma_i, {\bf 1}, i\gamma_5\gamma_i, i\gamma_i\gamma_j), \label{eq:conn} \\ \overbrace{\bar q_I(\vec{x},t)\Gamma q_I}(\vec{x},t) \overbrace{\bar q_J(\vec{y},0)\Gamma q_J}(\vec{y},0)}\right\rangle\nonumber\\ &&\ \ \ \ \ \ \ \ =-\sum_{\vec{x},\vec{y}}\left\langle\left\{{\rm Tr} \left[ G_{II}(\vec{x},t;\vec{x},t)\Gamma\right] -\sum_{\vec{x'}}\left\langle{\rm Tr} \left[ G_{II}(\vec{x'},t;\vec{x'},t)\Gamma\right]\right\rangle \right\} \right. \nonumber\\ && \ \ \ \ \ \ \ \ \ \ \ \ \ \times \left.\left\{{\rm Tr}\left[G_{JJ}(\vec{y},0;\vec{y},0)\Gamma\right] -\sum_{\vec{y'}}\left\langle{\rm Tr} \left[ G_{JJ}(\vec{y'},0;\vec{y'},0)\Gamma\right] \right\rangle\right\} \right\rangle\nonumber\\ &&\ \ \ \ \ \ \ \ \ \ \ \ \ \ (\Gamma=i\gamma_5, i\gamma_i, {\bf 1}, i\gamma_5\gamma_i, i\gamma_i\gamma_j) \label{eq:disc}. \end{eqnarray} Here $G_{IJ}(\vec{x},t;\vec{y},t')$ is the propagator of the four dimensional quark field \begin{eqnarray} &&G_{LL}^{c,\alpha;c',\alpha'}(\vec{x},t;\vec{y},t') = \left[D^{-1}(\vec{x},t;\vec{y},t')\right]^{c,\alpha;c',\alpha'}, \\ &&G_{LS}^{c,\alpha;c',\alpha'}(\vec{x},t;\vec{y},t') = \sum_{c''}\sum_{\vec{x'}}\left[D^{-1}(\vec{x},t;\vec{x'},t')\right]^{c,\alpha;c'',\alpha'}F^{c'',c'}(\vec{x'},\vec{y}), \\ &&G_{SL}^{c,\alpha;c',\alpha'}(\vec{x},t;\vec{y},t') = \sum_{c''}\sum_{\vec{x'}}F^{c,c''}(\vec{x},\vec{x'})\left[D^{-1}(\vec{x'},t;\vec{y},t')\right]^{c'',\alpha;c',\alpha'}, \\ &&G_{SS}^{c,\alpha;c',\alpha'}(\vec{x},t;\vec{y},t') = \sum_{c'',c'''}\sum_{\vec{x'},\vec{y'}}F^{c,c''}(\vec{x},\vec{x'}) \left[D^{-1}(\vec{x'},t;\vec{y'},t')\right]^{c'',\alpha;c''',\alpha'}F^{c''',c'}(\vec{y'},\vec{y}), \end{eqnarray} where $D^{-1}$ is written in terms of the inverse of the five dimensional matrix $D_{\rm DWF}^{-1}$ (Eq. (<ref>)): \begin{eqnarray} &&D^{-1}(x,y)=\vev{q(x)\bar{q}(y)} \nonumber\\ \left({1+\gamma_5\over2}\delta_{s,0}+{1-\gamma_5\over2}\delta_{s,L_s-1}\right)\nonumber\\ &=&\sum_{s,s'}\left({1-\gamma_5\over2}\delta_{s,0}+{1+\gamma_5\over2}\delta_{s,L_s-1}\right)D_{\rm DWF}^{-1}(x,s;y,s') \left({1+\gamma_5\over2}\delta_{s,0}+{1-\gamma_5\over2}\delta_{s,L_s-1}\right). \label{eq:DWF_q_propagator} \end{eqnarray} $F(\vec{x},\vec{y})$ is the smearing function which is defined in Eq.(<ref>). $c,c',c'',c'''$ are the color indices and $\alpha, \alpha'$ are the spin indices. We apply the zero-momentum projection to obtain the meson mass from meson energy: $E_{\vec{p}}=\sqrt{m_{\text{meson}}^2+\vec{p}^2}\to m_{\text{meson}}$, by summing over spatial volume $\vec{x},\vec{x'},\vec{y},\vec{y'}$. In eq. (<ref>), the sum over $\vec{y}$ is stochastically evaluated by the ${\bf Z}_2$ noise source at while the sums over $\vec{x}$ and $\vec{y}$ in (<ref>) are evaluated by ${\boldsymbol Z}_2$ source spreads over the space-time volume, c.f. (<ref>) and (<ref>): \begin{eqnarray} &&~~~{1\over N_{\rm noise}}\sum_{n=1}^{N_{\rm noise}}\sum_{\vec{x},\vec{y},\vec{z}}\langle {\rm Tr}[\{G_{IJ}(\vec{x},t;\vec{y},0)\xi^{(n)}(\vec{y},0)\}\Gamma \gamma_5 \{G_{IJ}(\vec{x},t;\vec{z},0)\xi^{(n)}(\vec{z},0)\}^{\dag}\gamma_5\Gamma]\rangle\nonumber\\ &&={1\over N_{\rm noise}}\sum_{n=1}^{N_{\rm noise}}\sum_{\vec{x},\vec{y},\vec{z}}\langle {\rm Tr}[G_{IJ}(\vec{x},t;\vec{y},0)\Gamma \gamma_5 &&\to \sum_{\vec{x},\vec{y}}\langle {\rm Tr}[G_{IJ}(\vec{x},t;\vec{y},0)\Gamma \gamma_5 G^{\dag}_{IJ}(\vec{x},t;\vec{y},0)\gamma_5\Gamma]\rangle~~(N_{\rm noise}\to\infty),\nonumber\\ &&=\sum_{\vec{x},\vec{y}}\langle {\rm Tr}[G_{IJ}(\vec{x},t;\vec{y},0)\Gamma && ~~~{1\over N_{\rm noise}}\sum_{n=1}^{N_{\rm noise}}\sum_{\vec{x},\vec{y},{t'}}\langle {\rm Tr}[\xi^{(n)*}(\vec{x},t)\{G_{II}(\vec{x},t;\vec{y},t')\xi^{(n)}(\vec{y},t')\}\Gamma &&\to \sum_{\vec{x}}\langle{\rm Tr}[G_{II}(\vec{x},t;\vec{x},t)\Gamma ]\rangle~~(N_{\rm noise}\to\infty), \end{eqnarray} The dagger ($\dagger$) is taken only for color and spinor (and not for space-time) indices, and we use the $\gamma_5$ hermiticity, $\gamma_5 D^{-1}\gamma_5 =[ D^{-1} ]^\dagger$, of the propagator (<ref>) in (<ref>). The trace over color and spinor indices is exactly carried out by solving the quark propagator $3\times 4$ times each for a random source. §.§ Meson mass fit Throughout this paper, we assume that the one particle state is the ground state for quantum numbers $I$ and $J^{PC}$, for compatibility with to the interpolation operator in Table <ref>. This assumption is not entirely true for some cases. For example, a $\rho$ meson may decay into pions. In our simulation, quarks are heavy with the lightest quark mass about half the strange quark mass, and confined in a relatively small ($\sim 2$ fm)$^3$ box. Many of the decay processes would not occur in this setting since the decaying particles have energies above the threshold. Also we restrict ourselves to degenerate up and down quarks, $N_f=2$, so that a meson such as $a_0$ can not decay due to exact symmetry. To extract the meson masses, the following two analyses are carried out. Standard method: Only the ground state of mass $m_O$ is assumed to exist in the correlation function $\langle O_{S} O_{S}\rangle$, which is fitted by the hyperbolic cosine function reflecting the periodic boundary condition for a meson at $t=T$; \begin{equation} \sum_{\vec{x},\vec{y}}\langle O_I(\vec{x},t) O^{\dag}_I(\vec{y},0)\rangle = {V\over 2m_O}|\langle 0|O_I|O(\vec{p}=\vec{0})\rangle|^2 \left[ e^{-m_{O}t} + e^{-m_{O}(T-t)}\right],~~~~ (I=L, S) \label{eq:method-a} \end{equation} for sufficiently large $t$ and $T-t$. Although our main results will be obtained from the smeared-quark case, $I=S$, we also analyze local quark case to monitor the excited-state contamination. The fitting range of $t$ is determined so that the effective meson mass becomes independent of the time. We also avoid a too large $t$ for which the statistical error becomes large and the results become unreliable. Variational method [28, 29]: In this case, we also assume the first excited state of mass $m_{O^*}$. Both the local ($I,J=L$) and the smeared ($I,J=S$) interpolation fields are used to construct the correlation function $\langle O_{I} O_{J} \rangle$. The $2\times2$ matrix, \begin{eqnarray} \begin{array}{cc} \sum_{\vec{x},\vec{y}}\langle O_{L}(\vec{x},t) O^{\dag}_{L}(\vec{y},0) \rangle & \sum_{\vec{x},\vec{y}}\langle O_{L}(\vec{x},t) O^{\dag}_{S}(\vec{y},0) \rangle \cr \sum_{\vec{x},\vec{y}}\langle O_{S}(\vec{x},t) O^{\dag}_{L}(\vec{y},0) \rangle & \sum_{\vec{x},\vec{y}}\langle O_{S}(\vec{x},t) O^{\dag}_{S}(\vec{y},0) \rangle \end{array} \right), \end{eqnarray} is normalized at a reference time $t_0$ to reduce the statistical error, then is diagonalized as \begin{eqnarray} && X^{-1/2}(t_{0})X(t)X^{-1/2}(t_{0}) \stackrel{\rm diag.}{\longrightarrow} \left( \begin{array}{cc} \lambda_O(t,t_{0}) & 0 \cr 0 & \lambda_{O^{*}}(t,t_{0}) \end{array} \right)~~. \label{eq:variational_diag} \end{eqnarray} The eigenvalues are fit as a function of $t$, \begin{eqnarray} && \lambda_O(t,t_{0})={e^{-m_{O}t} + e^{-m_{O}(T-t)} \over e^{-m_{O}t_{0}} + e^{-m_{O}(T-t_{0})}} \left(\stackrel{t,t_{0}\ll T/2}{\to}e^{-m_{O}(t-t_{0})}\right), \label{eq:method-c}\\ && \lambda_{O^{*}}(t,t_{0})={e^{-m_{O^{*}}t} + e^{-m_{O^{*}}(T-t)} \over e^{-m_{O^{*}}t_{0}} + e^{-m_{O^{*}}(T-t_{0})}} \left(\stackrel{t,t_{0}\ll T/2}{\to}e^{-m_{O^{*}}(t-t_{0})}\right)~ \label{eq:variational_excited_eig} \end{eqnarray} to obtain the masses of the states. The second method, called the variational method, is employed to extract the ground-state energy precisely and to determine the amount of excited state contamination. To fit $\lambda(t,t_0)$ using eq. (<ref>), without unknown amplitudes in front of the exponentials, $t_0$ should be sufficiently large to ignore the higher excited states. By monitoring $\lambda(t,t_0)$, we verify, for our choice of $t_0$, that such contamination is not apparent within the current statistics. As an example $a_0$ case is shown in Fig. <ref>. $t_0=2$ (squares) is chosen for the final results as $\lambda(t,t_0)$ for $t_0=1$ (circles) can't be fit to a linear function of $t-t_0$ meaning the meson propagator is not a single exponential, while those of $t_0>2$ (diamonds, triangles) have much larger error bars. If the number of available configurations were larger, we would have observed the effect from the second excited state and should have calculated for more variations of interpolation field. This point may be important for future investigations with larger statistical sample. $t_0$ dependence of $a_0$ eigenvalue for $m_f=0.02$. We chose $t_0=2$ (squares) by determining the contamination from the higher excited states. Comparison of the effective mass of $a_0$ obtained by the two methods. Open (filled) circles show the results from the local-local (smeared-smeared) interpolation field in the standard method while squares show the effective mass obtained in the variational method. For another example, the effective mass of $a_0$, which we will define in (<ref>), obtained from the two methods is plotted in Fig. <ref>. The effective mass obtained from the variational method (squares) has the smallest statistical error, which is consistent with that of standard method using a smeared-smeared interpolation field (filled circles). For the variational method, the plateau appears after a smaller time distance when the excited-state is separated from the ground-state. The global fits to the plateaux are almost identical to each other. The clear signal of contamination from larger excited states for the local-local interpolation field (open circles) is observed. The identical central values and error bars from the standard single exponential fit and the variational method indicate that the effect from excited states is small for both methods with these settings. We analyze all masses by both methods and compare the results to estimate the systematic uncertainty due to higher excited-states. We also explore the first excited state for pseudoscalar and vector mesons, $\pi^*$ and $\rho^*$, using the variational method. §.§ Decay constant The leptonic decay constant can be obtained from the amplitude of the two-point correlation function of a meson. We analyze decay constants for a pion, $\pi^*$ and $\rho$ mesons. Their respective decay constants, $f_\pi$, $f_{\pi^*}$ and $f_\rho$ can be defined through the conserved axial and vector currents, ${\cal A}_\mu^b(x)$ and ${\cal V}_i^b(x)$, \begin{eqnarray} && f_O m_O = \langle 0 | {\cal A}_4^b(x) | O(\vec{p}=\vec{0}) \rangle = Z_A\langle 0 | {A}_4^b(x) | O(\vec{p}=\vec{0}) \rangle \ \ (O=\pi, \pi^*)\ \label{eq:fps_def} \\ && f_\rho m_\rho \epsilon_i = \langle 0 | {\cal V}_i^b(x) | \rho(\vec{p}=\vec{0}) \rangle = Z_V\langle 0 | {V}_i^b(x) | \rho(\vec{p}=\vec{0}) \rangle \ \ (i=1,2,3) \label{eq:fvec_def} \end{eqnarray} where $\epsilon_i$ is the polarization vector of the vector meson state, and $Z_A$ and $Z_V$ are the matching factors between the lattice local currents, \begin{eqnarray} A^b_\mu(x) &=& \bar q(x) \tau^b \gamma_\mu \gamma_5 q(x),\\ V^b_\mu(x) &=& \bar q(x) \tau^b \gamma_\mu q(x), \end{eqnarray} and an appropriate renormalization is used scheme in the continuum QCD, which, in our case, is $\overline{MS}$ at $\mu=2$ GeV. For $f_\pi$ and $f_{\pi^*}$, the first matrix element in (<ref>) can be related to pseudoscalar density $P^b(\vec{x},t) = \bar q(x) \tau^b \gamma_5 q(x)$ using the (flavor nonsinglet) AWTI, \begin{eqnarray} \partial_\mu \langle 0 | {\cal A}_\mu^b(x) O(0) | 0 \rangle = 2(m_f+\mres) \langle 0 | P^b(x) O(0)| 0 \rangle, \label{eq:singlet AWT} \end{eqnarray} which leads to \begin{eqnarray} f_O m_O^2 = 2(m_f+\mres)\langle 0 | P^b | O(\vec{p}=\vec{0}) \rangle \ \ (O=\pi,\pi^*). \label{eq:fpi} \end{eqnarray} The actual determination of the decay constants is performed by the standard method (C) for a pion and $\rho$ meson, and the variational method (D) for a pion and $\pi^*$ meson: Standard method In this case we assume the $\langle \pi_{L} \pi_{L}\rangle$ and $\langle \rho_{L} \rho_{L}\rangle$ correlation functions contain only propagation of the ground-state. $\langle \pi_{L} \pi_{L}\rangle$ and $\langle \rho_{L} \rho_{L}\rangle$ are fitted by a standard hyperbolic cosine function: \begin{eqnarray} \sum_{\vec{x},\vec{y}}\langle \pi_L(\vec{x},t) \pi^{\dag}_L(\vec{y},0)\rangle &=& {V\over 2m_{\pi}}|\langle 0 | P^a_L | \pi(\vec{p}=\vec{0}) \rangle|^2\left[ e^{-m_{\pi}t} + e^{-m_{\pi}(T-t)}\right] \nonumber\\ &=&{V f_{\pi}^2 m_{\pi}^3 \over 8(m_f+\mres)^2}\left[ e^{-m_{\pi}t} + e^{-m_{\pi}(T-t)}\right], \label{eq:method-C_fpi}\\ \sum_{\vec{x},\vec{y}}\langle \rho_L(\vec{x},t) \rho^{\dag}_L(\vec{y},0)\rangle &=& {V\over 2m_{\rho}}|\langle 0 | V^a_L | \pi(\vec{p}=\vec{0}) \rangle|^2\left[ e^{-m_{\rho}t} + e^{-m_{\rho}(T-t)}\right] \nonumber\\ &=&{V f_{\rho}^2 m_{\rho} \over 2Z_V^2}\left[ e^{-m_{\rho}t} + e^{-m_{\rho}(T-t)}\right], \label{eq:method-C_frho} \end{eqnarray} to extract the quantities $m_\pi$, $f_\pi$, $m_\rho$, and Variational method In this case, the second excited state, $\pi^*$, in the correlation function of the local meson operator, $\langle \pi_{L} \pi_{L}\rangle$, is also taken into account. $\langle \pi_{L} \pi_{L}\rangle$ is fitted by a double hyperbolic cosine function: \begin{eqnarray} &&\sum_{\vec{x},\vec{y}}\langle \pi_L(\vec{x},t) \pi^{\dag}_L(\vec{y},0)\rangle = {V\over 2m_{\pi}}|\langle 0 | P^a_L | \pi(\vec{p}=\vec{0}) \rangle|^2\left[ e^{-m_{\pi}t} + e^{-m_{\pi}(T-t)}\right] \nonumber \\ && \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ +{V\over 2m_{\pi^*}}|\langle 0 | P^a_L | \pi^*(\vec{p}=\vec{0}) \rangle|^2\left[ e^{-m_{\pi^*}t} + e^{-m_{\pi^*}(T-t)}\right] \nonumber\\ &&\!\!\!\!\!\!\!\!={V f_{\pi}^2 m_{\pi}^3 \over 8(m_f+\mres)^2}\left[ e^{-m_{\pi}t} + e^{-m_{\pi}(T-t)}\right] +{V f_{\pi^*}^2 m_{\pi^*}^3 \over 8(m_f+\mres)^2}\left[ e^{-m_{\pi^*}t} + e^{-m_{\pi^*}(T-t)}\right]. \label{eq:method-b2} \end{eqnarray} In this fitting procedure, we first determine $m_\pi$ and $m_{\pi^*}$ by the variational method, (B) in the previous subsection and then fit the two-point function data to (<ref>) to determine $f_{\pi}$ and $f_{\pi^*}$ using the results from the first fitting. §.§ Chiral Extrapolation To obtain the masses and decay constants of various mesons at the physical quark mass point, $m_f=m_{u,d}$ [25], we need to extrapolate the numerical value calculated at heavier quark mass points. As the number of simulation points is limited and the statistical error is too large, we do not use the fitting formula of chiral perturbation theory at the next leading order or higher in this work. As a crude estimation of the mass of $\eta'$ at the physical point, we examine the formula valid in the lowest-order approximation from the flavor singlet AWTI given by eq. (<ref>): \begin{eqnarray} && m_{\eta'}^2 = C_0+C_1(m_f+\mres) \ \ {\text{(AWTI\ type)}}, \label{eq:chiral_sqrt_formula} \end{eqnarray} We also examine the simplest linear extrapolation for all meson masses as well as the decay constants, \begin{eqnarray} && O = C_0+C_1(m_f+\mres) \ \ {\text{(linear\ type)}}~ \label{eq:chiral_lin_formula} \end{eqnarray} where $O$ is either a meson mass or a decay constant. § NUMERICAL RESULTS §.§ Mass of $\rho$ meson and lattice scale First we analyze the mass of a $\rho$ meson using the methods (A) and (B) and determine the lattice scale from $m_\rho$ assuming that it is a stable particle, which is true for the relatively heavy quark in the small box used in our simulation. In Fig. <ref>, the effective mas of a $rho$ meson, taken from the damping rate between meson propagators at two neighboring times, $m_{\rho,IJ}^{\rm eff}(t+1/2)$, which is defined as \begin{eqnarray} {\sum_{\vec{x},\vec{y}}\vev{O_I(\vec{x},t) O_J^{\dag}(\vec{y},0)}\over\sum_{\vec{x},\vec{y}}\vev{O_I(\vec{x},t+1) O_J^{\dag}(\vec{y},0)}} = {e^{-m_{O, IJ}^{\rm eff}(t+{1\over2}) \ t} + e^{-m_{O, IJ}^{\rm eff}(t+{1\over2}) \ (T-t)} \over e^{-m_{O, IJ}^{\rm eff}(t+{1\over2}) \ (t+1)} + e^{-m_{O, IJ}^{\rm eff}(t+{1\over2}) \ (T-t-1)} }~, \label{eq:def-effmass} \end{eqnarray} is plotted in the top panels (method (A) and method (B) are shown in the left and right panels, respectively). The bottom panel shows an eigenvalue of the ground-state obtained from the variational method. The results of $m_\rho$ obtained from the standard hyperbolic cosine fit (method (A)) and the variational method (method (B)) are listed in Table <ref>. The masses obtained from both methods are consistent with each other within statistical error for all $m_f$, and the ground-state mass can be successfully extracted using the smeared operator. Effective mass of $\rho$ vs $t$ using method (A) (left) and method (B) (right), and $\rho$ eigenvalue vs $t-t_0$ (bottom). Lines show the globally fitted result with errors and the ranges of $t$. $m_{f}$ $m_{\rho}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.5741(39) 5 11 (A) 0.5729(41) 5 $t_0+1$ 12 (B) 0.5425(64) $^{\rm a}$ 5 16 (A) 0.03 0.5979(40) 7 14 (A) 0.5984(34) 5 $t_0+1$ 16 (B) 0.5946(58) $^{\rm a}$ 6 16 (A) 0.04 0.6385(39) 6 14 (A) 0.6379(35) 5 $t_0+1$ 16 (B) 0.6323(70) $^{\rm a}$ 7 16 (A) $^{(a)}$ These values are obtained from $(I,J)=(L,W)$ correlation functions and quoted by [25]. We perform linear extrapolation for both results and obtain $m_\rho$ at the physical quark mass point $(m_f=m_ {u,d})$. The result of the chiral extrapolation is shown in Fig. <ref> and Table <ref>. The values obtained from both methods at the physical quark mass point are consistent within statistical error; we choose the value from method (B) as our main value. The lattice scale determined from $m_\rho$=775.49 MeV [1] is \begin{eqnarray} a_{m_\rho}^{-1}=1.537(26) \ \text{GeV}~. \label{eq:a_value} \end{eqnarray} We have measured the potential energy between static quarks and extracted the Sommer scale $r_0$ from the potential $r_0/a=4.278(54)$ [25]. Using $a_{m_\rho}$ we obtained \begin{eqnarray} r_0^{\text{phys}}=0.5491(93)\ \text{fm}~~, \end{eqnarray} which is somewhat larger than previously estimated values by $\sim$ 10%. Although $r_0$ is one of the most precisely determined dimensionful quantities in the lattice QCD, its experimental value is not known; thus, we could not judge whether our larger value is close to the physical value in QCD or whether it reflects some systematic errors, which we discuss in a later section. By increasing the statistical sample size, the lattice scale changed from that we reported in our previous paper[25]. Accordingly, the physical quark mass point, $m_f=m_{u,d}$, may change. However, we use the old value of $m_{u,d}$ as the physical quark mass point in this paper. This is because the number of quark mass points newly obtained in this work is not sufficient to repeat the same analysis as before in which we used the formula of ChPT up to the next to the leading order. We will discuss the decay constant and the excited-state meson, $\rho^*$, in later $m_{\rho}$ vs $m_f$ $m_{\rho}$ at the physical quark mass point $(m_f= m_{u,d})$. $a_{m_\rho}^{-1}$ [GeV] $a_{m_\rho}$ [fm] method 0.5073(85) 1.528(26) 0.1291(22) (A) 0.5044(85) 1.537(26) 0.1284(22) (B) §.§ Pion mass In Fig. <ref>, we plot the effective mass of a pseudoscalar meson obtained by method (A) on the left, that obtained by method (B) on the right, and the ground-state eigenvalue obtained using method (B) on the bottom panel. Table <ref> summarizes the values of the pion mass obtained by both methods. By using 5-10 times more statical samples than in the previous analysis and extracting the ground-state information from the meson propagator over shorter time distance, which becomes possible using smeared operators, the statistical errors decrease to approximately half of those in the previous results. The fact that the reduction of the error size is closer to or even larger than that expected from the increase in the number of statistical samples, $1/2 > 1/\sqrt{\text{5-10}}$, suggests that the smearing itself does not necessarily cause the smaller statistical error for a pseudoscalar meson. Rather, we could determine the extent of the excited-state contamination using smeared operators with different overlaps with the states. In fact, our new results are consistent within statistical error with the previous results. We will discuss the decay constant and the excited-state meson, $\pi^*$, later. Pion effective mass vs. $t$ using method (A) (left) and method (B) (right), and pion eigenvalue vs. $t-t_0$ (bottom). Lines show fit values, errors and ranges. $m_{f}$ $m_{\pi}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.2940(14) 5 14 (A) 0.2934(13) 5 $t_0+1$ 14 (B) 0.2902(28) $^{\rm a}$ 9 16 (A) 0.03 0.3596(11) 6 16 (A) 0.3581(10) 5 $t_0+1$ 16 (B) 0.3575(19) $^{\rm a}$ 9 16 (A) 0.04 0.4075(11) 7 16 (A) 0.4092(11) 5 $t_0+1$ 16 (B) 0.4094(25) $^{\rm a}$ 9 16 (A) $^{(a)}$ These values are obtained from $(I,J)=(L,W)$ correlators and quoted by [25]. §.§ Mass of $a_0$ From experiments, there are two flavor-non-singlet scalar mesons, $a_0(980) $ and $a_0(1450)$, in nature. Although these are unstable particles in the more realistic $N_f=2+1$ case, we assume a stable one-particle state to be the ground state in the scalar meson sector in our $N_f=2$ case with a relatively heavy quark and small space-time. $a_0$ meson spectrum results previously obtained by lattice QCD calculation seem to fall roughly into two categories[43], studies reporting lighter masses of $\sim 1$ GeV[44, 42] and those reporting heavier masses $\sim 1.5$ GeV [46, 26, 45, 43]. Previous RBC results[26] are $m_{a_0}=1.58(34)$ GeV by the analysis of unitary points and 1.51 (19) GeV by partially quenched analysis. Fig. <ref> (left) shows the effective mass of $a_0$, Fig. <ref> (right) shows the eigenvalue of the ground-state using the variational method (B). The numerical values are listed in Table <ref>, in which we also quote the previous RBC Our new results for the mass of $a_0$ are significantly lighter than those of previous results, as shown in Table <ref>. Since the QCD ensemble used in both investigations is the same, this discrepancy must originate from the difference in measuring the meson operator. In the previous calculation the meson interpolation field was constructed from quark fields at a point. Although the point operator was convenient for theoretical investigation in the previous study, it is not necessarily optimal for extracting the ground state. In fact, as shwon in the left panel of Fig. <ref>, the effective mass of the point operator (open symbols) is very large at a short distance, which implies a large amount of excited-state contamination in the point operator. On the other hand, the effective mass obtained using the smeared operator (filled symbols) reaches plateaux earlier in time and coincides to that obtained from the ground-state eigenvalue by the variational method, shown in right panel. Note that the size of the statistical sample is increased by a factor of five or more in this work compared with that in the previous report. Effective mass of $a_0$ vs $t$ using method (A) (left) and method (B) (right). Lines show fitted values, errors, and ranges. $m_{f}$ $m_{a_0}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.750(28) $^{\rm a}$ 2 6 (A) 0.747(28) 2 $t_0+1$ 10 (B) 0.92(9) $^{\rm b}$ 4 10 exponential fit 0.03 0.816(17) $^{\rm a}$ 2 6 (A) 0.807(14) 2 $t_0+1$ 6 (B) 0.99(10) $^{\rm b}$ 5 10 exponential fit 0.04 0.814(19) $^{\rm a}$ 4 9 (A) 0.811(17) 4 $t_0+1$ 10 (B) 0.94(5) $^{\rm b}$ 5 12 exponential fit $^{(a)}$ These values are obtained by uncorrelated fitting. $^{(b)}$ These values are obtained from $(I,J)=(L,L)$ correlators and quoted in [26]. Fig. <ref> shows the results of extrapolation by linear fitting and Table <ref> shows $m_{a_0}$ at the physical quark mass point. Since both methods (A) and (B) are consistent with each other we choose \begin{eqnarray} m_{a_{0}}^{\text{phys}}=1.111(81) \ \text{GeV} \end{eqnarray} from method (B) as our final value in this work. $m_{a_0}$ vs $m_f$. The asterisk on the left shows the experimental values[1]. $m_{a_0}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{a_0}$ $m_{a_0}^{\text{phys}}$ [MeV] $m_{a_0} r_0$ method 0.721(54) 1,108(85) 3.08(23) (A) 0.723(51) 1,111(81) 3.10(22) (B) To clarify the discrepancies among results obtained from lattice calculations and experiments, further investigations including the calculation multiparticle scattering states and the strange sea quark effects ($N_f=2+1$) are §.§ Mass of $\eta'$ Before presenting the mass spectrum results of the flavor singlet pseudoscalar meson, $\eta'$, we check whether the theoretical expectation discussed in <ref> is realized. The ratio of the correlation function between disconnected quark loops, $D_{\gamma_5}(t)$, to the connected correlation function, $C_{\gamma_5}(t)$, was shown to approach unity for a large time separation, which is clearly different from the expectation of the linear growth in the quenched QCD case In the discussion, only the pion and $\eta'$ states were considered when coupling to the $I(J^P)=0(0^-)$ operator, leading to \begin{eqnarray} {N_f D_{\gamma_5}(t) \over C_{\gamma_5}(t)} = 1 - B{e^{-m_{\eta'}t} + e^{-m_{\eta'}(T-t)} \over e^{-m_{\pi}t} + e^{-m_{\pi}(T-t)} } \stackrel{ (T-t) \gg 1 }{\longrightarrow} \label{eq:eta_ratio_again} \end{eqnarray} In Fig. <ref>, the ratio extracted using the smeared operator, $\eta'_S$, is plotted. Indeed the ratio asymptotically approaches one for the two lighter quark masses (circles, squares), although it is statistically uncertain at a large time distance. However the heaviest quark mass point (diamonds) seems to approach to a value lower than one. The mass difference between $\eta'$ and $\pi$ is smaller for the heavier quark mass, and is close to zero for the heaviest quark mass (as we will discuss later); thus, the ratio only approaches unity at a very large $t$ from (<ref>). Moreover, this deviation from the simplest theoretical explanation might be due to the omission of the excited states such as the $\pi^*$ or $0^{-+}$ glueball state, which may play a more significant role in the heavier-quark-mass region. From the current results, we can not conclude whether the deviation from unity for the heaviest quark can be explained by the above-mentioned arguments or is due to other reasons, for example, insufficient sampling of the different topological $N_f D_{\gamma_5}(t)/C_{\gamma_5}(t)$ vs $t$. We now describe the $\eta'$ spectrum obtained using methods (A) and (B). Figure <ref> shows the effective mass (left: method (A), right: method (B)), and the ground-state eigenvalue, and their numerical values and their fitted ranges are given in Table <ref>. We did not use a propagator from longer distance, where the statistics are too poor and the standard error analysis would not be reliable, although the inclusion of a few more data points does not change the fitted results for most of the masses. Method (B) produces flatter plateaux than method (A) for this meson. As a consistency check, we also examined the temporal exponent of the ratio (<ref>) to extract the mass of $\eta'$. We have evaluated the effect of the finiteness of the lattice in the temporal direction by using the fitting formula (<ref>), and found the results to be unchanged. Combining the measured pion mass in Table <ref>, the values obtained are $m_{\eta'}=0.458(58),$ 0.571(48), and 0.461(15) for $m_f=0.02$, 0.03, and 0.04, respectively. These estimations are slightly smaller than the results in Table <ref>. One reason for this may be that the time range used in fitting the ratio is too short and the pion mass is overestimated, which causes the estimation for the mass of $\eta'$ to be smaller than its actual value. Because of this possibility, we won't use the results obtained from the ratio fitting in our main results. The mass of $\eta'$ has only slight dependence on the quark mass, as shown in Fig. <ref>: all three masses are consistent within two to three standard deviations of statistical error. Their central value fluctuates nonmonotonically in quark mass order. Before being convinced of this nonmonotonicity, we should question the reliability of the error estimation and other systematic uncertainties such as insufficient sampling over the topological charge since $\eta'$ is likely to depend strongly on the topological charge strongly. In our simulation we use DBW2 gauge action to reduce the size of the residual chiral symmetry breaking, $\mres$, sacrificing the configuration mobility among different topological sectors to some extent. Effective mass of $\eta'$ vs $t$ using method (A) (left) and method (B) (right), and $\eta'$ eigenvalue vs $t-t_0$ (bottom). $m_{f}$ $m_{\eta'}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.477(40) 2 5 (A) 0.473(50) 2 $t_0+1$ 6 (B) 0.03 0.571(60) 3 5 (A) 0.600(44) 2 $t_0+1$ 6 (B) 0.04 0.497(17) 2 5 (A) 0.492(15) 2 $t_0+1$ 5 (B) Although the quark mass dependence has not been resolved sufficiently clearly, we extrapolate the measured masses by the eqs. (<ref>) and (<ref>) to estimate the mass of $\eta'$ at the physical quark mass point. The results are shown in Fig. <ref> and Table <ref>. The central values of the estimation differ from each other by 15% but are within statistical error. Our main estimation for the mass of $\eta'$ at the physical quark mass point is obtained from the variational method (B) and chiral extrapolation using the lowest order of ChPT (<ref>), and is given by \begin{eqnarray} m_{\eta'}^{\text{phys}}=819(127) \ \text{MeV}~. \end {eqnarray} This is the first estimation of the mass of $\eta'$ performed with the two flavors of a dynamical (approximately) chiral fermion, which is certainly heavier than a pion, which is thought to be related to the chiral $U(1)_A$ anomaly. Apart from the large statistical error and the various systematic errors discussed above, the main results are close to the experimentally obtained mass of $\eta'$, which suggests that further improvements can be made by especially calculation using an $N_f=2+1$ ensemble. $m_{\eta'}^2$ vs $m_f$ (left), and $m_{\eta'}$ vs $m_f$ (right). The open circles and squares are extrapolated values at the physical quark mass point using (<ref>) and (<ref>), The asterisk on the left shows the experimental values[1]. $m_{\eta'}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{\eta'}$ $m_{\eta'}^{\text{phys}}$ [MeV] $\metap r_0$ method and chiral extrapolation 0.480(78) 738(121) 2.05(33) (A) AWTI type (<ref>) 0.487(78) 748(120) 2.08(33) (A) linear type (<ref>)0.532(82) 819(127) 2.28(35) (B) AWTI type (<ref>)0.560(89) 862(130) 2.40(36) (B) linear type (<ref>) §.§ Mass of $\omega$ We also examine the flavor singlet vector meson, $\omega$, using a similar procedure to that for $\eta'$. Fig. <ref> shows the effective mass of $\omega$ (left: method (A), right: method (B)), which is also listed in Table <ref>. We are able to extract a non-zero signal, but from a shorter time distance; thus there may be a significant distortion from the excited states. The results for the lightest point, $m_f=0.02$, has a particularly poor signal. Effective mass of $\omega$ vs $t$ using method (A) (left) and method (B) (right). $m_{f}$ $m_{\omega}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.464(48) 3 6 (A) 0.616(124) 2 $t_0+1$ 4 (B) 0.03 0.636(24) 2 5 (A) 0.651(28) 2 $t_0+1$ 4 (B) 0.04 0.717(23) 2 5 (A) 0.699(29) 2 $t_0+1$ 5 (B) We estimated the extrapolated mass at the physical quark mass point, as shown in Fig. <ref> and Table <ref>. The fitting formula used is the linear extrapolation (<ref>). Since the statistical error for the lightest point $m_f=0.02$ is large, as mentioned above, we examine two ways of the chiral extrapolation: using all three masses or the heaviest two points. In Table <ref>, one can see that the results obtained from the “method (A) 3-masses fitting” are significantly different from those of the “method (A) 2-masses fitting”. At the physical quark mass point, $m_\omega$ is obtained from the "method (B) 3 masses fitting", \begin{eqnarray} m_{\omega}^{\text{phys}} = 790 (194) \ \text{MeV}~. \end{eqnarray} Our estimation for $\omega$ is consistent with the experimental value, but with a large statistical error $\sim$ 25%. $m_{\omega}$ vs $m_f$. The asterisk on the left shows the experimental value[1]. Estimation of $m_{\omega}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{\omega}$ $m_{\omega}^{\text{phys}}$ [MeV] $m_{\omega} r_0$ fit method0.285(80) 439(123) 1.22(34) (A) 3 masses fit 0.394(117) 605(180) 1.68(50) (A) 2 masses fit 0.514(126) 790(194) 2.20(54) (B) 3 masses fit 0.509(141) 782(217) 2.18(60) (B) 2 masses fit We also calculated the propagators of the flavor singlet meson scalar, $f_0$, using the same quark propagator for $\eta'$ and $\omega$, and found that they are too noisy to extract the spectrum for all values of $m_f$. §.§ Pseudovector meson ($a_1$, $b_1$, $f_1$, $h_1$) spectra Figs. <ref>, <ref>, <ref>, and <ref> show the effective mass obtained using method (A) (left) and method (B) (right), and Tables <ref>, <ref>-<ref> list the results of fits for $a_1$, $b_1$, $f_1$, and $h_1$, respectively. Except for the $h_1$ meson propagator at $m_f=0.03$, the fitting procedure converges. Effective mass of $a_1$ vs $t$ using method (A) (left) and method (B) (right). Effective mass of $b_1$ vs $t$ using method (A) (left) and method (B) (right). Effective mass of $f_1$ vs $t$ using method (A) (left) and method (B) (right). Effective mass of $h_1$ vs $t$ using method (A) (left) and method (B) (right). $m_{f}$ $m_{a_1}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.816(12) 4 7 (A) 0.808(15) 4 $t_0+1$ 9 (B) 0.03 0.894(16) 5 8 (A) 0.880(11) 4 $ t_0+1$ 9 (B) 0.04 0.898(11) 4 8 (A) 0.895(12) 4 $ t_0+1$ 11 (B) $m_{f}$ $m_{b_1}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.849(17) 4 7 (A) 0.848(21) 4 $t_0+1$ 8 (B) 0.03 0.892(14) 4 8 (A) 0.898(15) 4 $t_0+1$ 8 (B) 0.04 0.889(26) 6 9 (A) 0.925(13) 4 $t_0+1$ 9 (B) $m_{f}$ $m_{f_1}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.801(41) 2 5 (A) 0.798(40) 2 $t_0+1$ 5 (B) 0.03 0.895(34) 2 4 (A) 0.893(35) 2 $t_0+1$ 4 (B) 0.04 0.925(43) 2 4 (A) 0.935(41) 2 $t_0+1$ 5 (B) $m_{f}$ $m_{h_1}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.818(78) 2 4 (A) 0.814(78) 2 $t_0+1$ 4 (B) 0.04 0.834(49) 2 4 (A) 0.832(49) 2 $t_0+1$ 4 (B) These meson masses are extrapolated linearly to the physical quark mass point, $m_f=m_{u,d}$, and are shown in Figs. <ref>-<ref>. The numerical values are summarized in Tables <ref>-<ref> for $a_1$, $b_1$, $f_1$, and $h_1$, respectively. As the masses are independent of the method used, within statistical error, we choose \begin{eqnarray} &&m_{a_1}^{\text{phys}}=1.140(51) \ \text{GeV}~, \\ &&m_{b_1}^{\text{phys}}=1.203(64) \ \text{GeV}~, \\ &&m_{f_1}^{\text{phys}}=1.033(137) \ \text{GeV}~, \\ &&m_{h_1}^{\text{phys}}=1.225(250) \ \text{GeV} \end{eqnarray} from method (B) as our main values. $m_{a_1}$ vs $m_f$. The asterisk on the left shows the experimental value[1]. $m_{b_1}$ vs $m_f$. The asterisk on the left shows the experimental value[1]. $m_{f_1}$ vs $m_f$. The asterisk on the left shows the experimental value[1]. $m_{h_1}$ vs $m_f$. The asterisk ont he left shows the experimental value[1]. $m_{a_1}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{a_1}$ $m_{a_1}^{\text{phys}}$ [MeV] $m_{a_1} r_0$ method 0.745(26) 1,146(45) 3.19(12) (A) 0.742(31) 1,140(51) 3.17(14) (B) $m_{b_1}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{b_1}$ $m_{b_1}^{\text{phys}}$ [MeV] $m_{b_1} r_0$ method 0.807(43) 1,241(70) 3.45(19) (A) 0.783(40) 1,203(64) 3.35(17) (B) $m_{f_1}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{f_1}$ $m_{f_1}^{\text{phys}}$ [MeV] $m_{f_1} r_0$ method 0.689(90) 1,058(139) 2.95(39) (A) 0.672(88) 1,033(137) 2.87(38) (B) $m_{h_1}$ at the physical quark mass point $(m_f= m_{u,d})$. $m_{h_1}$ $m_{h_1}^{\text{phys}}$ [MeV] $m_{h_1} r_0$ method 0.802(164) 1,233(252) 3.43(70) (A) 0.797(162) 1,225(250) 3.41(69) (B) These numbers may be compared with the experimental results for $b_1(1235)$, $h_1(1170)$, $a_1(1260)$, and $f_1(1285)$, the first two of which are in good agreement with the numerical results. However further investigations based on realistic settings are clearly needed for more detailed comparisons. §.§ Excited meson ($\pi^*$, $\rho^*$) masses In this subsection, the second excited states of the pion and $\rho$ meson are discussed. Using method (B), we extract the eigenvalue for the second excited state, $\lambda_{O^*}(t), O=\pi,\rho$ eq. (<ref>), which is plotted in the right panels of Fig. <ref> and <ref>, respectively. Although we only use two different operators for each meson, and $\lambda_{O^*}(t)$ may have a significant contribution from the higher excited state, we fit $\lambda_{O^*}(t)$ to extract the temporal exponent, $m_{O^*}$, or the mass of the excited states using eq. (<ref>). The results of the fitting are shown in Tables <ref> and <ref>. We checked that the results for $t_0=5$ and $t_0=6$ are consistent with each other. Effective mass of $\pi^*$ and eigenvalue as functions of $t$ and $t-t_0$. Effective mass of $\rho^*$ and eigenvalue as functions of $t$ and $t-t_0$. $m_{f}$ $m_{\pi^*}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 1.215(50) 5 $t_0+1$ 8 (B) 0.03 1.211(27) 5 $t_0+1$ 8 (B) 0.04 1.242(26) 5 $t_0+1$ 8 (B) $m_{f}$ $m_{\rho^*}$ $t_{0}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 1.375(43) 5 $t_0+1$ 8 (B) 0.03 1.361(25) 5 $t_0+1$ 8 (B) 0.04 1.402(31) 5 $t_0+1$ 8 (B) We performed linear extrapolation using eq. (<ref>) to the physical quark mass point, and found that \begin{eqnarray} &&m_{\pi^*}^{\text{phys}}=1.791(138) \ \text{GeV}~, \\ &&m_{\rho^*}^{\text{phys}}=2.028(131) \ \text{GeV} \end{eqnarray} (see Figs. <ref> and <ref>, Table <ref> and <ref>). These states may be interpreted as $\pi(1300) $, and $\rho(1450) $ or $\rho(1700) $. $m_{\pi^*}$ vs. $m_f$. The left most star symbols show the experimental values[1] in the real world. $m_{\rho^*}$ vs. $m_f$ The left most star symbols show the experimental values[1] in the real world. $m_{\pi^*}$ at the physical quark mass point $(m_f= m_{u,d})$ $m_{\pi^*}$ $m_{\pi^*}^{\text{phys}}$ [MeV] $m_{\pi^*} r_0$ method 1.165(88) 1,791(138) 4.98(38) (B) $m_{\rho^*}$ at the physical quark mass point $(m_f= m_{u,d})$ $m_{\rho^*}$ $m_{\rho^*}^{\text{phys}}$ [MeV] $m_{\rho^*} r_0$ method 1.319(82) 2,028(131) 5.64(36) (B) §.§ Decay constants As the last set of numerical results, we present the leptonic decay constant in this subsection. The decay constant of the ground-state pion, $f_\pi$, is determined using method (C), and fitting the smeared two-point function to formula (<ref>). We also fitted the same two-point functions to the double exponential formula (<ref>) using the values of $m_\pi$ and $m_{\pi^*}$ determined from the variational method (method (B)) to investigate the decay constant for the second excited state, $f_{\pi^*}$, using method (D). Table <ref> shows the results for each simulated quark mass. The pion mass and decay constants are consistent with those reported in the previous paper[25] within statistical error. $f_{\pi}$ and $f_{\pi^*}$. $m_{f}$ $m_{\pi}$ $f_\pi$ $m_{\pi^*}$ $f_{\pi^*}$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.2936(13) 0.09561(40) — — 7 14 (C) 0.2934(13)(fixed) 0.09540(43) 1.215(50)(fixed) 0.02244(54) 4 14 (D) 0.2938(18) $^{\rm a}$ 0.09494(62) $^{\rm a}$ — — 9 16 (C) 0.03 0.3598(15) 0.10350(46) — — 10 16 (C) 0.3581(10)(fixed) 0.10370(44) 1.211(27)(fixed) 0.03236(65) 4 16 (D) 0.3610(18) $^{\rm a}$ 0.10253(56) $^{\rm a}$ — — 9 16 (C) 0.04 0.4098(12) 0.11002(39) — — 8 16 (C) 0.4092(11)(fixed) 0.10964(40) 1.242(26)(fixed) 0.04362(61) 4 16 (D) 0.4087(16) $^{\rm a}$ 0.11059(57) $^{\rm a}$ — — 9 16 (C) $^{\rm (a)}$ These values are quoted in the previous paper[25]. Although the $\pi^*$ decay constant is poorly numerically determined, an interesting theoretical prediction can be made. AWTI, (<ref>), for $\pi^*$ describes the equation for its decay constant, \begin{eqnarray} f_{\pi^*} ={2(m_f+\mres) \over m_{\pi^*}^2} \langle 0 | P^a | \pi^* \rangle~. \end{eqnarray} If $m_{\pi^*}$ is not an NG boson, so $m_{\pi^*}$ remains nonzero, the right-hand side vanishes at the chiral limit, $(m_f\to-\mres)$. This prediction was checked on a lattice QCD using Wilson fermions[35]. and $f_{\pi^*}$ was consistent to be zero at the chiral limit. Figure <ref> and Table <ref> show the linear extrapolation of $f_{\pi^*}$. At the chiral limit, the $\pi^*$ decay constant is also consistent with the theoretical prediction, i.e. , $f_{\pi^*}\to0$. $f_{\pi^*}$ vs $m_f$. $f_{\pi^*}$ at the physical quark mass point $(m_f= m_{u,d})$ and the chiral limit $m_f$ $f_{\pi^*}$ $f_{\pi^*}^{\text{phys}}$ [MeV] $f_{\pi^*} r_0$ method $m_{u,d}$ 0.0013(12) 20(19) 0.0057(53) (D) $-m_{\text{res}}$ $-0.0003(13)$ $-05(20)$ $-0.0013(57)$ (D) Next we discuss the $\rho$ meson decay constant, $f_{\rho}$. The result of the fitting using eq. (<ref>) is shown in Table <ref>. The mass of the $\rho$ extracted by this fitting is consistent with those obtained from methods (A) and (B) within statistical error for all $m_f$. $m_{f}$ $m_{\rho}$ $f_\rho/Z_V$ $t_{\rm min}$ $t_{\rm max}$ method 0.02 0.5730(96) 0.2011(66) 9 13 (C) 0.03 0.6035(64) 0.2025(50) 10 14 (C) 0.04 0.6448(51) 0.2164(37) 9 14 (C) Then the decay constant at the physical quark mass point is obtained as \begin{eqnarray} f_\rho^{\text{phys}}=210 (15) \ \text{MeV} \end {eqnarray} by linear extrapolation (see Table <ref>). The renormalization factor, $Z_V$, which converts the lattice operator into the one in the continuum for $\overline{\rm MS}$ at $\mu=2$ GeV is necessary to obtain a physical value for the decay constants. We use $Z_A=0.75734(55)$, which was determined in the previous paper[25], and the relation $Z_V=Z_A$, assuming the good chiral symmetry of the current simulation. $f_{\rho}/Z_V$ vs $m_f$. $f_{\rho}$ at the physical quark mass point $(m_f= m_{u,d})$ $f_\rho/Z_V$ $f_{\rho}$ $f_{\rho}^{\text{phys}}$ [MeV] $f_{\rho} r_0$ method 0.1800(123) 0.1363(94) 210(15) 0.0583(41) (C) §.§ Systematic uncertainties So far we have mainly discussed error due to the limited size of the statistical sample. Our numerical results were obtained only at one lattice scale, in one space-time volume, for three quark masses heavier than the physical values, and the strange sea quark was neglected. In this section, various sources of systematic errors are listed and some of their magnitudes are very roughly estimated to compare our results with those of experiments. * Approximating the continuous space-time by a discrete lattice results in a discretization error. Using DWF, the error starts with ${\cal{O}}(\mres a)+{\cal{O}}(a^2\Lambda_{\text{QCD}}^2)$ . The value of $\mres a$ is negligibly small in our simulation compared with the large statistical error involved except in the case of the pion. Our results are closer to their continuum values than those obtained using a Wilson-type fermion on similar lattice scale. For quenched DWF QCD, the physical values of $f_\pi$, $f_K$, and $f_K/f_\pi$ shift by $\sim$5%, 3%, and 2%, respectively, when the lattice scale changes from $a^{-1}=2$ GeV to continuum limit[36], which are equal or less than current statistical error. * Because of the limited number of quark mass points calculated in our simulation, we restricted ourselves to using the simplest linear chiral extrapolations (<ref>) and that obtained from the AWTI (<ref>). A more appropriate extrapolation based on a larger number of quark mass points is the chiral fitting formula from the (partially quenched) chiral perturbation theory. While the mass of $\eta'$, which is investigated as the main topic in this work, shows little dependence on quark mass, a more precise chiral extrapolation to the physical quark mass point using lighter quark masses is needed to obtain more reliable results. * Although our assumption, that the ground state is a one-particle state is certainly wrong for some quantum numbers, some of the decay channel in nature are prohibited in simulations using degenerate up and down quarks with heavier mass in a relatively small spatial box (2 fm)$^3$ without a strange quark. More sophisticated investigations such as calculating the scattering amplitudes between multiparticles are needed to verify our spectrum results for the decaying meson. * Without results obtained from a larger volume, it is difficult to estimate the finite-volume effect, although it might be smaller than that for baryons. * Strange sea quark effect: The number of quark flavors that play dynamical roles in the $\eta'$ meson may be very important as seen in the WV relation , $m_{\eta'}^2\propto N_f$. By increasing $N_f$ from 2 to 3 by including strange quark, the WV prediction for $m_{\eta'}$ becomes $\sim$ 20% larger. A strange quark is, however, heavier than up/down quarks, and the mass of $\eta'$ in the $N_f=2+1$ QCD is likely to be in between the results of $N_f=2$ and 3. * Topological charge distribution and its effects to $\eta'$ meson: In our simulation, we deliberately used a special gauge action, DBW2, for good chiral symmetry. However, the autocorrelation time of the topological charge in the simulation becomes longer. The samples taken in our simulation may not be sufficiently long for the reliable estimation of the autocorrelation time for $Q_\text{top}$. The growth of the binned-jackknife error for $\vev{Q_\text{top}}$ with increasing bin size was monitored, and we estimated the autocorrelation time of roughly $\sim$ 300 trajectories for the $m_f=0.02$ ensemble and $\sim$ 200 trajectories for $m_f=0.03, 0.04$. Because of the less frequent tunneling between different topological sectors, the charge distribution sampled in our simulation may be statistically skewed. In fact, $\vev{Q_\text{top}}=-0.7(7), 1.4(6),$ and 1.8(4) for $m_f=0.02, 0.03,$ and 0.04, respectively. Note that the central value for $m_f=0.04$ is more than four standard deviations away from zero. It is conceivable that this poor sample of the topological sectors causes significant systematic errors in $\eta'$ spectrum, particularly for the $m_f=0.04$ ensemble. Figure <ref> shows the topological susceptibility, $\chi_\text{top}$ in (<ref>) as a function of quark mass. The fact that the susceptibility for all three masses is constant within two standard deviations implies that the simulation points are far from the lighter-quark-mass region, where the susceptibility may vanish as a linear function of quark mass. It is also possible the tunneling between different topological sectors does not occur sufficiently frequently, as shown in Figure <ref>; thus, the estimation for the susceptibility has a larger systematic error. Circles show the measured $\chi_{\text{top}}$ as a function of $m_f$ [25, 53] while squares show values calculated from $m_{\eta'}$ and $m_\pi$, as described in the next section. The horizontal line shows the value obtained from a pure SU(3) YM simulation[4]. The dotted line shows the prediction from chiral perturbation theory. History of the topological charge in the same simulation as that used for Fig. <ref>. Of course, more reliable estimation of the magnitude of these systematic errors may be carried out by future simulations on a finer and larger lattice using lighter quark masses with the strange sea quark effect, and with a larger statistical sample size. § SUMMARY AND DISCUSSIONS We have measured light meson propagators in all channel (flavor nonsinglet/singlet pseudoscalar, vector, scalar, pseudovector, and tensor meson $\pi$, $\rho$, $a_0$, $a_1$, $b_1$; $\eta'$, $\omega$, $f_0$, $f_1$, $h_1$) and estimated the ground state meson masses and some of leptonic decay constants, as well as, the excited state mass, in two flavors of domain wall QCD. The size of the statistical sample used in the calculation is increased by five to ten times higher than that reported previously [25]. By applying the gauge-invariant Wuppertal smearing because of quark operators for their better overlap with the ground-state, the statistical error of the pion and $\rho$ masses is reduced by approximately 50% and the reduction for $\eta'$ is more than 100%, i.e., we were only able to obtain the nonzero signal by using smeared field. To extract values for the meson mass and decay constant by fitting the propagators, we use two methods, the standard and variational methods. The results of these methods are consistent with each other, which indicates that excited-state contamination of the ground-state is controlled by the smearing. The systematic uncertainties discussed in the previous section are difficult to estimate; thus, we only quote results with statistical errors. Our results linearly extrapolated to the physical quark mass point are \begin{eqnarray*} &&a_{m_\rho}^{-1}=1.537(26) \ {\text {GeV} }~,\\ &&r_0=0.5491(93)\ {\text{fm}} \end{eqnarray*} for quantities directly related to the lattice scale, \begin{eqnarray*} &&f_{\rho}=210(15) \ {\text {MeV} }~,\\ &&f_{\pi^*}=20(19) \ {\text {MeV } } \end{eqnarray*} for decay constants, and \begin{eqnarray*} &&m_{a_{0}}=1.111(81) \ {\text {GeV} }~,\\ &&m_{\eta'}=819(127) \ {\text {MeV} }~,\\ &&m_{\omega}=790(194) \ {\text {MeV} }~,\\ &&m_{\pi^*}=1.791(138) \ {\text {GeV} }~,\\ &&m_{\rho^*}=2.028(131) \ {\text {GeV} }~, \\ &&m_{a_1}=1.140(51) \ {\text {GeV} }~, \\ &&m_{b_1}=1.203(64) \ {\text {GeV} }~, \\ &&m_{f_1}=1.033(137) \ {\text {GeV} }~, \\ &&m_{h_1}=1.225(250) \ {\text {GeV} }~, \end{eqnarray*} for the mass spectrum. The lattice scale is set from $m_\rho=775.49$ MeV. In Fig. <ref>, the meson masses obtained in this work are compared with the experimental values[1]. Horizontal bars show the experimental values and filled circles show the simulation results. The error bars indicate statistical errors only. Comparison of simulation results with experimental values[1] in the real world. Horizontal bars show the experimental values and filled circles show the simulation results. The error bars indicate statistical errors only. The squares show the quantities used to set the lattice spacing and the physical quark mass point. The decay constant of the excited pseudoscalar meson turns out to be consistent with zero at the chiral limit as expected: \begin{equation} f_{\pi^*} = -05(20) \ {\text {MeV }}~. \end{equation} The flavor singlet scalar meson, ${f_0}$, was too noisy to obtain its mass in our data. In this paper we chose the simplest noise method, complex ${\boldsymbol Z}_2$, for evaluating the quark loop amplitudes. More elaborate and/or sophisticated methods [37, 38, 39, 40, 41] may improve the statistical accuracy of the calculation. The recalculated pion mass is consistent with the previous result, but the results for $\rho$ and $a_0$ meson masses are significantly different. The central value of $m_\rho$ is 10% larger, i.e., $a^{-1}_{m_\rho}$ is 10% smaller, and the error bar is reduced by 50% compared with the previous results [25]. Both the central value and the error bar of $m_{a_0}$ are 25% smaller than those in previous results[26]. We confirm that the flavor singlet pseudoscalar meson, $\eta'$, is not an NG boson, and $m_{\eta'}$ is not likely to be zero at the chiral limit, which is consistent with the standard understanding of the axial anomaly. Assuming the WV relation (<ref>) is exact at $N_c=3$, one can calculate the mass gap, $m_0^2$, and topological susceptibility, $\chi_\text{top}$, from our values of $m_{\eta'}$ and $m_\pi$, \begin{equation} \chi_\text{top}(\text{WV}) = {f^2_\pi\over 2 N_f} m_0^2,~~ m_0^2 = m_\eta'^2-m_\pi^2 pp \end{equation} The value of $\chi_\text{top}(\text{WV})$ for $N_f=2$ are plotted in Fig. <ref> as squares. The horizontal line is $\chi_\text{top}$ obtained from a pure $SU(3)$ YM simulation[4]. For $m_f=0.02$ and 0.03, $\chi_\text{top}(\text{WV})$ is consistent with the quenched value, while $m_f=0.04$ point $\chi_\text{top}(\text{WV})$ undershoots the line significantly. By linearly extrapolating to the chiral limit, we obtained $m_0^2=(808 (129){\text{ MeV}})^2$ and $\chi_{\text{top}}(\text{WV})= (193 (15){\text{ MeV}})^4$, which is consistent with the quenched value[4] $(191 (5){\text{ MeV}})^4$. The agreement, which may imply only small $1/N_c$ correction, is interesting and deserves further investigation in future. These results are susceptible to various systematic errors. First, we have only two flavors of dynamical quarks. The omission of the strange quark and antiquark pairs in vacuum, whose mass is comparable to the dynamical scale of the QCD, may skew our results significantly. The limited number of quark masses, three unitary points, restricted us to examining only the simplest function for the quark mass dependence of the physical results. Thus, the chiral extrapolation has a systematic error due to the omission of curvature resulting from the chiral logarithms and higher order terms although many of our results show little dependence on quark mass. The ensemble was generated only on a $16^3\times 32$ lattice with periodic boundary condition in the space directions; thus, all the meson spectrum is affected by the “mirror” images located $\sim 2$ fm away from the original image in each of the three spatial directions. The effects may be as large as $\sim$ 10% for the lightest quark mass points. The lattice discretization error in this study is small, ${\cal{O}}(\mres a)+{\cal{O}}(a^2\Lambda_{\text{QCD}}^2)\sim {\cal O}(1{\rm\%})$. The previous careful studies[25] on the scaling violation show a $\sim$ 5% level shift for $a\sim 0.1$ fm lattices. The omission of the isospin violation due to the differences in quark mass and electric charge is likely to be negligible compared with other sources of errors, but this issue can also be studied nonperturbatively using Despite the significant statistical error and the various remaining systematic uncertainties, this study should serve a benchmark calculation for the statistical features of difficult physical quantities, disconnected diagrams, and computational feasibility tests. The results for the mass of $\eta'$ were close to the experimental value, indicates that further improvements can be made particularly calculation using an $N_f=2+1$ DWF ensemble[33, 34]. § ACKNOWLEDGEMENTS We thank RIKEN, Brookhaven National Laboratory, and the U.S. Department of Energy for providing the facilities essential for the completion of this work. We are grateful to members of the RBC collaboration, especially to T. Blum, N. Christ, C. Dawson, R. Mawhinney, K. Orginos, and A. Soni for their various contributions in the early stages of this work and their continuous The QCDOC supercomputer at the RIKEN-BNL Research Center (RBRC) was used for the numerical calculations in this work. K.H. thanks RBRC for its hospitality while this work was partly performed. We are grateful to the authors and maintainers of the CPS[54], which was used in this work. This work is supported in part by the Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology (No. 17750050). [1] W. M. Yao et al. [Particle Data Group], J. Phys. G 33, 1 (2006). [2] E. Witten, Nucl. Phys. B 156, 269 (1979). [3] G. Veneziano, Nucl. Phys. B 159, 213 (1979). [4] L. Del Debbio, L. Giusti and C. Pica, Phys. Rev. Lett. 94, 032003 (2005) [5] S. Itoh, Y. Iwasaki and T. Yoshie, Phys. Rev. D 36, 527 (1987). [6] Y. Kuramashi, M. Fukugita, H. Mino, M. Okawa and A. Ukawa, Phys. Rev. Lett. 72, 3448 (1994). [7] M. Fukugita, Y. Kuramashi, M. Okawa and A. Ukawa, Phys. Rev. D 51, 3952 (1995). [8] V. I. Lesk et al. [CP-PACS Collaboration], Phys. Rev. D 67, 074503 (2003) [9] C. McNeile and C. Michael [UKQCD Collaboration], Phys. Lett. B 491, 123 (2000) [Erratum-ibid. B 551, 391 (2003)] [10] C. R. Allton et al. [UKQCD Collaboration], Phys. Rev. D 70, 014501 (2004) [11] T. Struckmann et al. [TXL Collaboration], Phys. Rev. D 63, 074503 (2001) [12] K. Schilling, H. Neff and T. Lippert, Lect. Notes Phys. 663, 147 (2005) [13] S. Aoki et al. [JLQCD Collaborations], PoS LAT2006, 204 (2006) [14] L. Venkataraman and G. Kilcup, [15] E. B. Gregory, A. C. Irving, C. M. Richards and C. McNeile, PoS LAT2006, 176 (2006) [16] S. R. Sharpe, PoS LAT2006, 022 (2006) [17] M. Creutz, arXiv:0708.1295 [hep-lat]. [18] D. B. Kaplan, Phys. Lett. B 288, 342 (1992) [19] Y. Shamir, Nucl. Phys. B 406, 90 (1993) [20] V. Furman and Y. Shamir, Nucl. Phys. B 439, 54 (1995) [21] P. Hernandez, K. Jansen and M. Luscher, Nucl. Phys. B 552 (1999) 363. [22] T. Takaishi, Phys. Rev. D 54, 1050 (1996). [23] P. de Forcrand et al. [QCD-TARO Collaboration], Nucl. Phys. B 577, 263 (2000) [24] Y. Aoki et al., Phys. Rev. D 69, 074504 (2004) [25] Y. Aoki et al., Phys. Rev. D 72, 114505 (2005) [26] S. Prelovsek, C. Dawson, T. Izubuchi, K. Orginos and A. Soni, Phys. Rev. D 70, 094503 (2004) [27] S. Gusken, Nucl. Phys. Proc. Suppl. 17, 361 (1990). [28] C. Michael, Nucl. Phys. B 259, 58 (1985). [29] M. Luscher and U. Wolff, Nucl. Phys. B 339, 222 (1990). [30] T. Blum, T. Doi, M. Hayakawa, T. Izubuchi and N. Yamada, arXiv:0708.0484 [hep-lat]. [31] V. Gadiyak and O. Loktik, Phys. Rev. D 72, 114504 (2005) [32] C. Dawson, T. Izubuchi, T. Kaneko, S. Sasaki and A. Soni, Phys. Rev. D 74, 114502 (2006) [33] D. J. Antonio et al. [RBC and UKQCD Collaborations], Phys. Rev. D 75, 114501 (2007) [34] C. Allton et al. [RBC and UKQCD Collaborations], Phys. Rev. D 76, 014504 (2007) [35] C. McNeile and C. Michael [UKQCD Collaboration], Phys. Lett. B 642, 244 (2006) [36] Y. Aoki et al., Phys. Rev. D 73, 094507 (2006) [37] J. Foley, K. Jimmy Juge, A. O'Cais, M. Peardon, S. M. Ryan and J. I. Skullerud, Comput. Phys. Commun. 172, 145 (2005) [38] A. Duncan and E. Eichten, Phys. Rev. D 65, 114502 (2002) [39] T. A. DeGrand and S. Schaefer, Comput. Phys. Commun. 159, 185 (2004) [40] C. Michael and J. Peisa [UKQCD Collaboration], Phys. Rev. D 58, 034506 (1998) [41] S. Collins, G. Bali and A. Schafer, arXiv:0709.3217 [hep-lat]. [42] C. McNeile and C. Michael [UKQCD Collaboration], Phys. Rev. D 74, 014508 (2006) [43] K. F. Liu, Prog. Theor. Phys. Suppl. 168, 160 (2007) [arXiv:0706.1262 [hep-ph]]. [44] A. Hart, C. McNeile and C. Michael [UKQCD Collaboration], Nucl. Phys. Proc. Suppl. 119, 266 (2003) [45] T. Burch, C. Gattringer, L. Y. Glozman, C. Hagen, C. B. Lang and A. Schafer, Phys. Rev. D 73, 094505 (2006) [46] W. A. Bardeen, A. Duncan, E. Eichten, N. Isgur and H. Thacker, Phys. Rev. D 65, 014509 (2001) [47] E. B. Gregory, A. C. Irving, C. M. Richards and C. McNeile, arXiv:0709.4224 [hep-lat]. [48] E. B. Gregory, A. Irving, C. M. Richards, C. McNeile and A. Hart, arXiv:0710.1725 [hep-lat]. [49] C. Michael and C. Urbach [ETM Collaboration], PoS LAT2007, 122 (2007) [arXiv:0709.4564 [hep-lat]]. [50] T. Izubuchi, S. Aoki, K. Hashimoto, Y. Nakamura, T. Sekido and G. Schierholz, PoS LAT2007, 106 (2007) [arXiv:0802.1470 [hep-lat]]. [51] Y. Shamir, Nucl. Phys. B 417, 167 (1994) [52] T. Blum et al., Phys. Rev. D 66, 014504 (2002) [53] F. Berruto, T. Blum, K. Orginos and A. Soni, Phys. Rev. D 73, 054509 (2006) [54] ${\tt http://qcdoc.phys.columbia.edu/chulwoo_index.html}$

arxiv-papers

2024-09-04T02:48:54.092651

0803.0310

# Direct observation of a Fermi surface and superconducting gap in LuNi2B2C P. Starowicz Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA M. Smoluchowski Institute of Physics, Jagiellonian University, Reymonta 4, 30-059 Kraków, Poland C. Liu Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA R. Khasanov Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA Physik-Institut der Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland T. Kondo Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA G. Samolyuk Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA D. Gardenghi Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA Bob Jones University, Greenville, SC 29614, USA Y. Lee Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA T. Ohta Advanced Light Source, Berkeley National Laboratory, Berkeley, CA 94720, USA B. Harmon Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA P. Canfield Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA S. Bud’ko Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA E. Rotenberg Advanced Light Source, Berkeley National Laboratory, Berkeley, CA 94720, USA A. Kaminski Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA ###### Abstract We measured the Fermi surface (FS), band dispersion and superconducting gap in LuNi2B2C using Angle Resolved Photoemission Spectroscopy. Experimental data were compared with the tight-binding version of the Linear Muffin-Tin Orbital (LMTO) method and Linearized Augmented Plane-Wave (LAPW) calculations. We found reasonable agreement between the two calculations and experimental data. The measured FS exhibits large parallel regions with a nesting vector that agrees with a previous positron annihilation study and calculations of the generalized susceptibility. The measured dispersion curves also agree reasonably well with the TB-LMTO calculations, albeit with some differences in the strength of the hybridization. In addition, the spectrum in the superconducting state revealed a 2meV superconducting gap. The data also clearly shows the presence of a coherent peak above the chemical potential, $\mu$ that originates from thermally excited electrons above the energy of 2$\Delta$. This feature was not previously observed in the Lu-based material. ###### pacs: 74.70.Dd, 71.18.+y, 71.20.-b, 71.27.+a ## I Introduction Figure 1: (Color online) Comparison between the Fermi surface maps measured by ARPES and the linear muffin-tin orbital (TB-LMTO) calculation. (a) Sketch of the first Brillouin zone for LuNi2B2C. (b-d) ARPES mapping at the chemical potential for incident photon energies of 128.13 eV, 119.44 eV, and 102.98 eV, respectively. (e-g) The Fermi surface maps obtained by TB-LMTO calculations for constant $k_{z}$ values equal to (e) 0.2, (f) 0.15 and (g) 0.8 expressed in the units of $\Gamma$ \- Z distance. Rare earth nickel borocarbides RNi2B2C (R - rare earth) constitute an interesting class of materials Cava1994 ; Nagarajan1994 ; Canfield1998 ; MullerNarozhnyi2001 ; MazumdarNagarajan2005 , in which there is a competition and coexistence between superconductivity and magnetism. Amongst these compounds, nonmagnetic LuNi2B2C has the highest superconducting critical temperature of 16.6 K Cava1994 . The borocarbides exhibit a peculiar anisotropy of the superconducting gap, the character of which is still under debate. It is believed that the gap is highly anisotropic in the two non- magnetic compounds LuNi2B2C and YNi2B2C Boaknin2001 ; Bobrov2005 ; Maki2002 ; Raychaudhuri2004 ; MartinezSamper2003 ; Izawa2002 ; Yokoya2000 . Its symmetry was proposed to be s \+ g Maki2002 , which is consistent with certain experimental results Raychaudhuri2004 but an anisotropic s-wave symmetry has also been considered MartinezSamper2003 . Other experimental data indicate that the gap in YNi2B2C has point nodes along the (100) and (010) directions Izawa2002 . LuNi2B2C crystallizes in a body-centered tetragonal structure with lattice parameters a = 3.4639 Å, and c = 10.6313 ÅSiegrist1994 . Its crystal structure consists of Lu-C layers with Ni2B2 sheets in between. Previously calculations reveal that LuNi2B2C is characterised by a large density of states (DOS) at the Fermi energy ($E_{F}$) originating mainly, but not exclusively, from Ni d electrons Mattheiss1994 ; PickettSingh1994 ; Coehoorn1994 . Another interesting feature is a flat band along the $\Gamma$-X direction just above $E_{F}$. The Fermi surface (FS) topography of LuNi2B2C was studied by ab-initio calculations Rhee1995 ; Kim1995 ; Dugdale1999 . Band structure calculations Rhee1995 revealed a pronounced maximum in the generalized electronic susceptibility at ($\sim$0.6a*, 0, 0), where a* $\equiv 2\pi/a$ and most likely arising from large nested regions of the FS. Moreover, phonon softening was observed in LuNi2B2C by means of inelastic neutron scattering for a range of wave vectors around (0.5a*, 0, 0) Dervenagas1995 . Interestingly enough the magnetic ordering, which was found in RNi2B2C compounds with magnetic atoms R = Er, Ho, Tb and Gd manifest similar modulation vector usually close to (0.55a*,0,0)MullerNarozhnyi2001 . The first experimental studies of the LuNi2B2C (RNi2B2C) Fermi surface were performed by means of two-dimensional angular correlation of electron-positron annihilation radiation (2D-ACAR) and the data were compared to the Linear Muffin-Tin Orbital (LMTO), local density approximation (LDA) calculations Dugdale1999 . Nested parts of the FS were found with a nesting vector corresponding to both the phonon softening and the magnetic modulation vectors. The fraction of the FS participating in nesting was determined to be 4.4 $\pm$ 0.5% Dugdale1999 . That study was however limited only to a rough “calipering” of the Fermi surface. Knowledge of the experimental band structure, Fermi surface and quasiparticle properties is deemed essential to understand the interplay of the various interactions in these materials, as it may shed new light on other phenomena such as anisotropic superconductivity, the role of phonon softening and the relationship between the superconductivity and magnetic ordering in the borocarbides. It is also a pre-requisit for direct determination of the alleged anisotropy of the superconducting gap in these materials. In this report we present angle resolved photoelectron spectroscopy (ARPES) measurements of the band dispersion, Fermi surface and supercondcuting gap in the borocarbide with the highest $T_{c}$, LuNi2B2C. The experimental results were compared with the tight-bounding LMTO (TB-LMTO) method and the full potential LAPW calculations. We found reasonable agreement with theory. The most significant difference between the calculations and experimental data is the strength of the hybridization. We also determined the superconducting gap to be 2.58 meV (extrapolated for T=0), in good agreement with the gap expected from the superconducting transition temperature (${{2\Delta}/{k_{B}T_{C}}}=2.78$). ## II Experimental LuNi2B2C single crystals were grown at Ames Laboratory by means of a high- temperature flux techniqueCanfield1998 ; Fisher2001 . The plate-like crystals were cleaved in situ at pressures better than 3 x 10-11 Tr to reveal and maintain fresh a-b surfaces. The Fermi surface and band structure mapping were performed at the 7.0.1 beamline at the Advanced Light Source, using a Scienta R4000 analyzer. The energy and angle resolution were set at $\sim$ 30 meV and $\sim$ 0.5 deg, respectively. The energy gap was measured with a Scienta 2002 analyser and He-I photon source (h$\nu$ = 21.2 eV), in which the overall energy resolution was set at 2 meV. The normal state data were measured at the Synchrotron Radiation Center using the PGM beamline and Scienta 2002 endstation, with the energy and angular resolution set at $\sim$ 13 meV and 0.25 deg, respectively. Tight-binding linear muffin-tin orbital calculations were performed by the TB-LMTO program, version 47 LMTO , and the Full- Potential Linearized Augmented Plane-Wave (LAPW) calculations were performed using the Wien2k package Wien . Figure 2: Dispersion of the conduction bands obtained with a photon energy of 102.98 eV (same as Fig. 1d), compared with the TB-LMTO calculation. a) Fermi surface map with momentum cuts indicated by the solid lines, in which cut (c) and (i) pass through the $\Gamma$ and X points, respectively. b) Fermi surface contours obtained by the TB-LMTO calculation for the value of $k_{z}$ corresponding to the data in panel (a). Panels (c-i): measured band dispersion along the cuts indicated in (a). Panels (j-p): calculated band dispersion along the cuts marked in panel (b). ## III Results and discussion Band structure and semi-planar Fermi Surface cuts were determined for incident photon energies 128.13 eV, 119.44 eV and 102.98 eV (Fig. 1b-d), where the $\Gamma$ point in the Brillouin zone (Fig. 1a) corresponds to normal emission of electrons along the (001) direction. A $\sin r/r$ correction term ($r$ is the distance from the $\Gamma$ point) was used to account for mapping of the momentum space onto the angular distribution of photoelectrons. The ARPES process in 3D materials leaves some ambiguity as to the $k_{z}$ component of the momentum (perpendicular to the sample surface), because it is not conserved in the photoemission process. This is due to jump of the potential at the sample surface. From conservation of energy and remaining components of the momentum one can calculate the relative changes of the $k_{z}$ for various photon energies. To obtain the offset one needs to seek guidance from the band structure calculations and identify the high symmetry points in the data.HUFNER This allows estimation of the $k_{z}$ offset. The change of the wave vector component $k_{z}$ (parallel to the c axis) between scans in Fig. 1b ($h\nu$=128.13) and 1c ($h\nu$=119.44) was calculated from momentum and energy conservation to be 0.33 of the $\Gamma$-Z distance. Similarly the change of the wave vector component $k_{z}$ between scans in Fig. 1c ($h\nu$=119.44) and 1d ($h\nu$=102.98) was 0.66 of the $\Gamma$-Z. The calculated Fermi surfaces were obtained for constant $k_{z}$ values by means of the TB-LMTO method and are shown in Fig. 1 panels e-g. We estimated the values of the inner potential, $V_{0}$ = 9.4 eV and the work function $\phi$ = 4.6 eV, by comparing the high symmetry points between the calculated Fermi surfaces and the experimental data. This allowed us to determine the offset of the photon energy that corresponds to $k_{z}$ = 0. The Fermi surface maps for the incident photon energies of 128.13 eV and 119.44 eV reveal large parallel parts of the FS with essentially the same nesting vector (spacing between the linear sections): $k_{n}$ = (0.59 $\pm$ 0.04)a* for Fig. 1b and $k_{n}$ = (0.58 $\pm$ 0.04)a* for Fig. 1c. $k_{z}$ is expressed in the units of the $\Gamma$-Z distance, where the $\Gamma$ point corresponds to $k_{z}$ = 0. Although the full three dimensional FS was not determined in great detail, a similar nesting vector was found for different $k_{z}$ values which indicates that the FS likely has considerable nesting properties for a wide range of $k_{z}$ values. The constancy of the value of the nesting vector between $k_{z}$=0.15 and -0.2 is also consistent with results of calculations. The spacing between the parallel segments of the Fermi surface predicted by TB-LMTO calculation is between 0.54a* and 0.55a*, the LAPW calculation results (not shown) are 0.50a* and 0.57a*, respectively. The detected $k_{n}$ is very close to the theoretically predicted value obtained from the generalized susceptibility Rhee1995 . Our results also agree reasonably well with the nesting vector previously determined via 2D-ACAR Dugdale1999 . The Fermi surface map obtained at 102.98 eV very closely resembles the calculated Fermi surface for $k_{z}$ = 0.8a*. The overall shapes of the measured and calculated Fermi surface sheets (Fig. 1d and 1g) are very similar, however there is one significant difference. In the calculations the four oval parts of the Fermi surface centered about $\Gamma$-Z are well separated in momentum space (Fig. 1g), while the data reveals that they actually are connected at the edges (Fig. 1d). A lack of separation in the experimental data may indicate that the hybridization gap is overestimated in the calculations. These oval parts arise from the intersection of the electron and hole-like bands. Interestingly enough at the edges along the diagonal directions (e. g. 110) the bottom of the electron band and the top of the hole band appear to be pinned at the chemical potential, resulting in a characteristic “flower” shape. In Fig. 2 we plot the band dispersion data along a few selected cuts in momentum space obtained at an incident photon energy of 102.98 eV (Figs. 2c-i), along with a calculated (TB-LMTO method) band dispersion for $k_{z}\sim$ 0.8a* (Figs. 2j-p). The agreement between the measured and calculated band dispersion is rather good, especially in the proximity of the chemical potential. In the corresponding TB-LMTO calculations (Figs. 2j-p), the same overall features are well reproduced, which shows the validity of the calculation in this material to a certain extent. This agreement also validates the assignment of $k_{z}$ values to the cuts measured at various photon energies, which is very important when studying 3D materials with ARPES. The most significant difference is the hybridization gap, which is quite large in the calculations but its signatures are for the most part absent in the measured data. For example in Fig. 2 panels (k) and (l) the high and low energy branches form hybridization gap of about 200 meV at E = -0.3 eV, while in the corresponding measured data (panels (d) and (e)) the bands appear to disperse without a signature of the hybridization gap. One should consider if the observed features have any relation to the superconducting gap asymmetry and observed phonon softening in LuNi2B2C. It should be noted that band structure calculations PickettSingh1994 ; Kim1995 show a flat band lies very close to, but slightly above, the Fermi level. This feature was unfortunately not observed in our data due to the Fermi function cut-off. However, a higher DOS near the Fermi level would explain the large number of scattered electrons observed with k vectors along (110) and phonon softening for the discussed wave vectors. Consequently this may lead to an anisotropy of the superconducting order parameter. This is in agreement with the results proposing for YNi2B2C that the superconducting gap is larger just at (110) and diminishes or even has nodes along the (100) and (010) directions Izawa2002 . Given the above concern, we measured the energy gap in LuNi2B2C by partial angle-integrated photoelectron spectroscopy and compared with the normal state Fermi surface. The opening of the superconducting gap is clearly shown in Fig. 3. In order to determine the magnitude of the gap, the Dynes function Dynes1978 was fitted to the symmetrized Norman1998 spectrum (Fig. 3b). The fitted function yields the gap value of $\Delta$ = 1.5 meV for the sample at T = 11 K with the $\Gamma$ parameter equal to 0.05 meV. The striking feature in Fig. 3c is the pronounced peak above the chemical potential. This peak arises from thermal excitation of electrons above the 2$\Delta$. This points to high DOS just above $\mu$ which is consistent with the idea that the flat band along (110) direction, a large part of which is slightly over the Fermi energy playing an important role in this anisotropic superconductivity. Similar peaks were recently reported in Y based borocarbides YOKOYA2007 . According to BCS theory the energy gap value at zero temperature ($\Delta_{0}$) is 2.45 meV for $T_{c}$ = 16 K superconductor. Our $\Delta$ value of 1.5 meV obtained at T=11K corresponds to ($\Delta_{0}$)=2.6 meV, in excellent agreement with the BCS predictions. Figure 3: (Color online) Superconducting gap of LuNi2B2C a) measured at T = 11$\pm$1 K, compared with the normal state at T = 40 K. b) The Dynes function (solid red line) with the parameters $\Delta$ = 1.5 meV and $\Gamma$ = 0.05 meV fitted to the symmetrised spectrum (solid black circles). c) enlarged portion of superconducting spectra from (a) close to the chemical potential. ## IV Conclusions We have performed measurements of the Fermi surface, band dispersion and supercondcuting gap for highest $T_{c}$ rare earth nickel borocarbide superconductor LuNi2B2C. The experimental data were compared with two different density functional calculations. The overall agreement between theory and measurement is good. In the experiment, large parallel FS parts spaced with the vector $k_{n}$ = 0.59a* have been found for two different incident photon energies, which is a confirmation of the previous theoretical predictions Rhee1995 and earlier experimental studies Dugdale1999 . The calculated FS confirms the existence of large nested parts, with a nesting vector in good agreement with the ARPES results presented here. The superconducting gap was measured and we also observed a coherent peak above the chemical potential. This peak arises due to electrons being thermally excited above the energy of 2$\Delta$. ## V Acknowledgments This work was supported by Director Office for Basic Energy Sciences, US DOE. Work at Ames Laboratory was supported by the Department of Energy - Basic Energy Sciences under Contract No. DE-AC02-07CH11358. Advanced Light Source is operated by the U.S. DOE under Contract No. DE-AC03-76SF00098. Synchrotron Radiation Center is supported by the National Science Foundation under award No. DMR-0537588. R. K. gratefully acknowledges support of K. Alex Müller Foundation. ## References * (1) R. J. Cava, H. Takagi, H. W. Zandbergen, J. J. Krajewski, W. F. Peck Jr, T. Siegrist, B. Batlogg, R. B. van Dover, R. J. Felder, K. Mizuhashi, J. O. Lee, H. Eisaki, and S. Uchida, Nature 367, 252 (1994). * (2) R. Nagarajan, C. Mazumdar, Z. Hossain, S.K. Dhar, K. V. Gopalakrishnan, L. C. Gupta, C. Godart, B. D. Padalia, and R. Vijayaraghavan, Phys. Rev. Lett. 72, 274 (1994). * (3) P. C. Canfield, P. L. Gammel, and D. J. Bishop, Physics Today 51, 40 (1998). * (4) K.-H. Müller and V. N. Narozhnyi, Rep. Prog. Phys. 64, 943 (2001). * (5) C. Mazumdar and R. Nagarajan, Current Science 88, 83 (2005). * (6) T. Yokoya, T. Kiss, T. Watanabe, S. Shin, M. Nohara, H. Takagi, and T. Oguchi, Phys. Rev. Lett. 85, 4952 (2000). * (7) E. Boaknin, R. W. Hill, C. Proust, C. Lupien, and L. Taillefer, Phys. Rev. Lett. 87, 237001 (2001). * (8) N. L. Bobrov, S. I. Beloborod’ko, L. V. Tyutrina, I. K. Yanson, D. G. Naugle, and K. D. D. Rathnayaka, Phys. Rev. B 71, 014512 (2005). * (9) K. Maki, P. Thalmeier, and H. Won, Phys. Rev. B 65, 140502 (2002). * (10) P. Raychaudhuri, D. Jaiswal-Nagar, Goutam Sheet, S. Ramakrishnan, and H. Takeya, Phys. Rev. Lett. 93, 156802 (2004). * (11) P. Martinez-Samper, H. Suderow, S. Vieira, J. P. Brison, N. Luchier, P. Lejay, and P. C. Canfield, Phys. Rev. B 67, 014526 (2003). * (12) K. Izawa, K. Kamata, Y. Nakajima, Y. Matsuda, T. Watanabe, M. Nohara, H. Takagi, P. Thalmeier, and K. Maki, Phys. Rev. Lett. 89, 137006 (2002). * (13) T. Siegrist, H. W. Zandbergen, R. J. Cava, J. J. Krajewski, and W. F. Peck Jr, Nature 367, 254 (1994). * (14) L. F. Mattheiss, Phys. Rev. B 49, 13279 (1994). * (15) W. E. Pickett and D. J. Singh, Phys. Rev. Lett. 72 , 3702 (1994). * (16) R. Coehoorn, Physica C 228, 5671 (1994). * (17) J. Y. Rhee, X. Wang, and B. N. Harmon, Phys. Rev. B 51, 15585 (1995). * (18) H. Kim, C.-D. Hwang, and J. Ihm, Phys. Rev. B 52, 4592 (1995). * (19) S. B. Dugdale, M. A. Alam, I. Wilkinson, R. J. Hughes, I. R. Fisher, P. C. Canfield, T. Jarlborg, and G. Santi, Phys. Rev. Lett. 83, 4824 (1999). * (20) P. Dervenagas, M. Bullock, J. Zarestky, P. Canfield, B. K. Cho, B. Harmon, A. I. Goldman, and C. Stassis, Phys. Rev. B 52, 9839 (1995). * (21) P. C. Canfield, I. R. Fisher, J. Crystal Growth, 225, 155-161 (2001) . * (22) O. Jepsen and O. K. Anderson, Solid State Commun. 9, 1763 (1971) and Phys. Rev. B 29, 5965 (1984); P. Blöchl, O. Jepsen and O. K. Anderson, Phys. Rev. B 49, 16223 (1994). * (23) P. Blaha, K. Schwarz, G. Madsen, D. Kvasnicka and J. Luitz, (2001) _WIEN2k, An Augmented Plane Wave_ \+ _Local Orbitals Program for Calculating Crystal Properties_ (Karlheinz Schwarz, Tech. Univ. Wien, Austria). * (24) S. Hufner, Photoelectron Spectroscopy, pp 268-270 (Springer, Berlin 1995). * (25) R. C. Dynes, V. Narayanamurti, and J. P. Garno, Phys. Rev. Lett. 41, 1509 (1978). * (26) M. R. Norman, H. Ding, M. Randeria, J. C. Campuzano, T. Yokoya, T. Takeuchi, T. Takahashi, T. Mochiku, K. Kadowaki, P. Guptasarma, and D. G. Hinks, Nature 392, 157 (1998). * (27) T. Baba, T. Yokoya, S. Tsuda, T. Kiss, T. Shimojima, K. Ishizaka, H. Takeya, K. Hirata, T. Watanabe, M. Nohara, H. Takagi, N. Nakai, K. Machida, T. Togashi, S. Watanabe, X.-Y. Wang, C. T. Chen, and S. Shin, Phys. Rev. Lett. 100, 017003 (2008).

arxiv-papers

2024-09-04T02:48:54.102257

0803.0359

# Stochastic Hard-Sphere Dynamics for Hydrodynamics of Non-Ideal Fluids Aleksandar Donev Lawrence Livermore National Laboratory, P.O.Box 808, Livermore, CA 94551-9900 Berni J. Alder Lawrence Livermore National Laboratory, P.O.Box 808, Livermore, CA 94551-9900 Alejandro L. Garcia Department of Physics, San Jose State University, San Jose, California, 95192 ###### Abstract A novel stochastic fluid model is proposed with non-ideal structure factor consistent with compressibility, and adjustable transport coefficients. This Stochastic Hard Sphere Dynamics (SHSD) algorithm is a modification of the Direct Simulation Monte Carlo (DSMC) algorithm and has several computational advantages over event-driven hard-sphere molecular dynamics. Surprisingly, SHSD results in an equation of state and pair correlation function identical to that of a deterministic Hamiltonian system of penetrable spheres interacting with linear core pair potentials. The fluctuating hydrodynamic behavior of the SHSD fluid is verified for the Brownian motion of a nano- particle suspended in a compressible solvent. With the increased interest in nano- and micro-fluidics, it has become necessary to develop tools for hydrodynamic calculations at the atomistic scale Noguchi et al. (2007); Fabritiis et al. (2007). Of particular interest is the modeling of flexible polymers in a flowing solvent for both biological (e.g., cell membranes) and engineering (e.g., micro-channel DNA arrays) applications. Typically the polymer chains are modeled using Molecular Dynamics (MD). For many applications, a realistic representation of the solvent and bidirectional coupling between the flow and the polymer motion is needed, for example, in the modeling of turbulent drag reduction. Previously, we introduced the Stochastic Event-Driven MD (SEDMD) algorithm that uses Direct Simulation Monte Carlo (DSMC) for the solvent coupled to deterministic EDMD for the polymer chain Donev et al. (2008). However, DSMC is limited to perfect gases. Efforts have been undertaken to develop solvents that have a _non-ideal_ EOS, and that also have greater computational efficiency than brute-force molecular dynamics. Examples include the Lattice-Boltzmann (LB) method Luo (2000), Dissipative Particle Dynamics (DPD) Pagonabarraga and Frenkel (2000), and Multi-Particle Collision Dynamics (MPCD) Ihle et al. (2006), each of which has its own advantages and disadvantages Noguchi et al. (2007). The _Stochastic Hard Sphere Dynamics_ (SHSD) algorithm described in this Letter is based on successive stochastic collisions of variable hard- sphere diameters and is thermodynamically consistent (i.e., the direct calculation of compressibility from density fluctuations agrees with the density derivative of pressure). SHSD modifies previous algorithms for solving the Enskog kinetic equation Frezzotti (1997); Montanero and Santos (1997) while maintaining good efficiency. In the SHSD algorithm randomly chosen pairs of approaching particles that lie less than a given diameter of each other undergo collisions as if they were hard spheres of diameter equal to their actual separation. The SHSD fluid is shown to be non-ideal, with structure and equation of state equivalent to that of a fluid mixture where spheres effectively interact with a repulsive linear core pairwise potential. We theoretically demonstrate this correspondence at low densities. Remarkably, we numerically find that this effective interaction potential, similar to the quadratic core potential used in many DPD variants, is valid at all densities. Therefore, the SHSD fluid, as DPD, is _intrinsically_ thermodynamically-consistent, while non-ideal MPCD is only _numerically_ thermodynamically-consistent for tuned choices of the parameters Ihle et al. (2006); Tüzel et al. (2006). As an algorithm, SHSD is similar in nature to DPD and has a similar computational complexity. In DPD, momentum is also stochastically exchanged between particles closer than a given distance. The essential difference is that DPD has a continuous-time formulation (a system of stochastic ODEs), where as the SHSD dynamics is discontinuous in time. This is similar to the difference between MD for continuous potentials and discontinuous potentials. Just as DSMC is a stochastic alternative to hard-sphere MD for low-density gases, SHSD is a stochastic modification of hard-sphere MD for dense gases. On the other hand, DPD is a modification of MD for smooth potentials to allow for larger time-steps and a hydrodynamically-consistent thermostat. The SHSD algorithm is not as efficient as DSMC at a comparable collision rate. However, when low compressibility is desired, SHSD is several times faster than EDMD for hard spheres, the fastest available deterministic alternative. Low compressibility, for example, is desirable so that flows are kept subsonic even for high Reynolds number flows. Furthermore, SHSD has several important advantages over EDMD, in addition to its simplicity: (1) SHSD has several controllable parameters that can be used to change the transport coefficients and compressibility, while EDMD only has density; (2) SHSD is time-driven rather than event-driven thus allowing for easy parallelization; (3) SHSD can be more easily coupled to continuum hydrodynamic solvers, just like ideal-gas DSMC Williams et al. (2008). Strongly-structured particle systems, such as fluids with strong interparticle repulsion (e.g., hard spheres), are more difficult to couple to hydrodynamic solvers Delgado-Buscalioni and Fabritiis (2007) than ideal fluids, such as MPCD or DSMC, or weakly-structured fluids, such as DPD or SHSD fluids. The standard DSMC Alexander and Garcia (1997) algorithm starts with a time step where particles are propagated advectively, $\mathbf{r}_{i}^{{}^{\prime}}=\mathbf{r}_{i}+\mathbf{v}_{i}\Delta t$, and sorted into a grid of cells. Then, a certain number $N_{coll}\sim\Gamma_{sc}N_{c}(N_{c}-1)\Delta t$ of _stochastic collisions_ are executed between pairs of particles randomly chosen from the $N_{c}$ particles inside the cell. The conservative stochastic collisions exchange momentum and energy between two particles $i$ and $j$ that is not correlated with the actual positions of the particles. Typically the probability of collision is made proportional to the magnitude of the relative velocity $v_{r}=\left|\mathbf{v}_{ij}\right|$ by using a conventional rejection procedure. DSMC, unlike MD, is not microscopically isotropic and does not conserve angular momentum, leading to an anisotropic stress tensor. To avoid such grid artifacts, all collision partners within a collision diameter $D$ must be considered even if they are in neighboring cells, and, if angular momentum conservation is required, only radial momentum should be exchanged in collisions as for hard spheres. This grid-free variant will be called Isotropic DSMC (I-DSMC). The cost is that is the computational efficiency is reduced by a factor of $2-3$ due to the need to perform neighbor searches. Note that a pairwise Anderson thermostat proposed within the context of MD/DPD in Ref. Lowe (1999) essentially adds (thermostated) I-DSMC collisions to ordinary MD and has very similar computational behavior. As in I-DSMC, in SHSD we consider particles in neighboring cells as collision partners in order to ensure isotropy of the collisional (non-ideal) component of the pressure tensor. The virial $\left\langle\Delta\mathbf{v}_{ij}\cdot\Delta\mathbf{r}_{ij}\right\rangle$ vanishes in I-DSMC giving an ideal-gas pressure. In order to introduce a non- trivial equation of state it is necessary to either give an additional displacement to the particles that is parallel to $\Delta\mathbf{v}_{ij}$, or to bias the momentum exchange $\Delta\mathbf{v}_{ij}$ to be (statistically) aligned to $\Delta\mathbf{r}_{ij}$. The former approach has already been investigated in the Consistent Boltzmann Algorithm (CBA) Alexander et al. (1995); however, CBA is not thermodynamically consistent since it modifies the compressibility without affecting the density fluctuations (i.e., the structure of the fluid is still that of a perfect gas). A fully consistent approach is to require that the particles collide as if they are elastic hard spheres of diameter equal to the distance between them at the time of the collision. Such collisions produce a positive virial only if the particles are approaching each other, $v_{n}=-\mathbf{v}_{ij}\cdot\hat{\mathbf{r}}_{ij}>0$, therefore, we reject collisions among particles that are moving apart. Furthermore, as for hard spheres, it is necessary to collide pairs with probability that is _linear_ in $v_{n}$, which requires a further increase of the rejection rate and thus decrease of the efficiency. Without rejection based on $v_{n}$ or $v_{r}$, fluctuations of the local temperature $T_{c}$ would not be consistently coupled to the local pressure $p_{c}\sim\left\langle\Delta\mathbf{v}_{ij}\cdot\Delta\mathbf{r}_{ij}\right\rangle_{c}\sim\Gamma_{sc}\sqrt{T_{c}}$ because $p_{c}$ would be $\sim\sqrt{T_{c}}$ instead of the necessary $p_{c}\sim T_{c}$. For DSMC the collisional rules can be manipulated arbitrarily to obtain the desired transport coefficients, however, for non- ideal fluids thermodynamic requirements eliminate some of the freedom. This important observation has not been taken into account in other algorithms that randomize hard-sphere MD Ge and Li (2003). Note that one can in fact add I-DSMC collisions to SHSD in order to tune the viscosity without affecting the compressibility. For sufficiently small time steps, the SHSD fluid can be considered as a simple modification of the standard hard-sphere fluid. Particles move ballistically in-between collisions. When two particles $i$ and $j$ are less than a diameter apart, $r_{ij}\leq D$, there is a probability rate $(3\chi/D)v_{n}\Theta(v_{n})$ for them to collide as if they were elastic hard spheres with a variable diameter $D_{S}=r_{ij}$. Here $\Theta$ is the Heaviside function, and $\chi$ is a dimensionless parameter determining the collision frequency. The prefactor $3/D$ has been chosen so that for an ideal gas the average collisional rate would be $\chi$ times larger than that of a low-density hard-sphere gas with density (volume fraction) $\phi=\pi ND^{3}/(6V)$. In order to understand properties of the SHSD fluid as a function of $\phi$ and $\chi$, we consider the equilibrium pair correlation function $g_{2}$ at low densities, where correlations higher than pairwise can be ignored. We consider the cloud of point walkers $ij$ representing the $N(N-1)/2$ pairs of particles, each at position $\mathbf{r}=\mathbf{r}_{i}-\mathbf{r}_{j}$ and with velocity $\mathbf{v}=\mathbf{v}_{i}-\mathbf{v}_{j}$. At equilibrium, the distribution of the point walkers in phase space will be _$f(\mathbf{v},\mathbf{r})=f(v_{r},r)\sim g_{2}(r)\exp(-mv_{n}^{2}/4kT)$._ Inside the core $r<D$ this distribution of pair walkers satisfies a kinetic equation $\frac{\partial f}{\partial t}+v_{n}\frac{\partial f}{\partial r}=v_{n}\Gamma_{0}f,$ where $\Gamma_{0}=3\chi/D$ is the collision frequency. At equilibrium, $\partial f/\partial t=0$ and $v_{n}$ cancels, consistent with choosing collision probability linear in $\left|v_{n}\right|$. Thus $dg_{2}/dx=3\chi g_{2}\Theta(1-x),$ with solution $g_{2}(x)=\exp\left[3\chi(x-1)\right]$ for $x\leq 1$ and $g_{2}(x)=1$ for $x>1$, where $x=r/D$. Indeed, numerical experiments confirmed that at sufficiently low densities the equilibrium $g_{2}$ for the SHSD fluid has this exponential form inside the collision core. This low density result is equivalent to $g_{2}^{U}=\exp[-U(r)/kT]$, where $U(r)/kT=3\chi(1-x)\Theta(1-x)$ is an effective _linear core_ pair potential similar to the quadratic core potential used in DPD. Remarkably, it was found _numerically_ that this repulsive potential can predict exactly $g_{2}(x)$ at _all_ liquid densities. Figure 1 shows a comparison between the pair correlation function of the SHSD fluid on one hand, and a Monte Carlo calculation using the linear core pair potential on the other, at several densities. Also shown is a numerical solution to the hyper-netted chain (HNC) integral equations for the linear core system, inspired by its success for the Gaussian core model Louis et al. (2000). The excellent agreement at all densities permits the use of the HNC result in practical applications, notably the calculation of the transport coefficients. Figure 1: (Color online) Equilibrium pair correlation function of the SHSD fluid (solid symbols), compared to MC (open symbols) and HNC calculations (solid lines) for the linear core system, at various densities and $\chi=1$. Interestingly, in the limit $\chi\rightarrow\infty$ the SHSD algorithm reduces to hard-sphere (HS) molecular dynamics. In fact, if the density $\phi$ is smaller than the freezing point for the HS system, the structure of the SHSD fluid approaches, as $\chi$ increases, that of the HS fluid. For higher densities, if $\chi$ is sufficiently high, crystallization is observed in SHSD, either to the usual hard-sphere crystals if $\phi$ is lower than the close-packing density, or if not, to an unusual partially ordered state with multiple occupancy per site, typical of weakly repulsive potentials. An exact BBGKY-like hierarchy of Master equations for the $s$-particle distribution functions of the SHSD fluid is given in Ref. Lachowicz and Pulvirenti (1990). For the first equation of this BBGKY hierarchy, valid at low densities, we can neglect correlations other than pair ones and approximate $f_{2}(\mathbf{r}_{1},\mathbf{v}_{1},\mathbf{r}_{2},\mathbf{v}_{2})=g_{2}(\mathbf{r}_{12})f_{1}(\mathbf{r}_{1},\mathbf{v}_{1})f(\mathbf{r}_{2},\mathbf{v}_{2})$. With this assumption we obtain a stochastic Enskog equation similar to a revised Enskog equation for hard spheres but with a smeared distribution of hard-sphere diameters, as studied in Ref. Polewczak and Stell (2002). The Chapman-Enskog expansion carried out in Ref. Polewczak and Stell (2002) produces the equation of state (EOS) $p=PV/NkT$, and approximations to the self-diffusion coefficient $\zeta$, the shear $\eta$ and bulk $\eta_{B}$ viscosities, and thermal conductivity $\kappa$ of the SHSD fluid. The expressions ultimately give the transport coefficients in terms of various integer moments of $g_{2}(x)$, $x_{k}=\int_{0}^{1}x^{k}g_{2}(x)dx$, specifically, $p-1=12\phi\chi x_{3}$, $\zeta/\zeta_{0}=\sqrt{\pi}/(48\phi\chi x_{2})$, $\eta_{B}/\eta_{0}=48\phi^{2}\chi x_{4}/\pi^{3/2}$, and $\eta/\eta_{0}\mbox{ or }\kappa/\kappa_{0}=\frac{c_{1}}{\sqrt{\pi}\chi x_{2}}(1+c_{2}\phi\chi x_{3})^{2}+c_{3}\eta_{B}/\eta_{0},$ where $\zeta_{0}=D\sqrt{kT/m}$, $\eta_{0}=D^{-2}\sqrt{mkT}$ and $\kappa_{0}=kD^{-2}\sqrt{kT/m}$ are natural units, and $c_{1}=5/48$, $c_{2}=24/5$ and $c_{3}=3/5$ for $\eta$, while $c_{1}=25/64$, $c_{2}=24/5$ and $c_{3}=3/5$ for $\kappa$. The above formula for the pressure is exact and is equivalent to the virial theorem for the linear core potential, and thus thermodynamic consistency between $g_{2}(x)$ and $p(\phi)$ is guaranteed. In the inset in the top part of Fig. 2, we directly demonstrate the thermodynamic consistency of SHSD by comparing the compressibility calculated from the EOS, $S_{c}=(p+\phi dp/d\phi)^{-1}$, to the structure factor at the origin $S_{0}=S(\omega=0,k=0)$. Furthermore, good agreement is found between the adiabatic speed of sound $c_{s}^{2}=S_{0}^{-1}+2p^{2}/3$ and the location of the Brilloin lines in the dynamic structure factor $S(\omega;k)$ for small $k$ values. In Fig. 2, we also compare the theoretical predictions for $\eta$ utilizing the HNC approximation for $g_{2}$ to the ones directly calculated from SHSD. Surprisingly, good agreement is found for the shear viscosity at all densities. The corresponding results for $\zeta$ show significant ($\sim 25\%$) deviations for the self-diffusion coefficient at higher densities because of corrections due to higher-order correlations. Figure 2: (Color online) Comparison between numerical results for SHSD at several collision frequencies (different symbols) with predictions based on the stochastic Enskog equation using the HNC $g_{2}(x)$ (solid lines). The low-density approximations are also indicated (dashed lines). (_Top_) Normalized equation of state. The inset compares the compressibility (pressure derivative, dashed lines) to the structure factor at the origin $S(k\rightarrow 0)$ (symbols), measured using a direct Fourier transform of the particle positions for small $k$ and extrapolating to $k=0$. (_Bottom_) The shear viscosity at high and low densities (inset), as measured using an externally-forced Poiseuille flow. There are significant corrections (Knudsen regime) for large mean free paths (i.e., at low densities and low collision rates). As an illustration of the correct hydrodynamic behavior of the SHSD fluid and the significance of compressibility, we study the velocity autocorrelation function (VACF) $C(t)=\left\langle v_{x}(0)v_{x}(t)\right\rangle$ for a single neutrally-buoyant hard sphere of mass $m$ and radius $R$ suspended in an SHSD fluid of mass density $\rho$. This problem is relevant to the modeling of polymer chains or (nano)colloids in solution, and led to the discovery of a long power-law tail in $C(t)$ Padding and Louis (2006); Heemels et al. (2000). Here the solvent-solvent particles interact as in SHSD. The solvent-solute interaction is treated as if the SHSD particles are hard spheres of diameter $D_{s}$, chosen to be somewhat smaller than their interaction diameter with other solvent particles (specifically, we use $D_{s}=D/4$) for computational efficiency reasons, using an event-driven algorithm Donev et al. (2008). Upon collision the relative velocity of the solvent particle is reversed in order to provide a no-slip condition at the surface of the suspended sphere Padding and Louis (2006); Donev et al. (2008) (slip boundaries give qualitatively identical results). For comparison, an ideal solvent of comparable viscosity is also simulated. Theoretically, $C(t)$ has been calculated from the linearized (compressible) fluctuating Navier-Stokes (NS) equations Padding and Louis (2006). The results are analytically complex even in the Laplace domain, however, at short times an inviscid compressible approximation applies. At large times the compressibility does not play a role and the incompressible NS equations can be used to predict the long-time tail. At short times, $t<t_{c}=2R/c_{s}$, the major effect of compressibility is that sound waves generated by the motion of the suspended particle carry away a fraction of the momentum, so that the VACF quickly decays from its initial value $C(0)=kT/m$ to $C(t_{c})\approx kT/M$, where $M=m+2\pi R^{3}\rho/3$. At long times, $t>t_{visc}=4\rho R_{H}^{2}/3\eta$, the VACF decays as in an incompressible fluid, with an asymptotic power-law tail $(kT/m)(8\sqrt{3\pi})^{-1}(t/t_{visc})^{-3/2}$, in disagreement with predictions based on the Langevin equation (Brownian dynamics), $C(t)=(kT/m)\exp\left(-6\pi R_{H}\eta t/m\right)$. We have estimated the effective (hydrodynamic) colloid radius $R_{H}$ from numerical measurements of the Stokes friction force $F=-6\pi R_{H}\eta v$. Figure 3: (Color online) The velocity autocorrelation function for a neutrally buoyant hard sphere suspended in a non-ideal SHSD ($\chi=1$) solvent at two densities (symbols), as well as an ideal I-DSMC solvent ($\phi=0.5$, $\chi=0.62$, symbols), at short and long times (inset). For the more compressible (less viscous) fluids the long time tails are statistically measurable only up to $t/t_{visc}\approx 5$. The theoretical predictions based on the inviscid, for short times, or incompressible, for long times, Navier- Stokes equations are also shown (lines). The diameter of the nano-colloidal particle is only $2.5D$, although we have performed simulations using larger spheres as well with very similar results. Since periodic boundary conditions were used we only show the tail up to about the time at which sound waves generated by its periodic images reach the particle, $t_{L}=L/c_{s}$. In Fig. 3 numerical results for the VACF for an I-DSMC solvent and an SHSD solvent at two different densities are compared to the theoretical predictions. It is seen, as predicted, that the compressibility or the sound speed $c_{s}$, determines the early decay of the VACF. The exponent of the power-law decay at large times is also in agreement with the hydrodynamic predictions. The coefficient of the VACF tail agrees reasonably well with the hydrodynamic prediction for the less dense solvents, however, there is a significant deviation of the coefficient for the densest solvent, perhaps due to ordering of the fluid around the suspended sphere, not accounted for in continuum theory. In closing, we should point out that for reasonable values of the collision frequency ($\chi\sim 1$) and density ($\phi\sim 1$) the SHSD fluid is still relatively compressible compared to a dense liquid, $c_{s}^{2}<10$. Indicative of this is that the diffusion coefficient is large relative to the viscosity as in typical DPD simulations, so that the Schmidt number $S_{c}=\eta(\rho\zeta)^{-1}$ is less than 10 instead of being on the order of 100-1000. Achieving higher $c_{s}$ or $S_{c}$ requires high collision rates (for example, $\chi\sim 10^{4}$ is used in Ref. Lowe (1999)) and appropriately smaller time steps to ensure that there is at most one collision per particle per time step, and thus a similar computational effort as in molecular dynamics. The advantage of SHSD is its simplicity, easy parallelization, and simpler coupling to continuum methods such as fluctuating hydrodynamics Williams et al. (2008). This work performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344 (LLNL- JRNL-401745). We thank Salvatore Torquato, Frank Stillinger, Ard Louis, Andres Santos, and Jacek Polewczak for their assistance and advice. ## References * Noguchi et al. (2007) H. Noguchi, N. Kikuchi, and G. Gompper, Europhysics Letters 78, 10005 (2007). * Fabritiis et al. (2007) G. D. Fabritiis, M. Serrano, R. Delgado-Buscalioni, and P. V. Coveney, Phys. Rev. E 75, 026307 (2007). * Donev et al. (2008) A. Donev, A. L. Garcia, and B. J. Alder, J. Comp. Phys. 227, 2644 (2008). * Luo (2000) L.-S. Luo, Phys. Rev. E 62, 4982 (2000). * Pagonabarraga and Frenkel (2000) I. Pagonabarraga and D. Frenkel, Molecular Simulation 25, 167 (2000). * Ihle et al. (2006) T. Ihle, E. Tüzel, and D. M. Kroll, Europhys. Lett. 73, 664 (2006). * Frezzotti (1997) A. Frezzotti, Phys. Fluids 9, 1329 (1997). * Montanero and Santos (1997) J. M. Montanero and A. Santos, Phys. Fluids 9, 2057 (1997). * Tüzel et al. (2006) E. Tüzel, T. Ihle, and D. M. Kroll, Math. and Comput. in Simul. 72, 232 (2006). * Williams et al. (2008) S. A. Williams, J. B. Bell, and A. L. Garcia, SIAM Multiscale Modeling and Simulation 6, 1256 (2008). * Delgado-Buscalioni and Fabritiis (2007) R. Delgado-Buscalioni and G. D. Fabritiis, Phys. Rev. E 76, 036709 (2007). * Alexander and Garcia (1997) F. J. Alexander and A. L. Garcia, Computers in Physics 11, 588 (1997). * Lowe (1999) C. P. Lowe, Europhysics Letters 47, 145 (1999). * Alexander et al. (1995) F. J. Alexander, A. L. Garcia, and B. J. Alder, Phys. Rev. Lett. 74, 5212 (1995). * Ge and Li (2003) W. Ge and J. Li, Chemical Engineering Science 58, 1565 (2003). * Louis et al. (2000) A. A. Louis, P. G. Bolhuis, and J. P. Hansen, Phys. Rev. E 62, 7961 (2000). * Lachowicz and Pulvirenti (1990) M. Lachowicz and M. Pulvirenti, Archive for Rational Mechanics and Analysis 109, 81 (1990). * Polewczak and Stell (2002) J. Polewczak and G. Stell, J. Stat. Phys. 109, 569 (2002). * Padding and Louis (2006) J. T. Padding and A. A. Louis, Phys. Rev. E 74, 031402 (2006). * Heemels et al. (2000) M. W. Heemels, M. H. J. Hagen, and C. P. Lowe, J. Comp. Phys. 164, 48 (2000).

arxiv-papers

2024-09-04T02:48:54.106293

0803.0415

# A series whose sum range is an arbitrary finite set Jakub Onufry Wojtaszczyk Department of Mathematics, Computer Science and Mechanics University of Warsaw ul. Banacha 2, 02-097 Warsaw, Poland email: onufry@duch.mimuw.edu.pl (March 21, 2004) ###### Abstract In finitely-dimensional spaces the sum range of a series has to be an affine subspace. It is long known this is not the case in infinitely dimensional Banach spaces. In particular in 1984 M.I. Kadets and K. Woz̀niakowski obtained an example of a series the sum range of which consisted of two points, and asked whether it is possible to obtain more than two, but finitely many points. This paper answers the question positively, by showing how to obtain an arbitrary finite set as the sum range of a series in any infinitely dimensional Banach space. ## 1 Introduction For a finitely-dimensional linear space $X$ the well-known Steinitz theorem states that for any conditionally convergent series the set of all possible limits of the series (called the sum range) is a affine subspace of $X$. In the ”Scottish Book” S. Banach posed the problem whether the same holds for infinitely dimensional Banach spaces. The problem was solved negatively in the same book by J. Marcinkiewicz. In his example the sum range is the set $M$ of all integer-valued functions in $L_{2}[0,1]$. The next example, due to M. I. Ostrovskii, showed that the sum range does not have to be a closed set - the sum range of Ostrovskii’s series was of the form $M+\sqrt{2}M$. Finally M. I. Kadets constructed an example in which the sum range consisted of two points, disproving, in particular, H. Hadwiger’s conjecture that the sum range has to be the coset of some additive subgroup of $X$. The justification of the example was obtained independently by K. Woz̀niakowski and P. A. Kornilov in 1986. It is still unknown what sets can be the sum ranges of series. In this paper it is shown that any finite subset of $X$ can be the sum range of a conditionally convergent series, which solves the problem posed by M. I. Kadets along with his two-point example (the problem is stated in [S91] in the general case, and in [U02] for $X=C(\Delta)$ and $n=3$). The example is an extension of the 2-point example of M. I. Kadets as given in [S91]. As far as possible I shall try to keep the notation consistent with the notation given there, although the lack of suitable letters in the latin alphabet will force me to abandon the notation in a few places. Everywhere all spaces are considered with the $L_{1}$ norm, i.e. $||f||_{X}=\int_{X}|f(x)|dx$. Frequently it is assumed it is obvious on which space the norm is taken, and only $||f||$ is written. ## 2 The results of K. Woz̀niakowski The work in this paper is strongly inspired by the 2-point example of M. I. Kadets and the proof by K. Woz̀niakowski. In this paper not only the final result of Woz̀niakowski’s work will be used, but also multiple technical facts than can be found in the proof. Rather than force the reader to search for those in the original paper, I shall reiterate here Woz̀niakowski’s work, at times formulating the results in a way that will make them easier to use in the subsequent sections. The whole content of this section in based on [S91], and a reader familiar with this work may probably skip to the next section. Let $Q=[0,1]^{\omega}$ be the infinite dimensional cube, i.e. the product of a countable number of unit segments, equipped with the standard product topology and measure. By $x=(x_{1},x_{2},\ldots)$ we shall denote the variable on $Q$. Suppose we have two sequences of functions on the cube: $a_{m}^{n}$ and $b_{m,j}^{n}$, where $n\in{\mathbb{N}}$ and for a given $n$ both $m$ and $j$ belong to some finite sets $M_{n}$ and $J_{n}=M_{n+1}$ respectively. By $A_{n}$ we shall denote the set $\\{a_{m}^{n}:m\in M_{n}\\}$, and by $B_{n}$ the set $\\{b_{m,j}^{n}:m\in M_{n},j\in J_{n}\\}$. For convenience if $X$ is a set of functions, by $\tilde{X}$ we shall denote the sum of the functions from $X$. We shall assume the following properties of the functions $a_{m}^{n}$ and $b_{m,j}^{n}$: $\displaystyle\tilde{A_{n}}(x)$ $\displaystyle=$ $\displaystyle 1\qquad\forall_{n\in{\mathbb{N}}}\forall_{x\in Q}$ (1) $\displaystyle||a_{m}^{n}||$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{n}|}$ (2) $\displaystyle\lim_{n{\rightarrow}\infty}|M_{n}|$ $\displaystyle=$ $\displaystyle\infty$ (5) The function $a_{m}^{n}$ depends only on the variable $x_{n}$ The functions $a_{m}^{n}$ assume only values 0 and 1 $\displaystyle b_{m,j}^{n}$ $\displaystyle=$ $\displaystyle-a_{m}^{n}\cdot a_{j}^{n+1}$ (6) We shall this collection of properties the Kadets properties on the cube $Q$. These properties mean that for each $n$ the interval $[0,1]$ is divided into $|M_{n}|$ sets $V_{m}^{n}$ of equal measure, and $a_{m}^{n}(x_{1},x_{2},\ldots)=1$ iff $x_{n}\in V_{m}$. The functions $b_{m,j}^{n}$ are negative, and are supported on rectangles $(x_{n},x_{n+1})\in V_{m}^{n}\times V_{j}^{n+1}$. From the Kadets properties we can easily deduce another few properties, mainly about the behaviour of $b_{m,j}^{n}$ based on properties 1 and 6: $\displaystyle a_{m}^{n}$ $\displaystyle=$ $\displaystyle-\sum_{j\in J_{n}}b_{m,j}^{n},$ (7) $\displaystyle a_{j}^{n+1}$ $\displaystyle=$ $\displaystyle-\sum_{m\in M_{n}}b_{m,j}^{n}$ (8) $\displaystyle\tilde{B_{n}}(x)$ $\displaystyle=$ $\displaystyle-{\mathbf{1}}\forall_{n\in{\mathbb{N}}}$ (9) $\displaystyle||b_{m,j}^{n}||$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{n}\times J_{n}|}$ (14) The function $b_{m,j}^{n}$ depends only on the variables $x_{n}$ and $x_{n+1}$ The functions $b_{m,j}^{n}$ assume only values 0 and -1 $a_{m}^{n}$ and $a_{m^{\prime}}^{n}$ have almost disjoint supports for $m\neq m^{\prime}$ These properties follow easily from the Kadets properties. In property 14 by almost disjoint supports we mean that the intersection of two supports is of measure zero, we can obviously modify $a_{m}^{n}$ so that the Kadets properties still hold and the sets $\\{x:a_{m}^{n}(x)>0\\}$ are disjoint for any constant $n$ and any two different values of $m$. Let $c_{k},k\in{\mathbb{N}}$ be any ordering of all the functions $a_{m}^{n}$ and $b_{m,j}^{n}$. Following Woz̀niakowski we shall investigate the convergence of any reordering $c_{\sigma(k)}$ of $c_{k}$. ###### Proposition 2.1. For any family of functions $c_{k}$ having the Kadets properties there exist such two permutations $\sigma$ and $\tau$ of ${\mathbb{N}}$ that $\sum c_{\sigma(k)}{\rightarrow}{\mathbf{0}}$ and $\sum c_{\tau(k)}{\rightarrow}{\mathbf{1}}$. ###### Proof. For $\sigma$ it is enough to order the functions $a_{m}^{n}$ lexicographically, i.e. $a_{m}^{n}$ appears before $a_{m^{\prime}}^{n^{\prime}}$ iff $n<n^{\prime}$ or $n=n^{\prime}$ and $m<m^{\prime}$, and then immediately after each $a_{m}^{n}$ to put the whole set $\\{b_{m,j}^{n}:j\in J_{n}\\}$. Then the sum of each block consisting of a single function $a_{m}^{n}$ and the functions $b_{m,j}^{n}$ following it sums up to zero due to property 7, so the norm of each partial sum is the norm of the currently open block, which converges to zero due to properties 2, 14 and 5. To get $\tau$ we order the functions $a_{m}^{n}$ in the same way, but we follow each function $a_{m}^{n}$ for $n>1$ by the set $\\{b_{l,m}^{n-1}:l\in M_{n-1}\\}$, the functions $a_{m}^{1}$ are not followed by anything (as there are no functions $b_{m,j}^{0}$). Then the functions $a_{m}^{1}$ sum up to the constant function 1 due to property 1. The following blocks again sum up to zero, this time due to property 8, so the norm of the difference between 1 and a particular partial sum is equal to the norm of the currently open block, which again converges to zero due to properties 2, 14 and 5.∎ ###### Remark 1. The series of functions from Proposition 2.1 converge not only in the $L_{1}$ norm, but also in any $L_{p}$ norm for any $p<\infty$. ###### Proof. Again it is only a question of investigating the norm of any given block, as the sum of the previous blocks is zero. Functions $a_{m}^{n}$ assume only values 0 and 1 and have disjoint supports for a set $n$ from properties 5 and 14. Functions $b_{m,j}^{n}$ for a given $n$ have disjoint supports (this follows from properties 6 and 14) and assume values $0$ and $-1$ (from 14). Thus for any $f$ being a sum of any set of functions $a_{m}^{n}$ and $b_{m,j}^{n}$ for a fixed $n$ (or $a_{m}^{n}$ and $b_{m,j}^{n-1}$ for a fixed $n$ in the case of $\tau$) we have $\|f\|_{\infty}\leq 1$. This implies for any $1\leq p<\infty$ $\|f\|_{p}=(\int|f|^{p})^{1/p}=(\int|f|\cdot|f|^{p-1})^{1/p}\leq(\|f\|_{1}\cdot\|f^{p-1}\|_{\infty})^{1/p}\leq\|f\|_{1}^{1/p}\cdot 1=\|f\|_{1}^{1/p}.$ Thus if the sum of the series tended to zero in the $L_{1}$ norm with $n$ tending to infinity, it also tends to zero in any $L_{p}$ norm for $p<\infty$.∎ ###### Proposition 2.2. If a reordering $c_{\sigma(k)}$ of a family $c_{k}$ having the Kadets properties converges, it converges to a constant integer function. ###### Proof. Due to properties 5 and 14 and the finiteness of the sets $M_{n}$ and $J_{n}$ only finitely many of the functions $c_{\sigma(k)}$ depend on a given variable $x_{l}$ \- precisely the functions belonging to $A_{l}$, $B_{l}$ and $B_{l-1}$. Moreover the sum of all these functions equals to the constant function $-1$ due to properties 1 and 9. Thus for some integer $K_{0}$ the function $\sum_{k=1}^{K}c_{\sigma(k)}$ is constant with regard to $x_{l}$ for $K\geq K_{0}$, and thus the limit of the series also has to be constant with regard to $x_{l}$. As this applied to an arbitrary $l$, the limit simply has to be constant. As the functions $c_{k}$ are integer-valued (properties 5 and 14), their sums also have to be integer-valued. Thus all the partial sums of the series are integer-valued, and so the limit is also integer-valued, which ends the proof.∎ The next step will be to show that 0 and 1 are the only possible limits of a rearrangement of a family of functions with the Kadets property. We shall set a fixed rearrangement $c_{\sigma(k)}$ of a given Kadets family, and we shall assume that the sum $\sum_{k}c_{\sigma(k)}$ converges to some constant integer $C\neq 1$ (we know $C=1$ can be achieved, it remains to prove that under these assumptions $C=0$). Take an arbitrary $\delta>0$ and fix $K_{0}=K_{0}(\delta)$ such that for any $K>K_{0}$, $\big{|\big{|}}C-\sum_{k=1}^{K}c_{\sigma(k)}\big{|\big{|}}\leq\delta$ (15) and for any $m>l>K_{0}$ the Cauchy condition holds, i.e. $\big{|\big{|}}\sum_{k=l}^{m}c_{\sigma(k)}\big{|\big{|}}\leq\delta.$ (16) In addition to the sets $A_{n}$ and $B_{n}$ introduced earlier we shall also consider $V_{n}=\bigcup_{k=1}^{n}(A_{k}\cup B_{k})$. Let $M$ be any integer such that $c_{\sigma(k)}\in V_{M}\cup A_{M+1}\qquad\mbox{for any $k\leq K$.}$ (17) Let $c_{k}^{*}=c_{\sigma(k)}$ if $c_{\sigma(k)}\in V_{M}\cup A_{M+1}$, 0 otherwise. Similarly let $\bar{c}_{k}=c_{\sigma(k)}$ if $c_{\sigma(k)}\in B_{M+1}$, 0 otherwise. By $c^{*}$ we shall denote $\sum_{k=K_{0}+1}^{\infty}c_{k}^{*}$, while by $c$ we shall denote $\sum_{k=1}^{K_{0}}c_{\sigma(k)}$. The sum $c+c^{*}$ is equal to $\tilde{V}_{M}+\tilde{A}_{M+1}=0+1=1$. Hence $||c^{*}||=||1-c||\geq||1-C||-||C-c||\geq 1-\delta$. Let $k_{0}=K_{0}$ and $k_{j+1}=\min\big{\\{}k:\frac{1}{4}-\frac{5\delta}{4}\leq{\big{|}\big{|}}\sum_{i=k_{j}+1}^{k}c_{k}^{*}{\big{|}\big{|}}\leq\frac{1}{4}-\frac{\delta}{4}\big{\\}}.$ (18) The indices $k_{j}$ are well defined for $j$ from 1 to 4 because the total norm of the sum $c^{*}$ is at least $1-\delta$ and each single $c_{k}^{*}$ has norm $\leq\delta$ due to the Cauchy condition (16). For $j=0,1,2,3$ define the following functions: $c_{j+1}^{**}=\sum_{k=k_{j}+1}^{k_{j+1}}c_{k}^{*},\qquad\bar{\bar{c}}_{j+1}=\sum_{k=k_{j}+1}^{k_{j+1}}\bar{c}_{k},\qquad\hat{c}_{j+1}=\sum_{k=k_{j}+1}^{k_{j+1}}c_{\sigma(k)},$ and for $j=1,2,3,4$ set $r_{j}=\hat{c}_{j}-\bar{c}_{j}-c^{**}_{j}$. In plain words this means that we divide the functions $c_{k}$ for $k_{j}<k\leq k_{j+1}$ into three sets - those from $A_{n}$ for $n\leq M+1$ or $B_{n}$ for $n\leq M$ (these add up to $c_{j}^{**}$), those from $B_{M+1}$ (these add up to $\bar{\bar{c}}_{j}$) and the rest (these add up to $r_{j}$). We will show that the functions from $B_{M+1}$ are placed in $c_{k}$ in similar proportions as the functions from $V_{M}\cup A_{M+1}$ — if, say, about a half of the functions from $V_{M}\cup A_{M+1}$ appeared in $c_{k}$ (that happens at $k_{2}$) then about a half of the functions from $B_{M+1}$ must have appeared, too. We shall need to estimate the norm of two sums, which we would like to be negligible: $||r_{j}||$ and $||\sum_{k=k_{4}+1}^{\infty}c_{k}^{*}||$. We know that the sum of all $c_{k}$ up to $k_{j}$ is negligible, thus if the high-$n$ functions ($r_{j}$) are negligible, the functions from $V_{M}\cup A_{M+1}$ and $B_{M+1}$ have to approximately cancel each other out. This motivates the following proposition: ###### Proposition 2.3. For a Kadets family of functions $c_{k}$, its rearrangement $c_{\sigma(k)}$ converging to some $C\neq 1$, an arbitrary $\delta$ and an arbitrary $M>K_{0}(\delta)$ as above, with the notation as above we have $\sum_{j=1}^{4}||r_{j}||\leq 18\delta$. ###### Proof. As $c_{j}^{**}$ is integer-valued (being a sum of some functions from a Kadets family), the condition $||c_{j}^{**}||\leq\frac{1}{4}$ implies $|{\rm supp}c_{j}^{**}|\leq\frac{1}{4}$. Thus we can use lemma 1 (from the section ”Auxiliary lemmas”) to get $||c_{j}^{**}+r_{j}||\geq||c_{j}^{**}||+(1-2|{\rm supp}c_{j}^{**}|)||r_{j}||=||c_{j}^{**}||+\frac{1}{2}r_{j}.$ Of course $||\hat{c}_{j}||\leq\delta$ from the Cauchy condition (16). We thus have $1\geq\sum_{j=1}^{4}||\bar{\bar{c}}_{j}||=\sum_{j=1}^{4}||\hat{c}_{j}-c_{j}^{**}-r_{j}||\geq\sum_{j=1}^{4}||c^{**}_{j}+r_{j}||-\sum_{j=1}^{4}||\hat{c}_{j}||\geq$ $\geq\sum_{j=1}^{4}(||c^{**}_{j}||+\frac{1}{2}||r_{j}||)-4\delta\geq 1-5\delta+\frac{1}{2}\sum_{j=1}^{4}||r_{j}||-4\delta,$ which gives us the sought estimate upon $||r_{j}||$, namely $\sum_{j=1}^{4}||r_{j}||\leq 18\delta$. In particular, of course, each $||r_{j}||$ is bounded by $18\delta$. ∎ ###### Corollary 2.4. With the notation and assumptions as above, $||\bar{\bar{c}}_{j}+c^{**}_{j}||\leq 19\delta$ ###### Proof. $||\bar{\bar{c}}_{j}+c^{**}_{j}||=||\hat{c}_{j}-r_{j}||\leq||\hat{c}_{j}||+||r_{j}||\leq\delta+18\delta=19\delta$.∎ ###### Proposition 2.5. For a Kadets family of functions $c_{k}$, its rearrangement $c_{\sigma(k)}$ converging to some $C\neq 1$, an arbitrary $\delta$ and an arbitrary $M>K_{0}(\delta)$ as above, with the notation as above we have $||\sum_{k=k_{4}+1}^{\infty}c_{k}^{*}||\leq 11\delta$. ###### Proof. We have $||\bar{\bar{c}}_{j}||=||\hat{c}_{j}-c_{j}^{**}-r_{j}||\geq||c_{j}^{**}+r_{j}||-||\hat{c}_{j}||\geq||c_{j}^{**}||+\frac{1}{2}||r_{j}||-||\hat{c}_{j}||\geq||c_{j}^{**}||-\delta\geq\frac{1}{4}-\frac{9\delta}{4}.$ Take any index $k^{\prime}>k_{4}$. If the norm $||\sum_{k=k_{4}+1}^{k^{\prime}}c_{k}^{*}||$ were greater then $11\delta$, then there would exist some $k_{5}\in(k_{4},k^{\prime}]$ such that $12\delta\geq||\sum_{k=k_{4}+1}^{k_{5}}c_{k}^{*}||>11\delta$. Then by a similar argument ($||\bar{\bar{c}}_{5}||\geq||c_{5}^{**}||+(1-24\delta)||r_{5}||-||\hat{c}_{5}||\geq 11\delta-\delta$) the norm of $\sum_{k=k_{4}+1}^{k_{5}}\bar{c}_{k}$ would be larger then $10\delta$ — but all the functions $\bar{c}_{k}$ are negative, so $||\sum\bar{c}_{k}||=\sum||\bar{c}_{k}||$, which in this case gives $1\geq||\sum_{k=k_{0}}^{k_{5}}\bar{c}_{k}||=\sum_{j=1}^{4}||\bar{c}_{k}||+||\sum_{k=k_{4}+1}^{k_{5}}\bar{c}_{k}||>1-9\delta+10\delta$, a contradiction. Thus the norm $||\sum_{k=k_{4}+1}^{\infty}c_{k}^{*}||$ has to be no greater than $11\delta$ (the sum is convergent, as it is in fact the sum of a finite number of functions, all coming from $V_{M+1}$). Let us denote this sum by $c_{5}^{**}$.∎ Now we can prove the main theorem of Woz̀niakowski’s work: ###### Theorem 2.6. For a Kadets family of functions $c_{k}$ and some rearrangement $c_{\sigma(k)}$ converging to $C\neq 1$ we have $|C-\frac{1}{2}|\leq\frac{1}{2}$, which (due to lemma 2.2) implies $C=0$. ###### Proof. Consider any $\delta$, and the partial sum $S=\sum_{k=1}^{k_{4}}c_{\sigma(k)}$ with the notation as above. As $k_{4}>K_{0}$, from assumption 15 we know that $||S-C||\leq\delta$, so it will suffice to estimate $||S-\frac{1}{2}||$. We have $||S-\frac{1}{2}||={\big{|}\big{|}}c+\sum_{j=1}^{4}c_{j}^{**}+\sum_{j=1}^{4}\bar{\bar{c}}_{j}+\sum_{j=1}^{4}r_{j}+c_{5}^{**}-c_{5}^{**}-\frac{1}{2}{\big{|}\big{|}}=$ $={\big{|}\big{|}}c+c^{*}-\frac{1}{2}+\sum_{j=1}^{4}\bar{\bar{c}}_{j}+\sum_{j=1}^{4}r_{j}-c_{5}^{**}{\big{|}\big{|}}\leq{\big{|}\big{|}}\frac{1}{2}+\sum_{j=1}^{4}\bar{\bar{c}}_{j}{\big{|}\big{|}}+{\big{|}\big{|}}\sum_{j=1}^{4}r_{j}{\big{|}\big{|}}+{\big{|}\big{|}}c_{5}^{**}{\big{|}\big{|}}.$ The function $\sum_{j=1}^{4}\bar{\bar{c}}_{j}$ is a sum of functions from $B_{M+1}$, which means assumes only the values 0 and $-1$, thus $|\frac{1}{2}+\sum_{j=1}^{4}\bar{\bar{c}}_{j}|$ is always equal to $\frac{1}{2}$. Inserting this and the bounds upon $r_{j}$ and $c_{5}^{**}$ we get $||S-\frac{1}{2}||\leq\frac{1}{2}+18\delta+11\delta=\frac{1}{2}+29\delta.$ As $||S-C||\leq\delta$ we get $||C-\frac{1}{2}||\leq\frac{1}{2}+30\delta$. As $\delta$ was chosen arbitrarily, we get the thesis.∎ ###### Corollary 2.7. The sum range of any Kadets family consists of two points, the constant functions 0 and 1, in any $L_{p}$ norm for $1\leq p<\infty$ ###### Proof. From Proposition 2.1 and Remark 1 we know that the two constant functions belong to the sum range. From the Proposition 2.2 we know that all functions in the sum range in the $L_{1}$ norm are constant integer functions, and from Theorem 2.6 we know that only the two functions $0$ and $1$ are eligible. If any permutation of the series converged to some function $g$ in some $L_{p}$ norm, then $\|S_{n}-g\|_{p}$ would tend to zero. But from the Hölder inequality we know that $\|S_{n}-g\|_{p}\geq\|S_{n}-g\|_{1}$ (as the measure of the whole space is 1), which would imply that the series $S_{n}$ converges also in the $L_{1}$ norm, contradicting Theorem 2.6.∎ ## 3 The 3-point series Denote by $Q_{i}=[0,1]^{\omega},i=1,2,3$ the infinite dimensional cube, i.e., the product of a countable number of unit segments equipped with the standard product probability measure. The example will be constructed in $L_{1}(Q_{1}\cup Q_{2}\cup Q_{3})$. In the whole paper $t=(t_{1},t_{2},\ldots)$ will denote the variable on $Q_{1}$, $u=(u_{1},u_{2},\ldots)$ will denote the variable on $Q_{2}$ and $v=(v_{1},v_{2},\ldots)$ will denote the variable on $Q_{3}$. Our series will consist of functions of three kinds. The functions of the first kind are defined as follows: $f_{m}^{n}(t)=\begin{cases}1&\mbox{if $\frac{m-1}{n}<t_{n}<\frac{m}{n}$}\\\ 0&\mbox{otherwise.}\end{cases}$ $f_{m}^{n}(u)=f_{m}^{n}(v)=0$ for $n\in{\mathbb{N}},m\in\\{1,2,\ldots,n\\}$. The second kind of functions is defined on all three cubes: $\displaystyle g_{m,j}^{n}(t)$ $\displaystyle=$ $\displaystyle\begin{cases}-1&\mbox{if $\frac{m-1}{n}<t_{n}<\frac{m}{n}$ and $\frac{j-1}{n+1}<t_{n+1}<\frac{j}{n+1}$}\\\ 0&\mbox{otherwise}\end{cases}$ $\displaystyle g_{m,j}^{n}(u)$ $\displaystyle=$ $\displaystyle\begin{cases}\frac{1}{n+1}&\mbox{if $\frac{m-1}{n}<u_{n}<\frac{m}{n}$}\\\ 0&\mbox{otherwise}\end{cases}$ $\displaystyle g_{m,j}^{n}(v)$ $\displaystyle=$ $\displaystyle\begin{cases}1&\mbox{if $\frac{(m-1)(n+1)+j-1}{n(n+1)}<v_{n}<\frac{(m-1)(n+1)+j}{n(n+1)}$}\\\ 0&\mbox{otherwise}\end{cases}$ for $n\in{\mathbb{N}},m\in\\{1,2,\ldots,n\\},j\in\\{1,2,\ldots,n+1\\}$. The functions of the third kind are defined on $Q_{2}$ and $Q_{3}$: $\displaystyle h_{m,j,k}^{n}(t)$ $\displaystyle=$ $\displaystyle 0$ $\displaystyle h_{m,j.k}^{n}(u)$ $\displaystyle=$ $\displaystyle\begin{cases}-\frac{1}{(n+1)^{2}(n+2)}&\mbox{if $\frac{m-1}{n}<u_{n}<\frac{m}{n}$}\\\ 0&\mbox{otherwise}\end{cases}$ $\displaystyle h_{m,j,k}^{n}(v)$ $\displaystyle=$ $\displaystyle\begin{cases}-1&\mbox{if $\frac{(m-1)(n+1)+j-1}{n(n+1)}<v_{n}<\frac{(m-1)(n+1)+j}{n(n+1)}$ and $\frac{k-1}{(n+1)(n+2)}<v_{n+1}<\frac{k}{(n+1)(n+2)}$}\\\ 0&\mbox{otherwise}\end{cases}$ for $n\in{\mathbb{N}},m\in\\{1,2,\ldots,n\\},j\in\\{1,2,\ldots,n+1\\},k\in\\{1,2,\ldots,(n+1)(n+2)\\}$. These functions have properties we want to generalize. Suppose we have three families of indices: $M_{n}$, $J_{n}$ and $K_{n}$, with $J_{n}=M_{n+1}$ and $K_{n}=M_{n+1}\times J_{n+1}$ (here $M_{n}=\\{1,2,\ldots,n\\}$ and the mapping between $\\{1,2,\ldots,n\\}\times\\{1,2,\ldots,n+1\\}$ and $\\{1,2,\ldots,n(n+1)\\}$ is given by $(m,j)\mapsto(m-1)(n+1)+j$). We have three families of functions: the first kind $\\{f_{m}^{n}:n\in{\mathbb{N}},m\in M_{n}\\}$, the second kind $\\{g_{m,j}^{n}:n\in{\mathbb{N}},m\in M_{n},j\in J_{n}\\}$ and the third kind $\\{h_{m,j,k}^{n}:n\in{\mathbb{N}},m\in M_{n},j\in J_{n},k\in K_{n}\\}$ defined on the union $Q_{1}\cup Q_{2}\cup Q_{3}$ of Hilbert cubes. The families $f$ and $g$ form a Kadets family on $Q_{1}$, while the functions $h$ disappear on $Q_{1}$. On $Q_{3}$ the functions $g$ and $h$ form a Kadets family (with $M_{n}\times J_{n}$ being the first index set and $K_{n}$ the second), while functions $f$ disappear. The properties of the functions on $Q_{2}$ are different, as follows: $\displaystyle\sum_{m\in M_{n}}\sum_{j\in J_{n}}g_{m,j}^{n}$ $\displaystyle=$ $\displaystyle{\mathbf{1}}$ (19) $\displaystyle\sum_{m\in M_{n}}\sum_{j\in J_{n}}\sum_{k\in K_{n}}h_{m,j,k}^{n}$ $\displaystyle=$ $\displaystyle-{\mathbf{1}},$ (20) $\displaystyle g_{m,j}^{n}$ $\displaystyle=$ $\displaystyle-\sum_{k\in K_{n}}h_{m,j,k}^{n},$ (21) $\displaystyle\sum_{m^{\prime}\in M_{n+1}}g_{m^{\prime},j^{\prime}}^{n+1}$ $\displaystyle=$ $\displaystyle-\sum_{m\in M_{n}}\sum_{j\in J_{n}}\sum_{m^{\prime}\in M_{n+1}}h_{m,j,(m^{\prime},j^{\prime})}^{n}.$ (22) $\displaystyle\sum_{j\in J_{n}}g_{m,j}$ assumes only values 0 and 1 (23) $\displaystyle\int_{Q_{2}}g_{m,j}^{n}$ $\displaystyle=$ $\displaystyle\int_{Q_{3}}g_{m,j}^{n}$ (24) $\displaystyle\int_{Q_{2}}h_{m,j,k}^{n}$ $\displaystyle=$ $\displaystyle\int_{Q_{3}}h_{m,j,k}^{n}$ (25) $\displaystyle||g_{m,j}^{n}||$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{n}\times J_{n}|}$ (26) $\displaystyle||h_{m,j,k}^{n}||$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{n}\times J_{n}\times K_{n}|}$ (28) The functions $g_{m,j}^{n}$ and $h_{m,j,k}^{n}$ on $Q_{2}$ depend only on $u_{n}$ Such a family of functions will be called a 3-Kadets family. It is easy (although maybe a bit tedious) to check that the family defined at the beginning of the section is a 3-Kadets family. We shall denote by $F_{n}$ the set $\\{f_{m}^{n}:m\in M_{n}\\}$, by $G_{n}$ the set $\\{g_{m,j}^{n}:m\in M_{n};j\in J_{n}\\}$ and by $H_{n}$ the set $\\{h_{m,j,k}^{n}:m\in M_{n},j\in J_{n};k\in K_{n}\\}$. Also, by $V_{M}$ we shall denote $\bigcup_{k=1}^{M}F_{k}\cup G_{k}\cup H_{k}$. Denote by $d_{n}$ any set enumeration of the whole 3-Kadets family. We are investigating the possible limits of $\sum_{n=1}^{\infty}d_{\sigma(n)}$ for all permutations $\sigma$ of ${\mathbb{N}}$. If a given rearrangement $d_{\sigma(n)}$ of a 3-Kadets family converges, it converges on each of the cubes separately. On $Q_{1}$ and $Q_{3}$ we have Kadets families of functions, so the series on each of these cubes converges either to ${\mathbf{0}}$ or to ${\mathbf{1}}$ due to theorem 2.6. The new part is the behaviour on $Q_{2}$. Same as in the first part of Proposition 2.2 only finitely many functions depend on a given variable $u_{n}$ – the functions $g_{m,j}^{n}$ and $h_{m,j,k}^{n}$ – and their sum is constant, equal to zero due to property (21) applied to each $j$ separately. Thus again the limit of the series $\sum d_{\sigma(n)}$ on $Q_{2}$ has to be a constant function. As $\int_{Q_{2}}d_{n}=\int_{Q_{3}}d_{n}$ for any $d_{n}$ (it is 0 for functions of the first kind and follows from properties 24 and 25 for the second and third kind), we get $\int_{Q_{2}}\sum_{n=1}^{N}d_{\sigma(n)}=\int_{Q_{3}}\sum_{n=1}^{N}d_{\sigma(n)}$. As the integral is a continuous functional on $L_{1}(Q_{2})$ and $L_{1}(Q_{3})$ we get that the integrals of the limits have to be equal – but we know that the limit of $\sum d_{\sigma(n)}$ on both $Q_{2}$ and $Q_{3}$ is a constant function, so the equality of integrals implies the equality of the limits. Thus the limit of the whole series is described by a pair of integers - the value on $Q_{1}$ and the value on $Q_{3}$. Let us denote the limit function by $d_{\infty}$. We are to show that it is possible to obtain exactly three different sums – precisely we can obtain $({\mathbf{0}},{\mathbf{0}}),({\mathbf{1}},{\mathbf{0}})$ and $({\mathbf{1}},{\mathbf{1}})$. To obtain any of these limits we first arrange the functions $f$ and $g$ as by Proposition 2.1 for a Kadets family on $Q_{1}$, and then after each $g$ we put the $h$ functions as by Proposition 2.1 for the cube $Q_{3}$. It remains to be seen if we get convergence on $Q_{2}$. In the case of $({\mathbf{0}},{\mathbf{0}})$ after a given $f_{m}^{n}$ there appear the all functions $g_{m,j}^{n}$ and $h_{m,j,k}^{n}$ with the same $m$ and $n$. The sum of all these functions on $Q_{2}$ is equal to ${\mathbf{0}}$ due to property (21) for each $j$ separately. Thus the norm of the partial sum on $Q_{2}$ is equal to the norm of the functions appearing after the last $f$, and this tends to zero due to properties 26, 28 and 5 (all the functions have the same index $m$, so the sum of their norms is equal to $\frac{2}{|M_{n}|}{\rightarrow}0$). In the case of $({\mathbf{1}},{\mathbf{0}})$ after a given $f_{m}^{n}$ we get the functions $g_{l,m}^{n-1}$ and $h_{l,m,k}^{n-1}$. The sum of all these functions on $Q_{2}$ is again ${\mathbf{0}}$ due to property 21, this time applied to each $l$ separately. Again the norm of the difference between the partial sum and $({\mathbf{1}},{\mathbf{0}})$ is the norm of the part after the last $f$, and that again tends to 0. In the case of $({\mathbf{1}},{\mathbf{1}})$ after a given $f_{m}^{n}$ we get the functions $g_{l,m}^{n-1}$ and $h_{l^{\prime},m^{\prime},(l,m)}^{n-2}$. Their sum is ${\mathbf{0}}$ due to property 22 applied to them all. Again the norm of the difference between the partial sum and ${\mathbf{1}}$ tends to 0. Again it is easy to check that the convergence occurs not only in the $L_{1}$ norm, but also in any $L_{p}$ norm for $p<\infty$ in the same way as in Remark 1 — on each of the cubes the $L_{\infty}$ norm of the partial sums is bounded by 1. One may wonder why the same arguments will not imply the convergence of the series arranged by rows in $G_{n}$ and columns in $H_{n-1}$ to $({\mathbf{0}},{\mathbf{1}})$. The answer is we lack the equivalent of property 22 for this arrangement. To illustrate this let us look at the 3-Kadets family given at the beginning of the section arranged in this natural way. The sum $\sum_{j=1}^{n+1}g_{m,j}^{n}$ on $Q_{2}$ is equal to $1$ on $\frac{m-1}{n}<u_{n}<\frac{m}{n}$, while the sum of the appropriate column of $H_{n-1}$, $\sum_{j=1}^{n+1}\sum_{m^{\prime}=1}^{n-1}\sum_{j^{\prime}=1}^{n}h_{m^{\prime},j^{\prime},(m-1)(n+1)+j}^{n-1}$ is equal to $-\frac{1}{n}$ on the whole cube $Q_{2}$. Thus the partial sums before each function of the first kind do not disappear as they did in the previous three cases, and when half of these functions from a given $F_{n}$ have appeared, the norm of the partial sum on $Q_{2}$ is $\frac{1}{2}$ regardless of $n$ – thus this particular series does not converge. Of course we still have to prove this is true for any rearrangement – but this example shows the nature of the reason why only three and not four possible limits exist. ## 4 Auxiliary lemmas Before we begin the main part of this paper – i.e. the proof that our series cannot converge to $({\mathbf{0}},{\mathbf{1}})$ – we shall need three auxiliary lemmas: ###### Lemma 1. (Lemma given without proof in [O89]) Let $(X,\mu)$ and $(Y,\nu)$ be measure spaces with probability measures. Let $f(x,y)$ and $g(x,y)$ be functions in $L_{1}(X\times Y)$, each of which depends on only one variable: $f(x,y)=\tilde{f}(x),g(x,y)=\tilde{g}(y)$. Then $||f+g||\geq||f||+||g||[1-2\mu({\rm supp}\tilde{f})].$ ###### Proof. $||f+g||=\int_{X\times Y}|f+g|=\int_{{\rm supp}\tilde{f}\times Y}|f+g|+\int_{(X\setminus{\rm supp}\tilde{f})\times Y}|g|\geq\int_{{\rm supp}(\tilde{f})\times Y}|f|-\int_{{\rm supp}(\tilde{f})\times Y}|g|+(1-\mu({\rm supp}\tilde{f}))||g||=||f||-\mu({\rm supp}\tilde{f})||g||+(1-\mu({\rm supp}\tilde{f}))||g||=||f||+||g||[1-2\mu({\rm supp}\tilde{f})].$∎ ###### Lemma 2. Let $A,B,C$ be arbitrary spaces equipped with probabilistic measures and let $X=A\times B\times C$ be equipped with the standard product measure. Suppose $f,g$ are bounded functions defined on $X$ of the form $f(a,b,c)=\tilde{f}(a,b)=\sum_{k=1}^{N}s_{k}\chi_{A_{k}\times B_{k}}$ and $g(a,b,c)=\tilde{g}(b,c)=\sum_{l=1}^{N}t_{l}\chi_{B_{l}\times C_{l}}$, and $\|f-g\|\leq{\varepsilon}$. Then there exists a function $h(a,b,c)=\tilde{h}(b)$ such that $\|h-g\|\leq 2{\varepsilon}$ and $\|h-f\|\leq 2{\varepsilon}$. Moreover if $f$ is integer-valued then $h$ can also be chosen to be integer-valued, and if for a family of sets $B_{\alpha}$ we have $\forall_{\alpha}\forall_{b_{1},b_{2}\in B_{\alpha}}\forall_{a\in A}f(a,b_{1},c)=f(a,b_{2},c)$, then we can choose a function $h$ constant on any set $B_{\alpha}$. ###### Proof. For any given $b\in B$ we take $\tilde{h}(b)$ such that $\int_{A}|\tilde{f}(a,b)-\tilde{h}(b)|da=\inf_{x\in{\mathbb{R}}}\\{\int_{A}|\tilde{f}(a,b)-x|da\\}.$ This is well defined, as $f$ is bounded, and thus in fact the $\inf$ is taken over a bounded, and thus compact set. For such an $h$ we have $\|h-f\|=\int_{X}|f(a,b,c)-\tilde{h}(b)|=\int_{C}\int_{B}\int_{A}|\tilde{f}(a,b)-\tilde{h}(b)|=\int_{C}\int_{B}\inf\\{\int_{A}|\tilde{f}(a,b)-x(b)|\\}\leq$ $\leq\int_{C}\int_{B}\int_{A}|\tilde{f}(a,b)-\tilde{g}(b,c)|\leq\int_{C}\int_{B}\int_{A}|f(a,b,c)-g(a,b,c)|=\|f-g\|\leq{\varepsilon}.$ As $\|h-f\|\leq{\varepsilon}$ and $\|f-g\|\leq{\varepsilon}$, we immediately have $\|g-h\|\leq 2{\varepsilon}$. As for the additional assumptions, if $f$ and $g$ are integer-valued, we can take the $\inf$ in the definition of $\tilde{h}$ to be taken only over integers, with the same result. Regardless of that which option we choose, if $f$ is constant with regard to $b$ on any $B_{\alpha}$, then from the definition $h$ also can be chosen to be constant on that set.∎ ###### Lemma 3. Let $A,B$ be arbitrary spaces equipped with probabilistic measures and $X=A\times B$ equipped with the standard product measure. Suppose $f,g,h$ are integer-valued functions defined on $X$ fulfilling $f(a,b)=\tilde{f}(a)$ and $h(a,b)=\tilde{h}(b)$ for some $\tilde{f},\tilde{h}$. Suppose too that the function $g$ assumes only two adjacent values (i.e. $k$ and $k+1$ for some $k$) . Finally suppose that $\|f+g+h\|<\delta<\frac{1}{9}$. Then either $f$ or $h$ is a constant function equal some integer $c$ on a set of measure $\geq 1-2\sqrt{\delta}$. Furthermore the function satisfies $\|f-c\|<3\sqrt{\delta}$ (or $\|h-c\|<3\sqrt{\delta}$, respectively). ###### Proof. The sets $F_{n}=\tilde{f}^{-1}((-\infty,n])$ and $H_{n}=\tilde{h}^{-1}((-\infty,n])$ form two increasing families, the sum of each is the whole space $X$ and the intersection of each is empty. The measures $|F_{n}|$ thus form an ascending sequence with elements arbitrarily close to 0 when $n{\rightarrow}-\infty$ and arbitrarily close to 1 when $n{\rightarrow}\infty$. As $F_{n}\setminus F_{n-1}=\tilde{f}^{-1}(n)$, if $\tilde{f}$ is not constant on any set of measure $\geq 1-2\sqrt{\delta}$, then at least one element of the sequence $|F_{n}|$, say $F_{n_{f}}$, has to fall into the interval $[\sqrt{\delta},1-\sqrt{\delta}]$. Similarly if $\tilde{h}$ is constant on no set of measure $\geq 1-2\sqrt{\delta}$, then for some $n_{h}$ we have $\sqrt{\delta}\geq|H_{n_{h}}|\geq 1-\sqrt{\delta}$. Then on the set $X_{1}=F_{n_{f}}\times H_{n_{h}}$ we have $f(a,b)+h(a,b)\leq n_{h}+n_{f}$, while on $X_{2}=(A\setminus F_{n_{f}})\times(B\setminus H_{n_{h}})$ we have $f(a,b)+h(a,b)\geq n_{h}+n_{f}+2$. As $g$ assumes two adjacent values, it is either $\leq-(n_{h}+n_{f}+1)$ or $\geq-(n_{h}+n_{f}+1)$ on the whole space $X$. Thus on one of the sets $X_{1},X_{2}$ we have $|f+g+h|\geq 1$, call it $X_{i}$. As both $X_{1}$ and $X_{2}$ are products of two sets of measure $\geq\sqrt{\delta}$, we have $\|f+g+h\|=\int_{X}|f(a,b)+g(a,b)+h(a,b)|\geq\int_{X_{i}}|f(a,b)+g(a,b)+h(a,b)|\geq|X_{i}|\geq\delta$, which contradicts the assumptions of the lemma. Thus one of the functions has to be constant on a large set. Without the loss of generality we may assume it is $h$, and that it is equal to some integer $c$. Let us examine the function $f$, taking into account that all the functions are integer-valued, and thus if their sum is non-zero, it is at least one : $\delta>\|f+g+h\|\geq\|f+g+c\|_{A\times h^{-1}(c)}\geq|\\{\tilde{f}(a)\not\in\\{-k-c,-k-c-1\\}\\}\times h^{-1}(c)|=$ $=|\\{\tilde{f}(a)\not\in\\{-k-c,-k-c-1\\}\\}|\cdot(1-2\sqrt{\delta}),$ which implies $\tilde{f}(a)\in\\{-k-c,-k-c-1\\}$ on a set of measure at least $1-\frac{\delta}{1-2\sqrt{\delta}}$. Denote this set by $A^{\prime}$. Now we return to the function $h$: $\|h-c\|_{X}\leq\frac{1}{1-2\sqrt{\delta}}\|h-c\|_{A^{\prime}\times B}=\frac{1}{1-2\sqrt{\delta}}\|h-c\|_{A^{\prime}\times(B\setminus h^{-1}(c))}.$ On the set $A^{\prime}$ the function $f+g+c$ assumes values of absolute value $\leq 1$, so by substituting $f+g$ for $-c$ we shall decrease the norm at most by $1\cdot|A^{\prime}\times(B\setminus h^{-1}(c))|\leq(1-\frac{\delta}{1-2\sqrt{\delta}})(2\sqrt{\delta})\leq 2\sqrt{\delta},$ thus giving the inequality $\|h-c\|_{X}\leq\frac{1}{1-2\sqrt{\delta}}\|h+f+g\|_{A^{\prime}\times(B\setminus h^{-1}(c))}+2\sqrt{\delta}\leq\frac{1}{1-2\sqrt{\delta}}\|f+g+h\|_{X}+2\sqrt{\delta}\leq\frac{\delta}{1-2\sqrt{\delta}}+2\sqrt{\delta}.$ As $\delta\leq\frac{1}{9}$, we have $\frac{\delta}{1-2\sqrt{\delta}}\leq\sqrt{\delta}$, and thus $||h-c||\leq 3\sqrt{\delta}$. ∎ ## 5 The fourth point Now we can begin to prove the main theorem of the paper: ###### Theorem 5.1. The function $d_{\infty}=({\mathbf{0}},{\mathbf{1}})$ does not belong to the sum range of any 3-Kadets family series. ###### Proof. Suppose we have a rearrangement of some 3-Kadets family $d_{\sigma(n)}$ the sum of which converges to $d_{\infty}$. Again, take an arbitrarily small $\delta>0$ (we shall need $927\sqrt{\delta}<\frac{1}{4}$, i.e. $\delta<\frac{1}{13749264}$) and an integer $K$ satisfying inequalities (15) and (16), i.e. the tails and Cauchy sums are smaller than $\delta$ for $N>K$. Then, again, we take any $M$ satisfying (17), i.e. such that $V_{M}$ contains the first $K$ elements of our series. Then we take an $N_{0}$ such that $V_{M}\subset\\{d_{\sigma(1)},d_{\sigma(2)},\ldots,d_{\sigma(N_{0})}\\}.$ (29) Consider any fixed $N>N_{0}$. We will prove that $\int_{Q_{3}}\sum_{n=1}^{N}d_{\sigma(n)}<\frac{1}{4}.$ Of course this suffices to prove that our series does not converge to $1$ on $Q_{3}$, which contradicts the assumption the rearrangement converged to $({\mathbf{0}},{\mathbf{1}})$. Denote for any $L,k\in{\mathbb{Z}}$ by $D_{k}$ the set $\\{d_{\sigma(1)},\ldots,d_{\sigma(k)}\\}$, and by $F^{k}_{L},G^{k}_{L},H^{k}_{L}$ and $V^{k}_{L}$ the intersections of sets $F_{L},G_{L},H_{L}$ or $V_{L}$, respectively, with the set $D_{k}$. First we shall prove the following lemma: ###### Lemma 4. If functions $f^{n}_{m},g_{m,j}^{n}$ and $h_{m,j,k}^{n}$ are a 3-Kadets family on the cubes $Q_{1},Q_{2}$ and $Q_{3}$ and their set permutation $d_{\sigma(n)}$ tends to $0$ on $Q_{1}$ and $1$ on $Q_{2}$ and $Q_{3}$, and for a given $L$ we have $\int_{Q_{3}}\tilde{G}^{N}_{L}\geq\frac{1}{2}+38\delta$, where $N>N_{0}$ as above, then there exists a $P\subset[0,1]$ such that $|P|=\frac{1}{2}$ and $[(\tilde{H}_{L}^{N})^{-1}(0)]\cap\\{v:v_{L}\in P\\}\subset Q_{3}$ has measure $\leq 450\delta$. ###### Remark 2. What this lemma really tells us is: if up to the $N$th element of the series at least half plus something $(38\delta)$ of the $G_{L}$ functions have appeared, then at least half minus something $(450\delta)$ of the $H_{L}$ functions had to appear. Moreover, the $H_{L}$ functions do not appear in a haphazard fashion - we know that at least half minus something rows had to appear (a row is the set of the functions $h_{m,j,k}^{L}$ with fixed $m$ and $j$ and varying $k$). ###### Proof. If $L\leq M$ then our thesis is automatically fulfilled – all functions from $H_{L}$ belong to the set $D_{N}$, thus we can take any set of measure $\frac{1}{2}$ for $P$ and the set $(\tilde{H}_{L}^{N})^{-1}(0)$ will be empty, so $P$ will satisfy the required conditions. Now consider the case $L>M$. The numbers $K$ and $L-1$ satisfy the conditions (15), (16) and (17) (as $L>M$ and $M$ satisfied (17)). Thus we know there exist numbers $n_{i}$ satisfying (18). We shall prove that $N\geq n_{2}$. We know that $\int_{Q_{3}}\tilde{G}^{N}_{L}=-\int_{Q_{1}}\tilde{G}^{N}_{L}$ (as all $g_{m,j}^{n}$ are of the same constant sign on each cube, the absolute value of the integral is equal to the norm, and the norms on each cube are equal) . If $N<n_{2}$, then $\|\tilde{G}_{L}^{N}\|_{Q_{1}}\leq\|\tilde{G}_{L}^{n_{2}}\|_{Q_{1}}=\|\bar{\bar{d}}_{1}+\bar{\bar{d}}_{2}\|\leq\|d^{**}_{1}\|+19\delta+\|d^{**}_{2}\|+19\delta<\frac{1}{2}+38\delta,$ which contradicts our assumption (the first inequality follows from the fact, that $g_{m,j}^{n}$ are all non-positive functions on $Q_{1}$, the second inequality from corollary 2.4). Thus $N>n_{2}$. Consider $\tilde{V}_{L-1}^{n_{2}}+\tilde{F}_{L}^{n_{2}}$ on $Q_{1}$. This function is dependent on variables $t_{1},t_{2},\ldots,t_{L}$, while $\tilde{G}_{L}^{n_{2}}=\bar{\bar{d}}_{1}+\bar{\bar{d}}_{2}$ on $Q_{1}$ depends on $t_{L}$ and $t_{L+1}$. From property (15) and Corollary 2.4 we get $\|\tilde{V}_{L-1}^{n_{2}}+\tilde{F}_{L}^{n_{2}}+\tilde{G}_{L}^{n_{2}}\|\leq\|\tilde{D}_{k}\|+\|d^{**}_{1}+\bar{\bar{d}}_{1}\|+\|d^{**}_{2}+\bar{\bar{d}}_{2}\|\leq\delta+19\delta+19\delta=39\delta.$ We can thus use lemma 2 for functions $-\tilde{V}_{L-1}^{n_{2}}-\tilde{F}_{L}^{n_{2}}$ and $\tilde{G}_{L}^{n_{2}}$ to get that on $Q_{1}$ both these functions are closer than $39\delta$ to some integer-valued function $\tilde{A}$ depending only on $t_{L}$. Each function $f_{m}^{n}$ depends only on $t_{n}$ and assumes values $0$ and $1$ only (properties 5 and 5), so it is in fact the characteristic function of a set $\\{t:t_{n}\in S_{m}^{n}\\}$ for some $S_{m}^{n}\subset[0,1]$. As the $f_{m}^{n}$ functions have disjoint support for a fixed $n$, they are all constant on any given $S_{m}^{n}$. The $g$ functions are also constant with regard to $t_{n}$ on the $S_{m}^{n}$ due to property 6, and all the other functions are constant with regard to $t_{n}$ on the whole interval. Thus the functions $-\tilde{V}_{L-1}^{n_{2}}-\tilde{F}_{L}^{n_{2}}$ and $\tilde{G}_{L}^{n_{2}}$ are constant with respect to $t_{L}$ on sets $\\{t_{L}\in S_{m}^{L}\\}$ we can choose $\tilde{A}$ to be constant on those sets. Thus $\tilde{A}$ coincides on $Q_{1}$ with the sum of some of the rows of $G_{L}$, i.e. $\tilde{A}$ corresponds to some subset $A$ of $G_{L}$ such that for a fixed $m$ either all or none of the functions $g_{m,j}^{L}$ belong to $A$. Define $\tilde{A}$ on $Q_{2}$ and $Q_{3}$ as the sum of all the elements of $A$ as well, which agrees with our notation that $\tilde{U}$ is the sum of all the elements of $U$ for an arbitrary set of functions. We know from (18) and Proposition 2.5 that $\|\sum_{n=n_{2}+1}^{\infty}d_{n}^{*}\|_{Q_{1}}\leq\frac{1-\delta}{4}+\frac{1-\delta}{4}+11\delta\leq\frac{1}{2}+11\delta$. Remark that $(\tilde{V}^{n_{2}}_{L-1}+\tilde{F}_{L}^{n_{2}}+\sum_{n=n_{2}+1}^{\infty}d^{*}_{n})|_{Q_{1}}=(\tilde{V}_{L-1}+F_{L})|_{Q_{1}}={\mathbf{1}}|_{Q_{1}}$, so $\|\tilde{V}_{L-1}^{n_{2}}+\tilde{F}_{L}^{n_{2}}\|_{Q_{1}}\geq\frac{1}{2}-11\delta$. On the other hand $\|\tilde{V}_{L-1}^{n_{2}}+\tilde{F}_{L}^{n_{2}}\|_{Q_{1}}=\|\tilde{D}_{K}+d_{1}^{**}+d^{**}_{2}\|_{Q_{1}}\leq\delta_{\frac{1-\delta}{4}}+\frac{1-\delta}{4}\leq\frac{1}{2}+\delta$. As $\|\tilde{V}_{L-1}^{n_{2}}+\tilde{F}_{L}^{n_{2}}-\tilde{A}\|_{Q_{1}}\leq 39\delta$, taking into account the equality $\|\tilde{A}\|_{Q_{1}}=\|\tilde{A}\|_{Q_{2}}$ we can estimate that $\frac{1}{2}-50\delta\leq\|\tilde{A}\|_{Q_{2}}\leq\frac{1}{2}+40\delta.$ (30) Distinct functions from $G_{L}$ have disjoint supports on $Q_{1}$ (this follows from the properties 14 and 6 of Kadets families), and each has the same norm $\psi=\frac{1}{|M_{L}\times J_{L}|}$. Thus if the distance between two functions corresponding to two subsets of $G_{L}$ on $Q_{1}$ is smaller than $n\psi$, then at most $n$ functions belong to the symmetric difference of those two subsets. If at most $n$ functions belong to the symmetric difference, then the distance between the two functions on $Q_{2}$ is at most $n\psi$ (as on $Q_{2}$ the norm of a single function is also equal $\psi$ by property 26). Thus, in general, if $B,C\subset G_{L}$, then $\|\tilde{B}-\tilde{C}\|_{Q_{1}}\geq\|\tilde{B}-\tilde{C}\|_{Q_{2}}$. In particular $\tilde{G}_{L}^{n_{2}}$ is at most $39\delta$ distant from $\tilde{A}$ on $Q_{2}$. Now consider what happens on $Q_{2}$. From (23) the restriction of $\tilde{A}$ to $Q_{2}$ is equal to $1$ on some set (on intervals $t_{L}\in[\frac{m-1}{L},\frac{m}{L}]$ for $m$ such that $g^{L}_{m,j}\in A$) and 0 on the rest. From (15), as $n_{2}>K$, we have $\|\tilde{D}_{n_{2}}-1\|_{Q_{2}}\leq\delta$. If we substitute $\tilde{A}$ for $\tilde{G}_{L}^{n_{2}}$, we will be at most $40\delta$ distant from zero, precisely $\|\tilde{D}_{n_{2}}-1-\tilde{G}^{n_{2}}_{L}+\tilde{A}\|_{Q_{2}}\leq 40\delta.$ However as only $G_{L}$ and $H_{L}$ depend on $u_{L}$, this sum is composed of two parts - the part $\tilde{A}+\tilde{H}_{L}^{n_{2}}$ dependent on $u_{L}$ and the whole rest (i.e. $\tilde{D}_{n_{2}}-(\tilde{G}_{L}^{n_{2}}+\tilde{H}_{L}^{n_{2}})$) dependent on other variables. Thus we can apply a simplified version of lemma 2, with $f=\tilde{A}+\tilde{H}_{L}^{n_{2}}$, $g=-(\tilde{D}_{n_{2}}-V_{L}^{n_{2}})$, and a trivial one-point space as $B$. We learn that both our functions are within $80\delta$ from a function $c$ dependent on $b$ – but as $B$ was a one- point space, $c$ is a constant function. As $\tilde{A}$ assumes values $0$ and $1$, and $\tilde{H}_{L}^{n_{2}}\in[-1,0]$, their sum is non-negative on ${\rm supp}\tilde{A}$ and non-positive on the remainder of $Q_{2}$. From (30) we know that $|{\rm supp}\tilde{A}|\geq\frac{1}{2}-50\delta$, thus $\tilde{A}+\tilde{H}_{L}^{n_{2}}$ is non-negative on a set of measure $\geq\frac{1}{2}-50\delta$. If $c$ is positive, then (as $\delta<\frac{1}{200}$) $80\delta\geq\|\tilde{A}+\tilde{H}_{L}^{n_{2}}-c\|\geq c(\frac{1}{2}-50\delta)\geq\frac{c}{4},$ which implies $c\leq 320\delta$. Similarly if $c$ is negative, we know from (30) that $|Q_{2}\setminus{\rm supp}\tilde{A}|\geq\frac{1}{2}-40\delta$, yielding again $c>-\frac{800}{3}\delta$. Thus $|c|<320\delta$, so $\|\tilde{A}+\tilde{H}_{L}^{n_{2}}\|\leq\|\tilde{A}+\tilde{H}_{L}^{n_{2}}-c\|+|c|\leq 80\delta+320\delta=400\delta$. Thus $\tilde{H}_{L}^{n_{2}}$ is within $400\delta$ of a function with values 0 and -1 on $Q_{2}$ – the function $-\tilde{A}$. Remark, that $-\tilde{A}=-\tilde{A}^{\prime}$ on $Q_{2}$ for a subset $A^{\prime}$ of $H_{L}$ with the property that for a given $m$ either all of the functions $h_{m,j,k}^{L}$ belong to $A^{\prime}$, or none of the functions belongs to $A^{\prime}$ (if a given $g_{m,j}^{L}$ belongs to $A$, then all $h_{m,j,k}^{L}$ belong to $A^{\prime}$) . If $\tilde{A^{\prime}}$, where $A^{\prime}\subset H_{L}$, is a function assuming only values $0$ and $1$ on $Q_{2}$ and $B\subset H_{L}$, then $\displaystyle\|\tilde{A^{\prime}}-\tilde{B}\|_{Q_{2}}$ $\displaystyle=$ $\displaystyle\|\tilde{A^{\prime}}-\tilde{B}\|_{{\rm supp}\tilde{A^{\prime}}}+\|\tilde{A^{\prime}}-\tilde{B}\|_{{\mathbb{Q}}_{2}\setminus{\rm supp}\tilde{A^{\prime}}}$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{L}\times J_{L}\times K_{L}|}|\\{h:h\in A^{\prime}\wedge h\not\in B\\}|+\frac{1}{|M_{L}\times J_{L}\times K_{L}|}|\\{h:h\not\in A^{\prime}\wedge h\in B\\}|$ $\displaystyle=$ $\displaystyle\frac{1}{|M_{L}\times J_{L}\times K_{L}|}|A\bigtriangleup B|=\|\tilde{A^{\prime}}-\tilde{B}|_{Q_{3}}.$ Let us take any subset $A^{\prime\prime}$ of $H_{L}$ depending only on $m$ and $j$ with exactly half of the elements of $H_{L}$ and containing $A^{\prime}$ or contained in $A^{\prime}$. If $B\subset C\subset H_{L}$ or $C\subset B\subset H_{L}$, then $\|\tilde{C}-\tilde{B}\|=|\|\tilde{C}\|-\|\tilde{B}\||$, because all the the functions in $H_{L}$ are non-positive. As $\tilde{A^{\prime}}=-\tilde{A}$ on $Q_{2}$ and from (30) $|\|\tilde{A}\|_{Q_{2}}-\frac{1}{2}|\leq 50\delta$, we get $\|\tilde{A^{\prime}}-\tilde{A^{\prime\prime}}\|_{Q_{2}}\leq 50\delta$, and thus $\|H_{L}^{n_{2}}-\tilde{A^{\prime\prime}}\|_{Q_{3}}=\|H_{L}^{n_{2}}-\tilde{A^{\prime\prime}}\|_{Q_{2}}\leq 450\delta$. Now consider what happens on $Q_{3}$. As $\tilde{H}_{L}^{n_{2}}$ and $\tilde{A^{\prime\prime}}$ are both integer-valued on $Q_{3}$, this means they differ on a set of measure at most $450\delta$, and thus their difference can be positive on a set of measure at most $450\delta$. When we increase $n$ from $n_{2}$ to $N$ the set where the difference is positive can only decrease. Thus $|\\{H_{L}^{N}-\tilde{A^{\prime\prime}}>0\\}|\leq 450\delta$. Now for $P$ we take ${\rm supp}\tilde{A}^{\prime\prime}$. The set $[(\tilde{H}_{N}^{L})^{-1}(0)]\cap\\{v:v_{L}\in P\\}$ is the set where $H_{N}^{L}$ is equal to zero and $\tilde{A}^{\prime\prime}$ is negative — thus their difference is positive, so the set has to have measure smaller than $450\delta$, which is what we had to prove. ∎ Now the main proof. Assume $d_{\infty}=({\mathbf{0}},{\mathbf{1}})$, i.e. our series converges to ${\mathbf{1}}$ on $Q_{2}$ and $Q_{3}$ and to ${\mathbf{0}}$ on $Q_{1}$. We shall prove by induction upon $L$ that $\int_{Q_{3}}\tilde{V}_{L}^{N}\leq\frac{1}{4}$. As $\sum_{n=1}^{N}d_{\sigma(n)}$ is finite, its elements are contained in some $V_{L}$, thus if the thesis is true, we get $\int_{Q_{3}}\sum_{n=1}^{N}d_{\sigma(n)}\leq\frac{1}{4}$, which is what we had to prove. For $L<M$ we have $V_{L}\subset D_{N}$ and from property (7) $\int_{Q_{3}}\tilde{V}_{L}^{N}=0\leq\frac{1}{4}$. Now suppose we have the thesis for $L-1$ and attempt to prove it for $L$. Denote by $P_{1}$ the function $(\tilde{V}_{L-1}^{N}+\tilde{G}_{L}^{N})|_{Q_{3}}$ and by $P_{2}$ the function $\sum_{n>L}\tilde{G}_{n}^{N}+\tilde{H}_{n}^{N}|_{Q_{3}}$. Consider the function $\tilde{H}_{L}^{N}|_{Q_{3}}$. It depends on variables $v_{L}$ and $v_{L+1}$. The function $P_{1}$ depends on $v_{1},\ldots,v_{L}$, while $P_{2}$ depends on $v_{L+1},\ldots,v_{Z}$ for some $Z\in{\mathbb{Z}}$. The function $H_{L}^{N}|_{Q_{3}}$ assumes only values 0 and -1, all three functions – $H_{L}^{N}|_{Q_{3}}$, $P_{1}$ and $P_{2}$ are integer-valued, and from (15) their sum is less then $\delta$ distant from ${\mathbf{1}}$ on $Q_{3}$. Thus by taking $P_{1}^{\prime}=P_{1}-1$ we have three functions fulfilling the assumptions of lemma 3. Thus either $P_{1}$ or $P_{2}$ is within $3\sqrt{\delta}$ of a constant function. In each of these cases the proof will also depend on whether $\int_{Q_{3}}\tilde{G}_{L}^{N}\leq\frac{1}{2}+38\delta$ or $\int_{Q_{3}}\tilde{G}_{L}^{N}>\frac{1}{2}+38\delta$. Thus we have in total four cases to consider. Suppose first that $P_{2}$ is within $3\sqrt{\delta}$ of a constant function. As $\|P_{1}+P_{2}+\tilde{H}_{L}^{N}-1\|\leq\delta$, this means that $P_{1}+\tilde{H}_{L}^{N}$ is within $3\sqrt{\delta}+\delta\leq 4\sqrt{\delta}$ of a constant function. If $\int_{Q_{3}}\tilde{G}_{L}^{N}\leq\frac{1}{2}+38\delta$, then $\int_{Q_{3}}\tilde{V}_{L}^{N}=\int_{Q_{3}}\tilde{V}_{L-1}^{N}+\tilde{G}_{L}^{N}+\tilde{H}_{L}^{N}\leq\frac{1}{4}+(\frac{1}{2}+38\delta)+0=\frac{3}{4}+38\delta$. But this function is equal $P_{1}+\tilde{H}_{L}^{N}$, and so is within $4\sqrt{\delta}$ of some constant integer $c$ and its integral also has to be within $4\sqrt{\delta}$ of $c$. As $4\sqrt{\delta}+38\delta<\frac{1}{4}$, we get $c\leq 0$, thus $\int_{Q_{3}}\tilde{V}_{L}^{N}\leq c+4\sqrt{\delta}\leq\frac{1}{4}$. If $P_{2}$ is within $3\sqrt{\delta}$ of a constant function, and $\int_{Q_{3}}\tilde{G}_{L}^{N}>\frac{1}{2}+38\delta$, then again $P_{1}+\tilde{H}_{L}^{N}$ is within $4\sqrt{\delta}$ from a constant integer $c$. From lemma 4 we have in particular that $\int_{Q_{3}}\tilde{H}_{L}^{N}\leq-\frac{1}{2}+450\delta$. Obviously $\int_{Q_{3}}\tilde{G}_{L}^{N}\leq 1$, thus $\int_{Q_{3}}V_{L}^{N}=\int_{Q_{3}}V_{L-1}^{N}+\tilde{G}_{L}^{N}+\tilde{H}_{L}^{N}\leq\frac{1}{4}+1-\frac{1}{2}+450\delta=\frac{3}{4}+450\delta$. As this is supposed again to within $4\sqrt{\delta}$ of $c$, we have $c\leq 0$ as $450\delta+4\sqrt{\delta}\leq\frac{1}{4}$. Again thus $\int_{Q_{3}}\tilde{V}_{L}^{N}\leq c+4\sqrt{\delta}\leq\frac{1}{4}$. In the third case we suppose that $P_{1}^{\prime}$, and thus also $P_{1}$ is within $3\sqrt{\delta}$ of a constant function and $\int_{Q_{3}}\tilde{G}_{L}^{N}\leq\frac{1}{2}+38\delta$. As $\int_{Q_{3}}\tilde{V}_{L-1}^{N}\leq\frac{1}{4}$ from the inductive assumption, we have $\int_{Q_{3}}P_{1}\leq\frac{3}{4}+38\delta$. As $P_{1}$ is supposed to be within $3\sqrt{\delta}$ of some constant integer $c$, its integral also has to be within $3\sqrt{\delta}$ of $c$, which again implies $c\leq 0$ and $\int_{Q_{3}}P_{1}\leq 3\sqrt{\delta}$. As $\tilde{V}_{L}^{N}=P_{1}+\tilde{H}_{L}^{N}$ and $\tilde{H}_{L}^{N}\leq 0$, we get $\int_{Q_{3}}\tilde{V}_{L}^{N}\leq 3\sqrt{\delta}\leq\frac{1}{4}$. The last case is when $P_{1}$ is within $3\sqrt{\delta}$ of a constant integer $c$ and $\int_{Q_{3}}\tilde{G}_{L}^{N}>\frac{1}{2}+38\delta$. In this case from lemma 4 we know there exists a set $P^{\prime}\subset Q_{3}$ dependent only on $v_{L}$ such that $|P^{\prime}|=\frac{1}{2}$ and $\int_{P^{\prime}}\tilde{H}_{L}^{N}\leq-\frac{1}{2}+450\delta$. If $P_{1}$ is within $3\sqrt{\delta}$ of a constant integer function and $P_{1}+P_{2}+\tilde{H}_{L}^{N}$ is within $\delta$ of $1$ (from 15) then $P_{2}+\tilde{H}_{L}^{N}$ is within $3\sqrt{\delta}+\delta\leq 4\sqrt{\delta}$ of some constant integer function $C$. Taking $P_{2}^{\prime}=P_{2}-C$ we arrrive in the situation of lemma 2: $\tilde{H}_{L}^{N}$ depends on $v_{L}$ and $v_{L+1}$ while $P_{2}^{\prime}$ depends on $v_{L+1},v_{L+2},\ldots,v_{Z}$. This means that each of them is within $8\sqrt{\delta}$ of some integer function $P_{3}$ dependent only on $v_{L+1}$. As $\int_{P^{\prime}}\tilde{H}_{L}^{N}\leq-\frac{1}{2}+450\delta$ and $\|\tilde{H}_{L}^{N}-P_{3}\|\leq 8\sqrt{\delta}$, we gather that $\int_{P^{\prime}}P_{3}\leq-\frac{1}{2}+450\delta+8\sqrt{\delta}\leq-\frac{1}{2}+458\sqrt{\delta}$. As $P^{\prime}$ depends only on $v_{L}$ and $P_{3}$ only on $v_{L+1}$ and $|P^{\prime}|=|Q_{3}\setminus P^{\prime}|$ , $\int_{Q_{3}}P_{3}=\int_{P^{\prime}}P_{3}+\int_{Q_{3}\setminus P^{\prime}}P_{3}=2\int_{P^{\prime}}P_{3}\leq-1+916\sqrt{\delta}.$ Returning to $\tilde{H}_{L}^{N}$ we get $\int_{Q_{3}}\tilde{H}_{L}^{N}\leq\int_{Q_{3}}P_{3}+8\sqrt{\delta}\leq-1+924\sqrt{\delta}$. As $\int_{Q_{3}}\tilde{G}_{L}^{N}\leq 1$ and $\int_{Q_{3}}\tilde{V}_{L-1}^{N}\leq\frac{1}{4}$ we get $\int_{Q_{3}}P_{1}\leq\frac{5}{4}$. As before, $\int_{Q_{3}}P_{1}$ has to be within $3\sqrt{\delta}$ of the integer $c$, implying $c\leq 1$ and $\int_{Q_{3}}P_{1}\leq 1+3\sqrt{\delta}$. We have $\int_{Q_{3}}\tilde{V}_{L}^{N}=\int_{Q_{3}}P_{1}+\tilde{H}_{L}^{N}\leq 1+3\sqrt{\delta}-1+924\sqrt{\delta}\leq 927\sqrt{\delta}\leq\frac{1}{4}$. Thus in all four cases we have completed the induction step, which proves in a finite number of steps that $\int_{Q_{3}}\tilde{D}_{N}\leq\frac{1}{4}$. This holds for an arbitrary $N>N_{0}$, and would thus have to hold for the limit function, $\int_{Q_{3}}d_{\infty}\leq\frac{1}{4}$, which obviously contradicts the assumption that $d_{\infty}|_{Q_{3}}={\mathbf{1}}$.∎ ###### Corollary 5.2. A 3-Kadets series has a 3-point sum range, consisting of the functions $({\mathbf{0}},{\mathbf{0}})$, $({\mathbf{1}},{\mathbf{0}})$ and $({\mathbf{1}},{\mathbf{1}})$. As previously, this holds for any $L_{p}$ with $1\leq p<\infty$ ## 6 More points From the previous section we know how to make 3 points out of 2. The same mechanism can be applied to make $r+1$ points out of $r$. ###### Theorem 6.1. For any $r>1$ there exist a family $d_{k}$ of functions defined on a union of cubes $Q_{1},\ldots,Q_{N}$ with an $r$-point sum range. Additionally we can distinguish two disjoint subsets ${\mathcal{F}}$ and ${\mathcal{G}}$ of $\\{d_{k}:k\in{\mathbb{N}}\\}$ which form a Kadets family on $Q_{N}$, while all other functions $d_{k}$ disappear on $Q_{N}$. Moreover one function in the sum range of $d_{k}$ is equal to ${\mathbf{1}}$ on $Q_{N}$ and all the other functions from the sum range disappear on $Q_{N}$. Finally there exist rearrangements convergent to any point of the sum range in which the sets ${\mathcal{F}}$ and ${\mathcal{G}}$ are arranged as in Proposition 2.1. ###### Proof. We shall prove the thesis by induction upon $r$. For $r=2$ the original Kadets example with $N=1$ satisfies the given conditions. Suppose we have an appropriate family for $r-1$. We add two cubes to the domain of $d_{k}$: $Q_{N+1}$ and $Q_{N+2}$. Denote by $x=(x_{1},x_{2},\ldots)$ the variable on $Q_{N+1}$ and by $y=(y_{1},y_{2},\ldots)$ the variable on $Q_{N+2}$. All the functions except ${\mathcal{G}}$ will disappear on these cubes. For each $n$ we divide the unit interval $[0,1]$ into $|M_{n}|$ sets $S_{m}^{n},m\in M_{n}$ of measure $\frac{1}{|M_{n}|}$ each. We define $g_{m,j}^{n}$ to be equal $\frac{1}{|J_{n}|}$ if $x_{n}\in S_{m}^{n}$, 0 otherwise. Next we define $K_{n}=M_{n+1}\times J_{n+1}$ and divide the unit interval $[0,1]$ into $|K_{n}|$ sets $T_{k}^{n}$ of equal measure, and on $Q_{N+2}$ define $g_{m,j}^{n}$ to be equal to 1 if $y_{n}\in T_{(m,j)}^{n-1}$, 0 otherwise. Finally to the functions $d_{k}$ we add a set of functions ${\mathcal{H}}=\\{h_{m,j,k}^{n}\\}$ which disappear on the cubes $Q_{1}$ to $Q_{N}$, and satisfy $h_{m,j,k}^{n}=-\frac{1}{|K_{n}|}g_{m,j}^{n}$ on $Q_{N+1}$ and $h_{m,j,k}^{n}=-g_{m,j}^{n}\cdot g_{k}^{n+1}$ on $Q_{M+2}$. It is again easy, although tedious, to check that ${\mathcal{F}}$, ${\mathcal{G}}$ and the new functions ${\mathcal{H}}$ form a 3-Kadets family on $Q_{N},Q_{N+1},Q_{N+2}$. We claim that the set $\\{d_{k}\\}\cup{\mathcal{H}}$ satisfies the conditions given in the theorem. The sets ${\mathcal{G}}$ and ${\mathcal{H}}$ form a Kadets family on $Q_{N+2}$, all other functions disappear on $Q_{N+2}$. We have to check the sum ranges. Let us fix any convergent rearrangement $e_{k}$ of $\\{d_{k}\\}\cup{\mathcal{H}}$. From the properties of 3-Kadets families given in section 3 we know that the limit on $Q_{N+1}$ and $Q_{N+2}$ is going to be the same, and equal either ${\mathbf{0}}$ or ${\mathbf{1}}$. From theorem 5.1 we know that if the series converges to ${\mathbf{0}}$ on $Q_{M}$, it has to converge to ${\mathbf{0}}$ on $Q_{N+1}$ and $Q_{N+2}$. Thus at most $r+1$ limits can be achieved - the functions with ${\mathbf{0}}$ on $Q_{N}$ generate one each (by the 0-extension onto $Q_{N+1}\cup Q_{N+2}$), while the single function with ${\mathbf{1}}$ on $Q_{N}$ can be extended by either ${\mathbf{0}}$ or ${\mathbf{1}}$ to $Q_{N+1}\cup Q_{N+2}$. This also satisfies the condition that only one of the points in the sum range is ${\mathbf{1}}$ on $Q_{N+2}$, while the other points disappear on $Q_{2}$. We can of course attain all the desired points in the sum range with ${\mathcal{G}}$ and ${\mathcal{H}}$ ordered as in Proposition 2.1 by taking the rearrangements with ${\mathcal{F}}$ and ${\mathcal{G}}$ ordered as in the proposition and inserting ${\mathcal{H}}$ as in section 3. ∎ Thus it is possible to attain a affine-independent finite set of any size $r$ as a sum range of a conditionally convergent series. Again, this works for any $L_{p}$, $1\leq p<\infty$. To attain full generality on $L_{p}$ we would attain arbitrary sum ranges, and not only the affine-independent sum range given above. We will do that according to the scheme from [K90], as follows: ###### Lemma 5. Let $\Omega$ be an arbitrary probability space, $c_{n}\in{\mathbb{R}}$, $c_{n}{\rightarrow}0$ and let $f_{n}\in L_{2}(\Omega)$ be a sequence of integer-valued functions. Then the series $\sum_{n=1}^{\infty}(f_{n}+c_{n})$ converges if and only if both $\sum_{n=1}^{\infty}f_{n}$ and $\sum_{n=1}^{\infty}c_{n}$ converge. ###### Proof. The “if” part is obvious. For the “only if” part it is enough to prove that if $\sum c_{n}$ diverges, then $\sum(f_{n}+c_{n})$ has to diverge as well. In fact if $\sum c_{n}$ diverges then there exists an ${\varepsilon}\in(0,1/4)$ such that for any $N\in{\mathbb{N}}$ we have a large Cauchy sum above $N$, i.e. for some $l>k>N$ we have $|\sum_{n=k}^{l}c_{n}|>{\varepsilon}$. As $c_{n}{\rightarrow}0$ we can take $N$ large enough to ensure $|c_{j}|<{\varepsilon}$ for $j>N$. Thus we can select $l=l(k)$ such that ${\varepsilon}<\sum_{n=k}^{l(k)}c_{n}<2{\varepsilon}<\frac{1}{2}$. But then $\|\sum_{n=k}^{l(k)}(f_{n}+c_{n})\|\geq{\varepsilon}$ as a sum of an integer- valued function and a constant $c\in({\varepsilon},1/2)$, which ensures the divergence of $\sum(f_{n}+c_{n})$.∎ Now let us apply this lemma to our example from Theorem 6.1. We have a series $d_{k}$ with an $r+2$-point sum range $D$ defined on $\Omega=\bigcup_{i=1}^{2r+1}Q_{i}$ of cubes. We consider it as a series defined on $L_{2}(\Omega)$. Let us denote $X={\rm lin}\\{\chi_{Q_{1}},\chi_{Q_{2}},\ldots,\chi_{Q_{2r+1}}\\}$, i.e. the subspace of the piece-wise constant functions on $\Omega$. Let $P:L_{2}(\Omega){\rightarrow}X$ be the orthogonal projection onto $X$. Denote by $Y$ the subspace of $X$ consisting of those piecewise constant functions $(f_{i})_{i=1}^{2r+1}$, where $f_{i}$ is the value of $f$ on $Q_{i}$, that $f_{2j}=f_{2j+1}$ for $j=1,2,\ldots,r$. Recall that $\int_{Q_{2}j}d_{k}d\mu=\int_{Q_{2j+1}}d_{k}d\mu$ for $j=1,2,\ldots,r$. Thus for any $d_{k}$ we have $P(d_{k})\in Y$, and thus $P(D)$ is in fact a subset of $Y$. Recall also that for odd indices $j$ the functions $d_{k}$ are integer-valued. Let $T:Y{\rightarrow}Y$ be an arbitrary linear operator. Put $d_{k}^{\prime}=d_{k}+TP(d_{k})$. ###### Theorem 6.2. The sum range $D^{\prime}$ of the series $\sum d_{k}^{\prime}$ equal $(I+T)(D)$. ###### Proof. The inclusion $(I+T)(D)\subset D^{\prime}$ is evident. To prove the inverse inclusion consider an arbitrary arrangement $(b_{k}^{\prime})$ of $(d_{k}^{\prime})$ and the corresponding rearrangement $(b_{k})$ of $(d_{k})$. If $(b_{k}^{\prime})$ converges to some point $b^{\prime}\in D^{\prime}$, then its restrictions to $Q_{j}$ for odd indices $j$ satisfy the conditions of the lemma. Thus the restrictions to $Q_{j}$ for odd $j$ of $TP(b_{k})$ converge. Now the restrictions of $TP(b_{k})$ to $Q_{j-1}$ are equal to the corresponding restrictions to $Q_{j}$, so the whole series $TP(b_{k})$ converges. Then $\sum b_{k}=\sum(b_{k}^{\prime}-TP(b_{k}))$ also has to converge. The sum of this series $b$ belongs to $D$, hence $b^{\prime}=b+TP(b)$ belongs to $(I+T)(D)$.∎ This example can be transferred to any infinite-dimensional Banach space $Y$ using the results of V.M. Kadets. In [S91], Theorem 7.2.2 states: Let $X$ and $Y$ be Banach spaces, $X\stackrel{{\scriptstyle f}}{{\Rightarrow}}Y$. Suppose that $X$ has a basis $\\{e_{k}\\}_{k=1}^{\infty}$ and let $\sum_{k=1}^{\infty}x_{k}$ be a series in $X$ such that SR$(\sum_{k=1}^{\infty}x_{k})$ is not a linear set. Then for any monotone sequence of positive numbers $\\{a_{k}\\}_{k=1}^{\infty}$ with $a_{k}{\rightarrow}\infty,k{\rightarrow}\infty$, there exists a series $\sum_{k=1}^{\infty}y_{k}$ in $Y$ such that SR$(\sum_{k=1}^{\infty}y_{k})$ is not a linear set and $\|y_{k}\|\leq a_{k}\|x_{k}\|$ for all $k\in{\mathbb{N}}$, Corollary 7.2.1 points out that if $X$ is $l_{2}$ then by Dvoretzky’s theorem $X\stackrel{{\scriptstyle f}}{{\Rightarrow}}Y$, and Corollary 7.2.2 states that In any infinite-dimensional Banach space there are series whose sum range consists of two points. This is achieved by applying the two-point example in $L_{2}$ to Corollary 7.2.1 and following the proof of Theorem 7.2.2 to see that no new points appear and all the old ones are transferred to the space $Y$. We have an $n$-point example in $L_{2}$ which can be in the same manner, through obvious modifications in the proof of Theorem 7.2.2 transferred to any Banach space $Y$. Finally for any finite- dimensional subspaces $H_{1},H_{2}$ of a infinitely dimensional Banach space $Y$ and any isomorphism $f:H_{1}{\rightarrow}H_{2}$ there exists an isomorphism $\tilde{f}:Y{\rightarrow}Y$ extending $f$. Thus having any $n$ points satisfying some linear equations as a sum range of $y_{k}$ in $Y$ we can take an $f$ transferring them to any other $n$ points satisfying the same linear equations and then transfer the whole series by $\tilde{f}$. ## References * [S91] M. I. Kadets and V. M. Kadets, Series in Banach Spaces, Conditional and Unconditional Convergence, Birkhäuser Verlag, 1991. * [O89] M. I. Kadets and K. Woz̀niakowski, On series whose permutations have only two sums, Bull. Polish Acad. Sci. Math. 37 (1989), no. 1–6, 15–21. * [U02] P. L. Ulyanov, On interconnections between the research of russian and polish mathematicians in the theory of functions, Banach Center Publ, Polish Acad. Sci. Math 56 (2002), p. 122–128 * [K90] V. M. Kadets, How many points can the sum-set of a series in a Banach space contain?, Teor. Funktsij, Funktsional. Anal. i Prilozhen. 54(1990), p. 54–57. (Russian)

arxiv-papers

2024-09-04T02:48:54.112501

0803.0433

# The square negative correlation property for generalized Orlicz balls Jakub Onufry Wojtaszczyk Department of Mathematics, Computer Science and Mechanics University of Warsaw ul. Banacha 2, 02-097 Warsaw, Poland email: onufry@duch.mimuw.edu.pl (September 11, 2005) ###### Abstract Recently Antilla, Ball and Perissinaki proved that the squares of coordinate functions in $l_{p}^{n}$ are negatively correlated. This paper extends their results to balls in generalized Orlicz norms on ${\mathbb{R}}^{n}$. From this, the concentration of the Euclidean norm and a form of the Central Limit Theorem for the generalized Orlicz balls is deduced. Also, a counterexample for the square negative correlation hypothesis for 1-symmetric bodies is given. ## 1 Introduction Given a convex, central-symmetric body $K\subset{\mathbb{R}}^{n}$ of volume 1, consider the random variable $X=(X_{1},X_{2},\ldots,X_{n})$, uniformly distributed on $K$. We are interested in determining whether the vector has the square negative correlation, i.e. if ${\rm cov}(X_{i}^{2},X_{j}^{2}):={\mathbb{E}}(X_{i}^{2}X_{j}^{2})-{\mathbb{E}}X_{i}^{2}{\mathbb{E}}X_{j}^{2}\leq 0.$ We assume that $K$ is in isotropic position, i.e. that ${\mathbb{E}}X_{i}=0\hbox{\ \ \ and \ \ \ }{\mathbb{E}}X_{i}\cdot X_{j}=L_{K}^{2}\delta_{ij},$ where $\delta_{ij}$ is the Kronecker delta and $L_{K}$ is a positive constant. Since any convex body not supported on an affine subspace has an affine image which is in isotropic position, this is not a restrictive assumption. The motivation in studying this problem comes from the so-called central limit problem for convex bodies, which is to show that most of the one-dimensional projections of the uniform measure on a convex body are approximately normal. It turns out that the bounds on the square correlation can be crucial to estimating the distance between the one-dimensional projections and the normal distribution (see for instance [ABP03], [MM05]). A related problem is to provide bounds for the quantity $\sigma_{K}$, defined by $\sigma_{K}^{2}=\frac{{\rm Var}(|X|^{2})}{nL_{K}^{4}}=\frac{n{\rm Var}(|X|^{2})}{({\mathbb{E}}|X|^{2})^{2}},$ where $X$ is uniformly distributed on $K$. It is conjectured (see for instance [BK03]) that $\sigma_{K}$ is bounded by a universal constant for any convex symmetric isotropic body. Recently Antilla, Ball and Perissinaki (see [ABP03]) observed that for $K=l_{p}^{n}$ the covariances of $X_{i}^{2}$ and $X_{j}^{2}$ are negative for $i\neq j$, and from this deduced a bound on $\sigma_{K}$ in this class. In this paper we shall study the covariances of $X_{i}^{2}$ and $X_{j}^{2}$ (or, more generally, of any functions depending on a single variable) on a convex, symmetric and isotropic body. We will show a general formula to calculate the covariance for given functions and $K$, and from this formula deduce the covariance of any increasing functions of different variables, in particular of the functions $X_{i}^{2}$ and $X_{j}^{2}$, has to be negative on generalized Orlicz balls. Then we follow [ABP03] to arrive at a concentration property and [MM05] to get a Central Limit Theorem variant for generalized Orlicz balls. The layout of this paper is as follows. First we define notations which will be used throughout the paper. In Section 2 we transform the formula for the square correlation into a form which will be used further on. In Section 3 we use the formula and the Brunn-Minkowski inequality to arrive at the square negative correlation property for generalized Orlicz balls. In Section 4 we show the corollaries, in particular a central-limit theorem for generalized Orlicz balls. Section 5 contains another application of the formula from Section 2, a simple counterexample for the square negative correlation hypothesis for 1-symmetric bodies. ### Notation Throughout the paper $K\subset{\mathbb{R}}^{n}$ will be a convex central- symmetric body of volume 1 in isotropic position. Recall that by isotropic position we mean that for any vector $\theta\in S^{n-1}$ we have $\int_{K}\left\langle\theta,x\right\rangle^{2}dx=L_{K}^{2}$ for some constant $L_{K}$. For $A\subset{\mathbb{R}}^{n}$ by $|A|$ we will denote the Lebesgue volume of $A$. For $x\in{\mathbb{R}}^{n}$, $|x|$ will mean the Euclidean norm of $x$. We assume that ${\mathbb{R}}^{n}$ is equipped with the standard Euclidean structure and with the canonic orthonormal base $(e_{1},\ldots,e_{n})$. For $x\in{\mathbb{R}}^{n}$ by $x_{i}$ we shall denote the $i$th coordinate of $x$, i.e. $\left\langle e_{i},x\right\rangle$. We will consider $K$ as a probability space with the Lebesgue measure restricted to $K$ as the probability measure. If there is any danger of confusion, then $\mathbb{P}_{K}$ will denote the probability with respect to this measure, ${\mathbb{E}}_{K}$ will denote the expected value with respect to $\mathbb{P}_{K}$, and so on. By $X$ we will usually denote the $n$-dimensional random vector equidistributed on $K$, while $X_{i}$ will denote its $i$th coordinate. By the covariance ${\rm cov}(Y,Z)$ for real random variables $Y$, $Z$ we mean ${\mathbb{E}}(YZ)-{\mathbb{E}}Y{\mathbb{E}}Z$. By an 1-symmetric body $K$ we mean one that is invariant under reflections in the coordinate hyperplanes, or equivalently, such a body that $(x_{1},x_{2},\ldots,x_{n})\in X{\Longleftrightarrow}({\varepsilon}_{1}x_{1},{\varepsilon}_{2}x_{2},\ldots,{\varepsilon}_{n}x_{n}\in X)$ for any choice of ${\varepsilon}_{i}\in\\{-1,1\\}$. The parameter $\sigma_{K}$, as in [BK03], will be defined by $\sigma_{K}^{2}=\frac{{\rm Var}(|X|^{2})}{nL_{K}^{4}}=\frac{n{\rm Var}(|X|^{2})}{({\mathbb{E}}|X|^{2})^{2}}.$ For any $n\geq 1$ and convex increasing functions $f_{i}:[0,\infty){\rightarrow}[0,\infty)$, $i=1,\ldots,n$ satisfying $f_{i}(0)=0$ (called the Young functions) we define the generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ to be the set of points $x=(x_{1},\ldots,x_{n})$ satisfying $\sum_{i=1}^{n}f_{i}(|x_{i}|)\leq 1.$ This is easily proven to be convex, symmetric and bounded, thus $\|x\|=\inf\\{\lambda:x\in\lambda K\\}$ defines a norm on ${\mathbb{R}}^{n}$. In the case of equal functions $f_{i}$ the norm is called an Orlicz norm, in the general case a generalized Orlicz norm. Examples of Orlicz norms include the $l_{p}$ norms for any $p\geq 1$ with $f(t)=|t|^{p}$ being the Young functions. The generalized Orlicz spaces are also referred to as modular sequence spaces (I thank the referee for pointing this out to me). ## 2 The general formula We wish to calculate ${\rm cov}(f(X_{i}),g(X_{j}))$, where $f$ and $g$ are univariate functions, $i\neq j$ and $X_{i},X_{j}$ are the coordinates of the random vector $X$, equidistributed on a convex, symmetric and isotropic body $K$. For simplicity we will assume $i=1$, $j=2$ and denote $X_{1}$ by $Y$ and $X_{2}$ by $Z$. For any $(y,z)\in{\mathbb{R}}^{2}$ let $m(y,z)$ be equal to the $n-2$-dimensional Lebesgue measure of the set $(\\{(y,z)\\}\times{\mathbb{R}}^{n-2})\cap K$. We set out to prove: ###### Theorem 2.1. For any symmetric, convex body $K$ in isotropic position and any functions $f$, $g$ we have ${\rm cov}(f(Y),g(Z))=\int_{{\mathbb{R}}^{4},|y|>|{\bar{y}}|,|z|>|{\bar{z}}|}\big{(}m(y,z)m({\bar{y}},{\bar{z}})-m(y,{\bar{z}})m({\bar{y}},z)\big{)}\big{(}f(y)-f({\bar{y}})\big{)}\big{(}g(z)-g({\bar{z}})\big{)}.$ Furthermore, for 1-symmetric bodies and symmetric functions we will have the following corrolary: ###### Corollary 2.2. For any symmetric, convex, uncondtitional body $K$ in isotropic position and symmetric functions $f$, $g$ we have ${\rm cov}(f(Y),g(Z))=16\ \int_{{\mathbb{R}}^{4},y>{\bar{y}}>0,z>{\bar{z}}>0}\big{(}m(y,z)m({\bar{y}},{\bar{z}})-m(y,{\bar{z}})m({\bar{y}},z)\big{)}\big{(}f(y)-f({\bar{y}})\big{)}\big{(}g(z)-g({\bar{z}})\big{)}.$ The corollary is a simple consequence of the fact that for symmetric functions $f$ and $g$ and an 1-symmetric body $K$ the integrand is invariant under the change of the sign of any of the variables, so we may assume all of them are positive. As concerns the sign of ${\rm cov}(f,g)$, which is what we set out to determine, we have the following simple corollary: ###### Corollary 2.3. For any central-symmetric, convex, 1-symmetric body $K$ in isotropic position and symmetric functions $f$, $g$ that are non-decreasing on $[0,\infty)$ if for all $y>{\bar{y}}>0$, $z>{\bar{z}}>0$ we have $m(y,{\bar{z}})m({\bar{y}},z)\geq m(y,z)m({\bar{y}},{\bar{z}}),$ (1) then ${\rm cov}(f,g)\leq 0.$ Similarly, if the opposite inequality is satisfied for all $y>{\bar{y}}>0$ and $z>{\bar{z}}>0$, then the covariance is non-negative. ###### Proof. The second and third bracket of the integrand in Corollary 2.2 is positive under the assumptions of Corollary 2.3. Thus if we assume the first bracket is negative, then the whole integrand is negative, which implies the integral is negative, and vice-versa.∎ ###### Proof of Theorem 2.1. We have ${\rm cov}(f(Y),g(Z))={\mathbb{E}}f(Y)g(Z)-{\mathbb{E}}f(Y){\mathbb{E}}g(Z).$ From the Fubini theorem we have ${\mathbb{E}}f(Y)g(Z)=\int_{R^{2}}m(y,z)f(y)g(z),$ and similar equations for ${\mathbb{E}}f(Y)$ and ${\mathbb{E}}g(Z)$. For any function $h$ of two variables $a,b\in A$ we can write $\int_{A^{2}}h(a,b)=\int_{A^{2}}h(b,a)=\frac{1}{2}\int_{A^{2}}h(a,b)+h(b,a)$. We shall repeatedly use this trick to transform the formula for the covariance of $f$ and $g$ into the required form: $\displaystyle{\mathbb{E}}f(Y){\mathbb{E}}g(Z)$ $\displaystyle=$ $\displaystyle\int_{{\mathbb{R}}^{2}}m(y,z)f(y)\int_{{\mathbb{R}}^{2}}m({\bar{y}},{\bar{z}})g({\bar{z}})$ $\displaystyle=$ $\displaystyle\int_{{\mathbb{R}}^{4}}m(y,z)m({\bar{y}},{\bar{z}})f(y)g({\bar{z}})=\int_{{\mathbb{R}}^{4}}m({\bar{y}},{\bar{z}})m(y,z)f({\bar{y}})g(z)=$ $\displaystyle=$ $\displaystyle\frac{1}{2}\int_{{\mathbb{R}}^{4}}m({\bar{y}},{\bar{z}})m(y,z)\big{(}f({\bar{y}})g(z)+f(y)g({\bar{z}})\big{)}.$ We repeat this trick, exchanging $z$ and ${\bar{z}}$ (and leaving $y$ and ${\bar{y}}$ unchanged): $\displaystyle{\mathbb{E}}f(Y){\mathbb{E}}g(Z)$ $\displaystyle=$ $\displaystyle\frac{1}{4}\int_{{\mathbb{R}}^{4}}m({\bar{y}},{\bar{z}})m(y,z)\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)\big{)}+m({\bar{y}},z)m(y,{\bar{z}})\big{(}f(y)g(z)+f({\bar{y}})g({\bar{z}})\big{)}.$ We perform the same operations on the second part of the covariance. To get a integral over ${\mathbb{R}}^{4}$ we multiply by an ${\mathbb{E}}1$ factor (this in effect will free us from the assumption that the body’s volume is 1): $\displaystyle{\mathbb{E}}f(Y)g(Z){\mathbb{E}}1$ $\displaystyle=$ $\displaystyle\int_{{\mathbb{R}}^{4}}m(y,z)m({\bar{y}},{\bar{z}})f(y)g(z)$ $\displaystyle=$ $\displaystyle\frac{1}{4}\int_{{\mathbb{R}}^{4}}m(y,z)m({\bar{y}},{\bar{z}})\big{(}f(y)g(z)+f({\bar{y}})g({\bar{z}})\big{)}+m(y,{\bar{z}})m({\bar{y}},z)\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)\big{)}.$ Thus: $\displaystyle{\rm cov}(f(Y),g(Z))={\mathbb{E}}(f(Y)g(Z)){\mathbb{E}}1-{\mathbb{E}}f(Y){\mathbb{E}}g(Z)=$ $\displaystyle=$ $\displaystyle\frac{1}{4}\bigg{(}\int_{{\mathbb{R}}^{4}}m(y,z)m({\bar{y}},{\bar{z}})\big{(}f(y)g(z)+f({\bar{y}})g({\bar{z}})\big{)}+m(y,{\bar{z}})m({\bar{y}},z)\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)\big{)}-$ $\displaystyle-m({\bar{y}},{\bar{z}})m(y,z)\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)\big{)}-m({\bar{y}},z)m(y,{\bar{z}})\big{(}f(y)g(z)+f({\bar{y}})g({\bar{z}})\big{)}\bigg{)}=$ $\displaystyle=$ $\displaystyle\frac{1}{4}\int_{{\mathbb{R}}^{4}}\bigg{(}\big{(}m(y,{\bar{z}})m({\bar{y}},z)-m(y,z)m({\bar{y}},{\bar{z}})\big{)}\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)\big{)}+$ $\displaystyle+\big{(}m(y,z)m({\bar{y}},{\bar{z}})-m({\bar{y}},z)m(y,{\bar{z}})\big{)}\big{(}f(y)g(z)+f({\bar{y}})g({\bar{z}})\big{)}\bigg{)}=$ $\displaystyle=$ $\displaystyle\frac{1}{4}\int_{{\mathbb{R}}^{4}}\big{(}m(y,{\bar{z}})m({\bar{y}},z)-m(y,z)m({\bar{y}},{\bar{z}})\big{)}\big{(}f(y)g({\bar{z}})+f({\bar{y}})g(z)-f(y)g(z)-f({\bar{y}})g({\bar{z}})\big{)}=$ $\displaystyle=$ $\displaystyle\frac{1}{4}\int_{{\mathbb{R}}^{4}}\big{(}m(y,{\bar{z}})m({\bar{y}},z)-m(y,z)m({\bar{y}},{\bar{z}})\big{)}\big{(}f(y)-f({\bar{y}})\big{)}\big{(}g({\bar{z}})-g(z)\big{)}$ Finally, notice that if we exchange $y$ and ${\bar{y}}$ in the above formula, then the formula’s value will not change — the first and second bracket will change signs, and the third will remain unchanged. The same applies to exchanging $z$ and ${\bar{z}}$. Thus ${\rm cov}(f,g)=\int_{{\mathbb{R}}^{4},|y|>|{\bar{y}}|,|z|>|{\bar{z}}|}\big{(}m(y,z)m({\bar{y}},{\bar{z}})-m(y,{\bar{z}})m({\bar{y}},z)\big{)}\big{(}f(y)-f({\bar{y}})\big{)}\big{(}g(z)-g({\bar{z}})\big{)}.$ ∎ ## 3 Generalized Orlicz spaces Now we will concentrate on the case of symmetric, non-decreasing functions on generalized Orlicz spaces. We will prove the inequality (1): ###### Theorem 3.1. If $K$ is a ball in an generalized Orlicz norm on ${\mathbb{R}}^{n}$, then for any $y>{\bar{y}}>0$ and $z>{\bar{z}}>0$ we have $m(y,{\bar{z}})m({\bar{y}},z)\geq m(y,z)m({\bar{y}},{\bar{z}}).$ (2) From this Theorem and Corollary 2.3 we get ###### Corollary 3.2. If $K$ is a ball in an generalized Orlicz norm on ${\mathbb{R}}^{n}$ and $f,g$ are symmetric functions that are non-decreasing on $[0,\infty)$, then ${\rm cov}_{K}(f,g)\leq 0$. It now remains to prove the inequality (2). ###### Proof of Theorem 3.1. Let $f_{i}$ denote the Young functions of $K$. Let us consider the ball $K^{\prime}\subset{\mathbb{R}}^{n-1}$, being an generalized Orlicz ball defined by the Young functions $\Phi_{1},\Phi_{2},\ldots,\Phi_{n-1}$, where $\Phi_{i}(t)=f_{i+1}(t)$ for $i>1$ and $\Phi_{1}(t)=t$ — that is, we replace the first two Young functions of $K$ by a single identity function. For any $x\in{\mathbb{R}}$ let $P_{x}$ be the set $(\\{x\\}\times{\mathbb{R}}^{n-2})\cap K^{\prime}$, and $|P_{x}|$ be its $n-2$-dimensional Lebesgue measure. $K^{\prime}$ is a convex set, thus, by the Brunn-Minkowski inequality (see for instance [G02]) the function $x\mapsto|P_{x}|$ is a logarithmically concave function. This means that $x\mapsto\log|P_{x}|$ is a concave function, or equivalently that $|P_{tx+(1-t)y}|\geq|P_{x}|^{t}\cdot|P_{y}|^{1-t}.$ In particular, for given real positive numbers $a$, $b$, $c$ we have $|P_{a+c}|\geq|P_{a}|^{b/(b+c)}|P_{a+b+c}|^{c/(b+c)},$ $|P_{a+b}|\geq|P_{a}|^{c/(b+c)}|P_{a+b+c}|^{b/(b+c)},$ and as a consequence when we multiply the two inequalities, $|P_{a+b}|\cdot|P_{a+c}|\geq|P_{a}|\cdot|P_{a+b+c}|.$ (3) Now let us consider the ball $K$. Let us take any $y>{\bar{y}}>0$ and $z>{\bar{z}}>0$. Let $a=f_{1}({\bar{y}})+f_{2}({\bar{z}})$, $b=f_{1}(y)-f_{1}({\bar{y}})$, and $c=f_{2}(z)-f_{2}({\bar{z}})$. The numbers $a$, $b$ and $c$ are positive from the assumptions on $y$, $z$, ${\bar{y}}$ and ${\bar{z}}$ and because the Young functions are increasing. Then $m({\bar{y}},{\bar{z}})$ is equal to the measure of the set $\\{x_{3},x_{4},\ldots,x_{n}:f_{1}({\bar{y}})+f_{2}({\bar{z}})+\sum_{i=3}^{n}f_{i}(x_{i})\leq 1\\}=\\{x_{3},x_{4},\ldots,x_{n}:a+\sum_{i=2}^{n}\Phi_{i}(x_{i})\leq 1\\}=P_{a}.$ Similarly $m(y,{\bar{z}})=|P_{a+b}|$, $m({\bar{y}},z)=|P_{a+c}|$ i $m(y,z)=|P_{a+b+c}|$. Substituting those values into the inequality (3) we get the thesis: $m(y,{\bar{z}})m({\bar{y}},z)\geq m(y,z)m({\bar{y}},{\bar{z}}).$ ∎ ## 4 The consequences For the consequences we will take $f(t)=g(t)=t^{2}$. The first simple consequence is the concentration property for generalized Orlicz balls. Here, we follow the argument of [ABP03] for $l_{p}$ balls. ###### Theorem 4.1. For every generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ we have $\sigma_{K}\leq\sqrt{5}.$ ###### Proof. From the Cauchy-Schwartz inequality we have $n^{2}L_{K}^{4}=\bigg{(}\sum_{i=1}^{n}{\mathbb{E}}_{K}X_{i}^{2}\bigg{)}^{2}=\bigg{(}{\mathbb{E}}_{K}|X|^{2}\bigg{)}^{2}\leq{\mathbb{E}}_{K}|X|^{4}.$ On the other hand from Corollary 3.2 we have $\displaystyle{\mathbb{E}}_{K}|X|^{4}$ $\displaystyle=$ $\displaystyle{\mathbb{E}}_{K}\bigg{(}\sum_{i=1}^{n}X_{i}^{2}\bigg{)}^{2}=\sum_{i=1}^{n}{\mathbb{E}}_{K}X_{i}^{4}+\sum_{i\neq j}{\mathbb{E}}_{K}X_{i}^{2}X_{j}^{2}$ $\displaystyle\leq$ $\displaystyle\sum_{i=1}^{n}{\mathbb{E}}_{K}X_{i}^{4}+\sum_{i\neq j}{\mathbb{E}}_{K}X_{i}^{2}{\mathbb{E}}_{K}X_{j}^{2}$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{n}{\mathbb{E}}_{K}X_{i}^{4}+n(n-1)L_{K}^{4}.$ As for 1-symmetric bodies the density of $X_{i}$ is symmetric and log-concave, we know (see e.g. [KLO96], Section 2, Remark 5) ${\mathbb{E}}_{K}X_{i}^{4}\leq 6\bigg{(}{\mathbb{E}}_{K}X_{i}^{2}\bigg{)}^{2}=6L_{K}^{4},$ thence $n^{2}L_{K}^{4}\leq{\mathbb{E}}_{K}|X|^{4}\leq(n^{2}+5n)L_{K}^{4}.$ This gives us ${\rm Var}(|X|^{2})={\mathbb{E}}_{K}|X|^{4}-n^{2}L_{K}^{4}\leq 5nL_{K}^{4},$ and thus $\sigma_{K}^{2}=\frac{{\rm Var}|X|^{2}}{nL_{K}^{4}}\leq 5.$ ∎ ###### Corollary 4.2. For every generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ and for every $t>0$ we have $\mathbb{P}_{K}\bigg{(}\bigg{|}\frac{|X|^{2}}{n}-L_{K}^{2}\bigg{|}\geq t\bigg{)}\leq\frac{5L_{K}^{4}}{nt^{2}}$ and $\mathbb{P}_{K}\bigg{(}\bigg{|}\frac{|X|}{\sqrt{n}}-L_{K}\bigg{|}\geq t\bigg{)}\leq\frac{5L_{K}^{2}}{nt^{2}}$ ###### Proof. From the estimate on the variance of $|X|^{2}$ and Chebyshev’s inequality we get $t^{2}\mathbb{P}_{K}\bigg{(}\bigg{|}\frac{|X|^{2}}{n}-L_{K}^{2}\bigg{|}\geq t\bigg{)}\leq{\mathbb{E}}_{K}\bigg{(}\frac{|X|^{2}}{n}-L_{K}^{2}\bigg{)}^{2}\leq\frac{1}{n^{2}}{\rm Var}(|X|^{2})\leq\frac{5}{n}L_{K}^{4}.$ For the second part let $t>0$. We have $\displaystyle\mathbb{P}_{K}(|X|-\sqrt{n}L_{K}|\geq t\sqrt{n})$ $\displaystyle\leq$ $\displaystyle\mathbb{P}_{K}(|X|^{2}-nL_{K}^{2}|\geq tnL_{K})$ $\displaystyle\leq$ $\displaystyle\frac{5L_{K}^{4}}{t^{2}nL_{K}^{2}}=\frac{5L_{K}^{2}}{t^{2}n}.$ ∎ This result confirms the so-called concentration hypothesis for generalized Orlicz balls. The hypothesis, see e.g. [BK03], states that the Euclidean norm concentrates near the value $\sqrt{n}L_{K}$ as a function on $K$. More precisely, for a given ${\varepsilon}>0$ we say that $K$ satisfies the ${\varepsilon}$-concentration hypothesis if ${\mathbb{P}}_{K}\bigg{(}\bigg{|}\frac{|X|}{\sqrt{n}}-L_{K}\bigg{|}\geq{\varepsilon}L_{K}\bigg{)}\leq{\varepsilon}.$ From Corollary 4.2 we get that the class of generalized Orlicz balls satisfies the ${\varepsilon}$-concentration hypothesis with ${\varepsilon}=\sqrt{5}n^{-1/3}$. A more complex consequence is the Central Limit Property for generalized Orlicz balls. For $\theta\in S^{n-1}$ let $g_{\theta}(t)$ be the density of the random variable $\left\langle X,\theta\right\rangle$. Let $g$ be the density of ${\mathcal{N}}(0,L_{K}^{2})$. Then for most $\theta$ the density $g_{\theta}$ is very close to $g$. More precisely, by part 2 of Corollary 4 in [MM05] we get ###### Corollary 4.3. There exists an absolute constant $c$ such that $\sup_{t\in{\mathbb{R}}}\bigg{|}\int_{-\infty}^{t}\big{(}g_{\theta}(s)-g(s)\big{)}ds\bigg{|}\leq c\|\theta\|_{3}^{3/2}.$ ## 5 The counterexample for 1-symmetric bodies It is generally known that the negative square correlation hypothesis does not hold in general in the class of 1-symmetric bodies. However, the formula from section 2 allows us to give a counterexample without any tedious calculations. Let $K\subset{\mathbb{R}}^{3}$ be the ball of the norm defined by $\|(x,y,z)\|=|x|+\max\\{|y|,|z|\\}.$ The quantity $m(y,z)$ considered in Corollary 2.3, defined as the volume of the cross-section $({\mathbb{R}}\times\\{y,z\\})\cap K$ is equal to $2(1-\max\\{|y|,|z|\\})$ for $|y|,|z|\leq 1$ and $0$ for greater $|y|$ or $|z|$. To check the inequality (1) for $y>{\bar{y}}>0$ and $z>{\bar{z}}>0$ we may assume without loss of generality that $y\geq z$ (as $K$ is invariant under the exchange of $y$ and $z$). We have $\displaystyle m(y,{\bar{z}})m({\bar{y}},z)$ $\displaystyle-$ $\displaystyle m(y,z)m({\bar{y}},{\bar{z}})=$ $\displaystyle=$ $\displaystyle 4(1-\max\\{y,{\bar{z}}\\})(1-\max\\{{\bar{y}},z\\})-4(1-\max\\{y,z\\})(1-\max\\{{\bar{y}},{\bar{z}}\\})$ $\displaystyle=$ $\displaystyle 4(1-y)(1-\max\\{{\bar{y}},z\\})-4(1-y)(1-\max\\{{\bar{y}},{\bar{z}}\\})$ $\displaystyle=$ $\displaystyle 4(1-y)(\max\\{{\bar{y}},{\bar{z}}\\}-\max\\{{\bar{y}},z\\}).$ As $y\leq 1$ all we have to consider is the sign of the third bracket. However, as $z>{\bar{z}}$, the third bracket is never positive, and is negative when $z>{\bar{y}}$. Thus from Corollary 2.3 the covariance ${\rm cov}(f,g)$ is positive for any increasing symmetric functions $f(Y)$ and $g(Z)$, in particular for $f(Y)=Y^{2}$ and $g(Z)=Z^{2}$. ## References * [ABP03] M. Anttila, K. Ball and I. Perissinaki, The central limit problem for convex bodies. Trans. Amer. Math. Soc., 355 (2003), pp. 4723–-4735. * [BK03] S. G. Bobkov and A. Koldobsky, On the Central Limit Property of Convex Bodies. GAFA Seminar, Lecture Notes in Math. 1807 (2003), pp. 44–52. * [G02] R. J. Gardner, The Brunn-Minkowski Inequality, Bull. Amer. Math. Soc. 39 (2002), pp. 355-405 * [KLO96] S. Kwapień, R. Latała and K. Oleszkiewicz, Comparison of Moments of Sums of Independent Random Variables and Differential Inequalities. Journal of Functional Analysis, 136 (1996), pp. 258–268. * [MM05] E. Meckes and M. Meckes, The Central Limit Problem for Random Vectors with Symmetries. Preprint. Available at http://arxiv.org/abs/math.PR/0505618.

arxiv-papers

2024-09-04T02:48:54.118852

0803.0434

# The negative association property for the absolute values of random variables equidistributed on a generalized Orlicz ball Marcin Pilipczuk (malcin@duch.mimuw.edu.pl) Jakub Onufry Wojtaszczyk (onufry@duch.mimuw.edu.pl) Department of Mathematics, Computer Science and Mechanics University of Warsaw ul. Banacha 2, 02-097 Warsaw, Poland Partially supported by MEiN Grant no 1 PO3A 012 29 ###### Abstract Random variables equidistributed on convex bodies have received quite a lot of attention in the last few years. In this paper we prove the negative association property (which generalizes the subindependence of coordinate slabs) for generalized Orlicz balls. This allows us to give a strong concentration property, along with a few moment comparison inequalities. Also, the theory of negatively associated variables is being developed in its own right, which allows us to hope more results will be available. Moreover, a simpler proof of a more general result for $\ell_{p}^{n}$ balls is given. ###### Contents 1. 1 Introduction 1. 1.1 Notation 2. 1.2 Results 3. 1.3 Motivations 4. 1.4 Acknowledgements 2. 2 Easy facts 1. 2.1 Simplifying 2. 2.2 Simple proportion lemmas 3. 3 The $\ell_{p}^{n}$ ball case 4. 4 The generalized Orlicz ball case — preliminaries, the proper measure, lens sets 1. 4.1 Idea of the proof 2. 4.2 Definitions 3. 4.3 The generalized Orlicz ball lemmas 4. 4.4 $1/m$-concave functions and proper measures 5. 4.5 Lens sets and $\Theta$ functions 5. 5 The $\Theta$ theorem 1. 5.1 Preparations for divisibility 2. 5.2 Almost horizontal divisions 3. 5.3 ${\varepsilon}$-appropriateness of lens sets 6. 6 The transfinite induction 1. 6.1 Starting the transfinite induction 2. 6.2 The induction step for successor ordinals 3. 6.3 The induction step for limit ordinals 7. 7 $\Theta$ functions on Orlicz balls 1. 7.1 The one-dimensional case — the $\phi$ functions 2. 7.2 The general case — the $\psi$ function ## 1 Introduction ### 1.1 Notation We shall begin by introducing the notation used throughout the paper. For any set $A$ by ${\mathbf{1}}_{A}$ we shall denote the characteristic function of $A$. As usually, ${\mathbb{R}}$ and ${\mathbb{R}}_{+}$ will denote the reals and the non-negative reals respectively. By ${\mathbb{R}}^{k}$ we shall mean the $k$-dimensional Euclidean space equipped with the standard scalar product $\left\langle\cdot,\cdot\right\rangle$, the Lebesgue measure denoted by $\lambda$ or $\lambda_{k}$ and a system of orthonormal coordinates $x_{1},x_{2},\ldots,x_{k}$. By ${\mathbb{R}}_{+}^{k}$ we mean the generalized positive quadrant, that is the set $\\{(x_{1},\ldots,x_{k})\in{\mathbb{R}}^{k}:\forall_{i}\ x_{i}\geq 0\\}$. For a given set $K\subset{\mathbb{R}}^{k}$ by $K_{+}$ we shall denote the positive quadrant of $K$, that is $K\cap{\mathbb{R}}_{+}^{k}$. For a given set $A$ by ${\bar{A}}$ we will denote the complement of $A$. For a measure $\mu$ on ${\mathbb{R}}^{n}$ and an affine subspace $H\subset{\mathbb{R}}^{n}$, by the projection of $\mu$ onto $H$ we mean the measure $\mu_{H}$ defined by $\mu_{H}(C)=\mu(\\{x\in{\mathbb{R}}^{n}:P(x)\in C\\})$, where $P$ is the orthogonal projection onto $H$. If $\mu$ is given by a density function $m$ and $K\subset H\subset{\mathbb{R}}^{n}$, then by the restriction of $\mu$ to $K$ we mean the measure $\mu_{|K}$ on $H$ given with the density $m\cdot{\mathbf{1}}_{K}$. By the support of a function $m:X\to{\mathbb{R}}$, denoted ${\rm supp}m$, we mean ${\rm cl}\\{x\in X:m(x)\neq 0\\}$. If $\mu$ is a measure, then by ${\rm supp}\mu$ we mean the smallest closed set $A$ such that $\mu(\bar{A})=0$. In the cases we consider, when $\mu$ will be given by a density $m$, we will always have $supp\mu={\rm supp}m$. We shall call a set $K\subset{\mathbb{R}}^{n}$ a symmetric body if it is convex, bounded, central-symmetric (i.e. if $x\in K$ then $-x\in K$) and has a non-empty interior. A body $K\subset{\mathbb{R}}^{n}$ is called 1-symmetric if for any $({\varepsilon}_{1},\ldots,{\varepsilon}_{n})\in\\{-1,1\\}^{n}$ and any $(x_{1},\ldots,x_{n})\in K$ we have $({\varepsilon}_{1}x_{1},\ldots,{\varepsilon}_{n}x_{n})\in K$. Such a body is sometimes called unconditional. A function $f:{\mathbb{R}}_{+}{\rightarrow}{\mathbb{R}}_{+}\cup\\{\infty\\}$ is called a Young function if it is convex, $f(0)=0$ and $\exists_{x}:f(x)\neq 0$, $\exists_{x\neq 0}:f(x)\neq\infty$. If we have $n$ Young functions $f_{1},\ldots,f_{n}$, then the set $K=\\{(x_{1},\ldots,x_{n}):\sum_{i=1}^{n}f_{i}(|x_{i}|)\leq 1\\}$ is a 1-symmetric body in ${\mathbb{R}}^{n}$. Such a set is called a generalized Orlicz ball, also known in the literature as a modular sequence space ball. We shall call a Young function $f$ proper if it does not attain the $+\infty$ value and $f(x)>0$ for $x>0$. A generalized Orlicz ball is called proper if it can be defined by proper Young functions. If the coordinates of the space ${\mathbb{R}}^{n}$ are denoted $x_{1},x_{2},\ldots,x_{n}$, the appropriate Young functions will be denoted $f_{1},f_{2},\ldots,f_{n}$, with the assumption $f_{i}$ is applied to $x_{i}$. If some of the coordinates are denoted $x,y,z,\ldots$, the appropriate Young functions will be denoted $f_{x},f_{y},f_{z},\ldots$, with the assumption that $f_{x}$ is applied to $x$, $f_{y}$ to $y$ and so on. A function $f:{\mathbb{R}}{\rightarrow}{\mathbb{R}}$ is called increasing (decreasing) if $x\geq y$ implies $f(x)\geq f(y)$ ($f(x)\leq f(y)$) — we do not require a sharp inequality. A function $f:{\mathbb{R}}^{k}{\rightarrow}{\mathbb{R}}$ or $f:{\mathbb{R}}_{+}^{k}{\rightarrow}{\mathbb{R}}$ is called coordinate-wise increasing (decreasing), if for $x_{i}\geq y_{i}$, $i=1,2,\ldots,n$ we have $f(x_{1},\ldots,x_{k})\geq f(y_{1},\ldots,y_{n})$ ($f(x_{1},\ldots,x_{k})\leq f(y_{1},\ldots,y_{n})$). A set $A\subset{\mathbb{R}}_{+}^{k}$ is called a c-set, if for $x_{i}\geq y_{i}\geq 0$, $i=1,2,\ldots,n$ and $(x_{1},\ldots,x_{n})\in A$ we have $(y_{1},\ldots,y_{n})\in A$. For a coordinate-wise increasing function $f:{\mathbb{R}}_{+}^{k}{\rightarrow}{\mathbb{R}}$ the sets $f^{-1}((-\infty,t])$ are c-sets, and conversely the characteristic function of a c-set is a coordinate-wise decreasing function on ${\mathbb{R}}_{+}^{k}$. Similarily a function $f:{\mathbb{R}}_{+}^{k}{\rightarrow}{\mathbb{R}}$ is radius-wise increasing if $f(tx_{1},tx_{2},\ldots,tx_{n})\geq f(x_{1},x_{2},\ldots,x_{n})$ for $t>1$, and a set $A$ is a radius-set if its characteristic function is radius-wise decreasing. We say a function $f:{\mathbb{R}}^{n}{\rightarrow}{\mathbb{R}}_{+}$ is log- concave if $\ln f$ is concave. A measure $\mu$ on ${\mathbb{R}}^{n}$ is called log-concave if for any nonempty $A,B\subset{\mathbb{R}}^{n}$ and $t\in(0,1)$ we have $\mu(tA+(1-t)B)\geq\mu(A)^{t}\mu(B)^{1-t}$. A classic theorem by Borell (see [Bo74]) states that any log-concave density not concentrated on any affine hyperplane has a density function, and that function is log- concave. A random vector in ${\mathbb{R}}^{n}$ is said to be log-concave if its distribution is log-concave. A sequence of random variables $(X_{1},\ldots,X_{n})$ is said to be negatively associated, if for any coordinate-wise increasing bounded functions $f,g$ and disjoint sets $\\{i_{1},\ldots,i_{k}\\}$ and $\\{j_{1},\ldots,j_{l}\\}\subset\\{1,\ldots,n\\}$ we have ${\rm Cov}\big{(}f(X_{i_{1}},\ldots,X_{i_{k}}),g(X_{j_{1}},\ldots,X_{j_{l}})\big{)}\leq 0.$ (1.1.1) We say that the sequence $(X_{j})$ is weakly negatively associated if inequality (1.1.1) holds for $l=1$, and very weakly negatively associated if (1.1.1) holds for $l=k=1$. For a 1-symmetric body $K\subset{\mathbb{R}}^{n}$ we can treat the body, or its positive quadrant, as a probability space, with the normalized Lebesgue measure as the probability. Formally, we consider $\Omega=K$, the Borel subsets of $K$ as the $\sigma$-family and ${\mathbb{P}}=\frac{1}{\lambda(K)}\lambda$ as the probability measure. We do similarly for $K_{+}$. We also define $n$ random variables $X_{1},\ldots,X_{n}$, with $X_{i}$ being the $i$-th coordinate of a point $\omega\in K_{+}$ or $K$. ### 1.2 Results Our main subject of interest is to prove negative associacion type properties for some classes of symmetric bodies in ${\mathbb{R}}^{n}$. An straightforward approach is bound to fail due to the following proposition: ###### Proposition 1.1. If for a 1-symmetric body $K$ we consider the random vectors uniformly distributed on $K$ (not just on $K_{+}$) and the coordinate variables are very weakly negatively associated, then they are pairwise independent, and thus $K$ is a rescaled cube. ###### Proof. Take any $i,j\in\\{1,\ldots,n\\}$ and any increasing functions $f,g:{\mathbb{R}}{\rightarrow}{\mathbb{R}}$. Then $f^{\circ}(x)=-f(-x)$ is increasing too. $K$ is 1-symmetric, so $(X_{i},X_{j})$ has the same joint distribution as $(-X_{i},X_{j})$, so ${\rm Cov}(f^{\circ}(X_{i}),g(X_{j}))={\rm Cov}\big{(}-f(-X_{i}),g(X_{j}))=-{\rm Cov}(f(X_{i}),g(X_{j})\big{)}.$ If both ${\rm Cov}(f(X_{i}),g(X_{j}))$ and $-{\rm Cov}(f(X_{i}),g(X_{j}))$ are non-positive, then ${\rm Cov}(f(X_{i}),g(X_{j}))=0$. This holds for every $i,j,f,g$. In particular for every $a,b$ we have ${\mathbb{P}}(X_{i}\in[a,\infty)\cap X_{j}\in[b,\infty))-{\mathbb{P}}(X_{i}\in[a,\infty))\cdot{\mathbb{P}}(X_{j}\in[b,\infty))={\rm Cov}({\mathbf{1}}_{[a,\infty)},{\mathbf{1}}_{[b,\infty)})=0.$ A standard argument shows that $X_{i}$ and $X_{j}$ are independent, thus the density of ${\mathbf{1}}_{K}$ is a product density, so $K$ has to be a product of intervals. ∎ Thus, even very weak negative associacion for coordinate variables occurs only in the trivial case. The problem becomes more interesting if we look at the variables $|X_{i}|$ (or, equivalently, restrict ourselves to $X_{i}\geq 0$). K. Ball and I. Perissinaki in [BP98] prove the subindependence of coordinate slabs for $\ell^{p}$ balls, from which very weak negative association of $(|X_{1}|,\ldots,|X_{n}|)$ is a simple consequence. In the paper [W06] Corollary 3.2 states that the sequence of variables $(|X_{1}|,\ldots,|X_{n}|)$ is very weakly negatively associated for generalized Orlicz balls. In this paper we shall prove that for a generalized Orlicz ball the sequence of variables $(|X_{1}|,\ldots,|X_{n}|)$ is negatively associated: ###### Theorem 1.2. Let $K$ be an generalized Orlicz ball, and let $X_{i}$ be the coordinates of a random vector uniformly distributed on $K$. Then the sequence $|X_{i}|$ is negatively associated. We shall also prove an even stronger property of $\ell_{p}^{n}$ balls: ###### Theorem 1.3. Take any $p\in[1,\infty)$ and any $n\in{\mathbb{N}}$. Let $m:{\mathbb{R}}_{+}{\rightarrow}{\mathbb{R}}_{+}$ be any log-concave function and let $\mu$ be the measure on ${\mathbb{R}}^{n}$ with the density at $x$ equal to $m(\|x\|_{p}^{p})$ normalized to be a probability measure. Let $I=\\{i_{1},\ldots,i_{k}\\},J=\\{j_{1},\ldots,j_{l}\\}$ be two disjoint subsets of $\\{1,2,\ldots,n\\}$, and let $f:{\mathbb{R}}_{+}^{k}{\rightarrow}{\mathbb{R}}$, $g:{\mathbb{R}}_{+}^{l}{\rightarrow}{\mathbb{R}}$ be any radius-wise increasing functions bounded on ${\rm supp}\mu$. Let $X=(X_{1},X_{2},\ldots,X_{n})$ be the vector distributed according to $\mu$. Then ${\rm Cov}(f(|X_{i_{1}}|,|X_{i_{2}}|,\ldots,|X_{i_{k}}|),g(|X_{j_{1}}|,|X_{j_{2}}|,\ldots,|X_{j_{l}}|))\leq 0.$ This is an equivalent of the above theorem, but the uniform distribution is replaced by the class of distribution with the density being a log-concave function of the $p$-th power of the $p$-th norm, and the coordinate-wise increasing function replaced by radius-wise decreasing functions. Let us comment on the organization of the paper. In the following subsection we shall state the main results and show a few corollaries which motivate these results. Section 2 is a collection of general lemmas, which allow us to reformulate the problem in a simpler fashion. In Section 3 a simple proof for the $\ell_{p}^{n}$ result is given. Section 4 introduces the definitions used in dealing with the generalized Orlicz ball case and investigates the basic properties of the defined objects. Section 5 states the $\theta$-theorem, which is the main tool of the proof, and gives a part of the proof. Section 6 contains the second part of the proof, which is a large transfinite inductive construction. Finally Section 7 applies the $\theta$-theorem to obtain the result for generalized Orlicz balls. ### 1.3 Motivations This study was motivated by a desire to link the results achieved in convex geometry in [ABP03] for $\ell_{p}$ balls and in [W06] for generalized Orlicz balls with an established theory, which will hopefully allow us to avoid repeating proofs already made in a more general case. For example, a form of the Central Limit Theorem for negative associated variables was already known in 1984 (see [N84]). We also hope some new observations can be made using this approach. The negative association property is stronger then the sub-independence of coordinate slabs, which has been studied in the context of the Central Limit Theorem (see [ABP03], [BP98]). The statement of Theorem 1.3 was motivated by Theorem 6 of [BGMN05], where a proof of subindependence of coordinate slabs is given for a different class of measures with density dependent on the $p$-th norm, also including the uniform measure and the normalized cone measure on the surface. An example that can prove useful for applications in convex geometry is a pair of comparison inequalities due to Shao (see [S00]). First, notice that as $|X_{i}|$ are negatively associated, they remain negatively associated when multiplied by any non-negative scalars (which amounts to multiplying $X_{i}$ by any scalars) and after the addition of any constant scalars. Thus the vectors $|a_{i}X_{i}|-c_{i}$ are negatively associated for any $a_{i},c_{i}\in{\mathbb{R}}$. Shao’s inequalities, when applied to our case it will state the following: ###### Theorem 1.4. Let $K\subset{\mathbb{R}}^{n}$ be a generalized Orlicz ball, $(a_{i})_{i=1}^{n}$ be any sequence of reals and $(X_{i})_{i=1}^{n}$ be the coordinates of the random vector uniformly distributed on $K$. Then for any convex function $f:{\mathbb{R}}{\rightarrow}{\mathbb{R}}$ we have ${\mathbb{E}}f\Big{(}\sum_{i=1}^{n}|a_{i}X_{i}|\Big{)}\leq{\mathbb{E}}f\Big{(}\sum_{i=1}^{n}|a_{i}X_{i}^{\star}|\Big{)},$ where $X_{i}^{\star}$ denote independent random variables with $X_{i}$ and $X_{i}^{\star}$ having the same distribution for each $i$. Additionally, if $f$ is increasing, then for any sequence of reals $(c_{i})_{i=1}^{n}$ we have ${\mathbb{E}}f\Big{(}\max_{k=1,2,\ldots,n}\sum_{i=1}^{k}|a_{i}X_{i}|-c_{i}\Big{)}\leq{\mathbb{E}}f\Big{(}\max_{k=1,2,\ldots,n}\sum_{i=1}^{k}|a_{i}X_{i}^{\star}|-c_{i}\Big{)}.$ A more direct consequence is a moment comparision theorem suggested by R. Latała (note we compare the moments of the sums of variables, and not their absolute values): ###### Theorem 1.5. Let $K\subset{\mathbb{R}}^{n}$ be a generalized Orlicz ball, $(a_{i})_{i=1}^{n}$ be a sequence of reals and $(X_{i})_{i=1}^{n}$ be the coordinates of the random vector uniformly distributed on $K$. Then for any even positive integer $p$ we have ${\mathbb{E}}\Big{(}\sum_{i=1}^{n}a_{i}X_{i}\Big{)}^{p}\leq{\mathbb{E}}\Big{(}\sum_{i=1}^{n}a_{i}X_{i}^{\star}\Big{)}^{p},$ with $X_{i}^{\star}$ defined as before. ###### Proof. When we open the brackets in $(\sum a_{i}X_{i})^{p}$ the summands in which at least one $X_{i}$ appears with an odd exponent average out to zero, as $K$ is 1-symmetric. Thus what is left is a sum of elements of the form $(a_{i}X_{1})^{2\alpha_{1}}(a_{2}X_{2})^{2\alpha_{2}}\ldots(a_{n}X_{n})^{2\alpha_{n}}=|a_{1}X_{1}|^{2\alpha_{1}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}.$ If we put $f(a_{1}x_{1})=(a_{1}x_{1})^{2\alpha_{1}}$ and $g(a_{2}x_{2},\ldots,a_{n}x_{n})=(a_{2}x_{2})^{2\alpha_{2}}\cdot\ldots\cdot(a_{n}x_{n})^{2\alpha_{n}}$, applying negative association we get $\displaystyle{\mathbb{E}}|a_{1}X_{1}|^{2\alpha_{1}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}$ $\displaystyle\leq{\mathbb{E}}|a_{1}X_{1}|^{2\alpha_{1}}{\mathbb{E}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}$ $\displaystyle={\mathbb{E}}|a_{1}X_{1}^{\star}|^{2\alpha_{1}}{\mathbb{E}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}$ $\displaystyle={\mathbb{E}}|a_{1}X_{1}^{\star}|^{2\alpha_{1}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}.$ Repeating this process inductively we separate all the variables and get $\displaystyle{\mathbb{E}}\Big{(}\sum_{i=1}^{n}a_{i}X_{i}\Big{)}^{p}$ $\displaystyle=\sum_{\alpha_{1}+\ldots+\alpha_{n}=p/2}C_{\alpha_{1},\ldots,\alpha_{n}}{\mathbb{E}}|a_{1}X_{1}|^{2\alpha_{1}}|a_{2}X_{2}|^{2\alpha_{2}}\ldots|a_{n}X_{n}|^{2\alpha_{n}}\leq$ $\displaystyle\leq\sum_{\alpha_{1}+\ldots+\alpha_{n}=p/2}C_{\alpha_{1},\ldots,\alpha_{n}}{\mathbb{E}}|a_{1}X_{1}^{\star}|^{2\alpha_{1}}|a_{2}X_{2}^{\star}|^{2\alpha_{2}}\ldots|a_{n}X_{n}^{\star}|^{2\alpha_{n}}=$ $\displaystyle={\mathbb{E}}\Big{(}\sum_{i=1}^{n}a_{i}X_{i}^{\star}\Big{)}^{p}.$ ∎ Finally, we can apply Shao’s maximal inequality to get a exponential concentration of the euclidean norm. Theorem 3 in [S00] states: ###### Theorem 1.6. Let $(X_{i})_{i=1}^{n}$ be a sequence of negatively associated random variables with zero means and finite second moments. Let $S_{k}=\sum_{i=1}^{k}X_{i}$ and $B_{n}=\sum_{i=1}^{n}{\mathbb{E}}X_{i}^{2}$. Then for all $x>0$, $a>0$ and $0<\alpha<1$ ${\mathbb{P}}\Big{(}\max_{1\leq k\leq n}|S_{k}|\geq x\Big{)}\leq 2{\mathbb{P}}(\max_{1\leq k\leq n}|X_{k}|>a)+\frac{2}{1-\alpha}\exp\Bigg{(}-\frac{x^{2}\alpha}{2(ax+B_{n})}\cdot\Big{(}1+\frac{2}{3}\ln\Big{(}1+\frac{ax}{B_{n}}\Big{)}\Big{)}\Bigg{)}.$ We say $K\subset{\mathbb{R}}^{n}$ is in isotropic position if $\lambda_{n}(K)=1$ and ${\mathbb{E}}X_{i}^{2}=L_{K}^{2}$ for some constant $L_{K}$ (any bounded convex set with a non-empty interior can be moved into isotropic position by an affine transformation, for more on this subject see e.g. [MS86]). Notice that if $|X_{i}|$ are negatively associated and $f_{i}$ are increasing, then $f_{i}(|X_{i}|)$ are also negatively associated. Thus the sequence $(X_{i}^{2}-L_{K}^{2})_{i=1}^{n}$ for $X=(X_{i})_{i=1}^{n}$ uniformly distributed on a generalized Orlicz ball is also negatively associated. The moments of log-concave variables are comparable (see for instance [KLO96], Section 2, remark 5), thus we have ${\mathbb{E}}(X_{i}^{2}-L_{K}^{2})^{2}={\mathbb{E}}X_{i}^{4}+L_{K}^{4}-2L_{K}^{2}{\mathbb{E}}X_{i}^{2}={\mathbb{E}}X_{i}^{4}-L_{K}^{4}\leq 5L_{K}^{4}.$ If we put $\alpha=1/2$ and $x=nt$ in Shao’s inequality and apply the bound we got above for the variance we get ###### Corollary 1.7. Let $K\subset{\mathbb{R}}^{n}$ be a generalized Orlicz ball in isotropic position, and $(X_{i})_{i=1}^{n}$ be the coordinates of the random vector uniformly distributed on $K$. Then for any $t>0$, $a>0$ we have: $\displaystyle{\mathbb{P}}\Big{(}\max_{1\leq k\leq n}\Big{|}\sum_{i=1}^{k}(X_{i}^{2}-L_{K}^{2})\Big{|}>nt\Big{)}$ $\displaystyle\leq 2{\mathbb{P}}\Big{(}\max_{1\leq k\leq n}|X_{k}^{2}-L_{K}^{2}|>a\Big{)}+$ $\displaystyle+4\exp\Bigg{(}-\frac{nt^{2}}{4(at+5L_{K}^{4})}\cdot\bigg{(}1+\frac{2}{3}\ln\Big{(}1+\frac{at}{5L_{K}^{4}}\Big{)}\bigg{)}\Bigg{)}.$ To apply this result probably an idea on what order of convergence is possible to achieve with this formula would be needed. To this end we give the following corollary: ###### Corollary 1.8. Let $K\subset{\mathbb{R}}^{n}$ be a generalized Orlicz ball in isotropic position, and $(X_{i})_{i=1}^{n}$ be the coordinates of the random vector uniformly distributed on $K$. Then for any $t>0$ we have: ${\mathbb{P}}\Big{(}\Big{|}\frac{\sum_{i=1}^{n}X_{i}^{2}}{n}-L_{K}^{2}\Big{|}>t\Big{)}\leq Ce^{-cnt^{2}}+Cne^{-c\sqrt[3]{nt}},$ where $C$ and $c$ are universal constants independent of $t$, $n$ and $K$. For $t>t_{0}$ a better bound (of the order of $e^{-t\sqrt{n}}$) is due to Bobkov and Nazarov (see [BN03]). However, frequently a bound for $t{\rightarrow}0$ is needed — for instance the proof of the Central Limit Theorem for convex bodies uses bounds for the concentration of the second norm for small $t$ (see for instance [ABP03]). In full generality (ie. for an arbitrary log-concave isotropic measure and for arbitrary $t$) such a result is given in a very recent paper by Klartag (see [K07]) with worse exponents — the bound for the probabilty is of the order of $e^{t^{3.33}n^{0.33}}$. Previous proofs of such results (see [FGP07], [K07,2]) gave a logarithmic dependence of the exponent on $n$. The bound given in the corollary above is very rough, and in any particular case it is very likely it may be improved. However, we give it in order to show an explicit exponential bound in the concentration inequality which is uniform for all generalized Orlicz balls in a given dimension and applies for any $t>0$. ###### Proof. Obviously ${\mathbb{P}}\Bigg{(}\Bigg{|}\frac{\sum_{i=1}^{n}X_{i}^{2}}{n}-L_{K}^{2}\Bigg{|}>t\Bigg{)}\leq{\mathbb{P}}\Bigg{(}\max_{1\leq k\leq n}\Bigg{|}\sum_{i=1}^{k}(X_{i}^{2}-L_{K}^{2})\Bigg{|}>nt\Bigg{)},$ so we have only to bound the right hand side in Corollary 1.7. Put $a=\sqrt[3]{n^{2}t^{2}}$. We know (see [MP89]) that $L_{K}^{2}$ is bounded by some universal constant $L$ independent of $n$ and $K$ for any 1-symmetric body in ${\mathbb{R}}^{n}$. If $c$ is small enough and $C$ large enough, then for $a<L_{K}^{2}$ we have $Cne^{-c\sqrt[3]{nt}}=Cne^{-c\sqrt{a}}\geq 1.$ Thus we may consider only the case $a>L_{K}^{2}$. In this case $\displaystyle{\mathbb{P}}(\max_{1\leq k\leq n}|X_{k}^{2}-L_{K}^{2}|>a)$ $\displaystyle\leq n\max_{1\leq k\leq n}{\mathbb{P}}(|X_{k}^{2}-L_{K}^{2}|>a)=n\max_{k}{\mathbb{P}}(X_{k}^{2}>a+L_{K}^{2})$ $\displaystyle\leq n\max_{k}{\mathbb{P}}(X_{k}^{2}>a)=n\max_{k}{\mathbb{P}}(|X_{k}|>\sqrt{a}).$ Due to the Brunn-Minkowski inequality $X_{k}$ is log-concave (see for instance [Ga02]), we know that $Var(X_{k})\leq L_{K}^{2}<C$ and ${\mathbb{E}}X_{k}=0$, and thus $P(|X_{k}|>t)\leq c_{1}e^{-c_{2}t}$ for some universal constants $c_{1}$ and $c_{2}$ independent of the distribution of $X_{k}$ and of $t$ (Borell’s Lemma, see for instance [MS86]). Thus we get ${\mathbb{P}}\Big{(}\max_{1\leq k\leq n}|X_{k}^{2}-L_{K}^{2}|>a\Big{)}\leq c_{1}e^{-c_{2}\sqrt{a}}=c_{1}e^{-c_{2}\sqrt[3]{nt}}.$ In the second part we shall simply bound $\Bigg{(}1+\frac{2}{3}\ln\Big{(}1+\frac{at}{5L_{K}^{4}}\Big{)}\Bigg{)}\geq 1.$ Then $4\exp\Bigg{(}-\frac{nt^{2}}{4(at+5L_{K}^{4})}\cdot\bigg{(}1+\frac{2}{3}\ln\Big{(}1+\frac{at}{5L_{K}^{4}}\Big{)}\bigg{)}\Bigg{)}\leq 4\exp\big{(}-\frac{nt^{2}}{4n^{2/3}t^{5/3}+20L_{K}^{4}}\big{)}\leq Ce^{-c\sqrt[3]{nt}}+Ce^{-cnt^{2}}.$ ∎ ### 1.4 Acknowledgements We would very much like to thank Rafał Latała, who encouraged us to write the paper, was the first person to read it and check the reasoning, and helped improve the paper in innumerable aspects. He also taught us most of what we know in the subject. We would also like to thank prof. Stanisław Kwapień, who first suggested to us the idea of searching for negative-association type properties for convex bodies. ## 2 Easy facts ### 2.1 Simplifying We want to prove inequality (1.1.1) for various classes of functions (coordinate-wise increasing in the case of Theorem 1.2 and radius-wise increasing in the case of Theorem 1.3). We may assume $k+l=n$ by putting $\tilde{g}(x_{j_{1}},\ldots,x_{j_{l}},x_{r_{1}},\ldots,x_{r_{n-l-k}})=g(x_{j_{1}},\ldots,x_{j_{l}})$. For convienience we shall assume that the Lebesgue volume of $K_{+}$ is 1 (inequality (1.1.1) is invariant under homothety). It will be more convienient to work with c-sets or radius-sets than with functions, which motivates the following Lemma: ###### Lemma 2.1. Let $\mu$ be any probability measure on ${\mathbb{R}}_{+}^{n}$ and let $X=(X_{1},X_{2},\ldots,X_{n})$ be the random vector distributed according to $\mu$. Assume that for given $0\leq k,l\leq n$ we have two families of bounded functions $\mathcal{F}$ on ${\mathbb{R}}_{+}^{k}$ and $\mathcal{G}$ on ${\mathbb{R}}_{+}^{l}$. Let $\mathcal{A}=\\{f^{-1}(-\infty,t]:f\in\mathcal{F},t\in{\mathbb{R}}\\}$, and similarly $\mathcal{B}$ for $\mathcal{G}$. If for any $A\in\mathcal{A}$ and $B\in\mathcal{B}$ we have $\mu(A\times B)\mu({\bar{A}}\times{\bar{B}})\leq\mu(A\times{\bar{B}})\mu({\bar{A}}\times B),$ (2.1.1) then inequality (1.1.1) holds for $X$ and any $f\in\mathcal{F},g\in\mathcal{G}$. In particular, if inequality (2.1.1) holds for any $k$ and for any c-sets $A,B$, then the random variables $X_{1},X_{2},\ldots,X_{n}$ are negatively associated. ###### Proof. Let us take any two functions $\mathcal{F}\ni f:{\mathbb{R}}_{+}^{k}{\rightarrow}{\mathbb{R}}$ and $\mathcal{G}\ni g:{\mathbb{R}}_{+}^{l}{\rightarrow}{\mathbb{R}}$. As covariance is bilinear and is 0 if one of the functions is constant, we may assume without loss of generality that $f$ and $g$ are non-negative. For non-negative functions we have $f(x)=\int_{0}^{\infty}{\mathbf{1}}_{f^{-1}[t,\infty)}(x)\ dt.$ Thus (again, by the bilinearity of the covariance) we can restrict ourselves to functions $f$ and $g$ of the form $1-{\mathbf{1}}_{A}$ and $1-{\mathbf{1}}_{B}$, where $A\in\mathcal{A}$ and $B\in\mathcal{B}$. Since ${\rm Cov}(1-{\mathbf{1}}_{A},1-{\mathbf{1}}_{B})={\rm Cov}({\mathbf{1}}_{A},{\mathbf{1}}_{B})$, we have to prove that ${\rm Cov}({\mathbf{1}}_{A},{\mathbf{1}}_{B})\leq 0$. Let us denote by $\mathbf{X}$ the $k$-dimensional vector $(X_{i_{1}},\ldots,X_{i_{k}})$ on which $f$ is taken, and by $\mathbf{Y}$ the $l$-dimensional vector on which $g$ is taken. Then $\displaystyle{\rm Cov}\big{(}{\mathbf{1}}_{A}({\mathbf{X}}),{\mathbf{1}}_{B}({\mathbf{Y}})\big{)}$ $\displaystyle={\mathbb{E}}{\mathbf{1}}_{A}({\mathbf{X}}){\mathbf{1}}_{B}({\mathbf{Y}})-{\mathbb{E}}{\mathbf{1}}_{A}({\mathbf{X}}){\mathbb{E}}{\mathbf{1}}_{B}({\mathbf{Y}})=\mu(A\times B)-\mu(A\times{\mathbb{R}}^{l})\mu({\mathbb{R}}^{k}\times B)$ $\displaystyle=\mu(A\times B)\mu\big{(}(A\cup{\bar{A}})\times(B\cup{\bar{B}})\big{)}-\mu\big{(}A\times(B\cup{\bar{B}})\big{)}\mu\big{(}(A\cup{\bar{A}})\times B\big{)}$ $\displaystyle=\mu(A\times B)\mu({\bar{A}}\times{\bar{B}})-\mu(A\times{\bar{B}})\mu({\bar{A}}\times B),$ which is non-positive by (2.1.1). ∎ ### 2.2 Simple proportion lemmas During the course of further proofs we shall frequently need to compare two ratios of integrals of the same functions over different sets. In this subsection we will demonstrate some simple properties of ratios of integrals. ###### Fact 2.2. Let $a,b\geq 0$ and $c,d>0$. Then the following are equivalent: * • $\frac{a}{c}\geq\frac{b}{d}$, * • $\frac{a}{c}\geq\frac{a+b}{c+d}$, * • $\frac{a+b}{c+d}\geq\frac{b}{d}$. Whenever there is equality in one of the inequalities, all aforementioned fractions are equal. ###### Lemma 2.3. Let $\mu$ be a non-negative measure on ${\mathbb{R}}$ supported on the (possibly unbounded) interval $[l_{\mu},r_{\mu}]$. Suppose that $f,g,h:{\mathbb{R}}{\rightarrow}{\mathbb{R}}_{+}$ are functions bounded on ${\rm supp}\mu$, positive on the interior of their supports, satisfying: 1. 1. The support of any function $u\in\\{f,g,h\\}$ is an interval $[l_{u},r_{u}]$ (possibly unbounded), 2. 2. $\frac{f}{g}$ is a decreasing function where defined, and $r_{f}\leq r_{g}$, 3. 3. $h$ is an increasing function, Then: 1. (1a) For any $a<b<c$, $b\in(l_{\mu},r_{\mu})\cap(l_{g},r_{g})$ we have $\frac{\int_{a}^{b}f(x)d\mu}{\int_{a}^{b}g(x)d\mu}\geq\frac{f(b)}{g(b)}\hbox{ and }\frac{f(b)}{g(b)}\geq\frac{\int_{b}^{c}f(x)d\mu}{\int_{b}^{c}g(x)d\mu}$ whenever both sides of an inequality are defined. 2. (1b) Moreover, if for some $a<b<c$ we have two equalities in inequality (1a) then $\frac{f(x)}{g(x)}$ is constant on $(a,c)\cap{\rm supp}g\cap{\rm supp}\mu$ and for any $a\leq s<t\leq c$ $\frac{\int_{s}^{t}f(x)d\mu}{\int_{s}^{t}g(x)d\mu}$ is equal to $f(b)/g(b)$ if defined. 3. (2a) For any points $a,b,c,d$ satisfying $a<b\leq d$ and $a\leq c<d$ we have: $\frac{\int_{a}^{b}f(x)d\mu}{\int_{a}^{b}g(x)d\mu}\geq\frac{\int_{c}^{d}f(x)d\mu}{\int_{c}^{d}g(x)d\mu}$ whenever both sides are defined. 4. (2b) Moreover, if this inequality is an equality and either $\int_{a}^{c}g(x)d\mu(x)$ or $\int_{b}^{d}g(x)d\mu(x)$ is strictly positive, then $\frac{f}{g}$ is constant on $[a,d]$ where defined, and we have an equality for any $a\leq a^{\prime}\leq b^{\prime}\leq d^{\prime}\leq d$ and $c^{\prime}\in[a^{\prime},d^{\prime}]$ if both sides are defined. 5. (3) If $l_{g}=l_{f}$ the following inequality occurs for any interval $I$: $\frac{\int_{I}f(x)d\mu(x)}{\int_{I}g(x)d\mu(x)}\geq\frac{\int_{I}f(x)h(x)d\mu(x)}{\int_{I}g(x)h(x)d\mu(x)}$ if both sides are defined. ###### Proof. 1. (1a) Consider the first inequality. Let $a^{\prime}=\max\\{l_{\mu},l_{g},a\\}$ . We have $a\leq a^{\prime}<b$ (otherwise the denominator of the left-hand side would be undefined). Also $\int_{a}^{b}g(x)d\mu(x)=\int_{a^{\prime}}^{b}g(x)d\mu(x)>0$ and $g>0$ on $(a^{\prime},b]$ (it has to be positive in $b$ or the right-hand side would be undefined). Thus $\frac{\int_{a}^{b}f(x)d\mu(x)}{\int_{a}^{b}g(x)d\mu(x)}\geq\frac{\int_{a^{\prime}}^{b}f(x)}{\int_{a^{\prime}}^{b}g(x)}=\frac{\int_{a^{\prime}}^{b}g(x)\frac{f(x)}{g(x)}}{\int_{a^{\prime}}^{b}g(x)}\geq\frac{\int_{a^{\prime}}^{b}g(x)\frac{f(b)}{g(b)}}{\int_{a^{\prime}}^{b}g(x)}=\frac{f(b)}{g(b)},$ A similar reasoning with $c^{\prime}=\min\\{r_{\mu},r_{g},c\\}$ proves the second inequality (note $r_{f}\leq r_{g}$, so the first inequality in the reasoning above becomes an equality). 2. (1b) If equality occurs, then $\frac{f(x)}{g(x)}=\frac{f(b)}{g(b)}$ for almost all $x\in(a^{\prime},c^{\prime})$ as $g$ is strictly positive on $(a^{\prime},c^{\prime})$. As $\frac{f}{g}$ is decreasing, if it is constant on almost whole $(a^{\prime},c^{\prime})$, it is constant on the whole interval and thus $\frac{\int_{s}^{t}f(x)d\mu(x)}{\int_{s}^{t}g(s)d\mu(x)}=\frac{f(b)}{g(b)}$ if defined for any $s,t\in(a^{\prime},c^{\prime})$. We know $\int_{a}^{a^{\prime}}g(x)d\mu(x)=\int_{c^{\prime}}^{c}g(x)d\mu(x)=0$, so to have equalities we also have to have $\int_{a}^{a^{\prime}}f(x)d\mu(x)=\int_{c^{\prime}}^{c}f(x)d\mu(x)=0$, thus $\int_{s}^{t}f(x)d\mu(x)=\int_{(s,t)\cap(a^{\prime},c^{\prime})}f(x)d\mu(x)$ and similarly for $g$, thus the thesis. 3. (2a) Let $F(x,y)=\int_{x}^{y}f(t)$ and $G(x,y)=\int_{x}^{y}g(t)$. As the left-hand side is defined, $G(a,b)>0$ and thus $G(a,d)>0$. We apply (1a) to get: $\frac{F(a,b)}{G(a,b)}\geq\frac{F(b,d)}{G(b,d)}$ (2.2.1) if the right-hand side is defined and from Fact 2.2 we have $\frac{F(a,b)}{G(a,b)}\geq\frac{F(a,b)+F(b,d)}{G(a,b)+G(b,d)}=\frac{F(a,d)}{G(a,d)}.$ If the right-hand side in (2.2.1) was not defined, $G(b,d)=0$ and thus $F(b,d)=0$ as $r_{f}\leq r_{g}$, so $\frac{F(a,b)}{G(a,b)}\geq\frac{F(a,d)}{G(a,d)}$. Similarly from (1a) $\frac{F(a,c)}{G(a,c)}\geq\frac{F(c,d)}{G(c,d)}$ if the left-hand side is defined, and thus from Fact 2.2 $\frac{F(a,d)}{G(a,d)}\geq\frac{F(c,d)}{G(c,d)}.$ If the left-hand side was undefined, $G(a,d)=G(c,d)$ and obviously $F(a,d)\geq F(c,d)$, so we get the same inequality. Linking the two inequalities we get the thesis. 4. (2b) Suppose $G(b,d)>0$. As $\frac{F(a,b)}{G(a,b)}\geq\frac{F(a,d)}{G(a,d)}\geq\frac{F(c,d)}{G(c,d)}$ and the first and last expressions are equal, all inequalities are in fact equalities. Thus from the first one of them and Fact 2.2 we get $\frac{F(a,b)}{G(a,b)}=\frac{F(b,d)}{G(b,d)},$ and applying (1b) we get the thesis. 5. (3) Let $I^{\prime}=I\cap{\rm supp}g$. As ${\rm supp}f\subset{\rm supp}g$ all integrals in the thesis over $I$ are equal to the appropriate integrals over $I^{\prime}$. Consider the functions $h$ and $\frac{f}{g}$ on the interval ${\rm Int}I^{\prime}$ (note $\frac{f}{g}$ is defined on ${\rm Int}I^{\prime}$) taken with a measure with density $\frac{g(x)}{\int_{I^{\prime}}g(t)d\mu(t)}d\mu$ (this is defined as the left- hand side in the thesis was defined, so $\int_{I^{\prime}}g(t)d\mu(t)>0$). From the continuous Chebyshev sum inequality (that is, if $F$ is increasing and $G$ is decreasing, then $\int F\int G\geq\int FG\int 1$) we know $\displaystyle\int_{I^{\prime}}h(x)\frac{g(x)}{\int_{I^{\prime}}g(t)d\mu(t)}d\mu(x)$ $\displaystyle\int_{I^{\prime}}\frac{f(x)}{g(x)}\frac{g(x)}{\int_{I^{\prime}}g(t)d\mu(t)}d\mu(x)$ $\displaystyle\geq$ $\displaystyle\int_{I^{\prime}}h(x)\frac{f(x)}{g(x)}\frac{g(x)}{\int_{I^{\prime}}g(t)d\mu(t)}d\mu(x)\int_{I^{\prime}}\frac{g(x)}{\int_{I^{\prime}}g(t)d\mu(t)}d\mu(x).$ Multiplying both sides by $[\int_{I^{\prime}}g(t)d\mu(t)]^{2}$ we get the thesis. ∎ ###### Lemma 2.4. Let $\mu$ be a non-negative measure on $I\subset{\mathbb{R}}$. Suppose $f,g,p,q:I{\rightarrow}{\mathbb{R}}_{+}$ are functions satisfying $f(x)g(y)\geq f(y)g(x)$ for $x\geq y$ and $p(x)q(y)\leq p(y)q(x)$ for $x\geq y$. Then $\int_{I}p(x)f(x)d\mu(x)\int_{I}q(x)g(x)d\mu(x)\leq\int_{I}p(x)g(x)d\mu(x)\int_{I}q(x)f(x)d\mu(x).$ ###### Proof. Using Fubini’s theorem we have to prove $\int_{I}\int_{I}p(x)f(x)q(y)g(y)\ d\mu(y)\ d\mu(x)\leq\int_{I}\int_{I}p(y)f(x)q(x)g(y)\ d\mu(y)\ d\mu(x).$ Multiplying sides by two and changing names $x$ and $y$: $\int_{I}\int_{I}\big{[}p(x)f(x)q(y)g(y)+p(y)f(y)q(x)g(x)-p(x)f(y)q(y)g(x)-p(y)f(x)q(x)g(y)\big{]}\ d\mu(y)\ d\mu(x)\leq 0$ $\int_{I}\int_{I}\big{(}p(x)q(y)-p(y)q(x)\big{)}\big{(}f(x)g(y)-f(y)g(x)\big{)}\ d\mu(y)\ d\mu(x)\leq 0,$ which follows from the assumptions, as the integrand is always non-positive. ∎ ###### Lemma 2.5. Suppose $f,g:X{\rightarrow}{\mathbb{R}}_{+}$ are defined on any set $X$ with a measure $\mu$. Let $\\{D_{i}\\}_{i\in I}$ be a family of disjoint subsets of $X$. If $t\int_{D_{i}}g(x)d\mu(x)\geq\int_{D_{i}}f(x)d\mu(x)\geq s\int_{D_{i}}g(x)d\mu(x)$ for some $t,s\in{\mathbb{R}}\cup\\{-\infty,\infty\\}$, then $t\int_{\bigcup_{i}D_{i}}g(x)d\mu(x)\geq\int_{\bigcup_{i}D_{i}}f(x)d\mu(x)\geq s\int_{\bigcup_{i}D_{i}}g(x)d\mu(x).$ If $X=X_{1}\times X_{2}$ and $\mu=\mu_{1}\otimes\mu_{2}$, and for some set $D\subset X_{1}\times X_{2}$ and any $x_{1}\in X_{1}$ we have $t\int_{(\\{x_{1}\\}\times X_{2})\cap D}g(x)d\mu_{2}(x)\geq\int_{(\\{x_{1}\\}\times X_{2})\cap D}f(x)d\mu_{2}(x)\geq s\int_{(\\{x_{1}\\}\times X_{2})\cap D}g(x)d\mu_{2}(x),$ then $t\int_{D}g(x)d\mu(x)\geq\int_{D}f(x)d\mu(x)\geq s\int_{D}g(x)d\mu(x).$ ###### Proof. In the first case, we should add all the inequalities by sides. In the second case, we should not sum but integrate using Fubini’s theorem. ∎ ## 3 The $\ell_{p}^{n}$ ball case First we shall give the proof for $\ell_{p}^{n}$ balls. Recall the $\ell_{p}^{n}$ ball is the generalized Orlicz ball defined by the Young functions $f_{i}(x)=|x|^{p}$. We include this case for two reasons: first, it is much simpler than the Orlicz ball case, and serves as a good illustration of what is happening, and second, because we are able to achieve a stronger result, namely prove Theorem 1.3. Note that in particular we can take $m$ to be $c_{r}{\mathbf{1}}_{[0,r]}$ to get the result for the uniform measure on the $\ell_{p}^{n}$ ball. As any coordinate-wise increasing function is radius-wise increasing, this result is stronger than the negative associacion property we prove for generalized Orlicz balls. By a simple approximation argument we can also get the result above for $\mu$ being the cone measure on the surface of $\ell_{p}^{n}$. ###### Proof. Let $B_{p}^{n}$ denote the $\ell_{p}^{n}$ ball. Let $M(x_{1},x_{2},\ldots,x_{n})=(|x_{1}|,|x_{2}|,\ldots,|x_{n}|)$ and let $\tilde{\mu}$ be defined by $\tilde{\mu}(A)=\mu(M^{-1}(A))$. Notice $\tilde{\mu}$ describes the distribution of $(|X_{1}|,|X_{2}|,\ldots,|X_{n}|)$. As $\mu$ is 1-symmetric, we may equivalently define $\tilde{\mu}$ as $2^{n}$ times the restriction of $\mu$ to ${\mathbb{R}}_{+}^{n}$. Recall that the cone measure on $\partial B_{p}^{n}$ (that is, the boundary of $B_{p}^{n}$), which we shall denote $\nu$, is defined for $A\subset\partial B_{p}^{n}$ by $\nu_{n}(A)=\frac{\lambda_{n}(ta:t\in{\mathbb{R}},a\in A,ta\in B_{p}^{n})}{\lambda_{n}(B_{p}^{n})}.$ For this measure we have the polar integration formula: $\int_{{\mathbb{R}}^{n}}f(x)dx=n\lambda_{n}(B_{p}^{n})\int_{R_{+}}r^{n-1}\int_{\partial B_{p}^{n}}f(r\theta)d\nu_{n}(\theta)dr.$ Let $C_{n}=n\lambda_{n}(B_{p}^{n})$. Due to Lemma 2.1 we only need to prove inequality $\tilde{\mu}(A\times B)\tilde{\mu}({\bar{A}}\times{\bar{B}})\leq\tilde{\mu}(A\times{\bar{B}})\tilde{\mu}({\bar{A}}\times B)$ for any radius-sets $A,B$, which is equivalent to $\mu(A\times B)\mu({\bar{A}}\times{\bar{B}})\leq\mu(A\times{\bar{B}})\mu({\bar{A}}\times B)$. We have: $\displaystyle\mu(A\times B)$ $\displaystyle=\int_{{\mathbb{R}}^{k}}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{A}(x){\mathbf{1}}_{B}(y)m(\|x\|_{p}^{p}+\|y\|_{p}^{p})dxdy=$ $\displaystyle=\int_{{\mathbb{R}}_{+}}\int_{\partial B_{p}^{k}}\int_{{\mathbb{R}}^{n-k}}C_{k}r^{k-1}{\mathbf{1}}_{A}(r\theta){\mathbf{1}}_{B}(y)m(r^{p}+\|y\|_{p}^{p})d\nu_{k}(\theta)drdy$ $\displaystyle=\int_{{\mathbb{R}}_{+}}\bigg{[}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{B}(y)m(r^{p}+\|y\|_{p}^{p})dy\bigg{]}\bigg{[}\int_{\partial B_{p}^{k}}{\mathbf{1}}_{A}(r\theta)d\nu_{k}(\theta)\bigg{]}C_{k}r^{k-1}dr.$ Denote $f_{B}(r)=\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{B}(y)m(r^{p}+\|y\|_{p}^{p})dy$ and $g_{A}(r)=\int_{\partial B_{p}^{k}}{\mathbf{1}}_{A}(r\theta)d\nu_{k}(\theta)$. Let $\sigma_{1}$ be the measure on ${\mathbb{R}}_{+}$ with density $C_{k}r^{k-1}$. We can perform similar operations for the other three expressions in inequality (2.1.1). What we have to prove becomes the inequality $\int_{{\mathbb{R}}_{+}}f_{B}(r)g_{A}(r)d\sigma_{1}(r)\int_{{\mathbb{R}}_{+}}f_{{\bar{B}}}(r)g_{{\bar{A}}}(r)d\sigma_{1}(r)\leq\int_{{\mathbb{R}}_{+}}f_{{\bar{B}}}(r)g_{A}(r)d\sigma_{1}(r)\int_{{\mathbb{R}}_{+}}f_{B}(r)g_{{\bar{A}}}(r)d\sigma_{1}(r).$ Due to lemma 2.4 it is enough to prove the following two inequalities: $\displaystyle f_{B}(r_{1})f_{{\bar{B}}}(r_{2})\geq f_{B}(r_{2})f_{{\bar{B}}}(r_{1})\hbox{ for }r_{1}\geq r_{2},$ (3.0.1) $\displaystyle g_{A}(r_{1})g_{{\bar{A}}}(r_{2})\leq g_{A}(r_{2})g_{{\bar{A}}}(r_{1})\hbox{ for }r_{1}\geq r_{2}.$ (3.0.2) Inequality (3.0.2) is simple — ${\mathbf{1}}_{A}(r\theta)$ is decreasing as a function of $r$ for any fixed $\theta$, while ${\mathbf{1}}_{{\bar{A}}}(r\theta)$ is increasing, as $A$ is a radius-set. Thus $g_{A}(r)$ is decreasing, $g_{{\bar{A}}}$ is increasing, so $g_{A}(r_{1})\leq g_{A}(r_{2})$ and $g_{{\bar{A}}}(r_{2})\leq g_{{\bar{A}}}(r_{1})$. Inequality (3.0.1) will require a bit more work. We have: $\displaystyle f_{B}(r_{1})$ $\displaystyle=\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{B}(y)m(r_{1}^{p}+\|y\|_{p}^{p})dy$ $\displaystyle=\int_{{\mathbb{R}}_{+}}\int_{\partial B_{p}^{n-k}}C_{n-k}s^{n-k-1}{\mathbf{1}}_{B}(s\xi)m(r_{1}^{p}+s^{p})d\nu_{n-k}(\xi)dr$ $\displaystyle=\int_{{\mathbb{R}}_{+}}\bigg{[}m(r_{1}^{p}+s^{p})\bigg{]}\bigg{[}\int_{\partial B_{p}^{n-k}}{\mathbf{1}}_{B}(s\xi)d\nu_{n-k}(\xi)\bigg{]}C_{n-k}s^{n-k-1}ds.$ We are going to use Lemma 2.4 once again. Let $p_{r_{1}}(s)=m(r_{1}^{p}+s^{p})$ and $q_{B}(s)=\int_{\partial B_{p}^{n-k}}{\mathbf{1}}_{B}(s\xi)d\nu_{n-k}(\xi)$ and $\sigma_{2}$ the measure with density $C_{n-k}s^{n-k-1}$. We do the similar calculation for the other three expressions in inequality (3.0.1), and it becomes $\int_{{\mathbb{R}}_{+}}p_{r_{1}}(s)q_{B}(s)d\sigma_{2}(s)\int_{{\mathbb{R}}_{+}}p_{r_{2}}(s)q_{{\bar{B}}}(s)d\sigma_{2}(s)\geq\int_{{\mathbb{R}}_{+}}p_{r_{2}}(s)q_{B}(s)d\sigma_{2}(s)\int_{{\mathbb{R}}_{+}}p_{r_{1}}(s)q_{{\bar{B}}}(s)d\sigma_{2}(s).$ Applying Lemma 2.4 we have to prove $\displaystyle p_{r_{1}}(s_{1})p_{r_{2}}(s_{2})\leq p_{r_{2}}(s_{1})p_{r_{1}}(s_{2})\hbox{ for }s_{1}\geq s_{2},$ (3.0.3) $\displaystyle q_{B}(s_{1})q_{{\bar{B}}}(s_{2})\leq q_{{\bar{B}}}(s_{1})q_{B}(s_{2})\hbox{ for }s_{1}\geq s_{2}.$ (3.0.4) Inequality (3.0.4) is proved in the same way as inequality (3.0.2) — $q_{B}$ is decreasing and $q_{{\bar{B}}}$ is increasing. Inequality (3.0.3) means $m(r_{1}^{p}+s_{1}^{p})m(r_{2}^{p}+s_{2}^{p})\leq m(r_{2}^{p}+s_{1}^{p})m(r_{1}^{p}+s_{2}^{p}),$ which follows from the log-concavity of $m$. ∎ As we saw, this proof was quite simple. Unfortunately, it takes advantage of the fact that the Young function of the $\ell_{p}^{n}$ ball scales well with the radius, that is, that $f_{i}(tx_{i})=\phi(t)f_{i}(x_{i})$ for some function $\phi$. Of all Orlicz ball only the $\ell_{p}$ balls have this property, which makes it impossible to apply the same proof to the generalized Orlicz ball case. ## 4 The generalized Orlicz ball case — preliminaries, the proper measure, lens sets ### 4.1 Idea of the proof We would like to transfer the result given above for $\ell_{p}^{n}$ balls to the more general case of generalized Orlicz balls. In the generalized Orlicz ball cas the Young function does not, unfortunately, scale with the radius, and this creates the need for a different approach. Again by Lemma 2.1 we can restrict ourselves to characteristic functions of c-sets. As generalized Orlicz balls are 1-symmetric, we can restrict ourselves to the positive quadrant of our generalized Orlicz ball. We shall proceed in two steps. The first will be to prove that generalized Orlicz balls satisfy inequality (1.1.1) if one of the functions, say $g$, is univariate — in other words, to begin by proving weak negative association. This is equivalent to proving 2.1.1 for one of the sets, say $B$, being one- dimensional. Due to Lemma 2.3, part 1, we will simply need to prove that the function $\frac{\lambda_{n-1}(A\times\\{z\\}\cap K)}{\lambda_{n-1}({\bar{A}}\times\\{z\\}\cap K)}$ is decreasing with $z$. Thus, we take any $z_{2}>z_{1}\geq 0$ and concentrate on them. We want to prove $\frac{\lambda_{n-1}(A\times\\{z_{1}\\}\cap K)}{\lambda_{n-1}({\bar{A}}\times\\{z_{1}\\}\cap K)}\leq\frac{\lambda_{n-1}(A\times\\{z_{2}\\}\cap K)}{\lambda_{n-1}({\bar{A}}\times\\{z_{2}\\}\cap K)}.$ Switching the right denominator with the left numerator we get $\frac{\lambda_{n-1}((\\{z_{2}\\}\times A)\cap K)}{\lambda_{n-1}((\\{z_{1}\\}\times A)\cap K)}\geq\frac{\lambda_{n-1}((\\{z_{2}\\}\times{\bar{A}})\cap K)}{\lambda_{n-1}((\\{z_{1}\\}\times{\bar{A}})\cap K)}$ as the inequality we need to prove. We shall denote the proportion of the measure of $K_{z_{2}}$ to the measure of $K_{z_{1}}$ on a given set $D$ by $\theta(D)$. The second step will be to pass from the univariate case to the general case. It turns out that a very similar argument, using the proportion $\frac{\lambda(D\cap{\bar{B}})}{\lambda(D\cap B)}$ as $\theta(D)$ will allow us to do that. Thus, to avoid repetition (as the argument is quite long), we shall take the properties of both of these functions which make the similar arguments possible and call any function with such properties a $\Theta$-function, then attempt to prove $\theta(K\cap A)\geq\theta(K)\geq\theta(K\cap{\bar{A}})$ (4.1.1) for any $\Theta$-function $\theta$. Section 4 is devoted to defining the concepts used in the proof (subsection 4.2) and proving general lemmas about those concepts (subsections 4.3, 4.4 and 4.5). In particular, the properties defining a $\Theta$-function are given. Section 7 assumes inequality 4.1.1 and proves Theorem 1.2. Sections 5 and 6 are devoted to the proof of inequality 4.1.1. The idea of Section 7 is quite simple — a Brunn-Minkowski argument and a few approximations are enough to verify that the appropriate functions considered for generalized Orlicz balls are in fact $\Theta$-functions. The main line of the reasoning is similar to [W06]. To prove inequality 4.1.1 we shall attempt to divide the set $K_{+}$ into appropriately small convex subsets $D$ for which $\theta(D)=\theta(K)$. On each of these sets we will prove inequality (4.1.1) with $D$ substituted for $K$, which proves the thesis ($\theta$ is a proportion, so if it is attains some value on a family of disjoint sets, it attains the same value on the sum of this family). The problem, of course, is to prove the inequality (4.1.1) for any set $D$ (this is the aim of Section 5) and to construct a division into suitable sets $D$ (this is the aim of Section 6). For Section 5, the sets $D$ will have to be of the form $\tilde{D}\times{\mathbb{R}}^{n-2}$, where $\tilde{D}$ is 2-dimensional. Moreover, we will need $\tilde{D}$ to be “long and narrow”. This will allow us to take one direction (the one in which $\tilde{D}$ is “long”) to be a new coordinate, replacing the two coordinates of $\tilde{D}$, and to approximate the set $A$ and the function $\theta$ on $D$ with their approximations constant in the other, “narrow”, variable. If the approximation is good enough (and it turns out to be), we can inductively use the inequality (4.1.1) for the $n-1$ dimensional case for the approximating functions and then transfer the result to the original functions. We cannot reasonably expect the sets $D$ to have constant width in the “narrow” coordinate. This means that in the inductive step we shall have to consider weighted measures to take this into account. This motivates us to consider a more general theorem, in which the Lebesgue measure on $K$ will be replaced by a proper weighted measure. The argument in Section 6 is somewhat similar to the Kanaan–Lovasz–Simonovits localization lemma. However, we need the sets $D$ to satisfy additional assumptions, in particular to be “positively inclined” (this roughly means that the “long” coordinate axis has to be of the form $y=ax+b$, where $a$ is positive). We were unable to fit this into the localization lemma scheme, so the division is done by hand. We prove in Section 5 we can cut off a “good” set $D$ from our ball. Unfortunately, we have no control of the measure of the set we cut off (apart from the fact it is positive). Thus inductive cutting off good sets does not necassarily cover the whole $K$. This leads us to a transfinite inductive reasoning, where we cut off “good” sets in a transfinite fashion (that is, after cutting off countably many we see what is left and continue cutting). This approach leads to a number of technical problems associated with the limit step, and Section 6 is devoted to dealing with these problems and following through with the transfinite induction. ### 4.2 Definitions For the convienience of the reader all the basic definitions have been gathered in one place. So here we will just introduce the concepts required in the proof, and the next sections will be devoted to gaining a deeper understanding of those concepts. We shall usually consider a generalized Orlicz ball $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$. By $K_{x=u}$ we shall mean the section of $K_{+}$ with the hyperplane $x=u$, similarly for any other variable in ${\mathbb{R}}^{n}$. For a given set $D\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ by ${\tilde{D}}$ we shall denote the projection of $D$ to ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$. If not said otherwise, we shall assume $D={\tilde{D}}\times{\mathbb{R}}^{n-2}$. ###### Definition 4.1. A function $f:{\mathbb{R}}^{n}{\rightarrow}[0,\infty)$ is called $1/m$-concave if its support is a convex set and the function $f^{1/m}$ is concave on its support. ###### Definition 4.2. Let $K\subset{\mathbb{R}}^{n}$ be a generalized Orlicz ball. A measure $\mu$ on ${\mathbb{R}}^{n}$ is called a proper measure with respect to $K$ for ${\mathbb{R}}^{n}={\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ ($n\geq 2$) if the following conditions are satisfied: * • $\mu$ is a non-negative measure with density $f(x)g(y){\mathbf{1}}_{K_{+}}$. * • The functions $f$ and $g$ are $1/m$-concave for some $m>0$. * • If $K_{x=x_{0}}=\emptyset$ for a given $x_{0}$ then $f(x_{0})=0$, and if $K_{y=y_{0}}=\emptyset$ for a given $y_{0}$ then $g(y_{0})=0$. In the case $n=1$ a proper measure is a non-negative measure with a $1/m$-concave density $f$ for some $m>0$, satisfying ${\rm supp}f\subset K_{+}$. This definition describes the “proper weighted measures” which we will have to analyze in the subsequent induction steps of the proof outlined above. We shall denote the support of $f$ by $[x_{-},x_{+}]$ and the support of $g$ by $[y_{-},y_{+}]$. Of course $0\leq x_{-}\leq x_{+}$ and similarly for $y$. If we have a proper measure on ${\mathbb{R}}^{n}$ with respect to $K$ we can define a lens set. This definition describes the shape of a set, which will be one of the conditions of “not losing too much on approximation” and also will be a condition under which further dividing will be possible. ###### Definition 4.3. A set $D\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ is called a lens set if: * • $D$ is a convex set, * • $D=\tilde{D}\times{\mathbb{R}}^{n-2}$, * • for some $x_{-}\leq x_{1}<x_{2}\leq x_{+}$ and $y_{-}\leq y_{1}<y_{2}\leq y_{+}$, we have ${\tilde{D}}\subset[x_{1},x_{2}]\times[y_{1},y_{2}]$ and $(x_{1},y_{1})\in{\tilde{D}}$ and $(x_{2},y_{2})\in{\tilde{D}}$, * • $\mu(D)>0$. A lens set is said to be a strict lens set if $x_{-}<x_{1}<x_{2}<x_{+}$, $y_{-}<y_{1}<y_{2}<y_{+}$ and points $(x_{1},y_{1})$ and $(x_{2},y_{2})$ are the only points of ${\rm cl}{\tilde{D}}$ belogning to the boundary of the rectangle $[x_{1},x_{2}]\times[y_{1},y_{2}]$. Note that the boundary of the projection of a strict lens set onto ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ consists of an upper part, which is a graph of an concave, strictly increasing function, and a lower part, which is the graph of a convex, strictly increasing function. The boundary of a (non- strict) lens set may additionaly contain horizontal and vertical intervals adjacent to $(x_{1},y_{1})$ and $(x_{2},y_{2})$. We shall speak of the upper- left border and the lower-right border of a lens set. For a lens set $D$ we define the extremal points of ${\tilde{D}}$ to be two points $(x_{1},y_{1})$ and $(x_{2},y_{2})$. From the definition of a lens set, the extremal points belong to ${\tilde{D}}$. The extremal line of a lens set is the line connecting extremal points. By the width of a lens set we shall mean the length of its projection upon the line perpendicular to its extremal line in the plane ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$. ###### Definition 4.4. For a line $L$ in ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ the inclination of $L$ will denote measure of the angle between ${\mathbb{R}}_{x}$ and $L$ oriented so that the inclination of the line $\\{x=y\\}$ is $\pi/4$. A line is said to have positive inclination if its inclination belongs to $(0,\pi/2)$, and non-negative inclination if the inclination belongs to $[0,\pi/2]$. The inclination of a lens set $D$ is simply the inclination of its extremal line. By a positively inclined hyperplane in ${\mathbb{R}}^{n}$ we mean a hyperplane $H$ defined by $x_{i}=\lambda x_{j}+c$, where $\lambda\geq 0$. ###### Definition 4.5. For a given convex set $D$ and a proper measure $\mu$ by the relevant diameter of $D$ we mean the diameter of $D\cap{\rm supp}\mu$. ###### Definition 4.6. For a given generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ by its restriction to a positively inclined hyperplane $H$ we mean such a generalized Orlicz ball $K^{\prime}\subset{\mathbb{R}}^{n-1}$ such that $K_{+}\cap H$ is isometric to $K_{+}^{\prime}$. By Lemma 4.19 there exists such a generalized Orlicz ball $K^{\prime}$. ###### Definition 4.7. For a given generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ by its restriction to an interval $I\subset{\mathbb{R}}_{+}$ with respect to the coordinate $x_{i}$ we mean such a generalized Orlicz ball $K^{\prime}\subset{\mathbb{R}}^{n}$ that $K_{+}^{\prime}$ is isometric to $K\cap\\{x_{i}\in I\\}$. By Lemma 4.18 there exists such a generalized Orlicz ball $K^{\prime}$. When it is obvious in which coordinate the interval $I$ is taken we shall simply write that $K^{\prime}$ it a restriction of $K$ to $I$. ###### Definition 4.8. For a given generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ and a generalized Orlicz ball $K^{\prime}\subset{\mathbb{R}}^{m}$ we say that $K^{\prime}$ is a derivative of $K$ if there exists a sequence $K=K_{0},K_{1},\ldots,K_{n}=K^{\prime}$ of generalized Orlicz balls such that for each $i\in\\{1,\ldots,n\\}$ the ball $K_{i}$ is either a restriction of $K_{i-1}$ to some positively inclined hyperplane or a restriction of $K_{i-1}$ with respect to some variable $x_{k}$ to some interval $I\subset{\mathbb{R}}_{+}$. We can embed isometrically the positive quadrant of any derivative of $K$ into the positive quadrant of $K$. We shall identify without notice the positive quadrant of the derivative with the image of this embedding in the positive quadrant of $K$. In particular for a function $f$ defined on $K_{+}$ we shall speak of its restriction to $K_{+}^{\prime}$, meaning such a function $\tilde{f}$ that $\tilde{f}(x)=f(\phi(x))$, where $\phi$ is the embedding of $K_{+}^{\prime}$ into $K_{+}$. For the space ${\mathbb{R}}^{n}$ with a fixed orthonormal system $e_{1},\ldots,e_{n}$ by a coordinate-wise decompostion of ${\mathbb{R}}^{n}$ we mean a decompostion ${\mathbb{R}}^{n}={\mathbb{R}}^{k}\times{\mathbb{R}}^{l}$, where ${\mathbb{R}}^{k}={\rm span}\\{e_{i_{1}},\ldots,e_{i_{k}}\\}$ and ${\mathbb{R}}^{l}={\rm span}\\{e_{j_{1}},\ldots,e_{j_{l}}\\}$, with $i_{p}\neq j_{q}$ for any $p,q$. The main tool used in this proof will be the $\Theta$ functions. We define the $\Theta$ functions as follows: ###### Definition 4.9. For a given generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ and two functions $\eta_{1},\eta_{2}$ defined on $K_{+}$ we say that $\eta_{1}$ and $\eta_{2}$ define a $\Theta$ function on $K$ if the following properties are satisfied: 1. T1. The functions $\eta_{1}$ and $\eta_{2}$ are bounded. 2. T2. The functions $\eta_{1}$ and $\eta_{2}$ are coordinate-wise non-increasing. 3. T3. We have $\eta_{1}\geq\eta_{2}\geq 0$. 4. T4. For any derivative $K^{\prime}\subset{\mathbb{R}}^{m}$ of $K$, any proper measure $\mu$ on $K^{\prime}$ and any coordinate-wise decomposition ${\mathbb{R}}^{m}={\mathbb{R}}^{k}\times{\mathbb{R}}^{m-k}$ the function $\theta_{k}^{\mu}(\mathbf{x})=\frac{\int_{{\mathbb{R}}_{+}^{k}}\eta_{2}((\mathbf{y},\mathbf{x}))d\mu_{|{\mathbb{R}}^{k}}(\mathbf{y})}{\int_{{\mathbb{R}}_{+}^{k}}\eta_{1}((\mathbf{y},\mathbf{x}))d\mu_{|{\mathbb{R}}^{k}}(\mathbf{y})}$ is a coordinate-wise non-increasing function of $\mathbf{x}=(x_{j_{1}},\ldots,x_{j_{m-k}})$ where defined. Recall $\mu_{|{\mathbb{R}}^{k}}$ denotes the restriction of $\mu$ to ${\mathbb{R}}^{k}$. For a fixed proper measure $\mu$ on $K$ we define the function $\theta^{\mu}$ by $\theta^{\mu}(A)=\frac{\int_{A}\eta_{2}(x)d\mu(x)}{\int_{A}\eta_{1}(x)d\mu(x)}$ for any Borel set $A$ with $\mu(A)>0$. We shall say that $\theta^{\mu}$ is the $\Theta$ function defined for the measure $\mu$ by $\eta_{1}$ and $\eta_{2}$. For a fixed proper measure $\mu$ on $K$ and a fixed coordinate-wise decomposition ${\mathbb{R}}^{n}={\mathbb{R}}^{k}\times{\mathbb{R}}^{n-k}$ we shall also define $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k};A)=\frac{\int_{A}\eta_{2}(a_{1},a_{2},\ldots,a_{k},x_{k+1},x_{k+2},\ldots,x_{n})d\mu_{|\\{(a_{1},a_{2},\ldots,a_{k})\\}\times{\mathbb{R}}^{n-k}}}{\int_{A}\eta_{1}(a_{1},a_{2},\ldots,a_{k},x_{k+1},x_{k+2},\ldots,x_{n})d\mu_{|\\{(a_{1},a_{2},\ldots,a_{k})\\}\times{\mathbb{R}}^{n-k}}}$ for such sets $A$ and number $a_{1},a_{2},\ldots,a_{k}$ for which the denominator is positive. If $A={\mathbb{R}}^{n-k}$ we shall omit it and write $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k})$ for $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k};{\mathbb{R}}^{n-k})$, and if $\mathbf{a}=(a_{1},a_{2},\ldots,a_{k})$, we will write $\theta_{n-k}^{\mu}(\mathbf{a};A)$ or $\theta_{n-k}^{\mu}(\mathbf{a})$ for $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k};A)$ and $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k};{\mathbb{R}}^{n-k})$ respectively, which is consistent with the notation above. If $A\subset{\mathbb{R}}^{n}$ by $\theta^{\mu}_{n-k}(\mathbf{a};A)$ we mean $\theta^{\mu}_{n-k}(\mathbf{a};A\cap\\{\mathbf{a}\\}\times{\mathbb{R}}^{n-k})$. If there could be doubts as to what coordinate-wise decomposition is taken, we may write $\theta_{n-k}^{\mu}(x_{1}=a_{1},x_{1}=a_{2},\ldots,x_{k}=a_{k};A)$ for $\theta_{n-k}^{\mu}(a_{1},a_{2},\ldots,a_{k};A)$. ###### Fact 4.10. If $\eta_{1}$ and $\eta_{2}$ define a $\Theta$ function $\theta^{\mu}$ for a proper measure $\mu$ on a generalized Orlicz ball $K$, then the following are true: 1. T5. The function $\theta^{\mu}$ is continuous with respect to the symmetric difference distance, that is if $\theta^{\mu}$ is defined for all $C_{i}$ and $\mu(C_{0}\bigtriangleup C_{i}){\rightarrow}0$, then $\theta^{\mu}(C_{i}){\rightarrow}\theta^{\mu}(C_{0})$. 2. T6. If $D^{\prime}\subset D\subset{\mathbb{R}}^{n}$, $\theta^{\mu}(D)=\theta^{\mu}(D^{\prime})$ and $\theta^{\mu}$ is defined for $D\setminus D^{\prime}$, then $\theta^{\mu}(D\setminus D^{\prime})=\theta^{\mu}(D)$. 3. T7. If $K^{\prime}$ is a derivative of $K$, then the restrictions of $\eta_{1}$ and $\eta_{2}$ to $K^{\prime}$ define a $\Theta$ function on $K^{\prime}$. Further on, as the proper measure taken rarely changes, we omit the $\mu$ in the upper index and simply write $\theta$ for $\theta^{\mu}$. Note that as $\eta_{1}$ is positive on $K_{+}$ from property (T3) and ${\rm supp}\mu\subset K_{+}$, we know that $\theta^{\mu}(D)$ is well defined if and only if $\mu(D)>0$. ###### Definition 4.11. Functions $\eta_{1}$ and $\eta_{2}$ defining a $\Theta$ function on a generalized Orlicz ball $K$ are said to define a strict $\Theta$ function if the following extra conditions are satisfied: 1. S1. ${\rm supp}\eta_{2}\subset{\rm Int}_{{\mathbb{R}}_{+}}{\rm supp}\eta_{1}$, where ${\rm Int}_{{\mathbb{R}}_{+}}$ denotes the interior taken with respect to the space ${\mathbb{R}}_{+}$. 2. S2. The generalized Orlicz ball $K$ is proper. 3. S3. For any coordinate-wise decomposition ${\mathbb{R}}^{n}={\mathbb{R}}^{k}\times{\mathbb{R}}^{n-k}$ with $n>k$ the functions ${\tilde{\eta}}_{i}:{\mathbb{R}}^{k}{\rightarrow}{\mathbb{R}}$ defined by $x\mapsto\int_{{\mathbb{R}}^{n-k}}\eta_{i}(x,y)d\lambda_{n-k}(y)$ are continuous. 4. S4. $\eta_{1}>0$ on ${\rm Int}K_{+}$. ###### Definition 4.12. For $\eta_{1}$ and $\eta_{2}$ defining a $\Theta$ function $\theta$ on a generalized Orlicz ball $K$ by a derivative of $\theta$ we mean the function defined on a derivative $K^{\prime}$ of $K$ by the restrictions of $\eta_{1}$ and $\eta_{2}$ to $K^{\prime}$. Note that the derivatives of a $\Theta$ function are $\Theta$ functions. ###### Definition 4.13. For a given generalized Orlicz ball $K$ we say that $\eta_{1}$ and $\eta_{2}$ define a weakly non-degenerate $\Theta$ function on $K$ if for every ${\varepsilon}>0$ there exists a generalized Orlicz ball $K^{\prime}\subset K$ with $\lambda(K\setminus K^{\prime})\leq{\varepsilon}\lambda(K)$ and functions $\eta_{1}^{\prime}$ and $\eta_{2}^{\prime}$ defining a strict $\Theta$ function on $K^{\prime}$ with $\int|\eta_{i}-\eta_{i}^{\prime}|d\lambda\leq{\varepsilon}$. A $\Theta$ function is called non-degenerate if it is weakly non-degenerate and all its derivatives are weakly non-degenerate. Note that as the density of any proper measure is bounded, in all the bounds in the definition above we can replace $\lambda$ by any proper measure $\mu$. Frequently we shall take the same collection of assumptions for our theorems. To make reading the paper easier, we will use the following notation: ###### Definition 4.14. We shall speak of * • Standard assumptions if $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ is a generalized Orlicz ball, $\mu$ is a proper measure for $K$, $\eta_{1}$ and $\eta_{2}$ define a $\Theta$ function $\theta=\theta^{\mu}$ on $K$ for $\mu$ and $A$ is a c-set in ${\mathbb{R}}_{+}^{n}$, * • Non-degenerate assumptions if additionally we require the $\Theta$ function defined by $\eta_{1}$ and $\eta_{2}$ to be non-degenerate, and * • Strict assumptions if $K$ is a proper generalized Orlicz ball and $\eta_{1}$ and $\eta_{2}$ define a strict non-degenerate $\Theta$ function. ###### Definition 4.15. Under standard assumptions a set $D\subset{\mathbb{R}}^{n}$ will be called appropriate, if * • $\theta(D)$ is defined, * • $\theta(D\cap A)\geq\theta(K)\geq\theta(D\cap{\bar{A}})$ if the left-hand side and the right-hand side are defined, * • $\theta(D)=\theta(K)$. ###### Definition 4.16. Under standard assumptions let $\mu_{2}$ be the restriction of $\mu$ to ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times\\{0\\}$. For any ${\varepsilon}>0$ a set ${\tilde{D}}\times{\mathbb{R}}^{n-2}=D$ is called ${\varepsilon}$-appropriate, if * • $\theta(D)$ is defined, * • $\theta(D)=\theta(K)$, * • For each $U\in\\{A,{\bar{A}}\\}$ and each $i\in\\{1,2\\}$ there exists a number $C_{U,i}$ such that $\bigg{|}\int_{D\cap U}\eta_{i}(t)d\mu(t)-C_{U,i}\bigg{|}\leq{\varepsilon}\mu_{2}({\tilde{D}})$ and $\frac{C_{A,2}}{C_{A,1}}\geq\theta(D)\geq\frac{C_{{\bar{A}},2}}{C_{{\bar{A}},1}}.$ The definition of an appropriate set describes the properties we desire for the set into which we divide $K_{+}$. In fact, due to the approximation, we shall divide $K_{+}$ into ${\varepsilon}$-appropriate sets to prove it is ${\varepsilon}$-appropriate, and then take ${\varepsilon}{\rightarrow}0$. ### 4.3 The generalized Orlicz ball lemmas In this subsection we will prove a few lemmas about the structure generalized Orlicz balls. They show that the class of generalized Orlicz balls is closed under taking derivatives, and that proper generalized Orlicz balls are, in a sense, dense in the class of generalized Orlicz balls. These lemmas are the main reason the whole reasoning in this paper has to be done for generalized Orlicz balls, and not simply Orlicz balls — the class of Orlicz balls does not enjoy the same closedness propeties. ###### Fact 4.17. A product of intervals $\prod_{i=1}^{n}[a_{i},b_{i}]$ is isometric to the positive quadrant of the Orlicz ball $K\subset{\mathbb{R}}^{k}$ defined by the functions $\displaystyle f_{i}(x_{i})=\begin{cases}0&\hbox{ if $x_{i}\leq b_{i}-a_{i}$}\\\ \infty&\hbox{ if $x_{i}>b_{i}-a_{i}$}\end{cases}$ for $b_{i}>a_{i}$. ###### Lemma 4.18. If $K_{+}\subset{\mathbb{R}}^{n}$ is a generalized Orlicz ball positive quadrant and $0\leq x_{a}<x_{b}$, then $K_{+}\cap\\{x_{1}\in[x_{a},x_{b}]\\}$ is isometric to a generalized Orlicz ball positive quadrant or empty ###### Proof. Let $f_{1},f_{2},\ldots,f_{n}$ be the Young functions defining $K$. Let $K_{+}^{\prime}=K_{+}\cap\\{x_{1}\in[x_{a},x_{b}]\\}$. Let $c=f_{1}(x_{a})$. If $c=1$, then $K_{+}^{\prime}=\\{x:x_{1}=x_{a},\forall_{i>1}f_{i}(x_{i})=0\\}$, which is a product of intervals and thus isometric to a generalized Orlicz ball positive quadrant. If $c>1$ then $K_{+}^{\prime}$ is empty. If $c<1$ we define $\bar{f}_{1}$ by $\displaystyle\bar{f}_{1}(x_{1})=\begin{cases}\frac{f_{1}(x_{1}+x_{a})-f_{1}(x_{a})}{1-c}&\hbox{ for $x_{1}<x_{b}$}\\\ \infty&\hbox{ for $x_{1}>x_{b}$,}\end{cases}$ and $\bar{f}_{i}$ for $i>1$ by $\bar{f}_{i}(x_{i})=\frac{f_{i}(x_{i})}{1-c}.$ Now $(x_{1},x_{2},\ldots,x_{n})\in\bar{K}_{+}$ iff $(x_{1}+x_{a},x_{2},\ldots,x_{n})\in K_{+}^{\prime}$, where $\bar{K}_{+}$ is the positive quadrant of the Orlicz ball defined by $\bar{f}_{i}$. ∎ ###### Lemma 4.19. If $K\subset{\mathbb{R}}^{n}$ is a generalized Orlicz ball and $H=\\{x\in{\mathbb{R}}^{n}:x_{1}=\lambda x_{2}+c\\}$ is a positively inclined hyperplane in ${\mathbb{R}}^{n}$, then $K_{+}\cap H$ is the positive quadrant of some generalized Orlicz ball $L$ or an empty set. ###### Proof. As $H$ is positively inclined, $\lambda\geq 0$. If $\lambda=0$ and $c<0$ we have $K_{+}\cap H=\emptyset$. If $c<0$ and $\lambda>0$ we can transform the equation giving $H$ to $H=\\{x\in{\mathbb{R}}^{n}:x_{2}=\frac{1}{\lambda}x_{1}-\frac{c}{\lambda}\\}$. Thus we can assume $c\geq 0$. For $x_{2}\geq 0$ we have $x_{1}\geq c$ in $H$. Thus if $f_{1}(c)>1$, then for $x_{2}\geq 0$ we have $f_{1}(x_{1})>1$ for $x\in H,x_{1}\geq 0$, thus $H\cap K_{+}=\emptyset$. If $f_{1}(c)=1$ and $\lambda>0$, then $H\cap K_{+}$ is the set $\\{x:x_{1}=c,x_{2}=0,f_{i}(x_{i})=0$ for $i>2\\}$. This set is a cartesian product of intervals, and isometric to a generalized Orlicz ball positive quadrant. If $f_{1}(c)=1$ and $\lambda=0$, the situation is the same, except $x_{2}=0$ is replaced by $f_{2}(x_{2})=0$. Now we may assume $f_{1}(c)<1$. Let $x_{3},x_{4},\ldots,x_{n+1}$ be the coordinates on $H$, with $x_{1}=\lambda x_{n+1}+c$, $x_{2}=x_{n+1}$. Let us take $f_{n+1}(t)=f_{1}(\lambda t+c)+f_{2}(t)-f_{1}(c)$, then $f_{1}(x_{1})+f_{2}(x_{2})=f_{n+1}(x_{2})+f_{1}(c)$. The function $f_{n+1}$ is a sum of three convex functions, thus it is convex, and $f_{n+1}(0)=0$. The set $\\{(x_{i})_{i=3}^{n+1}:f_{i}(x_{i})<1-f_{1}(c)\\}\cap{\mathbb{R}}_{+}$ is equal to $K_{+}\cap H$. If we consider Young functions $\tilde{f}_{i}(t)=\frac{f_{i}(t)}{1-f_{1}(c)}$ for $i=3,4,\ldots,n+1$ we get the generalized Orlicz ball $L\subset H$ such that $K_{+}\cap H=L\cap{\mathbb{R}}_{+}^{n}$.∎ ###### Lemma 4.20. For any generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ and any ${\varepsilon}>0$ there exists a proper generalized Orlicz ball $K^{\prime}\subset K$ with $\lambda(K\setminus K^{\prime})<{\varepsilon}$. Furthermore if any Young function $f_{i}$ of $K$ is already a proper Young function, the same $f_{i}$ will be the appropriate Young function of $K^{\prime}$. ###### Proof. This lemma is easy to believe in, but somewhat technical to prove. An impatient reader might be well advised to skip the next two proofs (or prove the Lemmas her$/$himself, if desired) and go to the more crucial parts of the paper. As any generalized Orlicz ball is 1-symmetric, it suffices to prove $\lambda(K_{+}\setminus K^{\prime}_{\geq 0})\leq{\varepsilon}/2^{n}$. We shall thus consider only the points in ${\mathbb{R}}_{+}^{n}$ and decrease ${\varepsilon}$ to be $2^{n}$ times smaller. Recall that a proper Young function is such a Young function that $f(x)=0$ only for $x=0$ and $f(x)<\infty$. Thus we have to get rid of superfluous zeroes and of infinity values. First we shall take care of the zeroes. Let $f_{i}$ be Young functions defining $K$. Let $M$ be the largest of the $(n-1)$-dimensional measures of the projections of $K$ onto the hyperplanes $x_{i}=0$. Let $t_{i}=\inf\\{x_{i}:f_{i}(x_{i})>1/2n\\}$. Let $c=\inf_{i}\\{f_{i}^{\prime}(t_{i})\\}$. We shall prove that for $\delta<1/2n$ the set $U_{\delta}=\\{x:\sum f_{i}(x_{i})\in[1-\delta,1]\\}$ has measure no larger than $\frac{Mn\delta}{c}$. First note that $U_{\delta}=\bigcup U_{i}$, where $U_{i}=U_{\delta}\cap\\{x:f_{i}(x_{i})>1/2n\\}$, as at least one of $f_{i}(x_{i})$ has to be large for the sum to be large. We shall bound the measure of each $U_{i}$ separately. For each point $x=(x_{1},\ldots,x_{i-1},x_{i+1},\ldots,x_{n})$ the set of those $x_{i}$ that $(x,x_{i})\in U_{i}$ has length at most $\frac{\delta}{c}$. Thus, from Fubini’s theorem, the measure of $U_{i}$ can be bounded by $\frac{M\delta}{c}$, and summing over all $i$ we get the desired bound for $U_{\delta}$. Let us take $\delta=\frac{c{\varepsilon}}{2Mn^{2}}$. For each $i$ for which we have superfluous zeroes let us take $s_{i}=\inf\\{x_{i}:f_{i}(x_{i})>\delta\\}$ and replace $f_{i}$ by $g_{i}$ defined by $g_{i}(x_{i})=\begin{cases}f_{i}(x_{i})&\hbox{ for }x_{i}\geq s_{i}\\\ x_{i}\delta/s_{i}&\hbox{ for }x_{i}<s_{i}.\end{cases}$ We have $g_{i}(x_{i})\geq f_{i}(x_{i})$ and $g_{i}(x_{i})-f_{i}(x_{i})\leq\delta$. Thus if $K^{\prime}$ is the generalized Orlicz ball defined by $g_{i}$, we have $K^{\prime}\subset K$ and $K\setminus K^{\prime}\subset U_{n\delta}$, and thus $\lambda(K\bigtriangleup K^{\prime})\leq\frac{n^{2}M\delta}{c}={\varepsilon}/2$. Now we shall deal with the $\infty$ values. Let $\delta=\frac{{\varepsilon}}{2nM}$. Note that the shape of $K^{\prime}$ is determined by the values of $g_{i}$ only up to $g_{i}(x_{i})=1$. Thus we have to make some corrections to $g_{i}$ up to $g_{i}(x_{i})=1$, and then extend $g_{i}$ anyhow, say linearly. For each $i$ such that $g_{i}$ attains the $\infty$ value let $r_{i}=\inf\\{x_{i}:g_{i}(x_{i})=\infty\\}$, and let $v_{i}=\lim_{x_{i}{\rightarrow}v_{i}^{-}}g_{i}(x_{i})$. If $v_{i}\geq 1$, then all we have to do is to extend $g_{i}$ in a different way after $r_{i}$, and that does not change the ball $K^{\prime}$ defined by $g_{i}$. If, however, $v_{i}<1$, we define $h_{i}$ as follows: $h_{i}(x_{i})=\begin{cases}g_{i}(x_{i})&\hbox{ for }x_{i}<r_{i}-\delta\\\ 2&\hbox{ for }x_{i}=r_{i}\\\ \hbox{linear continuous extension}&\hbox{ otherwise.}\end{cases}$ Let $K^{\prime\prime}$ be the ball defined by $h_{i}$. Again, $K^{\prime\prime}\subset K^{\prime}$, as $h_{i}\geq g_{i}$ on the set where $g_{i}\geq 1$, from the convexity of $g_{i}$. The difference, however, is obviously contained in $\bigcup_{i}K^{\prime}\cap\\{x_{i}\in[r_{i}-\delta,r_{i}]\\}$, thus $\lambda(K^{\prime}\setminus K^{\prime\prime})\leq nM\delta={\varepsilon}/2$. Adding the two estimates together we get $\lambda(K\setminus K^{\prime\prime})\leq{\varepsilon}$. ∎ ###### Corollary 4.21. With the assumptions of Lemma 4.20 if we take any $y_{0}$ (where $y$ is any coordinate in ${\mathbb{R}}^{n}$), then we can take such a $K^{\prime}$ as before and $y_{1}$ that $\lambda_{n-1}((K\cap\\{y=y_{0}\\})\bigtriangleup(K^{\prime}\cap\\{y=y_{1}\\}))<{\varepsilon}$. ###### Proof. This, again, is easy to believe in, and actually simple if $f_{y}(y_{0})\neq 1$. The special case where $f_{y}(y_{0})=1$ could arguably be ignored (as it happens only on a set of measure zero), but to avoid omitting a set of measure zero in all other places of the proof, we shall go through the technicalities here. If $f_{y}(y_{0})>1$, we can simply take $y_{1}=y_{0}$, and $\lambda_{n-1}(K\cap\\{y=y_{0}\\})=\lambda_{n-1}(K^{\prime}\cap\\{y=y_{1}\\})=0$. If $f_{y}(y_{0})<1$, we need to control the Orlicz ball $\sum f_{i}(x_{i})=1-f_{y}(y_{0})$. This Orlicz ball $L$ is given by Young functions $f_{i}/(1-f_{y}(y_{0}))$. For this Orlicz ball we also calculate values of $M$ and $c$, and apply the reasoning in the proof of Lemma 4.20 taking the larger $M$ and the smaller $c$ of those calculated for the two balls. We thus get good approximations $K^{\prime}$ and $L^{\prime}$ of both $K$ and $L$. Now take such a $y_{1}$ that $f_{y}(y_{0})=h_{y}(y_{1})$, this can be done as $h_{y}$ is continuous. Now $K^{\prime}\cap\\{y=y_{1}\\}=L^{\prime}$, which proves the thesis. In the case $f_{y}(y_{0})=1$ if any of the other $f_{i}$ do not have superfluous zeroes, the measure of $K\cap\\{y=y_{0}\\}$ is 0, and thus taking $y_{1}=y_{0}+1$ we get the thesis. If, however, all the other $f_{i}$ have superfluous zeroes, the intersection $K\cap\\{y=y_{0}\\}$ is the cube ${\prod}_{i}f_{i}^{-1}(0)$. In this case we shall need a better approximation. Let $z_{i}=\sup\\{x_{i}:f_{i}(x_{i})=0\\}$. Let us, as before for ${\varepsilon}$, define $\delta^{\prime}=\frac{c{\varepsilon}^{\prime}}{2Mn^{2}}$ and $s_{i}^{\prime}=\inf\\{x_{i}:f_{i}(x_{i})>\delta^{\prime}\\}$. We need ${\varepsilon}^{\prime}$ to be so small that $s_{i}^{\prime}/z_{i}\leq 1+\frac{(1+{\varepsilon}/M)^{1/n}-1}{n}$ and smaller than ${\varepsilon}$. Note that as ${\varepsilon}^{\prime}{\rightarrow}0$ we have $\delta^{\prime}{\rightarrow}0$ and $s_{i}^{\prime}{\rightarrow}z_{i}$, so taking ${\varepsilon}^{\prime}$ small enough we can achieve the desired inequality for all $i$. Conduct the proof of Lemma 4.20 taking ${\varepsilon}^{\prime}$ instead of ${\varepsilon}$. Take $y_{1}$ such that $h_{y}(y_{1})=1-n\delta^{\prime}$. Note that if $x_{i}\leq s_{i}^{\prime}$ for all $i$, then $\sum h_{i}(x_{i})\leq n\delta^{\prime}$, and thus $x=(x_{i})\in K^{\prime}\cap\\{y=y_{1}\\}$. On the other hand if for any $i$ we have $x_{i}>s_{i}^{\prime}+(n-1)(s_{i}^{\prime}-z_{i})$, then $h_{i}(x_{i})\geq f_{i}(x_{i})$, and as $f_{i}(z_{i})=0$, $f_{i}(s_{i}^{\prime})\geq\delta^{\prime}$ and $f_{i}$ is convex, we have $f_{i}(x_{i})>n\delta^{\prime}$, and thus $\sum h_{i}(x_{i})>n\delta^{\prime}$ and $x\not\in K^{\prime}\cap\\{y=y_{1}\\}$. Thus $K\cap\\{y=y_{0}\\}={\prod}_{i}f_{i}^{-1}(0)\subset K^{\prime}\cap\\{y=y_{1}\\}\subset{\prod}_{i}[0,z_{i}+n(s_{i}^{\prime}-z_{i})].$ Now we have the following inequalities: $\displaystyle\frac{s_{i}^{\prime}}{z_{i}}$ $\displaystyle\leq 1+\frac{(1+{\varepsilon}/M)^{1/n}-1}{n}$ $\displaystyle\frac{n(s_{i}^{\prime}-z_{i})}{z_{i}}$ $\displaystyle\leq(1+{\varepsilon}/M)^{1/n}-1$ $\displaystyle\frac{z_{i}+n(s_{i}^{\prime}-z_{i})}{z_{i}}$ $\displaystyle\leq(1+{\varepsilon}/M)^{1/n}$ $\displaystyle\prod_{i}\frac{z_{i}+n(s_{i}^{\prime}-z_{i})}{z_{i}}$ $\displaystyle\leq 1+{\varepsilon}/M$ $\displaystyle\lambda({\prod}_{i}[0,z_{i}+n(s_{i}^{\prime}-z_{i})])$ $\displaystyle\leq\lambda({\prod}[0,z_{i}])+\frac{{\varepsilon}\lambda({\prod}[0,z_{i}])}{M}$ $\displaystyle\lambda(K^{\prime}\cap\\{y=y_{1}\\})-\lambda(K\cap\\{y=y_{0}\\})$ $\displaystyle\leq{\varepsilon}.$ The last inequality follows as ${\prod}[0,z_{i}]$ is a subset of the projection of $K$ onto $y=0$, and thus its measure is no bigger than $M$. This, along with the fact that $K\cap\\{y=y_{0}\\}\subset K^{\prime}\cap\\{y=y_{1}\\}$ gives the thesis. This reasoning can be extended to approximate any finite number of sections of $K$ along with $K$. ∎ ### 4.4 $1/m$-concave functions and proper measures Here we give a few elementary facts about $1/m$-concave functions and proper measures. Most facts are easily proved and quite a few are well known, so we skip some of the proofs. ###### Fact 4.22. If a function $f$ is $1/m$-concave for some $m>0$, then it is also $1/m^{\prime}$-concave for any $m^{\prime}>m$. ###### Fact 4.23. The product of $1/m$-concave functions is $1/2m$-concave. ###### Fact 4.24. From the Brunn-Minkowski inequality, if $K\subset{\mathbb{R}}^{n}$ is a convex set, then $(y_{1},y_{2},\ldots,y_{k})\mapsto\lambda_{n-k}(K\cap\\{\forall_{1\leq i\leq k}x_{i}=y_{i}\\})$ is a $1/(n-k)$-concave function, where $x_{i}$ are the coordinates on ${\mathbb{R}}^{n}$. Conversely, if we have a $1/m$-concave function on ${\mathbb{R}}^{n}$, then there exists a convex set $K\subset{\mathbb{R}}^{n+m}$ such that $f$ is the projection of the Lebesgue measure restricted to $K$ onto ${\mathbb{R}}^{n}$. As a corollary of these two facts the projection of a $1/m$ concave function on ${\mathbb{R}}^{n}$ onto ${\mathbb{R}}^{k}$ is a $1/(n+m-k)$-concave function. ###### Fact 4.25. The restriction of the Lebesgue measure to $K_{+}$ is a proper measure with respect to $K$. ###### Fact 4.26. The support of a proper measure $\mu$ is a convex set. ###### Lemma 4.27. If $\mu$ is a proper measure on ${\mathbb{R}}^{n}$ and $H=\\{x\in{\mathbb{R}}^{n}:x_{1}=\lambda x_{2}+c\\}$, with $\lambda\geq 0$ is a hyperplane in ${\mathbb{R}}^{n}$, then $\mu$ restricted to $H$ with coordinates $(v,x_{3},\ldots,x_{n})$ is a proper measure. ###### Proof. The density of $\mu$ is equal to $f(x)g(y){\mathbf{1}}_{K}$ for some $(1/m)$-concave functions $f$ and $g$ and some generalized Orlicz ball $K$. From Lemma 4.19 the set $K_{+}\cap H$ is the positive quadrant of some Orlicz ball $K^{\prime}$ with coordinates $(v,x_{3},\ldots,x_{n})$. The product $f\cdot g$ on $H$ varies with at most two variables, and is from, Fact 4.23, a $(1/2n)$-concave function with respect to $v$. ∎ ###### Lemma 4.28. If $\mu$ is a proper measure on ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times R^{n-2}$, then the restriction of $\mu$ to an interval $I$ with respect to any variable is also a proper measure. ###### Proof. Due to Lemma 4.18 if $K_{+}$ is the Orlicz ball quadrant for which $\mu$ is defined, $K_{+}^{\prime}=K_{+}\cap\\{t\in I\\}$ is also an Orlicz ball quadrant. Let $f$ and $g$ be the functions defining the density of $\mu$. To make them define a proper measure on $K_{+}^{\prime}$ we simply have to restrict them to the set $\\{x_{0}:\lambda_{n-1}(K^{\prime}_{x=x_{0}})>0\\}$ for $f$ and similarly for $g$, and additionaly to the interval $I$ if it was taken in $x$ or $y$ respectively. Both functions will have a convex support after this restriction, and as they were $1/m$-concave on a larger domain, they will still be $1/m$-concave. ∎ ### 4.5 Lens sets and $\Theta$ functions ###### Fact 4.29. Let $D$ be a lens set with extremal points $(x_{1},y_{1})$ and $(x_{2},y_{2})$. Let $x^{-}(y)=\inf\\{x:(x,y)\in D\\}$, $x^{+}(y)=\sup\\{x:(x,y)\in D\\}$ for $y\in[y_{1},y_{2}]$. Then $x^{-}$ and $x^{+}$ are increasing function on their domains, $x^{-}$ is convex, and $x^{+}$ is concave. ###### Lemma 4.30. Under standard assumptions consider a fixed $y_{0}$ and two intervals $[x_{a},x_{b}],[x_{c},x_{d}]$ with $x_{a}\leq x_{c}$ and $x_{b}\leq x_{d}$. Then we have $\theta^{\mu}_{n-1}(y_{0};[x_{a},x_{b}]\times{\mathbb{R}}^{n-2})\geq\theta^{\mu}_{n-1}(y_{0};[x_{c},x_{d}]\times{\mathbb{R}}^{n-2})$ if both sides are defined. The same applies when $x$ is exchanged with $y$. ###### Proof. From property (T4) we know that $\theta^{\mu}_{n-2}(x,y_{0})$ is a decreasing function of $x$. The domain of this function is a convex set, so its intersections with $[x_{a},x_{b}]$ and $[x_{c},x_{d}]$ are both intervals (they are non-empty, for $\theta^{\mu}_{n-1}$ is defined for both intervals). Applying Lemma 2.3, part 2, to $\int_{{\mathbb{R}}^{n-2}}\eta_{2}(x,y,t_{1},\ldots,t_{n-2})d\mu_{|(x,y)\times{\mathbb{R}}^{n-2}}(t_{1},\ldots,t_{n-2})$ and $\int_{{\mathbb{R}}^{n-2}}\eta_{1}(x,y,t_{1},\ldots,t_{n-2})d\mu_{|(x,y)\times{\mathbb{R}}^{n-2}}(t_{1},\ldots,t_{n-2})$ and the shortened intervals we get the thesis. ∎ ###### Lemma 4.31. Under standard assumptions for a given interval $I=[x_{a},x_{b}]$ the function $\theta^{\mu}_{n-1}(y;I\times{\mathbb{R}}^{n-2})$ is a decreasing function of $y$ on its domain. The same applies when $x$ is exchanged with $y$. ###### Proof. Take any $0\leq y_{1}\leq y_{2}$ in the domain. $K^{\prime}=K\cap\\{x\in I\\}$ is a derivative of $K$, and $\theta^{\mu}_{n-1}(y;I\times{\mathbb{R}}^{n-2})=\bar{\theta}_{n-1}^{\mu}(y)$, where $\bar{\theta}^{\mu}$ is defined by the restrictions of $\eta_{1}$ and $\eta_{2}$ to $K^{\prime}$. Thus from property (T4) we get the thesis. ∎ ###### Lemma 4.32. Under standard assumptions for a given lens set $D$ the domain of the function $y\mapsto\theta^{\mu}_{n-1}(y;D)$ is an interval and the function is decreasing. ###### Proof. Let $(x_{1},y_{1})$ and $(x_{2},y_{2})$ be the extremal points of ${\tilde{D}}$. Take any $y_{4}$ such that $\theta_{n-1}^{\mu}(y_{4};D)$ is defined and take any $y_{3}\in(y_{1},y_{4}]$. Thus $\theta_{n-2}^{\mu}(x,y_{4})$ is defined for more than one $x$ such that $(x,y_{4})\in{\tilde{D}}$ (actually, for a set of positive Lebesgue measure), let $x_{4}$ be any such $x$ except the smallest. We want to prove $\theta_{n-1}^{\mu}(y_{3};D)$ is defined. Note that ${\rm supp}\mu$ is a c-set on $x>x_{-}$, $y>y_{-}$ and ${\rm supp}\eta_{1}$ is also a c-set, thus their intersection is a c-set. Thus $\theta_{n-2}^{\mu}(x,y)$ is defined for any $x_{-}<x\leq x_{4}$ and $y_{-}<y\leq y_{4}$. As $D$ is a lens set, the set of $x\leq x_{4}$ such that $(x,y_{3})\in{\tilde{D}}$ has positive Lebesgue measure, thus $\theta_{n-1}^{\mu}(y_{3};D)$ is defined, which means that $\theta_{n-1}^{\mu}(y;D)$ is defined on some interval $(y_{1},y_{0})$ and undefined outside. Now we shall prove $\theta_{n-1}^{\mu}(y;D)$ is decreasing. Take $y_{3}\leq y_{4}$ from the domain. Let $[x_{3}^{-},x_{3}^{+}]$ be the interval ${\tilde{D}}\cap\\{y=y_{3}\\}$ and $[x_{4}^{-},x_{4}^{+}]$ the interval ${\tilde{D}}\cap\\{y=y_{4}\\}$. From the definition of a lens set $x_{3}^{-}\leq x_{4}^{-},x_{3}^{+}\leq x_{4}^{+}$. From Lemmas 4.31 and 4.30 (twice) we have $\displaystyle\theta_{n-1}^{\mu}(y_{3};D)$ $\displaystyle=\theta_{n-1}^{\mu}(y_{3};[x_{3}^{-},x_{3}^{+}]\times{\mathbb{R}}^{n-2})\geq\theta_{n-1}^{\mu}(y_{3};[x_{3}^{-},x_{4}^{+}]\times{\mathbb{R}}^{n-2})$ $\displaystyle\geq\theta_{n-1}^{\mu}(y_{4};[x_{3}^{-},x_{4}^{+}]\times{\mathbb{R}}^{n-2})\geq\theta_{n-1}^{\mu}(y_{4};[x_{4}^{-},x_{4}^{+}]\times{\mathbb{R}}^{n-2})=\theta_{n-1}^{\mu}(y_{4};D).$ Note that the last expression in the first line and the first in the second line are well defined, for the second argument is a superset of the second argument for $\theta_{n-1}^{\mu}(y_{3};D)$ and $\theta_{n-1}^{\mu}(y_{4};D)$ respectively. ∎ ###### Corollary 4.33. Under standard assumptions for a given lens set $D$ and a given $y_{0}$ in the domain of $\theta_{n-1}^{\mu}(y;D)$ we have $\theta^{\mu}(D\cap\\{y\leq y_{0}\\})\leq\theta_{n-1}^{\mu}(y_{0};D)\leq\theta^{\mu}(D\cap\\{y\geq y_{0}\\})$ and $\theta^{\mu}(D\cap\\{y\leq y_{0}\\})\leq\theta^{\mu}(D)\leq\theta^{\mu}(D\cap\\{y\geq y_{0}\\})$ if the left and right hand sides are defined. Moreover, if $\theta^{\mu}(D\cap\\{y\leq y_{0}\\})=\theta^{\mu}(D\cap\\{y\geq y_{0}\\})$ for any $y_{0}\in(y_{1},y_{2})$, then $\theta^{\mu}_{n-1}(y_{3};D),\theta^{\mu}(D\cap\\{y\geq y_{3}\\})$ and $\theta^{\mu}(D\cap\\{y\leq y_{3}\\})$ are all constant where defined and equal $\theta^{\mu}(D)$ for $y_{3}\in(y_{1},y_{2})$. ###### Proof. From Lemma 4.32 the function $\theta_{n-1}^{\mu}(y;D)$ is decreasing as a function $y$ on its domain, and its domain is an interval. We know that ${\rm supp}\eta_{2}\subset{\rm supp}\eta_{1}$, so we can apply Lemma 2.3, part 1a, to the appropriate integrals of $\eta_{2}$ and $\eta_{1}$ to get the first part of the thesis. The second part follows from the first and Fact 2.2. The third follows again from Lemma 2.3, part 1b. ∎ ###### Proposition 4.34. Under standard assumptions if $D$ is ${\varepsilon}$-appropriate for any ${\varepsilon}>0$, then $D$ is appropriate. ###### Proof. The third and first condition in Definition 4.15 follows from the definition of ${\varepsilon}$-appropriate for any ${\varepsilon}$. We have to check the second condition. Let $C_{U,i}^{\varepsilon}$ denote the numbers $C_{U,i}$ which show $D$ is and ${\varepsilon}$-appropriate set. We have $\theta^{\mu}(D)=\frac{C_{A,2}^{\varepsilon}}{C_{A,1}^{\varepsilon}}\leq\frac{\int_{D\cap A}\eta_{2}(t)d\mu(t)+{\varepsilon}\mu_{2}({\tilde{D}})}{\int_{D\cap A}\eta_{1}(t)d\mu(t)-{\varepsilon}\mu_{2}({\tilde{D}})}\rightarrow_{{\varepsilon}{\rightarrow}0}\theta^{\mu}(D\cap A),$ and similarly for the second inequality. ∎ ###### Proposition 4.35. Under standard assumptions if $D_{k}$ are ${\varepsilon}$-appropriate sets for $k\in K$, then $D=\bigcup_{k\in K}D_{k}$ is ${\varepsilon}$-appropriate. ###### Proof. We have $\theta^{\mu}(D)=\theta^{\mu}(K)$ from Lemma 2.5. For the third condition in Definition 4.16 we take $C_{U,i}^{D}=\sum_{k\in K}C_{U,i}^{D_{k}}$. These are good approximations as $\mu_{2}({\tilde{D}})=\sum_{k}\mu_{2}({\tilde{D}}_{k})$, and obviously satisfy the proportion inequality. ∎ ## 5 The $\Theta$ theorem ### 5.1 Preparations for divisibility In this section we shall prove the main theorem concerning $\Theta$ functions. Under standard assumptions, we shall consider $\mu$ to be a fixed proper measure on ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$. By $\mu_{2}$ we shall denote the restriction of $\mu$ to ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times\\{0\\}$. Note that as the support of $\mu$ is a c-set with respect to the $n-2$ variables of ${\mathbb{R}}^{n-2}$, the support of $\mu_{2}$ is the projection of the support of $\mu$. As we fix $\mu$, we shall omit the upper index when writing the $\Theta$ function and write $\theta$ or $\theta_{k}$ instead of $\theta^{\mu}$ or $\theta_{k}^{\mu}$. The main theorem we want to prove is: ###### Theorem 5.1. Under non-degenerate assumptions $\theta(A)\geq\theta(K)$ and $\theta(K)\geq\theta({\bar{A}})$, whenever both sides of an inequality are defined. This looks like a quite simple theorem, and we suspect there is a simpler proof than the one we present here. However, we were not able to find it (and would be interested to learn if anyone does). Notice that if $\theta(A)$ is undefined, then $\int_{A}\eta_{1}d\mu=0$, which implies $\int_{A}\eta_{2}d\mu=0$ from property (T3). Thus $\theta({\bar{A}})=\theta(K)$ and the Theorem is satisfied. Thus we assume $\theta(A)$ is defined. Similarly we may assume $\theta({\bar{A}})$ is defined. From Fact 2.2 it is enough to prove $\theta(A)\geq\theta(K)$ and the second inequality will follow. Thus, we concentrate on the first inequality. First, for technical reasons, we shall deal with the low-dimensional case: ###### Theorem 5.2. Under standard assumptions with $n\leq 2$ (that is, $K\subset{\mathbb{R}}$ or $K\subset{\mathbb{R}}^{2}$) we have $\theta(A)\geq\theta(K)$ and $\theta(K)\geq\theta({\bar{A}})$, whenever both sides of an inequality are defined. ###### Proof. For $n=1$ the set $A$ is one-dimensional, and thus (being a c-set) is an interval of the form $[0,a)$. We apply property (T4) to $K^{\prime}=K$, the measure $\mu$ and the decomposition ${\mathbb{R}}\times\\{0\\}$ and get that $\frac{\eta_{2}}{\eta_{1}}$ is a decreasing function. Thus from Lemma 2.3, part 1a, $\theta(A)\geq\theta({\bar{A}})$ and the thesis follows from Fact 2.2. For $n=2$ we shall approximate the set $A$ by a $l$-stair set. The $l$-stair set is defined as follows: ###### Definition 5.3. A $1$-stair set defined by $x_{1}=0$ and $a_{1}\geq 0$ (denoted $A(x_{1};a_{1})$) is the empty set. A $l$-stair set defined by $0=x_{1}\leq x_{2}\leq\ldots\leq x_{l}$ and $a_{1}\geq a_{2}\geq\ldots\geq a_{l}\geq 0$, denoted $A(x_{1},x_{2},\ldots,x_{l};a_{1},a_{2},\ldots,a_{l})$ is defined by $A(x_{1},x_{2},\ldots,x_{l};a_{1},a_{2},\ldots,a_{l})=\Big{(}A(x_{1},x_{2},\ldots,x_{l-1};a_{1},a_{2},\ldots,a_{l-1})\cap\\{x\leq x_{l}\\}\Big{)}\cup A(0,a_{l}).$ That means that a $l$-stair set consists of $l$ steps, the $k$-th step goes from $x_{k}$ to $x_{k+1}$ (the last one goes all the way to infinity) at height $a_{k}$. A proper $l$-stair set is a $l$-stair set with $a_{l}=0$ Notice that $\theta(A)=\theta(K\cap A)$ as ${\rm supp}\mu\subset K$. Thus we may assume $A\subset K$, and thus $A$ is bounded. Take $A_{n}=\\{(x,y):([xn]/n,y)\in A\\}$, where $[xn]$ denotes the integer part of $xn$. This is a proper stair set defined by $0,1/n,2/n,\ldots$ (a finite sequence as $A$ is bounded) and the sequence $a_{k}=\sup\\{y:(k/n,y)\in A\\}$. Notice also $A_{2^{n}}\supset A_{2^{n+1}}\supset A$ and $\mu(A_{2^{n}}\setminus A){\rightarrow}0$. Thus $\theta(A_{2^{n}}){\rightarrow}\theta(A)$, so it is enough to prove $\theta(A_{2^{n}})\geq\theta(K)$ and go to the limit. Thus, instead of considering all c-sets we may restrict ourselves to proper $l$-stair sets. The proof for $A$ being a proper $l$-stair set will be an induction upon $l$. For $l=1$ the set $A$ is empty and the thesis is obvious. For $l=2$ let $I=[0,x_{2}]$. From Lemma 4.31 the function $\theta_{1}(y;I)$ is decreasing where defined. Thus from Lemma 2.3 we have $\theta(I\times[0,a_{1}])\geq\theta_{1}(a_{1};I)\geq\theta(I\times[a_{1},\infty))$. Note that $A=I\times[0,a_{1}]$. Thus if $\theta(A)\geq\theta(K)$ or $\theta(A)$ is undefined, the thesis is satisfied. Otherwise, as $\theta(K)>\theta(I\times[0,a_{1}])$ we have $\theta(K)>\theta(I\times[a_{1},\infty))$ if defined, and thus from Lemma 2.5 $\theta(K)>\theta(I\times{\mathbb{R}}_{y})$. Now apply property (T4) to $K^{\prime}=K$, the decomposition ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ and the measure $\mu$ to get that $\theta_{1}(x)$ is a decreasing function. Again from Lemma 2.3 and Lemma 2.5 this implies $\theta(I\times{\mathbb{R}}_{y})\geq\theta(K)$, a contradiction. Thus the thesis is satisfied for $l=2$. For larger $l$ let $I=[x_{l-1},x_{l}]$. Again from Lemma 4.31 the function $\theta_{1}(y;I)$ is decreasing. Thus $\theta(I\times[0,a_{l-1}])\geq\theta(I\times[a_{l-1},a_{l-2}])$ (5.1.1) if both are defined. Note $A(x_{1},x_{2},\ldots,x_{l};a_{1},a_{2},\ldots,a_{l})\setminus(I\times[0,a_{l-1}])=A(x_{1},x_{2},\ldots,x_{l-1};a_{1},a_{2},\ldots,a_{l-2},0)$ and $A(x_{1},x_{2},\ldots,x_{l};a_{1},a_{2},\ldots,a_{l})\cup(I\times[a_{l-1},a_{l-2}])=A(x_{1},x_{2},\ldots,x_{l-2},x_{l};a_{1},a_{2},\ldots,a_{l-2},a_{l}).$ Suppose $\theta(A)<\theta(K)$. If $\theta(I\times[0,a_{l-1}])>\theta(A)$ or is undefined, then from Lemma 2.5 we have $\theta(A\setminus(I\times[0,a_{l-1}]))\leq\theta(A)$, but from the inductive assumption for $l-1$ we have $\theta(A\setminus(I\times[0,a_{l-1}]))\geq\theta(K)$, from which $\theta(A)\geq\theta(K)$. If, on the other hand, $\theta(I\times[0,a_{l-1}])\leq\theta(A)$, then from (5.1.1) $\theta(I\times[a_{l-1},a_{l-2}])\leq\theta(A)$ or is undefined, thus from Lemma 2.5 $\theta(A)\geq\theta(A\cup(I\times[a_{l-1},a_{l-2}]))$, and again from the inductive assumption $\theta(A)\geq\theta(K)$. Thus for any $l$ and for any $A$ being a proper $l$-stair set we have $\theta(A)\geq\theta(K)$, which ends the proof. ∎ ###### Proof of the Theorem 5.1. The proof will proceed by induction upon $n$. For $n\leq 2$ we use Theorem 5.2. For greater $n$ let $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$. Assume the thesis is true for all cases with $n^{\prime}<n$. If the theorem holds for strict $\Theta$ functions, then for any non-degenerate $\theta$ we take a sequence $\theta_{i}$ of strict $\Theta$ functions for ${\varepsilon}=1/i$. For any set $C$ for which $\theta(C)$ is defined, we have $\theta_{i}(C){\rightarrow}\theta(C)$, so as we had $\theta_{i}(A)\geq\theta_{i}(K_{i})\geq\theta_{i}({\bar{A}})$, we get the thesis when $i$ tends to infinity. Thus it is enough to restrict ourselves to strict assumptions. Note that under strict assumptions $\theta(U)$ is defined for any set $U$ with $\mu(U)>0$ as ${\rm supp}\mu\subset{\rm supp}\eta_{1}$. In particular, if $\mu_{2}({\tilde{U}})>0$, then $\theta(U\times{\mathbb{R}}^{n-2})$ is well defined. Also note that if $\theta(K_{+})=0$, then $\eta_{2}$ has to be zero $\mu_{2}$-almost everywhere, which means $\theta(U)=0$ for any $U$ such that it is defined, thus Theorem 5.1 holds. Thus we can assume $\theta(K_{+})>0$. We want to prove that for any ${\varepsilon}>0$ the quadrant $K_{+}$ is an ${\varepsilon}$-appropriate set. We shall frequently require the following property from various sets $D$: $\theta(D)=\theta(K),$ (5.1.2) or (for lower-dimensional sets) $\theta_{k}(\mathbf{a};D)=\theta(K).$ (5.1.3) We shall need to bound the diameter of the constructed sets from below. To this end consider the following sets: ${\tilde{S}}^{0}=\\{(x,y):\theta_{n-2}(x,y)=\theta(K)\\}$, ${\tilde{S}}^{+}=\\{(x,y):\theta_{n-2}(x,y)\geq\theta(K)\\}$, ${\tilde{S}}^{-}=\\{(x,y):\theta_{n-2}(x,y)\leq\theta(K)\\}$ and ${\tilde{S}}={\rm cl}{\tilde{S}}^{+}\cap{\rm cl}{\tilde{S}}^{-}$. We take a $\delta$-neighbourhood ${\tilde{S}}_{\delta}$ of ${\tilde{S}}$ with $\delta$ so small that $\mu({\tilde{S}}_{\delta}\setminus{\tilde{S}})\leq{\varepsilon}\mu_{2}({\tilde{K}})(\lambda_{n-2}(K\cap\\{x=0,y=0\\})\sup_{K}\eta_{1})^{-1}/3.$ Note that as from property (T4) the function $\theta_{n-2}(x,y)$ is coordinate-wise decreasing, the set ${\tilde{S}}\setminus{\tilde{S}}^{0}$ has measure 0. ###### Remark 5.4. Note that any ${\tilde{D}}\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ having property (5.1.2) must, from Fact 2.5, have a non-empty intersection both with ${\tilde{S}}^{+}$ and ${\tilde{S}}^{-}$ in some points where the density of $\mu_{2}$ is positive. Thus any convex set ${\tilde{D}}$ with property 5.1.2 will satisfy ${\tilde{D}}\cap{\rm supp}\mu_{2}\cap{\tilde{S}}\neq\emptyset$, as the set ${\tilde{D}}\cap{\rm supp}\mu_{2}$ is convex, and thus connected. Thus either ${\tilde{D}}\cap{\rm supp}\mu_{2}$ is contained in ${\tilde{S}}_{\delta}$ or it has diameter at least $\delta$. The main part of the proof will be an transfinite inductive construction of subsequent ${\varepsilon}$-appropriate strict lens sets by the following Theorem: ###### Theorem 5.5. Let $n>2$. Assume Theorem 5.1 holds under non-degenerate assumptions for any $n^{\prime}<n$. Then under strict assumptions if $\theta(K_{+})>0$ for any ordinal $\gamma$ there exists a division of the set $\tilde{K}_{+}$ into $\gamma+2$ sets $U(\gamma,\beta)$ for $0\leq\beta\leq\gamma+1$ satisfying: * • The set $U(\gamma,\gamma+1)$ is of $\mu_{2}$ measure at most ${\varepsilon}^{\prime}\mu_{2}({\tilde{K}})(\lambda_{n-2}(K\cap\\{x=0,y=0\\})\sup_{K}\eta_{1})^{-1}$. * • The set $U(\gamma,\gamma)$ is either an appropriate set, a strict lens set satisfying condition (5.1.2) or has $\mu_{2}$ measure 0. * • All sets $U(\gamma,\beta)$ for $\beta<\gamma$ are either ${\varepsilon}^{\prime}$-appropriate sets, empty, or have non-zero $\mu_{2}$ measure and satisfy $U(\gamma,\beta)\cap\tilde{K}_{+}\subset{\tilde{S}}_{\delta}$ * • If any $U(\gamma,\beta)$ is empty for $\beta<\gamma$, then $U(\gamma,\gamma)$ has measure 0. If we prove this Theorem, we can apply it to prove Theorem 5.1. By the inductive assumption we assume Theorem 5.1 holds for $n^{\prime}<n$. We take $\gamma=\omega_{1}$ and ${\varepsilon}^{\prime}={\varepsilon}/3$. As the measure of $\tilde{K}_{+}$ is finite, it cannot have $\omega_{1}$ disjoint subsets of non-zero measure, thus some of $U(\omega_{1},\beta)$ for $\beta<\omega_{1}$ are empty. Thus $U(\omega_{1},\omega_{1})$ has measure 0. Let ${\tilde{T}}$ be the sum of those $U(\omega_{1},\beta)$ which are subsets of ${\tilde{S}}_{\delta}$. For every point $(x,y)$ in ${\tilde{T}}\cap{\tilde{S}}^{0}$ we apply Theorem 5.1 to the restrictions of $K,A,\theta$ and $\mu$ to $(x,y)\times{\mathbb{R}}^{n-2}$. The conditions are satisfied — the restiction of $\theta$ is a derivative of $\theta$ and thus non-degenerate, the restriction of $K$ is an generalized Orlicz ball due to Lemma 4.19 and the restriction of $\mu$ is a proper measure due to 4.27, the restriction of a c-set is obviously a c-set. Thus for all $(x,y)\in{\tilde{S}}^{0}$ we have $\theta_{n-2}(x,y;A)\geq\theta_{n-2}(x,y)\geq\theta_{n-2}(x,y;{\bar{A}}),$ and as $\theta_{n-2}(x,y)=\theta(K)$ from the definition of ${\tilde{S}}^{0}$, from Lemma 2.5 we get $\theta\big{(}(({\tilde{T}}\cap{\tilde{S}}^{0})\times{\mathbb{R}}^{n-2})\cap A\big{)}\geq\theta(K)\geq\theta\big{(}(({\tilde{T}}\cap{\tilde{S}}^{0})\times{\mathbb{R}}^{n-2})\cap{\bar{A}}\big{)},$ and also $\theta(({\tilde{T}}\cap{\tilde{S}}^{0})\times{\mathbb{R}}^{n-2})=\theta(K)$, if only $\mu({\tilde{T}}\cap{\tilde{S}}^{0})>0$. Thus ${\tilde{T}}\cap{\tilde{S}}^{0}$ either has measure 0, or is an appropriate set. Meanwhile ${\tilde{T}}\setminus{\tilde{S}}^{0}$ has measure at most ${\varepsilon}(\lambda_{n-2}(K\cap\\{x=0,y=0\\})\sup_{K}\eta_{1})^{-1}\mu_{2}({\tilde{K}})/3$ from the definition of ${\tilde{S}}_{\delta}$. We therefore have a division of $\tilde{K}_{+}$ except a set of measure $2{\varepsilon}(\lambda_{n-2}(K\cap\\{x=0,y=0\\})\sup_{K}\eta_{1})^{-1}\mu_{2}({\tilde{K}})/3$ into $({\varepsilon}/3)$-appropriate sets. The sum of all the $({\varepsilon}/3)$-appropriate sets is by Remark 4.35 an $({\varepsilon}/3)$-appropriate set. As the integral of $\eta_{i}$ over the remaining set is at most $2{\varepsilon}\mu_{2}({\tilde{K}})/3$, the whole $\tilde{K}_{+}$ is an ${\varepsilon}$-appropriate set with the same $C_{U,i}$. As we can do this for any ${\varepsilon}>0$, by Lemma 4.34 $K$ is an appropriate set, which is the thesis of Theorem 5.1 ∎ ### 5.2 Almost horizontal divisions We shall prove that if we can divide a lens set with a horizontal, or even almost horizontal (under strict assumptions) line into two sets with equal $\theta$, then the lens set is appropriate. ###### Lemma 5.6. Assume Theorem 5.1 holds for $n^{\prime}<n$. Under strict assumptions if for a given lens set $D\subset{\mathbb{R}}^{n}$ satisfying (5.1.2) there exists a horizontal or vertical line $L$ in ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ dividing ${\tilde{D}}$ into two sets ${\tilde{D}}_{-}$ and ${\tilde{D}}_{+}$ of non-zero $\mu_{2}$-measure with $\theta({\tilde{D}}_{-}\times{\mathbb{R}}^{n-2})=\theta({\tilde{D}}_{+}\times{\mathbb{R}}^{n-2})$, then $D$ is an appropriate set. ###### Proof. Suppose, without loss of generality, the line is horizontal given by $y=a_{0}$. From Corollary 4.33 we know that for any $a$ we also have $\theta_{n-1}(y=a;D)=\theta(D)$ if defined. From Lemma 4.27 and property (T7) we know that the restriction of $\mu$ to $\\{y=a\\}$ is a proper measure and the restriction of $\theta$ is a non- degenerate $\Theta$ function. From the assumption we can apply Theorem 5.1, thus $\theta_{n-1}(y=a;D\cap A)\geq\theta_{n-1}(y=a;D)\geq\theta_{n-1}(y=a;D\cap{\bar{A}}).$ As $\theta_{n-1}(y=a;D)=\theta(D)$, which does not depend on $a$, we can apply Lemma 2.5 to get $\theta(D\cap A)\geq\theta(D)\geq\theta(D\cap{\bar{A}})$, and as $\theta(D)=\theta(K)$ this means $D$ is appropriate. ∎ ###### Lemma 5.7. Assume Theorem 5.1 holds for $n^{\prime}<n$. Under strict assumptions if for a given strict lens set $D$ satisfying (5.1.2) for every $\beta>0$ there exists a line $L_{\beta}$ with inclination between $0$ and $\beta$ (i.e. almost horizontal) or between $\frac{\pi}{2}-\beta$ and $\frac{\pi}{2}$ (i.e. almost vertical) dividing ${\tilde{D}}$ into two sets ${\tilde{D}}_{-}$ and ${\tilde{D}}_{+}$ of non-zero $\mu_{2}$-measure with $\theta(D_{-})=\theta(D_{+})=\theta(D)$, then $D$ is an appropriate set. ###### Proof. Assume $\theta(D)>0$ (otherwise the thesis is trivial). We choose a sequence of such lines $L_{i}$ with $\beta{\rightarrow}0$. We choose a subsequence such that all lines are almost vertical or all are almost horizontal (we shall assume without loss of generality that all are almost horizontal). From the compactness of the set of lines intersecting the closure of ${\tilde{D}}\cap{\rm supp}(\mu_{2})$ we can choose a subsequence of lines converging to some line $L$, which, of course, will be horizontal. If $L$ cuts off a non-zero $\mu_{2}$ measure both above and below it, then both the sets into which ${\tilde{D}}$ is divided have the same $\theta=\theta(D)$ from the continuity of $\theta$ with respect to the set and the thesis follows from Lemma 5.6. The case left to examine is when $L_{i}$ approaches the lowest or highest point $p$ of ${\tilde{D}}\cap{\rm supp}(\mu_{2})$. From the definition of a lens set we know that the only points of ${\tilde{D}}$ on which $\mu_{2}$ vanishes lie outside $\tilde{K}_{+}$. Thus the lowest point of ${\tilde{D}}\cap{\rm supp}(\mu_{2})$ is the lower extremal point of ${\tilde{D}}$. The highest point can be either the upper extremal point of ${\tilde{D}}$ or can lie on the boundary of ${\rm supp}\mu_{2}$. First consider the second, simpler case. As $D$ is a strict lens set and $K$ is proper, for any neighbourhood ${\tilde{U}}\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ of the highest point $p$ if we take a sufficiently horizontal line passing sufficiently close to $p$, the set it will cut off from ${\tilde{D}}$ will be a subset of ${\tilde{U}}$ (${\tilde{D}}\cap{\rm supp}\mu_{2}$ has no horizontal edges). We know that ${\rm supp}\eta_{2}\subset{\rm Int}\ {\rm supp}\mu$, so ${\rm cl}\ \widetilde{{\rm supp}\eta_{2}}\subset{\rm supp}\mu_{2}$. As $p$ lies on the boundary of ${\rm supp}\mu_{2}$, it lies outside ${\rm Int}\ {\rm supp}\mu_{2}$ and thus outside ${\rm cl}\ \widetilde{{\rm supp}\eta_{2}}$, so we can choose an open neighbourhood ${\tilde{U}}$ of $p$ on which $\eta_{2}$ is 0. This neighbourhood has non-zero $\mu_{2}$ measure, and as $\mu_{2}({\tilde{D}})>0$, $\mu_{2}({\tilde{D}}\cap{\tilde{U}})>0$. But $\eta_{2}$ on the whole set ${\tilde{U}}$ is zero, thus any line cutting off only a part of ${\tilde{U}}$ cannot satisfy $\theta(D_{+})=\theta(D)>0$. In the first case $(L_{i})$ approaches one of the extremal points of ${\tilde{D}}$. Assume it is the lower point. For any line $L_{i}$ the set ${\tilde{D}}^{L_{i}}_{-}$ is a lens set. From Lemma 2.5 there has to be a point $p_{i}\in D^{L_{i}}_{-}$ with $\theta(\\{p_{i}\\}\times{\mathbb{R}}^{n-2})\leq\theta(D^{L_{i}}_{-})=\theta(D)$. The lines $L_{i}$ tend to the horizontal line through $(x_{1},y_{1})$, the lower extremal point of $D$. Thus, the vertical coordinate of $p_{i}$ tends to $y_{1}$, and as ${\tilde{D}}$ has no horizontal edges, the horizontal coordinate of $p_{i}$ tends to $x_{1}$, meaning $p_{i}{\rightarrow}(x_{1},y_{1})$. From property (T4), as $\theta_{n-2}(\\{p_{i}\\}\times{\mathbb{R}}^{n-2})\leq\theta(D)$, for all points $p\in D$ except for $(x_{1},y_{1})$ we have $\theta_{n-2}(\\{p\\}\times{\mathbb{R}}^{n-2})\leq\theta(D)$. This, however, from Lemma 2.5 implies in particular, that for any horizontal line $M$ dividing ${\tilde{D}}$ into two sets of non-zero $\mu$-measure we have $\theta(D_{+})\leq\theta(D)$, which from Lemma 4.33 implies $\theta(D_{+})=\theta(D)$, which from Lemma 5.6 implies that $D$ is appropriate. ∎ ### 5.3 ${\varepsilon}$-appropriateness of lens sets This subsection puts down precisely what we meant by “long and narrow” in the idea of the proof, and show how to go from the “longness and narrowness” to ${\varepsilon}$-appropriateness. ###### Lemma 5.8. Let $C\subset{\mathbb{R}}^{n}$ be a convex set with $\lambda(C)>0$, let $I\subset C$ be an interval of length $a$, and let $L$ be the line containing $I$. Let $f:C{\rightarrow}(0,\infty)$ be a $1/m$-concave function. Let $P:{\mathbb{R}}^{m}{\rightarrow}L$ be the orthogonal projection onto $L$. Let $J\subset I$ be an subinterval of length $b$. Let $C^{\prime}\subset C$ be such a set that $P(C^{\prime})\subset J$. Then $\int_{C^{\prime}}f(x)dx\leq\Big{(}\Big{(}\frac{a+b}{a-b}\Big{)}^{n+m}-1\Big{)}\int_{C}f(x)dx$ and also $\int_{C^{\prime}}f(x)dx\leq\frac{2^{n+m+2}b}{a}\int_{C}f(x)dx.$ ###### Proof. Let $p(y)=\int_{x:P(x)=y}f(x)dx$ and let $I^{\prime}=\\{y\in L:p(y)>0\\}$. From Fact 4.24 the function $p(y)=\int_{x:P(x)=y}f(x)dx$ is a $(1/m+n-1)$-concave function on $L$, thus $I^{\prime}$ is an interval. As $f$ is positive and $C$ is convex and has positive measure, $p$ is positive on ${\rm Int}I$, thus the length of $I^{\prime}$ is at least $a$. If $J\cap I^{\prime}=\emptyset$, then $\int_{C^{\prime}}f(x)dx=0$ and the thesis is satisfied, so assume $J\cap I^{\prime}\neq\emptyset$. Then $I^{\prime}\setminus J$ is a sum of two intervals (one may be empty) of total length at least $a-b$. Thus it contains an interval $I^{\prime\prime}$ of length at least $\frac{a-b}{2}$, let $\\{y_{1}\\}={\rm cl}I^{\prime\prime}\cap{\rm cl}J$ and $y_{2}$ be the other end of $I^{\prime\prime}$. Let $y_{2}$ and $y_{3}$ be the ends of $I^{\prime}$ and let $T$ be such that $y_{3}=Ty_{1}+(1-T)y_{2}$ (as $y_{1}$ lies between $y_{2}$ and $y_{3}$ we know $T\geq 1$). As $p$ is $1/(n+m-1)$-concave, $p^{1/(n+m-1)}(ty_{1}+(1-t)y_{2})\geq tp^{1/(n+m-1)}(y_{1})+(1-t)p^{1/(n+m-1)}(y_{2})\geq tp^{1/(n+m-1)}(y_{1})$ for $t\in[0,1]$, which means $\int_{I^{\prime\prime}}p(y)dy\geq|I^{\prime\prime}|\int_{[0,1]}t^{n+m-1}p(y_{1})dt=|I^{\prime\prime}|\frac{1}{n+m}p(y_{1}).$ Similarly for $T\geq t\geq 1$ we have $p^{1/(n+m-1)}(ty_{1}+(1-t)y_{2})\leq tp^{1/(n+m-1)}(y_{1})+(1-t)p^{1/(n+m-1)}(y_{2})\leq tp^{1/(n+m-1)}(y_{1})$ for $t\in[1,T]$, which gives $\int_{J}p(y)\leq|I^{\prime\prime}|\int_{1}^{(|J|+|I^{\prime\prime}|)/|I^{\prime\prime}|}t^{n+m-1}p(y_{1})=|I^{\prime\prime}|\frac{1}{n+m}\Big{(}\Big{(}\frac{a+b}{a-b}\Big{)}^{n+m}-1\Big{)}p(y_{1}).$ Thus $\frac{\int_{C^{\prime}}f(x)dx}{\int_{C}f(x)dx}\leq\Big{(}\frac{a+b}{a-b}\Big{)}^{n+m}-1,$ which proves the first part of the Lemma. For the second part note that if $a/b\leq 2^{m+n+2}$, then the thesis is true, as $\int_{C^{\prime}}f(x)dx\leq\int_{C}f(x)dx$ because $C^{\prime}\subset C$. For $b/a\leq 2^{-(n+m+1)}$ we have $\Big{(}\frac{a+b}{a-b}\Big{)}^{n+m}-1=\Big{(}\frac{1+b/a}{1-b/a}\Big{)}^{n+m}-1\leq\frac{1+2^{n+m}b/a}{1-2^{n+m}b/a}-1\leq\frac{2^{n+m+1}b/a}{1/2}=2^{n+m+2}b/a.$ ∎ ###### Corollary 5.9. Let ${\varepsilon}>0$. Let $\mu$ be a measure on ${\mathbb{R}}^{2}$ with a $1/m$ concave density. Let ${\tilde{D}}\subset{\mathbb{R}}^{2}$ be a lens set. Let $L$ be the extremal line of ${\tilde{D}}$ and $p:{\mathbb{R}}^{2}{\rightarrow}L$ the orthogonal projection onto $L$. Let $A$ be a c-set in ${\mathbb{R}}^{2}$. Let $A^{\prime}=p^{-1}(A\cap L)$. Assume the relevant length of ${\tilde{D}}$ (that is, the length of $L\cap{\tilde{D}}\cap{\rm supp}\mu$) is at least $d>0$, the inclination of ${\tilde{D}}$ between $\beta$ and $\pi/2-\beta$ with $\beta>0$ and width at most $w=\frac{1}{2\max(\cot\beta,\tan\beta)}2^{-m-3}{\varepsilon}d$. Then $\mu((A\bigtriangleup A^{\prime})\cap{\tilde{D}})\leq{\varepsilon}\mu({\tilde{D}})$. ###### Proof. Let $p=(x_{p},y_{p})$ be the rightmost point on $L\cap A$ (and at the same time on $L\cap A^{\prime}$, from the definition of $A^{\prime}$). As both $A$ and $A^{\prime}$ are c-sets, we have $A\bigtriangleup A^{\prime}\subset\\{(x,y):x>x_{p},y<y_{p}\\}\cup\\{(x,y):x<x_{p},y>y_{p}\\}$. As $D$ has width at most $w$, the projection of $(A\bigtriangleup A^{\prime})\cap D$ onto $L$ has length at most $2w\max(\tan\beta,\cot\beta)$. From Lemma 5.8 we know that as $2w\max(\tan\beta,\cot\beta)<2^{-m-3}{\varepsilon}d$, we have $\mu((A\bigtriangleup A^{\prime})\cap D)\leq{\varepsilon}\mu(D)$. ∎ ###### Corollary 5.10. Let $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ be a generalized Orlicz ball with a proper measure $\mu$ and $D$ be a lens set of relevant length at least $d$, inclination between $\beta$ and $\pi/2-\beta$ and width at most $w=\frac{1}{2\max(\cot\beta,\tan\beta)}2^{-m-3}{\varepsilon}d(\lambda_{n-2}(K\cap\\{x=0,y=0\\}))^{-1}$, where $m$ is such that the density of $\mu$ is $1/m$ concave. Let $A$ be a c-set in ${\mathbb{R}}^{n}$ and let $A^{\prime}$ be defined as before. Then $\mu((A\bigtriangleup A^{\prime})\cap D)\leq{\varepsilon}\mu_{2}({\tilde{D}}).$ ###### Proof. For each $t\in{\mathbb{R}}^{n-2}$ we may apply Corollary 5.9, and integrate over $K\cap\\{x=0,y=0\\}$ to get a bound for the Lebesgue measure. ∎ Note the same argument works if $A$ is the complement of a c-set. ###### Corollary 5.11. Let $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ be a generalized Orlicz ball with a proper measure $\mu$. Let $D$ be a lens set of relevant length at least $d$, inclination between $\beta$ and $\pi/2-\beta$ and width at most $w=\frac{1}{2\max(\cot\beta,\tan\beta)}2^{-m-3}dM^{-1}{\varepsilon}(\lambda_{n-2}(K\cap\\{x=0,y=0\\}))^{-1}$ and $\phi:{\mathbb{R}}^{n}{\rightarrow}[0,M]$ be a coordinate-wise decreasing function with $\bar{\phi}(t):=\phi(p(t))$, where $p$ is the orthogonal projection onto $L\times{\mathbb{R}}^{n-2}$, $L$ being the extremal line of ${\tilde{D}}$. Then for any $U\subset D$ we have $\big{|}\int_{U}\phi(t)d\mu(t)-\int_{U}\bar{\phi}(t)d\mu(t)\big{|}<{\varepsilon}\mu_{2}({\tilde{D}}).$ ###### Proof. As $\phi$ is coordinate-wise decreasing, the sets $\phi^{-1}([s,\infty))$ are c-sets. By the integration by parts, $\int_{U}\phi(t)d\mu(t)=\int_{0}^{M}\mu(\phi^{-1}([s,\infty)\cap U)ds.$ The sets $(\bar{\phi})^{-1}([s,\infty))$ are formed from the sets $\phi^{-1}([s,\infty))$ as in Corollary 5.10. Thus for each $s$ we have $\displaystyle\Bigg{|}\mu\Big{(}\phi^{-1}([s,\infty)\cap U\Big{)}-\mu\Big{(}(\bar{\phi})^{-1}([s,\infty)\cap U\Big{)}\Bigg{|}$ $\displaystyle\leq\mu\Big{(}\big{(}\phi^{-1}([s,\infty)\cap U\big{)}\bigtriangleup\big{(}(\bar{\phi})^{-1}([s,\infty)\cap U\big{)}\Big{)}$ $\displaystyle\leq\mu\Big{(}\big{(}\phi^{-1}([s,\infty)\cap D\big{)}\bigtriangleup\big{(}(\bar{\phi})^{-1}([s,\infty)\cap D\big{)}\Big{)}\leq M^{-1}{\varepsilon}\mu_{2}({\tilde{D}}),$ which integrated over $[0,M]$ gives the thesis. ∎ ###### Lemma 5.12. Consider a generalized Orlicz ball $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ with a proper measure $\mu$ with both its defining functions $1/m$ concave, a strict non-degenerate $\Theta$ function, any ${\varepsilon}>0$ and any c-set $A$. Assume Theorem 5.1 holds for $n^{\prime}<n$. Let $D$ be a lens set satisfying $\theta(D)=\theta(K)$ of relevant length at least $\delta$, inclination between $\beta$ and $\frac{\pi}{2}-\beta$ and width at most $w=\frac{1}{2\max(\cot\beta,\tan\beta)}2^{-3m-4}d\min\\{1,(\sup_{K}\eta_{1})^{-1}\\}{\varepsilon}(\lambda_{n-2}(K\cap\\{x=0,y=0\\}))^{-1}.$ Then $D$ is an $8{\varepsilon}$-appropriate set. ###### Proof. Let $L$ be the extremal line of $D$. We switch coordinates in the plane ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$ to orthogonal coordinates $(u,v)$ such that $L=\\{v=0\\}$ and $u>0$ on the positive quadrant of ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$. Define for any set $U\in\\{A,{\bar{A}}\\}$ the set $U^{\prime}$ by $U\cap\\{v=0\\}$ and $U^{\prime\prime}$ by $U\times{\mathbb{R}}_{v}$. Let $K^{\prime}$ be a generalized Orlicz ball in ${\mathbb{R}}_{u}\times{\mathbb{R}}^{n-2}$ such that $K_{+}^{\prime}=K_{+}\cap\\{v=0\\}$ given by Lemma 4.19 and $K^{\prime\prime}=K^{\prime}\times{\mathbb{R}}_{v}$. Let $\eta_{i}^{\prime}$ be the restriction of $\eta_{i}$ to $\\{v=0\\}$ and $\eta_{i}^{\prime\prime}(u,v,t)=\eta_{i}(u,0,t)$. Let $\mu^{\prime}$ be the measure on $K^{\prime}$ with density $h(u)=\int_{{\mathbb{R}}_{v}}{\mathbf{1}}_{(u,v)\in{\tilde{D}}}f(u,v)g(u,v)$, where $f$ and $g$ are the density functions defining $\mu$, and $\mu^{\prime\prime}$ be the measure on ${\mathbb{R}}^{n}$ with density $f(x)g(y)$ (without restricting to $K$). We want to prove that $\int_{U^{\prime}\cap K^{\prime}}\eta_{i}^{\prime}d\mu^{\prime}$ is a good approximation of $\int_{U\cap D}\eta_{i}d\mu$, then check the assumptions for Theorem 5.1 on $K^{\prime}$ and apply it for $A^{\prime}$. First note that $\int_{K^{\prime}}\phi(u,t)d\mu^{\prime}(u,t)=\int_{K^{\prime\prime}\cap D}\phi(u,0,t)d\mu^{\prime\prime}(u,v,t)$ for any function $\phi$ defined on $K^{\prime}$. This follows directly from the definitions of $K^{\prime\prime}$, $\mu^{\prime}$ and $\mu^{\prime\prime}$. Let $M=\min\\{1,(\sup_{K}\eta_{1})^{-1}\\}$. We know $K_{+}$ is a c-set, thus $\mu((K\bigtriangleup K^{\prime\prime})\cap D)\leq{\varepsilon}\mu_{2}({\tilde{D}})$ by Corollary 5.10. Thus for any $\phi$ we have $\displaystyle\Bigg{|}\int_{K^{\prime}}\phi(u,t)d\mu^{\prime}(u,t)$ $\displaystyle-\int_{K\cap D}\phi(u,0,t)d\mu(u,v,t)\Bigg{|}=$ $\displaystyle\Bigg{|}\int_{K^{\prime\prime}\cap D}\phi(u,t)d\mu^{\prime\prime}(u,v,t)-\int_{K\cap D}\phi(u,0,t)d\mu^{\prime\prime}(u,v,t)\Bigg{|}<M{\varepsilon}\mu_{2}({\tilde{D}})\sup|\phi|.$ We repeat the same trick for $U\in\\{A^{\prime\prime},{\bar{A}}^{\prime\prime}\\}$, putting $\phi^{\prime}=\phi\cdot{\mathbf{1}}_{U}$ in the above inequality and applying Corollary 5.10 again to get $\displaystyle\Bigg{|}\int_{K^{\prime}\cap A^{\prime}}\phi(u,t)d\mu^{\prime}(u,t)-\int_{K\cap D\cap A}\phi(u,0,t)d\mu(u,v,t)\Bigg{|}<M{\varepsilon}\mu_{2}({\tilde{D}})\sup|\phi|.$ Finally, we insert $\eta_{i}$ for $\phi$ and apply Corollary 5.11 to get $\Bigg{|}\int_{K^{\prime}\cap A^{\prime}}\eta_{i}^{\prime}(u,t)d\mu^{\prime}(u,t)-\int_{K\cap D\cap A}\eta_{i}(u,v,t)d\mu(u,v,t)\Bigg{|}\leq 3{\varepsilon}\mu_{2}({\tilde{D}}),$ and the same for integration over $K\cap D\cap{\bar{A}}$ and $K\cap D$. Now we want to check assumptions for Theorem 5.1. $K^{\prime}$ is a generalized Orlicz ball due to Lemma 4.19. $A^{\prime}$ is a c-set in ${\mathbb{R}}_{u}\times{\mathbb{R}}^{n-2}$ because $L$ is positively inclined, thus an increase in $u$ translates to an increase in both $x$ and $y$. $\mu^{\prime}$ is a projection of the measure with the density $f(x)g(y){\mathbf{1}}_{D}$. The first two functions are $1/m$ concave, the third is $1/1$ concave as $D$ is convex. Thus from Facts 4.23 and 4.24 the density $h(u)$ of $\mu^{\prime}$ is a $1/(3m+1)$ concave function. Thus $\mu^{\prime}$ is a proper measure on $K^{\prime}$ (recall $\mu^{\prime}$ is restricted to $K^{\prime}$, thus the third point of the Definition 4.2 is satisfied). $\eta_{1}^{\prime}$ and $\eta_{2}^{\prime}$ are restrictions of $\eta_{1}$ and $\eta_{2}$ to $K^{\prime}$, which is a derivative of $K$, thus they define a non-degenerate $\Theta$-function on $K^{\prime}$. Let us apply Theorem 5.1. We get $\frac{\int_{K^{\prime}\cap A^{\prime}}\eta_{2}^{\prime}(u,t)d\mu^{\prime}(u,t)}{\int_{K^{\prime}\cap A^{\prime}}\eta_{1}^{\prime}(u,t)d\mu^{\prime}(u,t)}\geq\frac{\int_{K^{\prime}}\eta_{2}^{\prime}(u,t)d\mu^{\prime}(u,t)}{\int_{K^{\prime}}\eta_{1}^{\prime}(u,t)d\mu^{\prime}(u,t)}\geq\frac{\int_{K^{\prime}\cap{\bar{A}}^{\prime}}\eta_{2}^{\prime}(u,t)d\mu^{\prime}(u,t)}{\int_{K^{\prime}\cap{\bar{A}}^{\prime}}\eta_{1}^{\prime}(u,t)d\mu^{\prime}(u,t)}.$ (5.3.1) We need to make the middle expression equal to $\theta(D)$, so for any $u_{0},t_{0}$ we define $\bar{\eta}_{i}^{\prime}(u_{0},t_{0})=\frac{\int_{K\cap D}\eta_{i}(u,v,t)d\mu(u,v,t)}{\int_{K^{\prime}}\eta_{i}^{\prime}(u,t)d\mu^{\prime}(u,t)}\eta_{i}^{\prime}(u_{0},t_{0}).$ As $\bar{\eta}_{i}^{\prime}=C_{i}\eta_{i}^{\prime}$, we have inequalities (5.3.1) for functions $\bar{\eta}_{i}^{\prime}$ (although they do not necessarily define a $\Theta$ function on $K^{\prime}$). To bound the error we have $\displaystyle\int_{K^{\prime}}\big{|}\bar{\eta}_{i}^{\prime}(u,t)-\eta_{i}^{\prime}(u,t)\big{|}d\mu^{\prime}(u,t)$ $\displaystyle=\int_{K^{\prime}}\eta_{i}^{\prime}(u,t)\Big{|}\frac{\int_{K\cap D}\eta_{i}(u,v,t)d\mu(u,v,t)}{\int_{K^{\prime}}\eta_{i}^{\prime}(u,t)d\mu^{\prime}(u,t)}-1\Big{|}d\mu^{\prime}(u,t)$ $\displaystyle=\Bigg{|}\int_{K\cap D}\eta_{i}(u,v,t)d\mu(u,v,t)-\int_{K^{\prime}}\eta_{i}^{\prime}(u,t)d\mu^{\prime}(u,t)\Bigg{|}\leq 3{\varepsilon}\mu_{2}({\tilde{D}}).$ As we bounded the integral of errors, the error on $K^{\prime}\cap A^{\prime}$ and $K^{\prime}\cap{\bar{A}}^{\prime}$ is no larger than $3{\varepsilon}\mu_{2}({\tilde{D}})$. We can now for $U\in\\{A^{\prime},{\bar{A}}^{\prime}\\}$ and $i\in\\{1,2\\}$ put $C_{U,i}=\int_{K^{\prime}\cap U}\bar{\eta}_{i}^{\prime}d\mu^{\prime}$. Applying inequalities (5.3.1) to $\bar{\eta}_{i}^{\prime}$ we get $\frac{C_{A,2}}{C_{A,1}}\geq\frac{\int_{K^{\prime}}{\bar{\eta}}_{2}^{\prime}(u,t)d\mu^{\prime}(u,t)}{\int_{K^{\prime}}\bar{\eta}_{1}^{\prime}(u,t)d\mu^{\prime}(u,t)}=\frac{\int_{K\cap D}\eta_{2}(u,v,t)d\mu(u,v,t)}{\int_{K\cap D}\eta_{1}(u,v,t)d\mu(u,v,t)}=\theta(D)=\theta(K)\geq\frac{C_{{\bar{A}},2}}{C_{{\bar{A}},1}},$ and putting together all the estimates we made we get $|C_{U,i}-\int_{K\cap D\cap U}\eta_{i}d\mu|\leq 6{\varepsilon}\mu_{2}({\tilde{D}})$. ∎ ## 6 The transfinite induction What is left to prove is Theorem 5.5. We will prove by transfinite induction an extended version of Theorem 5.5, which will allow us to carry the information we need through the induction steps. The sets $U(\gamma,\beta)$ will have to satisfy the conditions of Theorem 5.5, and furthermore the following conditions: * • For any $\gamma>\beta$ we have $U(\gamma,\beta)=U(\beta+1,\beta)$. * • For any $\gamma$ we have $U(\gamma,\gamma+1)=U(0,1)$. * • If $\gamma$ is a successor ordinal and $U(\gamma,\gamma)$ has positive $\mu_{2}$ measure, the sets $U(\gamma,\gamma-1)$ and $U(\gamma,\gamma)$ are formed by dividing $U(\gamma-1,\gamma-1)$ with a straight line of positive inclination. * • If $\gamma$ is a limit ordinal, $U(\gamma,\gamma)=\bigcap_{\beta<\gamma}U(\beta,\beta)$. * • For any $\gamma$ if $U(\gamma,\gamma)$ has positive $\mu_{2}$ measure, then for all $\beta<\gamma$ the sets $U(\beta,\beta)$ are strict lens sets. Remark that this in fact means we carry out a transfinite inductive construction. The sets $U(\gamma,\beta)$ for $\beta<\gamma$ depend only on the second argument, once constructed. The set $U(\gamma,\gamma+1)$ is equal to $U(0,1)$. The set $U(\gamma,\gamma)$ in each step has a part cut off to make a new set $U(\gamma+1,\gamma+1)$. Note that if $\theta(K)=0$, then $K$ is appropriate (as any $U\subset K$ with $\mu_{2}(U)>0$ has $\theta(U)=0$). Thus by putting $U(\gamma,0)=\tilde{K}_{+}$ for any $\gamma$ and $U(\gamma,\beta)=\emptyset$ for $\gamma+1\geq\beta>0$ we satisfy the conditions of Theorem 5.5. Thus, further on, we assume $\theta(K)>0$. ### 6.1 Starting the transfinite induction First we need to define the sets $U(0,0)$ and $U(0,1)$ to start the induction. If we take $D=[x_{-},x_{+}]\times[y_{-},y_{+}]\times{\mathbb{R}}^{n-2}$, then $D$ is a lens set and satisfies condition (5.1.2). It is not, however, a strict lens set. The idea is to take two almost vertical lines — one close to the left edge of ${\tilde{D}}$ and the other close to the right edge, then look at the $\theta$ of the set they cut off. If $\theta$ is too large, we move the left line closer to the edge, if too small, we move the right line closer to the edge. When we have balanced $\theta$, we repeat the same for horizontal lines. By cutting off a bit from each edge we shall also ensure $[x_{1},x_{2}]\subset(x_{-},x_{+})$ and similarly for $y$. Below is a formalization of the argument. If $\tilde{K}_{+}$ is appropriate to begin with, we take $U(0,1)=\emptyset$ and $U(0,0)=\tilde{K}_{+}$. Thus we assume $\tilde{K}_{+}$ is not appropriate. Denote by $L^{-}(x,\beta)$ the line through $(x,y_{-})$ with inclination $\pi/2-\beta$ and by $L^{+}(x,\beta)$ the line through $(x,y_{+})$ with inclination $\pi/2-\beta$. Denote by ${\tilde{D}}^{-}(x,\beta)$ the subset of $[x_{-},x_{+}]\times[y_{-}\times y_{+}]$ to the left of $L^{-}(x,\beta)$ and by ${\tilde{D}}^{+}(x,\beta)$ the subset to the right of $L^{+}(x,\beta)$. Note that for $\beta\in(0,\pi/2)$ those sets have positive $\mu_{2}$ measure by the definition of a proper measure. Let $\phi^{-}(x,\beta)=\theta({\tilde{D}}^{-}(x,\beta)\times{\mathbb{R}}^{n-2})-\theta(K)$ and $\phi^{+}(x,\beta)=\theta({\tilde{D}}^{+}(x,\beta)\times{\mathbb{R}}^{n-2})-\theta(K)$. From property T5 these functions are continuous in both arguments. From Lemma 4.33 and Lemma 5.7 there is a $\beta_{0}>0$ such that for $\beta<\beta_{0}$ we have $\phi^{-}(x,\beta)>0$ and $\phi^{+}(x,\beta)<0$ for $x\in(x_{-},x_{+})$. Now start with any $x_{l}$, $x_{u}$ and $0<\beta_{l},\beta_{u}<\beta_{0}$ such that the sets ${\tilde{D}}^{-}(x_{l},\beta_{l})$ and ${\tilde{D}}^{+}(x_{u},\beta_{l})$ have measure no larger than ${\varepsilon}^{\prime}(\lambda_{n-2}(K\cap\\{x=0,y=0\\})\sup_{K}\eta_{1})^{-1}\mu(K)/4$ and do not intersect. Now if we fix $x_{u}$ and $\beta_{u}$ while letting $x_{l}$ tend to $x_{-}$ and $\beta_{l}$ to 0, then $\theta$ of the sum of the two sets will tend to $\theta({\tilde{D}}^{+}(x_{u},\beta_{u})\times{\mathbb{R}}^{n-2})$, which is strictly smaller than $\theta(K)$. If, on the other hand, we fix $x_{l}$ and $\beta_{l}$ and let $x_{u}$ tend to $x_{+}$ and $\beta_{u}$ to 0, the $\theta$ of the two sets will approach $\theta({\tilde{D}}^{-}(x_{l},\beta_{l})\times{\mathbb{R}}^{n-2})$, which is strictly greater than $\theta(K)$. Thus, from the Darboux property, for some values $x_{-}<x_{l}<x_{u}<x_{+}$ and $\beta_{l}$ and $\beta_{u}$ we have the function $\theta\Big{(}{\tilde{D}}^{+}(x_{u},\beta_{u})\times{\mathbb{R}}^{n-2}\cup{\tilde{D}}^{-}(x_{l},\beta_{l})\times{\mathbb{R}}^{n-2}\Big{)}=\theta(K).$ The set that remains is a lens set with no vertical boundaries and satisfies property (5.1.2). If it is appropriate, we have found our $U(0,0)$ and define $U(0,1)={\tilde{D}}^{-}(x_{l},\beta_{l})\cup{\tilde{D}}^{+}(x_{u},\beta_{u})$. If not, then we can repeat the same trick for $y$ (we needed the non- appropriateness to use Lemma 5.6), and achieve a lens set with no horizontal and no vertical boundaries and separated from $x_{-}$ and $x_{+}$, i.e. a strict lens set. Thus we define $U(0,1)={\tilde{D}}^{-}(x_{l},\beta_{l})\cup{\tilde{D}}^{+}(x_{u},\beta_{u})\cup{\tilde{D}}^{-}(y_{l},\alpha_{l})\cup{\tilde{D}}^{+}(y_{u},\alpha_{u})$ and and $U(0,0)=([x_{-},x_{+}]\times[y_{-},y_{+}])\setminus U(0,1)$. ###### Remark 6.1. Assume $U(0,0)$ is a strict lens set (otherwise the induction will be trivial). Recall $f$ and $g$ are $1/m$-concave functions defining the proper measure $\mu$. As $U(0,0)$ is a strict lens set, it is separated from the boundary of the support of $f\cdot g$. Thus (as $f$ and $g$ are continuous on the interior of their support), they both attain positive minimal values $f_{L}$ and $g_{L}$. Also, as they are continuous on their support and $1/m$ concave, they are bounded from above by some $f_{U}$ and $g_{U}$. Thus for any set $T\subset U(0,0)$ we have $f_{U}g_{U}\lambda_{2}(T)\geq\mu_{2}(T)\geq f_{L}g_{L}\lambda_{2}(T),$ and for any function $t$ on $T$ we have $f_{U}g_{U}\int_{T}t(p)d\lambda_{2}(p)\geq\int_{T}t(p)d\mu_{2}(p)\geq f_{L}g_{L}\int_{T}t(p)d\mu_{2}(p).$ ### 6.2 The induction step for successor ordinals For a successor ordinal $\gamma+1$ we have a division of $\tilde{K}_{+}$ for $\gamma$. We put $U(\gamma+1,\gamma+2)=U(\gamma,\gamma+1)$. If $U(\gamma,\gamma)$ is appropriate of positive measure, we put $U(\gamma+1,\gamma)=U(\gamma,\gamma)$ (as an appropriate set is an ${\varepsilon}$-appropriate set) and $U(\gamma+1,\gamma+1)=\emptyset$. If $U(\gamma,\gamma)$ has measure 0, we put $U(\gamma+1,\gamma)=\emptyset$ and $U(\gamma+1,\gamma+1)=U(\gamma,\gamma)$. The difficult case to deal with will be when $U(\gamma,\gamma)$ is a non-appropriate strict lens set. For brevity denote $U(\gamma,\gamma)$ by ${\tilde{D}}$. In this case from Lemma 5.7 there exists an angle $\alpha^{\prime}>0$ such that any positively inclinated line dividing ${\tilde{D}}$ into two sets of non-zero $\mu$-measure with equal $\theta$ has inclination greater than $\alpha^{\prime}$ and smaller than $\frac{\pi}{2}-\alpha^{\prime}$. If the inclination of ${\tilde{D}}$ is $\alpha^{\prime\prime}$, let $\alpha=\min\\{\alpha^{\prime},\alpha^{\prime\prime},\frac{\pi}{2}-\alpha^{\prime\prime}\\}$. We shall attempt to cut off a “long and narrow” lens set $U(\gamma+1,\gamma)$ from $U(\gamma,\gamma)$. We shall cut off a narrow set satisfying (5.1.2). From Remark 5.4 it will either be long, or be a subset of ${\tilde{S}}_{\delta}$, both of which satisfy us. Take a sufficiently small $w$ ($w<\frac{1}{2\max(\cot\alpha,\tan\alpha)}2^{-3m-4}\delta\min\\{1,(\sup_{K}\eta_{1})^{-1}\\}\frac{{\varepsilon}}{8}(\lambda_{n-2}(K\cap\\{x=0,y=0\\})^{-1}$, where $m$ is such that the density functions of $\mu$ are $1/m$-concave, will suffice). For any angle $\xi\in[0,\frac{\pi}{2}]$ we can find continuously a line $L_{\xi}$ of inclination $\xi$ such that the part ${\tilde{D}}_{+}(\xi)$ of ${\tilde{D}}\cap{\rm supp}\mu$ lying above and to the left of $L_{\xi}$ has width no larger than $w$. From Lemma 4.33 we have $\theta(D_{+}(0))\geq\theta(D)$ and $\theta(D_{+}(\pi/2))\leq\theta(D)$. From the Darboux property for some $\xi$ we have $\theta(D_{+}(\xi))=\theta(D)$. We take $U(\gamma+1,\gamma)={\tilde{D}}_{+}(\xi)$. Let $I_{\xi}$ denote the segment of $L_{\xi}$ intersecting ${\tilde{D}}$. The set $U(\gamma+1,\gamma+1)=U(\gamma,\gamma)\setminus U(\gamma+1,\gamma)$ is, of course, a strict lens set, satisfying condition (5.1.2), because the new edge has inclination between $\alpha$ and $\frac{\pi}{2}-\alpha$, and all the other edges come from the old set ${\tilde{D}}$. It remains to check that $U(\gamma+1,\gamma)$ satisfies the conditions of the transfinite induction. First let us check what is the inclination of $U(\gamma+1,\gamma)$. If both the ends $I_{\xi}$ fall upon the upper-left border of $U(\gamma,\gamma)$, then they are the extremal points of $U(\gamma+1,\gamma)$, and thus the inclination of $U(\gamma+1,\gamma)$ is the inclination of the segment, which is between $\alpha^{\prime}$ and $\frac{\pi}{2}-\alpha^{\prime}$. If one of them falls upon the lower-right border, then the extremal points of $U(\gamma+1,\gamma)$ are the end of $I_{\xi}$ on the upper-left border and one of the extremal points of $U(\gamma,\gamma)$, and the inclination of $U(\gamma+1,\gamma)$ is between the inclination of $U(\gamma+1,\gamma)$ and the inclination of the segment, which means it is between $\alpha$ and $\frac{\pi}{2}-\alpha$. If both ends fall upon the lower-right border, the extremal points of $U(\gamma+1,\gamma)$ are the extremal points of $U(\gamma,\gamma)$, which means $U(\gamma+1,\gamma)$ has inclination $\alpha^{\prime\prime}$. Thus, the inclination of $U(\gamma,\gamma)$ is between $\alpha$ and $\frac{\pi}{2}-\alpha$. If $U(\gamma+1,\gamma)\subset{\tilde{S}}_{\delta}$, the induction thesis is satisfied. Thus we may assume $U(\gamma+1,\gamma)$ sticks outside ${\tilde{S}}_{\delta}$. Note that as $\theta_{n-2}(p)$, $p\in{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}$, is a coordinate-wise increasing function from property (T4), one of the extremal points of $U(\gamma+1,\gamma)$ has to lie outside ${\tilde{S}}_{\delta}$, and at least one point of ${\tilde{S}}$ lies on the extremal line of $U(\gamma+1,\gamma)$. Thus, the length of the segment of the extremal line contained in $\tilde{K}_{+}$ is at least $\delta$. Thus $U(\gamma+1,\gamma)$ has relevant length at least $\delta$, width at most $w$ and inclination between $\alpha$ and $\frac{\pi}{2}-\alpha$. Thus from Lemma 5.12 we know that $U(\gamma+1,\gamma)$ is ${\varepsilon}$-appropriate, which means we completed the induction step. ### 6.3 The induction step for limit ordinals For limit ordinals $\gamma$ the set $U(\gamma,\gamma+1)=U(0,1)$, the sets $U(\gamma,\beta)$ for $\beta<\gamma$ are defined by $U(\gamma,\beta)=U(\beta+1,\beta)$, and from the inductive assumption the conditions for $U(\gamma,\beta)$ are met. We define $U(\gamma,\gamma)$ as the intersection $\bigcap_{\beta<\gamma}U(\beta,\beta)$. We have to check that $U(\gamma,\gamma)$ thus defined satisfies the induction thesis. If any of the sets $U(\gamma^{\prime},\beta),\beta<\gamma^{\prime}$ was empty, then from the inductive assumption $U(\gamma^{\prime}+1,\gamma^{\prime}+1)$ has $\mu_{2}$ measure 0 and thus $U(\gamma,\gamma)$ has $\mu_{2}$ measure 0, which satisfies the conditions. If $U(\beta,\beta)$ was not a strict lens set for some $\beta<\gamma$, then $U(\gamma,\gamma)$ has measure 0, again satisfying the conditions. The case to worry about is when $U(\gamma,\gamma)$ is a intersection of a descending family of strict lens sets satisfying condition (5.1.2) and has a positive $\mu_{2}$ measure. A descending intersection of lens sets is a lens set — the circumscribed rectangle is the intersection of circumscribed rectangles, the extremal points belong to the intersection, and the intersection is convex. A descending intersection of sets satisfying (5.1.2) with positive $\mu_{2}$ measure satisfies (5.1.2) by property (T5). We have to prove that the intersection is either a strict lens set, or appropriate. As $U(\gamma,\gamma)\subset U(0,0)$, it is separated from $x_{-},x_{+},y_{-}$ and $y_{+}$. Thus we only have to check it does not have a horizontal or vertical edge. Suppose $U(\gamma,\gamma)$ has a horizontal or vertical edge $I$. We may assume, without loss of generality, that $I$ is a horizontal edge. We will assume it is an upper horizontal edge. In the case of the lower one, the proof goes very similarily: every construction of new points is done centrally- symetric, and every inequality is opposite. In one place, where the proof significantly changes, we will say it explicitly. Let $(x_{0},y_{0})$ be the left end of $I$ and $(x_{1},y_{0})$ the right end. First we shall prove the following Lemma: ###### Lemma 6.2. With the notation as previously we have ${\rm cl}I\cap{\rm supp}\eta_{2}\neq\emptyset$. ###### Proof. We shall prove the Lemma by contradiction. Suppose that ${\rm cl}I\cap{\rm supp}\eta_{2}=\emptyset$. The idea of the proof is that at some moment, a line dividing some $U(\beta,\beta)$ into $U(\beta+1,\beta+1)$ and $U(\beta+1,\beta)$ lies above $I$ and cuts off only points that are above and to the right of the left end of $I$, or almost so, and thus only cuts off points, which do not belong to ${\rm supp}\eta_{2}$. Thus $\eta_{2}$ is zero on the set $U(\beta+1,\beta)$ which was cut off, $\theta(U(\beta+1,\beta))=0$, a contradiction. Now for a formal proof: As $\theta(U(\gamma,\gamma))>0$, some point of $U(\gamma,\gamma)$ has to lie inside ${\rm supp}\eta_{2}$, thus (as ${\rm supp}\eta_{2}$ is a c-set), the lower left extremal point of $U(\gamma,\gamma)$ lies in ${\rm supp}\eta_{2}$. Note, that as $\eta_{2}=0$ on $I$, $I$ has to be an upper edge, the lower edge case is trivial here. Let $x_{2}<x_{0}$ be such that $(x_{2},y_{0})\not\in{\rm supp}\eta_{2}$. Then let $y_{2}<y_{0}$ be a number so close to $y_{0}$ that $(x_{2},y_{2})\not\in{\rm supp}\eta_{2}$ and $(x_{2},y_{2})\not\in U(\gamma,\gamma)$. Take a $\beta<\gamma$ such that $(x_{2},y_{2})\not\in U(\beta,\beta)$. As $U(\beta,\beta)$ is a lens set, no points $(x_{2},y)$ with $y>y_{2}$ belong to $U(\beta,\beta)$. As $U(\beta,\beta)$ is a strict lens set, and $I\subset U(\beta,\beta)$, there exists a $y_{3}>y_{0}$ such that $(x_{1},y_{3})\in U(\beta,\beta)$. Take $y_{3}$ to be so small that $\frac{y_{3}-y_{0}}{x_{1}-x_{0}}<\frac{y_{0}-y_{2}}{x_{0}-x_{2}}.$ (6.3.1) Let $\beta^{\prime}$ be the smallest such ordinal that $(x_{1},y_{3})\not\in U(\beta^{\prime},\beta^{\prime})$. Of course $\beta^{\prime}>\beta$ and from the inductive assumption $\beta^{\prime}$ is a successor ordinal. Let $L$ be the line which divides $U(\beta^{\prime}-1,\beta^{\prime}-1)$ into $U(\beta^{\prime},\beta^{\prime})$ and $U(\beta^{\prime},\beta^{\prime}-1)$. $L$ intersects the interval $[(x_{1},y_{0}),(x_{1},y_{3})]$ and does not intersect $I$, so, from (6.3.1), $L$ intersects the line $x=x_{2}$ at some point above $(x_{2},y_{2})$. $U(\beta^{\prime},\beta^{\prime}-1)\subset U(\beta^{\prime}-1,\beta^{\prime}-1)\subset U(\beta,\beta)$, thus $U(\beta^{\prime},\beta^{\prime}-1)$ contains no points $(x_{2},y)$ with $y>y_{2}$. Thus all points from $U(\beta^{\prime},\beta^{\prime}-1)$ lie above and to the right of $(x_{2},y_{2})$. As ${\rm supp}\eta_{2}$ is a c-set and $(x_{2},y_{2})\not\in{\rm supp}\eta_{2}$, we have $U(\beta^{\prime},\beta^{\prime}-1)\cap{\rm supp}\eta_{2}=\emptyset$, thus $\theta(U(\beta^{\prime},\beta^{\prime}-1))=0$. But as we assumed $\theta(K)>0$ this means that $U(\beta^{\prime},\beta^{\prime}-1)$ is empty, a contradiction. ∎ Thus we know that ${\rm cl}I\cap{\rm supp}\eta_{2}\neq\emptyset$, and as ${\rm Int}\ {\rm supp}\eta_{1}\supset{\rm supp}\eta_{2}$, there is an interval $I^{\prime}\subset I\cap\tilde{K}_{+}$ of positive length, which means $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$ is defined. The idea of the proof in this case is to prove that $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})=\theta(K)$, which from Lemma 4.33 and Lemma 5.6 will imply $U(\gamma,\gamma)$ is appropriate. We prove this by selecting a moment at which the set $U(\beta+1,\beta)$ which is being cut off lies above $I$, and comparing its $\theta$ (which we know to be $\theta(K)$) to $\theta_{n-1}(x_{0};I\times{\mathbb{R}}^{n-2})$. The formal proof goes as follows: We assume $I$ is an upper horizontal edge. In the case of $I$ being a lower horizontal edge, the below construction works centrally-symetrically.Recall $(x_{0},y_{0})$ be the left end of $I$ and $(x_{1},y_{0})$ the right end. Take any $0<{\varepsilon}<|I|$. Take $x_{2}=x_{0}-{\varepsilon}$ and $y_{2}<y_{0}$ and close enough that $(x_{2},y_{2})\not\in U(\gamma,\gamma)$. Take $\beta_{1}<\gamma$ such that $(x_{2},y_{2})\not\in U(\beta_{1},\beta_{1})$. Next take a point $(x_{1},y_{3})$ with $y_{3}>y_{0}$ such that (6.3.1) is satisfied, and take $\gamma>\beta_{2}>\beta_{1}$ such that the upper extremal point of $U(\beta_{2},\beta_{2})$ lies below $y_{3}$. Again, as in the proof of Lemma 6.2, any line dividing some $U(\beta,\beta)$ for $\beta>\beta_{2}$ and crossing $x=x_{1}$ between $y_{3}$ and $y_{0}$ will exit $U(\beta,\beta)$ at some $x>x_{2}$. For $\gamma>\beta>\beta_{2}$ any line cutting off the upper extremal point $p$ of $U(\beta,\beta)$ will cross $x=x_{1}$ between $y_{3}$ and $y_{0}$ because $p$ will lie below $y_{3}$ (as $U(\beta,\beta)\subset U(\beta_{2},\beta_{2})$ and to the right of and above $(x_{1},y_{0})$ as $U(\beta,\beta)\subset U(\gamma,\gamma)$ and the line has to go below $p$ and above $(x_{1},y_{0})$ as $(x_{1},y_{0})\in U(\beta,\beta)$. Let us consider the functions ${\tilde{\eta}}_{i}(x,y)=\int_{{\mathbb{R}}^{n-2}}\eta_{i}(x,y,t)dt$ for $i=1,2$. The set $[x_{-},x_{+}]\times[y_{-},y_{+}]$ is compact and ${\tilde{\eta}}_{i}$ are continuous from property (S4) (recall $n>2$), thus we can find a $\tilde{\delta}>0$ such that $\|p_{1}-p_{2}\|<\tilde{\delta}\ \ {\Rightarrow}\ \ |{\tilde{\eta}}_{i}(p_{1})-{\tilde{\eta}}_{i}(p_{2})|<{\varepsilon}$ for $i=1,2$. Also, as $g$ (the density of $\mu$ with respect to $y$) is $1/m$-concave, it is continuous on the interior of its support, and thus we can take $\tilde{\delta}$ such that also $|g(p_{1})-g(p_{2})|<{\varepsilon}$. If ${\tilde{\eta}}_{1}((x_{1},y_{0}))>0$ take $\delta=\tilde{\delta}$. If not, then as ${\rm Int}\tilde{K}_{+}\supset{\rm supp}{\tilde{\eta}}_{2}$, there exists an interval $J^{\prime}\subset I\cap({\rm Int}\tilde{K}_{+}\setminus{\rm supp}{\tilde{\eta}}_{2})$ of positive length $c$. As ${\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\setminus{\rm Int}\tilde{K}_{+}$ and ${\rm supp}{\tilde{\eta}}_{2}$ are closed, we may take $\delta\leq\tilde{\delta}$ small enough, that there exists an interval $J\subset I$ of length at least $c/2$, such that $J\times[y_{0}-\delta,y_{0}+\delta]\subset{\rm Int}\tilde{K}_{+}\setminus{\rm supp}{\tilde{\eta}}_{2}.$ Take $\gamma>\beta_{3}>\beta_{2}$ such that the whole set $U(\beta_{3},\beta_{3})$ lies below the line $y=y_{0}+\delta$. Now let $(x_{4},y_{4})$ be the upper right extremal point of $U(\beta_{3},\beta_{3})$. Let $\beta_{4}$ be the first $\beta$ such that $(x_{4},y_{4})\not\in U(\beta_{4},\beta_{4})$. The ordinal $\beta_{4}$ has to be a successor, let $L^{\prime}$ be the line dividing $U(\beta_{4}-1,\beta_{4}-1)$ into $U(\beta_{4},\beta_{4}-1)$ and $U(\beta_{4},\beta_{4})$, and let $l$ be the inclination of $L^{\prime}$. Any tangent to the upper-left border of $U(\beta_{4},\beta_{4})$ has inclination no smaller than $l$. Let $\beta_{5}$ be the first ordinal greater than $\beta_{4}$ for which some tangent to the upper left edge of $U(\beta_{5},\beta_{5})$ has inlination strictly smaller than $l$. Again, $\beta_{5}$ has to be a successor ordinal. Let $L$ be the line dividing $U(\beta_{5}-1,\beta_{5}-1)$ into $U(\beta_{5},\beta_{5}-1)$ and $U(\beta_{5},\beta_{5})$. This line has to go above $I$, to become a part of the upper edge of $U(\beta_{5},\beta_{5})$. As the inclination of this line is smaller than the inclination of any tangent to the upper left edge of $U(\beta_{5}-1,\beta_{5}-1)$, the right end of $L\cap U(\beta_{5}-1,\beta_{5}-1)$ lies on the lower right edge of $U(\beta_{5}-1,\beta_{5}-1)$. It lies above $y_{0}$, as it goes above $I$ and has positive inclination, and lies to the right of $x_{1}$, as the lower right edge of $u(\beta_{5}-1,\beta_{5}-1)$ above $y_{0}$ lies to the right of $x_{1}$. Now we will prove some inequalities on $\theta$. In the case of $I$ being lower edge, the inequalities are simply reversed. Let ${\tilde{D}}={\tilde{D}}({\varepsilon})$ be the part of $U(\beta_{5},\beta_{5}-1)$ that lies to the left of $x=x_{1}$. As usual, $D=D({\varepsilon})={\tilde{D}}({\varepsilon})\times{\mathbb{R}}^{n-2}$. As $U(\beta_{5},\beta_{5}-1)$ is a lens set, from Lemma 4.33 we know $\theta({\tilde{D}}\times{\mathbb{R}}^{n-2})\leq\theta(U(\beta_{5},\beta_{5}-1)\times{\mathbb{R}}^{n-2})=\theta(K).$ Remark that the line $L^{\prime\prime}$ that cut $(\beta_{5},\beta_{5}-1)$ off contains the whole lower edge of ${\tilde{D}}$. Thus as the inclination of $L^{\prime\prime}$ is smaller than the inclination of the upper edge of ${\tilde{D}}$ the function $x\mapsto\lambda_{1}({\tilde{D}}_{x})$, where ${\tilde{D}}_{x}$ is the section of ${\tilde{D}}$ at $x$, is strictly increasing. If ${\tilde{D}}$ has $\mu_{2}$ measure 0, then the lower extremal point $p_{5}$ of $U(\beta_{5},\beta_{5}-1)$ lies above and to the right of any point of $U(\gamma,\gamma)$. However, from property (T4) $\theta_{n-2}(p_{5})\leq\theta(U(\beta_{5},\beta_{5}-1)\times{\mathbb{R}}^{n-2})=\theta(K),$ which means that from property (T4) for any point $p\in U(\gamma,\gamma)$ we have $\theta_{n-2}(p)\leq\theta_{n-2}(p_{5})\leq\theta(K).$ However, we know $\theta(U(\gamma,\gamma)\times{\mathbb{R}}^{n-2})=\theta(K)$, which, from Fact 2.5 implies that for almost all points in $U(\gamma,\gamma)$ we have $\theta_{n-2}(p)=\theta(K)$. Thus any horizontal line divides $U(\gamma,\gamma)$ into two sets with equal $\theta$, which from Lemma 5.6 implies $U(\gamma,\gamma)$ is appropriate. Hereafter we shall assume $\mu_{2}({\tilde{D}})>0$. Note that the whole set ${\tilde{D}}$ lies in the rectangle $[x_{2},x_{1}]\times[y_{0}-\delta,y_{0}+\delta]$. It lies to the left of $x_{1}$ from its definition. To the right of $x_{2}$ as $\beta_{5}>\beta_{2}$. Below $y_{0}+\delta$ because $\beta_{5}>\beta_{3}$. Above $y_{0}-\delta$ because its lower edge is the line $L^{\prime\prime}$ which passes above $(x_{0},y_{0})$, so if it dipped below $y_{0}-\delta$, it would also (as ${\varepsilon}<|I|$) have to reach above $y_{0}+\delta$. Now we want to estimate $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$ by $\theta({\tilde{D}}\times{\mathbb{R}}^{n-2})$. This will, unfortunately, involve quite a lot of technicalities. We begin with a lemma: ###### Lemma 6.3. There exist two numbers $c_{1},c_{2}>0$ independent of ${\varepsilon}$ such that for sufficiently small ${\varepsilon}>0$ and a set ${\tilde{D}}$ constructed as above for this ${\varepsilon}$ we have $\lambda_{2}({\tilde{D}}\cap\\{(x,y):{\tilde{\eta}}_{i}(x,y)>c_{1}\\})>c_{2}\lambda_{2}({\tilde{D}}),$ for $i=1,2$. ###### Proof. The proof for this lemma is a bit different for $I$ being a lower edge. First, let us prove it for an upper edge. First we prove the thesis for ${\tilde{\eta}}_{1}$. Suppose ${\tilde{\eta}}_{1}(x_{1},y_{0})>0$. Then supposing ${\varepsilon}<\frac{1}{2}{\tilde{\eta}}_{1}(x_{1},y_{0})$ for any $(x,y)\in{\tilde{D}}$ we have ${\tilde{\eta}}_{1}(x,y)\geq{\tilde{\eta}}_{1}(x_{1},y)\geq{\tilde{\eta}}_{1}(x_{1},y_{0})-{\varepsilon}>\frac{1}{2}{\tilde{\eta}}_{1}(x_{1},y_{0}),$ as $|y-y_{0}|<\delta$ and ${\tilde{\eta}}_{1}$ is decreasing as $\eta_{1}$ is decreasing, thus it is enough to have $c_{1}<\frac{1}{2}{\tilde{\eta}}_{1}(x_{1},y_{0})$ and $c_{2}<1$. In the case ${\tilde{\eta}}_{1}(x_{1},y_{0})=0$ let $b_{1}=\sup\\{x:{\tilde{\eta}}_{1}(x,y_{0})>0\\}$. Recall that we constructed an interval $J$ of length $c$ (independent of ${\varepsilon}$) such that $J\times[y_{0}-\delta,y_{0}+\delta]\subset{\rm supp}{\tilde{\eta}}_{1}\setminus{\rm supp}{\tilde{\eta}}_{2}$. Let $J=[j_{0},j_{1}]$. Now as ${\tilde{D}}\subset[x_{2},x_{1}]\times[y_{0}-\delta,y_{0}+\delta]$ for $x\in J$ and $(x,y)\in{\tilde{D}}$ we have ${\tilde{\eta}}_{2}(x,y)=0$ and ${\tilde{\eta}}_{1}(x,y)>0$, which means $j_{2}\leq b_{1}$. On the other hand $\theta({\tilde{D}})\geq\theta(K)>0$, thus ${\tilde{D}}$ contains points with positive $\eta_{2}$, and thus for these points $(x,y)$ we have $x<j_{0}$. Note that as $\lambda_{1}({\tilde{D}}_{x})$ is strictly increasing, so if ${\tilde{D}}$ condains some point to the left of $j_{0}$, then for every $x\in J$ the set ${\tilde{D}}_{x}$ has positive Lebesgue measure. Let $j=\frac{j_{0}+j_{1}}{2}$ be the midpoint of $J$. If ${\varepsilon}<\frac{1}{2}{\tilde{\eta}}_{1}(j,y_{0})$ we have $\displaystyle\lambda_{2}\Bigg{(}{\tilde{D}}\cap\Big{\\{}(x,y):{\tilde{\eta}}_{1}(x,y)>\frac{1}{2}{\tilde{\eta}}_{1}(j,y_{0})\Big{\\}}\Bigg{)}$ $\displaystyle\geq\lambda_{2}\Bigg{(}\Big{\\{}(x,y)\in{\tilde{D}}:{\tilde{\eta}}_{1}(x,y_{0})\geq{\tilde{\eta}}_{1}(j,y_{0})\Big{\\}}\Bigg{)}$ $\displaystyle\geq\lambda_{2}\Big{(}\big{\\{}(x,y)\in{\tilde{D}}:x<j\big{\\}}\Big{)}.$ Now we perform a similar operation as in Lemma 5.8. The function $p(x)=\lambda({\tilde{D}}_{x})$ is concave on its support, $p(j_{0})\geq 0$, thus for every $t\in[0,1]$ we have $p((1-t)j_{0}+tj)\geq tp(j)$ and for $t>1$ we have $p((1-t)j_{0}+tj)\leq tp(j)$. Thus $\lambda_{2}\Big{(}\\{(x,y)\in{\tilde{D}}:x<j\\}\Big{)}=\int_{x<j}p(x)\geq|j-j_{0}|\int_{0}^{1}tp(j)=\frac{j-j_{0}}{2}p(j).$ In a similar vein $\lambda_{2}\Big{(}\\{x,y)\in{\tilde{D}}:x\geq j\\}\Big{)}=\int_{x\geq j}p(x)\leq|j-j_{0}|\int_{1}^{\frac{x_{1}-j_{0}}{j-j_{0}}}tp(j)=\frac{j-j_{0}}{2}\Bigg{(}\frac{(x_{1}-j_{0})^{2}}{(j-j_{0})^{2}}-1\Bigg{)}p(j),$ which gives us: $\frac{\lambda_{2}({\tilde{D}})}{\lambda_{2}({\tilde{D}}\cap\\{(x,y):{\tilde{\eta}}_{i}(x,y)>\frac{1}{2}{\tilde{\eta}}_{1}(j,y_{0})\\})}\leq 1+\frac{\lambda_{2}(\\{x,y)\in{\tilde{D}}:x\geq j\\})}{\lambda_{2}(\\{x,y)\in{\tilde{D}}:x<j\\})}\leq 1+\frac{(x_{1}-j_{0})^{2}}{(j-j_{0})^{2}}-1=\frac{(x_{1}-j_{0})^{2}}{(j-j_{0})^{2}},$ which gives the thesis for $c_{1}\leq\frac{1}{2}{\tilde{\eta}}_{1}(j,y_{0})$ and $c_{2}\leq\frac{(j-j_{0})^{2}}{(x_{1}-j_{0})^{2}}$. To deal with ${\tilde{\eta}}_{2}$ first use Remark 6.1 to get $\int_{\tilde{D}}{\tilde{\eta}}_{2}(x,y)d\mu_{2}(x,y)=\theta({\tilde{D}}\times{\mathbb{R}}^{n-2})\int_{{\tilde{D}}}{\tilde{\eta}}_{1}(x,y)d\mu_{2}(x,y)\geq\theta(K)c_{1}c_{2}f_{L}g_{L}\lambda_{2}({\tilde{D}}).$ On the other hand ${\tilde{\eta}}_{2}$ is bounded from above on ${\rm supp}{\tilde{\eta}}_{2}$ by $M={\tilde{\eta}}_{2}(0,0)$, as it is continuous. We have $\displaystyle f_{L}g_{L}c_{1}c_{2}\theta(K)\lambda_{2}({\tilde{D}})$ $\displaystyle\leq\int_{D}{\tilde{\eta}}_{2}(x,y)d\mu_{2}(x,y)\leq f_{U}g_{U}\int_{D}{\tilde{\eta}}_{2}(x,y)d\lambda_{2}$ $\displaystyle\leq f_{U}g_{U}\big{(}M\lambda_{2}({\tilde{D}}\cap\\{{\tilde{\eta}}_{2}(x,y)>a\\})+a\lambda({\tilde{D}})\big{)}.$ The above holds for any $a$. Let us take $2a=\frac{c_{1}c_{2}\theta(K)f_{L}g_{L}}{f_{U}g_{U}}$. Then we have $\lambda_{2}({\tilde{D}}\cap\\{{\tilde{\eta}}_{2}(x,y)>a\\})\geq a\lambda_{2}({\tilde{D}})/M,$ which implies (with the assumption ${\varepsilon}<a/2$) $\lambda({\tilde{D}}\cap\\{{\tilde{\eta}}_{2}(x,y_{0})>a/2\\})\geq\lambda({\tilde{D}}\cap\\{{\tilde{\eta}}_{2}(x,y_{0})>a-{\varepsilon}\\})\geq(a/M)\lambda(D).$ Now, let us assume that $I$ is a lower horizontal edge. The proof is much easier in that case. Since $\theta(U(\gamma,\gamma))>0$, there is a segment $I^{\prime}\subset I$ starting at lower left end of $I$, such that $I^{\prime}\subset{\rm supp}{\tilde{\eta}}_{2}$. Moreover, we can take such $I^{\prime\prime}\subset I^{\prime}$, that on $I^{\prime\prime}$ we have ${\tilde{\eta}}_{2}>c$ for some $c$. Since $x\to\lambda({\tilde{D}}_{x})$ is decreasing on $I$, we have $\lambda_{2}({\tilde{D}}\cap\\{(x,y):{\tilde{\eta}}(x,y)>c\\})\geq\lambda_{2}({\tilde{D}}\cap I^{\prime\prime}\times{\mathbb{R}})\geq\lambda_{2}({\tilde{D}})\frac{|I^{\prime\prime}|}{|I|}.$ ∎ ###### Corollary 6.4. There exists a constant $c_{3}$ such that for all sufficiently small ${\varepsilon}$ we have $\int_{\tilde{D}}{\tilde{\eta}}_{i}(x,y)d\mu_{2}(x,y)\geq c_{3}\mu_{2}({\tilde{D}}).$ ###### Proof. $\int_{\tilde{D}}{\tilde{\eta}}_{i}(x,y)d\mu_{2}\geq\int_{\tilde{D}}{\tilde{\eta}}_{i}(x,y){\mathbf{1}}_{{\tilde{\eta}}_{i}(x,y)>c_{1}}d\mu_{2}\geq c_{1}c_{2}\lambda_{2}({\tilde{D}})\geq c_{1}c_{2}f_{L}g_{L}\mu_{2}({\tilde{D}}).$ ∎ The rest of the proof is independent of the fact, whether $I$ is lower or upper edge, we simply use already proven facts. Now to estimate $\theta(D({\varepsilon}))$. As $\beta_{5}>\beta_{2}$ we know $\|(x,y)-(x,y_{0})\|<\delta$, thus $|{\tilde{\eta}}_{i}(x,y)-{\tilde{\eta}}_{i}(x,y_{0})|<{\varepsilon}$. Thus we get: $\displaystyle\theta(D({\varepsilon}))$ $\displaystyle=\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y)d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y)d\mu_{2}(x,y)}\leq\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})+{\varepsilon}d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})-{\varepsilon}d\mu_{2}(x,y)}$ $\displaystyle=\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})+{\varepsilon}d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})d\mu_{2}(x,y)}\cdot\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})-{\varepsilon}d\mu_{2}(x,y)}\cdot\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})d\mu_{2}(x,y)}.$ The first and second fraction will both be bounded by 1 as ${\varepsilon}{\rightarrow}0$ from Corollary 6.4: $\displaystyle\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})+{\varepsilon}\ d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})d\mu_{2}(x,y)}-1=\frac{{\varepsilon}\int_{{\tilde{D}}({\varepsilon})}d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})d\mu_{2}(x,y)}\leq\frac{{\varepsilon}\mu_{2}({\tilde{D}}({\varepsilon}))}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y)-{\varepsilon}\ d\mu_{2}(x,y)}=\frac{{\varepsilon}}{c_{3}-{\varepsilon}},$ and (here we prove that the lower bound for the reciprocal converges to 1, which is equivalent) $\displaystyle\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})-{\varepsilon}\ d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})d\mu_{2}(x,y)}-1=\frac{-{\varepsilon}\int_{{\tilde{D}}({\varepsilon})}d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})d\mu_{2}(x,y)}\geq\frac{-{\varepsilon}\mu_{2}({\tilde{D}}({\varepsilon}))}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y)-{\varepsilon}\ d\mu_{2}(x,y)}=\frac{-{\varepsilon}}{c_{3}-{\varepsilon}}.$ The third fraction is the one that should converge to (or at least, for very small ${\varepsilon}$, be bounded by) $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$. Let $I_{\varepsilon}=[x_{0}-{\varepsilon},x_{1}]=[x_{2},x_{1}]$. As $\|(x,y)-(x,y_{0})\|<\delta$, we have: $\displaystyle\frac{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})d\mu_{2}(x,y)}{\int_{{\tilde{D}}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})d\mu_{2}(x,y)}=\frac{\int_{I_{\varepsilon}}\int_{{\tilde{D}}_{x}({\varepsilon})}{\tilde{\eta}}_{2}(x,y_{0})f(x)g(y)dxdy)}{\int_{I_{\varepsilon}}\int_{{\tilde{D}}_{x}({\varepsilon})}{\tilde{\eta}}_{1}(x,y_{0})f(x)g(y)dxdy)}\leq\frac{g(y_{0})+{\varepsilon}}{g(y_{0})-{\varepsilon}}\cdot\frac{\int_{I_{\varepsilon}}{\tilde{\eta}}_{1}(x,y_{0})f(x)\lambda({\tilde{D}}_{x})dx}{\int_{I_{\varepsilon}}{\tilde{\eta}}_{2}(x,y_{0})f(x)\lambda({\tilde{D}}_{x})dx}.$ The first of these fractions obviously tends to $1$ as $g(y_{0})\geq g_{L}>0$. The second can be bounded using Lemma 2.3, part 3: $\displaystyle\frac{\int_{I_{\varepsilon}}{\tilde{\eta}}_{1}(x,y_{0})f(x)\lambda({\tilde{D}}_{x})dx}{\int_{I_{\varepsilon}}{\tilde{\eta}}_{2}(x,y_{0})f(x)\lambda({\tilde{D}}_{x})dx}\leq\frac{\int_{I_{\varepsilon}}{\tilde{\eta}}_{1}(x,y_{0})f(x)dx}{\int_{I_{\varepsilon}}{\tilde{\eta}}_{2}(x,y_{0})f(x)dx}=\frac{\int_{I_{\varepsilon}}{\tilde{\eta}}_{1}(x,y_{0})f(x)g(y_{0})dx}{\int_{I_{\varepsilon}}{\tilde{\eta}}_{2}(x,y_{0})f(x)g(y_{0})dx}=\theta_{n-1}(y_{0};I_{\varepsilon}\times{\mathbb{R}}^{n-2})$ From property (T5) used for restrictions to $y=y_{0}$ we have $\theta_{n-1}(y_{0};I_{\varepsilon}\times{\mathbb{R}}^{n-2}){\rightarrow}\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$ when ${\varepsilon}{\rightarrow}0$. Putting all the estimates together we get $\theta(K)\leq\theta(D({\varepsilon}))\leq c({\varepsilon})\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$, where $c({\varepsilon}){\rightarrow}1$. Thus we can go with ${\varepsilon}$ to 0 to get $\theta(K)\leq\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})$. On the other hand from Lemma 4.33 we have $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})\leq\theta(U(\gamma,\gamma))=\theta(K)$, which means $\theta_{n-1}(y_{0};I\times{\mathbb{R}}^{n-2})=\theta(K)$. From Lemma 4.32 this means that for any horizontal line $L$ intersecting $U(\gamma,\gamma)$ we have $\theta(I)\leq\theta(U(\gamma,\gamma)\cap L)$, which, from Lemma 4.33 implies that any horizontal line divides $U(\gamma,\gamma)$ into two sets with equal $\theta$. Thus, from Lemma 5.6, $U(\gamma,\gamma)$ is appropriate. This finishes the proof of the inductive step in the limit ordinal case: the assumption $U(\gamma,\gamma)$ has positive measure and is not a strict lens set led us to the conclusion it is appropriate. ## 7 $\Theta$ functions on Orlicz balls Our main target is proving Theorem 1.2: Due to Lemma 2.1 we need to prove inequality (2.1.1) for any c-sets $A\subset{\mathbb{R}}^{k}$ and $B\subset{\mathbb{R}}^{n-k}$. We shall attempt to prove (2.1.1) using Theorem 5.1. ### 7.1 The one-dimensional case — the $\phi$ functions First we need to apply the Brunn-Minkowski theorem to get a $\Theta$-like condition: ###### Lemma 7.1. Let $K\subset{\mathbb{R}}_{x}\times{\mathbb{R}}_{y}\times{\mathbb{R}}^{n-2}$ be a generalized Orlicz ball. Let $0\leq x_{1}\leq x_{2}\in{\mathbb{R}}_{x}$, $0\leq y_{1}\leq y_{2}\in{\mathbb{R}}_{y}$. Let $K_{x_{i},y_{j}}=K\cap(\\{(x_{i},y_{j})\\}\times{\mathbb{R}}^{n-2})$ for $i,j\in\\{1,2\\}$. Let $\nu$ be a log-concave measure on ${\mathbb{R}}^{n-2}$. Then $\nu(K_{x_{1},y_{1}})\cdot\nu(K_{x_{2},y_{2}})\leq\nu(K_{x_{1},y_{2}})\cdot\nu(K_{x_{2},y_{1}}).$ ###### Proof. Let $f_{i}$, $i=1,2,\ldots,n$ be the Young functions of $K$, with $f_{1}$ defined on ${\mathbb{R}}_{x}$ and $f_{2}$ on ${\mathbb{R}}_{y}$. Let us consider the generalized Orlicz ball $K^{\prime}\in{\mathbb{R}}^{n-1}$, with the Young functions $\Phi_{i}=f_{i+1}$ for $i>1$ and $\Phi_{1}(t)=t$ — that is, we replace the first two functions with a single identity function. For any $x\in{\mathbb{R}}$ let $P_{x}$ denote the set $K^{\prime}\cap(\\{x\\}\times{\mathbb{R}}^{n-2})$, and $|P_{x}|=\nu(P_{x})$. As $K^{\prime}$ is a convex set, from the Brunn-Minkowski inequality (see for instance [Ga02]) the function $x\mapsto|P_{x}|$ is a log-concave function, which means that for any $t\in[0,1]$ we have $|P_{tx+(1-t)y}|\geq|P_{x}|^{t}|P_{y}|^{1-t}.$ In particular, for given real non-negative numbers $a,b,c$ we have $|P_{a+c}|\geq|P_{a}|^{b/(b+c)}|P_{a+b+c}|^{c/(b+c)},$ $|P_{a+b}|\geq|P_{a}|^{c/(b+c)}|P_{a+b+c}|^{b/(b+c)},$ and as a consequence when we multiply the two inequalities, $|P_{a+b}|\ |P_{a+c}|\geq|P_{a}|\ |P_{a+b+c}|.$ (7.1.1) Now let us take $a=f_{1}(x_{1})+f_{2}(y_{1})$, $b=f_{1}(x_{2})-f_{1}(x_{1})$ and $c=f_{2}(y_{2})-f_{2}(y_{1})$. As the Young functions are non-negative and increasing on $[0,\infty)$, the numbers $a,b,c$ are non-negative. From the definitions above we have: $K_{x_{1},y_{1}}=\\{(z_{3},\ldots,z_{n})\in{\mathbb{R}}^{n-2}:f_{1}(x_{1})+f_{2}(y_{1})+\sum_{i=3}^{n}f_{i}(z_{i})\leq 1\\}=\\{(z_{i})_{i=3}^{n}:\Phi_{1}(a)+\sum_{i=3}^{n}\Phi_{i-1}(z_{i})\leq 1\\}=P_{a}.$ Similarily we have $K_{x_{2},y_{1}}=P_{a+b}$, $K_{x_{1},y_{2}}=P_{a+c}$ and $K_{x_{2},y_{2}}=P_{a+b+c}$. Substituting those values into inequality (7.1.1) we get the thesis. ∎ First we consider $K\subset{\mathbb{R}}^{n-1}\times{\mathbb{R}}_{z}$. Take any $z_{2}>z_{1}>0$ and consider any c-set $B$ in ${\mathbb{R}}^{n-1}$. We define $\phi_{1}(x)={\mathbf{1}}_{K}(x,z_{1})$ and $\phi_{2}(x)={\mathbf{1}}_{K}(x,z_{2})$ for $x\in{\mathbb{R}}^{n-1}$. Let $K_{+}^{\prime}=(K_{+})_{z=z_{1}}$. By Lemma 4.19 $K_{+}^{\prime}$ is a positive quadrant of some generalized Orlicz ball $K^{\prime}$. ###### Lemma 7.2. If $\bar{K}^{\prime}$ is a derivative of $K^{\prime}$, then there exists a generalized Orlicz ball $\bar{K}$ such that $\phi_{j}(x)$ on $\bar{K}^{\prime}$ is equal to ${\mathbf{1}}_{\bar{K}}(x,z_{j})$ for $j\in\\{1,2\\}$. ###### Proof. We have a sequence $K^{\prime}=K_{0}^{\prime},K_{1}^{\prime},\ldots,K_{m}^{\prime}=\bar{K}^{\prime}$ where $K_{i+1}^{\prime}$ is some restriction of $K_{i}^{\prime}$. We can, taking identical restrictions (that is, restrictions to hyperplanes defined by the same equations or to the same intervals with respect to the same variables), construct a sequence $K=K_{0},K_{1},\ldots,K_{m}=\bar{K}$ such that $K_{i}^{\prime}=(K_{i})_{z=z_{1}}$. As $z$ was not a variable of ${\mathbb{R}}^{n-1}$ of which $K^{\prime}$ was a subset, on each step being a hyperplane restriction $z$ does not appear in the equation of the restriction hyperplane, thus we can speak of a $z$ variable in all $K_{i}$, and the isometric immersion $u:\bar{K}\hookrightarrow K$ maps $(\bar{K})_{z=z_{j}}$ into $K_{z=z_{j}}$. Thus ${\mathbf{1}}_{\bar{K}}(x,z_{j})={\mathbf{1}}_{K}(u(x,z_{j}))$, which (when, as always, we identify $\bar{K}$ with its image in $K$) gives the thesis. ∎ ###### Lemma 7.3. For any generalized Orlicz ball $K\subset{\mathbb{R}}^{m-1}\times{\mathbb{R}}_{z}$, any $z_{2}>z_{1}>0$, any coordinate-wise decomposition ${\mathbb{R}}^{m-1}={\mathbb{R}}^{k}\times{\mathbb{R}}^{m-k-1}$ and any proper measure $\mu$ on $K^{\prime}=K_{z=z_{1}}$ the function $\theta^{1}_{k}(y)=\frac{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,y,z_{2})d\mu_{|{\mathbb{R}}^{k}}(x)}{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,y,z_{1})d\mu_{|{\mathbb{R}}^{k}}(x)}$ is coordinate-wise decreasing on ${\mathbb{R}}^{m-k-1}$. ###### Proof. Let $l=m-k-1$. Select any coordinate variable $y_{i}$ from ${\mathbb{R}}^{l}$ and fix all other variables $\mathbf{y}$ in ${\mathbb{R}}^{l}$ at some $\mathbf{y}_{0}$. For $y_{1}\leq y_{2}$ we have to prove $\frac{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,\mathbf{y}_{0},y_{1},z_{2})d\mu_{|{\mathbb{R}}^{k}}(x)}{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,\mathbf{y}_{0},y_{1},z_{1})d\mu_{|{\mathbb{R}}^{k}}(x)}\geq\frac{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,\mathbf{y}_{0},y_{2},z_{2})d\mu_{|{\mathbb{R}}^{k}}(x)}{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,\mathbf{y}_{0},y_{2},z_{1})d\mu_{|{\mathbb{R}}^{k}}(x)}.$ The intersection $K_{\mathbf{y}=\mathbf{y}_{0}}$ is a generalized Orlicz ball from Lemma 4.19 and the restriction of $\mu$ is a proper measure from Lemma 4.27. Thus taking $K^{\prime\prime}=K_{\mathbf{y}=\mathbf{y}_{0}}$ we have to prove $\frac{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K^{\prime\prime}}(x,y_{1},z_{2})d\mu_{|{\mathbb{R}}^{k}}(x)}{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K^{\prime\prime}}(x,y_{1},z_{1})d\mu_{|{\mathbb{R}}^{k}}(x)}\geq\frac{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K^{\prime\prime}}(x,y_{2},z_{2})d\mu_{|{\mathbb{R}}^{k}}(x)}{\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K^{\prime\prime}}(x,y_{2},z_{1})d\mu_{|{\mathbb{R}}^{k}}(x)}.$ Note that even if the density of $\mu$ changes with $y$, it cancels out in both fractions, thus we can assume the density of $\mu$ changes only on ${\mathbb{R}}^{k}$. As a proper measure has a $1/m$-concave density, and thus a log-concave density, we can apply Lemma 7.1 to get the thesis. ∎ ###### Lemma 7.4. The functions $\phi_{1}$ and $\phi_{2}$ defined as above define a $\Theta$ function on $K^{\prime}$. ###### Proof. We have to check the four properties defining $\Theta$ functions. Property (T1) is obvious, both $\phi_{1}$ and $\phi_{2}$ are bounded by one. Note that $K$ is a c-set, as it is convex and 1-symmetric, which immediately gives properties (T2) and (T3). Condition (T4) is a consequence of Lemma 7.3. If $\bar{K}^{\prime}$ is any derivative of $K^{\prime}$, then from Lemma 7.2 we have some $\bar{K}$ such that $\phi_{j}$ restricted to $\bar{K}^{\prime}$ are equal to ${\mathbf{1}}_{\bar{K}}(\cdot,z_{j})$, and thus from Lemma 7.3 the appropriate ratio of integrals is coordinate-wise decreasing. ∎ ###### Lemma 7.5. If $K$ is a proper generalized Orlicz ball, then $\phi_{1}$ and $\phi_{2}$ define a strict $\Theta$ function. ###### Proof. The properties (S2) and (S4) are trivial. For property (S1) notice that as the Young functions are strictly increasing, ${\rm Int}K_{z=z_{1}}\supset K_{z=z_{2}}$. To check property (S3) we have to prove that $\int_{{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(x,y,z_{j})d\mu_{|{\mathbb{R}}^{k}}(x)=\mu_{|{\mathbb{R}}^{k}}(K_{y,z_{j}})$ is continuous in $y$ for $j=1,2$ and $k>0$. Let $\mu_{k}$ denote $\mu_{|{\mathbb{R}}^{k}}$. Let us take any sequence $y^{i}{\rightarrow}y^{\infty}$. First note that as the Young functions $f_{l}$ do not assume the value $+\infty$, they are continuous. Thus $\sum f_{l}(y_{l}^{i}){\rightarrow}\sum f_{l}(y_{l}^{i})$. Let $L_{a}=\\{x\in{\mathbb{R}}^{k}:\sum f_{i}(x_{i})\leq 1-a\\}$, let $a_{l}=\sum f_{l}(y_{l}^{i})+f_{z}(z_{j})$ and $a=\sum f_{l}(y_{l})+f_{z}(z_{j})$. We know $a_{l}{\rightarrow}a$, we want to prove $\mu_{k}(L_{a_{l}}){\rightarrow}\mu_{k}(L_{a})$. However, $\displaystyle\lim_{l{\rightarrow}\infty}\mu_{k}(L_{a_{l}})\leq\lim_{t{\rightarrow}0^{+}}\mu_{k}(L_{a+t})=\mu_{k}(\bigcap_{t>0}L_{a+t})=\mu_{k}(L_{a})$ as measure is continuous with respect to the set, and $\displaystyle\lim_{l{\rightarrow}\infty}\mu_{k}(L_{a_{l}})\geq\lim_{t{\rightarrow}0^{-}}\mu_{k}(L_{a+t})=\mu_{k}(\bigcap_{t<0}L_{a+t})=\mu_{k}(L_{a}),$ where we use the fact that $\mu_{k}(\\{x\in{\mathbb{R}}^{k}:\sum f_{i}(x_{i})=1-a\\})=0$, as $f_{i}$ are strictly increasing. Thus $\mu_{k}(K_{y_{l},z_{j}}){\rightarrow}\mu_{k}(L_{y,z_{j}})$, which proves property (S3). ∎ ###### Corollary 7.6. For any generalized Orlicz ball $K$ the functions $\phi_{1}$ and $\phi_{2}$ define a non-degenerate $\Theta$ function. ###### Proof. First we prove that $\phi_{1}$ and $\phi_{2}$ define a weakly non-degenerate $\Theta$ function. From Lemma 4.20 we can approximate $K$ with a proper generalized Orlicz ball $K^{\prime}$ satisfying $K^{\prime}\subset K$ and $\lambda(K\setminus K^{\prime})<{\varepsilon}/2$. Additionally, from Corollary 4.21 we may take $z_{1}^{\prime}$ and $z_{2}^{\prime}$ such that $K^{\prime}\cap\\{z=z_{j}^{\prime}\\}$ approximates $K\cap\\{z=z_{j}\\}$ up to a set of $\lambda$ measure ${\varepsilon}$. We take $\phi_{1}^{\prime}(x)={\mathbf{1}}_{K^{\prime}}(x,z_{1}^{\prime})$ and $\phi_{2}^{\prime}(x)={\mathbf{1}}_{K^{\prime}}(x,z_{2}^{\prime})$. As the intersections of $K^{\prime}$ at $z_{j}^{\prime}$ were good approximations of intersections of $K$ at $z_{i}$, we have $\int|\phi_{i}-\phi_{i}^{\prime}|d\lambda=\lambda(K_{z=z_{1}}\bigtriangleup K^{\prime}_{z=z_{1}^{\prime}})\leq{\varepsilon}$. From Lemma 4.20 we know $K^{\prime}$ is a proper generalized Orlicz ball and $K^{\prime}\subset K$. From Lemma 7.5 we know that $\phi_{1}^{\prime}$ and $\phi_{2}^{\prime}$ define a strict $\Theta$ function. Thus $\phi_{1}$ and $\phi_{2}$ define a weakly non-degenerate $\Theta$ function. As for the derivatives of the function defined by $\phi_{1}$ and $\phi_{2}$ by Lemma 7.2 they are constructed in the same manner on some derivative of $K$, and thus also define a weakly non-degenerate $\Theta$ function. Thus $\phi_{1}$ and $\phi_{2}$ define a non-degenerate $\Theta$ function. ∎ ###### Corollary 7.7. For any generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$ and any c-set $A\subset{\mathbb{R}}^{n-1}$ the function $z\mapsto\frac{\int_{\bar{A}}{\mathbf{1}}_{K}(z,x)d\mu(x)}{\int_{{\mathbb{R}}^{n-1}}{\mathbf{1}}_{K}(z,x)d\mu(x)}$ is a decreasing function of $z$ where defined. ###### Proof. From Corollary 7.6 we can apply Theorem 5.1 to the $\Theta$ function defined by $\phi_{1}$, $\phi_{2}$ to get for any $0\leq z_{1}<z_{2}$: $\frac{\int_{A}{\mathbf{1}}_{K}(x,z_{2})d\mu(x)}{\int_{A}{\mathbf{1}}_{K}(x,z_{1})d\mu(x)}\geq\frac{\int_{\bar{A}}{\mathbf{1}}_{K}(x,z_{2})d\mu(x)}{\int_{\bar{A}}{\mathbf{1}}_{K}(x,z_{1})d\mu(x)},$ (7.1.2) if both sides are defined. We can apply Fact 2.2 to make it $\frac{\int_{{\mathbb{R}}^{n-1}}{\mathbf{1}}_{K}(x,z_{2})d\mu(x)}{\int_{{\mathbb{R}}^{n-1}}{\mathbf{1}}_{K}(x,z_{1})d\mu(x)}\geq\frac{\int_{\bar{A}}{\mathbf{1}}_{K}(x,z_{2})d\mu(x)}{\int_{\bar{A}}{\mathbf{1}}_{K}(x,z_{1})d\mu(x)}.$ (7.1.3) Switching the left numerator with the right denominator we get the thesis. If the right-hand side denominator in inequality (7.1.2) is zero, the right- hand side numerator is also zero, as $z_{1}<z_{2}$ and $K_{+}$ is a c-set. Thus both for $z_{1}$ and $z_{2}$ our function is either zero or undefined. If the left-hand side denominator is zero and the right-hand side is defined, again the left-hand side numerator is zero, thus in inequality (7.1.3) we have an equality, which again gives the thesis. ∎ ### 7.2 The general case — the $\psi$ function Let $\lambda_{K}$ denote the Lebesgue measure restricted to $K_{+}$. Recall that we set out to prove $\lambda_{K}({\bar{A}}\times B)\cdot\lambda_{K}(A\times{\bar{B}})\geq\lambda_{K}(A\times B)\cdot\lambda_{K}({\bar{A}}\times{\bar{B}})$ for any c-sets $A\subset{\mathbb{R}}^{k}$ and $B\subset{\mathbb{R}}^{n-k}$. This is equivalent to $\lambda_{K}(A\times{\bar{B}})\cdot\lambda_{K}({\bar{A}}\times{\mathbb{R}}^{n-k})\geq\lambda_{K}({\bar{A}}\times{\bar{B}})\cdot\lambda_{K}(A\times{\mathbb{R}}^{n-k}).$ If either $\lambda_{K}(A\times{\mathbb{R}}^{n-k})$ or $\lambda_{K}({\bar{A}}\times{\mathbb{R}}^{n-k})$ is zero, then respectively either $\lambda_{K}(A\times{\bar{B}})$ or $\lambda_{K}({\bar{A}}\times{\bar{B}})$ is zero and the thesis is satisfied. Thus it suffices to prove $\frac{\lambda_{K}(A\times{\bar{B}})}{\lambda_{K}(A\times{\mathbb{R}}^{n-k})}=\frac{\int_{A}\int_{\bar{B}}{\mathbf{1}}_{K}(z,x)dzdx}{\int_{A}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{K}(z,x)dzdx}\geq\frac{\int_{\bar{A}}\int_{\bar{B}}{\mathbf{1}}_{K}(z,x)dzdx}{\int_{\bar{A}}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{K}(z,x)dzdx}=\frac{\lambda_{K}({\bar{A}}\times{\bar{B}})}{\lambda_{K}({\bar{A}}\times{\mathbb{R}}^{n-k})},$ when both sides are defined, which means it is enough to prove $\psi_{1}(x)=\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{K}(z,x)dz$ and $\psi_{2}(x)=\int_{\bar{B}}{\mathbf{1}}_{K}(z,x)dz$ define a non-degenerate $\Theta$ function on $K^{\prime}=K_{z=0}\subset{\mathbb{R}}^{k}$ and apply Theorem 5.1. ###### Lemma 7.8. If $\bar{K}^{\prime}$ is a derivative of $K^{\prime}$, then there exists a generalized Orlicz ball $\bar{K}$ such that $\psi_{1}(x)$ on $\bar{K}^{\prime}$ is equal to $\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{\bar{K}}(z,x)dz$ and $\psi_{2}(x)$ is equal to $\int_{\bar{B}}{\mathbf{1}}_{\bar{K}}(z,x)dz$. The proof is identical to the proof of Lemma 7.2. ###### Proposition 7.9. For any generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$, any coordinate- wise decomposition ${\mathbb{R}}^{n}={\mathbb{R}}^{k}\times{\mathbb{R}}^{n-k}$ and any c-set $B\subset{\mathbb{R}}^{n-k}$ the functions $\psi_{1}$ and $\psi_{2}$ define a $\Theta$ function on $K$. ###### Proof. Property T1 follows from the fact that $K$ is bounded. Property T2 follows from the fact $K_{+}$ is a c-set. Property T3 follows from the fact that $B\subset{\mathbb{R}}^{n-k}$. As before, the tricky part is to prove property T4. Consider any coordinate-wise decomposition ${\mathbb{R}}^{k}={\mathbb{R}}^{k_{1}}\times{\mathbb{R}}^{k_{2}}$. Choose any variable $v$ in ${\mathbb{R}}^{k_{1}}$ and fix all the others at some fixed $\mathbf{v}_{0}$. We have: $\frac{\int_{{\mathbb{R}}^{k_{2}}}\psi_{2}(v,\mathbf{v}_{0},y)d\mu(y)}{\int_{{\mathbb{R}}^{k_{2}}}\psi_{1}(v,\mathbf{v}_{0},y)d\mu(y)}=\frac{\int_{{\mathbb{R}}^{k_{2}}}\int_{\bar{B}}{\mathbf{1}}_{K}(v,\mathbf{v}_{0},y,z)d\mu(y)dz}{\int_{{\mathbb{R}}^{k_{2}}}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{K}(v,\mathbf{v}_{0},y,z)d\mu(y)dz}=\frac{\int_{{\mathbb{R}}^{k_{2}}\times{\bar{B}}}{\mathbf{1}}_{K}(v,\mathbf{v}_{0},y,z)d\mu(y)dz}{\int_{{\mathbb{R}}^{l_{2}}\times{\mathbb{R}}^{k}}{\mathbf{1}}_{K}(v,\mathbf{v}_{0},y,z)d\mu(y)dz}.$ We have to prove this function is decreasing in $v$ where defined. Let us restrict ourselves to the generalized Orlicz ball $\hat{K}=K_{\mathbf{v}=\mathbf{v}_{0}}$. Notice that ${\mathbb{R}}^{l_{2}}\times A$ is a c-set in ${\mathbb{R}}^{l_{2}}\times{\mathbb{R}}^{k}$ and $\mu\otimes\lambda$ is a proper measure in ${\mathbb{R}}^{l_{2}}\times{\mathbb{R}}^{k}$. We have to prove $\frac{\int_{{\mathbb{R}}^{l_{2}}\times{\bar{A}}}{\mathbf{1}}_{\hat{K}}(v,y,z)d(\mu\otimes\lambda)(y,z)}{\int_{{\mathbb{R}}^{l_{2}}\times{\mathbb{R}}^{k}}{\mathbf{1}}_{\hat{K}}(v,y,z)d(\mu\otimes\lambda)(y,z)}$ is decreasing in $v$, but this is exactly the thesis of Corollary 7.7. Again, as in Lemma 7.4, due to Lemma 7.8, the appropriate ratio is also decreasing for any derivative $\bar{K}$ of $K$. ∎ ###### Proposition 7.10. For any generalized Orlicz ball $K\subset{\mathbb{R}}^{n}$, any coordinate- wise decomposition ${\mathbb{R}}^{n}={\mathbb{R}}^{k}\times{\mathbb{R}}^{n-k}$ and any c-set $B\subset{\mathbb{R}}^{n-k}$ the functions $\psi_{1}$ and $\psi_{2}$ define a non-degenerate $\Theta$ function on $K$. ###### Proof. Again the derivatives of $\psi$ are again functions formed as in Lemma 7.8, so it is enough to prove $\psi$ is weakly non-degenerate. Take any ${\varepsilon}>0$. From Lemma 4.20 we may take a proper generalized Orlicz ball $\hat{K}\subset K$ with $\lambda(K\setminus\hat{K})<{\varepsilon}\min\\{\lambda(K),1\\}/2$ and $\lambda_{k}(K_{z=0}\setminus\hat{K}_{z=0})<{\varepsilon}\min\\{\lambda_{k}(K_{z=0}),1\\})$ from Lemma 4.21. Denote $\hat{K}_{z=0}$ by $\hat{K}^{\prime}$. Let $z_{1}$ be any coordinate in ${\mathbb{R}}^{n-k}$, take $B^{\prime}=B\cup(\\{\mathbf{z}:z_{1}<\delta\\}\cap K_{+})$, where $\delta$ is so small that the addition is of $\lambda_{n-k}$ measure less than ${\varepsilon}/2$. $B^{\prime}$ is a sum of two c-sets and thus a c-set. We define $\psi_{1}^{\prime}(x)=\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{\hat{K}}(z,x)dz$ and $\psi_{2}^{\prime}(x)=\int_{{\bar{B}}^{\prime}}{\mathbf{1}}_{\hat{K}}(z,x)dz$. We have $\lambda_{k}(K^{\prime}\setminus\hat{K}^{\prime})<{\varepsilon}\lambda_{k}(K^{\prime})$ from the definition of $\hat{K}$. Also $\psi_{1}^{\prime}$ and $\psi_{2}^{\prime}$ are indeed good approximations of $\psi_{1}$ and $\psi_{2}$, as $\int_{{\mathbb{R}}^{k}}|\psi_{1}(x)-\psi_{1}^{\prime}(x)|dx\leq\int_{{\mathbb{R}}^{k}}\int_{{\mathbb{R}}^{n-k}}|{\mathbf{1}}_{K}(x,z)-{\mathbf{1}}_{\hat{K}}(x,z)|dzdx=\lambda(K\bigtriangleup\hat{K})\leq{\varepsilon}/2,$ and $\displaystyle\int_{{\mathbb{R}}^{k}}|\psi_{2}(x)-\psi_{2}^{\prime}(x)|dx$ $\displaystyle=\int_{{\mathbb{R}}^{k}}\Big{|}\int_{{\mathbb{R}}^{n-k}}{\mathbf{1}}_{K}(x,z){\mathbf{1}}_{\bar{B}}(x,z)-{\mathbf{1}}_{\hat{K}}(x,z){\mathbf{1}}_{{\bar{B}}^{\prime}}(x,z)dz\Big{|}dx\leq\mu((K\cap{\bar{B}})\bigtriangleup(\hat{K}\cap{\bar{B}}^{\prime}))$ $\displaystyle\leq\lambda(K\setminus\hat{K})+\lambda_{K}({\bar{B}}\bigtriangleup{\bar{B}}^{\prime})=\lambda(K\setminus\hat{K})+\lambda_{K}(B\bigtriangleup B^{\prime})\leq{\varepsilon}.$ Thus we only have to prove that $\psi_{1}^{\prime}$ and $\psi_{2}^{\prime}$ define a strict $\Theta$ function on $\hat{K}$. Property (S2) is true as $\hat{K}$ is proper — $\hat{K}^{\prime}$ is defined by those Young functions of $\hat{K}$ which act on the variables of ${\mathbb{R}}^{k}$. Property (S4) is obvious from the definition of $\psi_{1}^{\prime}$. The function $\psi_{2}^{\prime}$ is 0 on the set $\sum f_{i}(x_{i})>1-f_{z_{1}}(\delta)$ from the definition of $B^{\prime}$ — any point in ${\bar{B}}^{\prime}$ has $z_{1}>\delta$, hence property (S1). Finally (S3) is checked exactly as in Lemma 7.5. ∎ Thus $\psi_{1}$ and $\psi_{2}$ do define a non-degenerate $\Theta$ function, which ends the proof of Theorem 1.2. ## References * [ABP03] M. Anttila, K. Ball and I. Perissinaki, The central limit problem for convex bodies. Trans. Amer. Math. Soc., 355 (2003), pp. 4723–-4735. * [BP98] K. Ball and I. Perissinaki, The subindependence of coordinate slabs in $\ell_{p}^{n}$ balls, Israel J. Math., 107 (1998), pp. 289-299. * [BGMN05] F. Barthe, O. Gudeon, S. Mendelson and A. Naor, A Probabilistic Approach to the Geometry of the $\ell_{p}^{N}$-ball, Annals of Probability, 33 (2005), pp. 480–513. * [Ga02] R. J. Gardner, The Brunn-Minkowski Inequality, Bull. Amer. Math. Soc. 39 (2002), pp. 355-405 * [BN03] S. G. Bobkov, F. L. Nazarov, On convex bodies and log-concave probability measures with unconditional basis. Geometric aspects of functional analysis, 53–69, Lecture Notes in Math., 1807, Springer, Berlin, 2003. * [Bo74] C. Borell, Convex measures on locally convex spaces. Ark. Mat. 12 (1974), 239–252. * [FGP07] B. Fleury, O. Guedon, G. Paouris, A stability result for mean width of $L_{p}$-centroid bodies. Preprint. Available at http://www.institut.math.jussieu.fr/$\tilde{\ }$guedon/Articles/06/FGP-Accepted.pdf * [Gi03] A. A. Giannopoulos, Notes on isotropic convex bodies, Institute of Mathematics, Polish Academy of Sciences, Warsaw (2003), available at http://users.uoa.gr/$\tilde{}$apgiannop/isotropic-bodies.ps. * [K07,2] B. Klartag, A central limit theorem for convex sets, Invent. Math., Vol. 168, (2007), 91–131. * [K07] B. Klartag, Power-law estimates for the central limit theorem for convex sets, J. Funct. Anal., Vol. 245, (2007), pp. 284–310. * [KLO96] S. Kwapień, R. Latała and K. Oleszkiewicz, Comparison of Moments of Sums of Independent Random Variables and Differential Inequalities. Journal of Functional Analysis, 136 (1996), pp. 258–268. * [MM05] E. Meckes and M. Meckes, The Central Limit Problem for Random Vectors with Symmetries. Preprint. Available at http://arxiv.org/abs/math.PR/0505618. * [MP89] V. D. Milman and A. Pajor, Isotropic position and inertia ellipsoids and zonoids of the unit ball of a normed $n$-dimensional space. Lecture Notes in Mathematics, 1376 (1989), pp. 64–104. * [MS86] V. Milmanc G. Schechtman, Asymptotic theory of finite-dimensional normed spaces. With an appendix by M. Gromov. Lecture Notes in Mathematics, 1200. Springer-Verlag, Berlin, 1986. * [N84] C. M. Newman, Asymptotic independence and limit theorems for positively and negatively dependent random variables. In Y. L. Tong (ed.), Inequalities in Statistics and Probability, Hayward, CA, pp. 127–140. * [S00] Qi-Man Shao, A comparison theorem on moment inequalities between negatively associated and independent random variables, J. Theoret. Probab. 13 (2000), 343-356. * [W06] J. O. Wojtaszczyk, The square negative correlation property for generalized Orlicz balls. Preprint, to be published in GAFA. Available at http://www.mimuw.edu.pl/~onufry/papers/Orlicz.pdf

arxiv-papers

2024-09-04T02:48:54.125577

0803.0441

# Multivariate integration in $C^{\infty}([0,1]^{d})$ is not strongly tractable Jakub Onufry Wojtaszczyk Department of Mathematics, Informatics and Mechanics University of Warsaw ul. Banacha 2, 02-097 Warsaw, Poland email: jakub.wojtaszczyk@zodiac.mimuw.edu.pl (June 9, 2003) ###### Abstract It has long been known that the multivariate integration problem for the unit ball in $C^{r}([0,1]^{d})$ is intractable for fixed finite $r$. H. Woźniakowski has recently conjectured that this is true even if $r=\infty$. This paper establishes a partial result in this direction. We prove that the multivariate integration problem, for infinitely differential functions all of whose variables are bounded by one, is not strongly tractable. ## 1 Introduction Multivariate integration is a classical problem of numerical analysis that has been studied for various normed spaces $F_{d}$ of functions of $d$ variables. In practical applications $d$ is frequently very large, even in the hundreds or thousands. Tractability and strong tractability of multivariate integration has been recently thoroughly analyzed. These concepts are defined as follows. We consider the worst case setting and define $n({\varepsilon},B_{d})$ as the minimal number of function values that are needed to approximate the integral of any $f$ from the unit ball of $F_{d}$ with an error threshold of ${\varepsilon}$. We want to know how $n({\varepsilon},B_{d})$ depends on $d$ and ${\varepsilon}^{-1}$. The problem is _tractable_ if $n({\varepsilon},B_{d})$ is bounded by a polynomial in ${\varepsilon}^{-1}$ and $d$, and _strongly tractable_ if the bound only depends polynomially on ${\varepsilon}^{-1}$, with no dependence on $d$. The dependence of $n({\varepsilon},B_{d})$ on ${\varepsilon}^{-1}$ has been studied for many years, and bounds for $n({\varepsilon},B_{d})$ in terms of ${\varepsilon}^{-1}$ are known for many $B_{d}$. For instance, let $F_{d}=C^{r}([0,1]^{d})$ be the space of $r$ times continuously differentiable functions defined on the $d$-dimensional unit cube with the norm given as the supremum of the absolute values of all partial derivatives up to the $r$th order. Bakhvalov proved in 1959, see [NW01], that there exist two positive numbers $c_{r,d}$ and $C_{r,d}$ such that $c_{r,d}\,{\varepsilon}^{-d/r}\,\leq\,n({\varepsilon},B_{d})\,\leq\,C_{r,d}\,{\varepsilon}^{-d/r}\qquad\forall\,{\varepsilon}\in(0,1).$ When $r$ is fixed this implies that multivariate integration in $C^{r}([0,1]^{d})$ is intractable. Indeed, $n({\varepsilon},B_{d})$ cannot possibly be bounded by a polynomial in ${\varepsilon}^{-1}$ and $d$ since the exponent of ${\varepsilon}^{-1}$ goes to infinity with $d$, and $n({\varepsilon},B_{d})$ is exponential in $d$. However if $r$ varies, that is, when we consider the spaces $F_{d}=C^{r(d)}([0,1]^{d})$ with $\sup_{d}d/r(d)<\infty$, the behavior of $n({\varepsilon},B_{d})$ is not known since we do not have sharp bounds on $c_{r,d}$ and $C_{r,d}$. In fact, the best known bounds on $c_{r,d}$ are exponentially small in $d$, while the best known bounds on $C_{r,d}$ are exponentially large in $d$, see again [NW01]. Thus we can neither claim nor deny tractability or strong tractability in the class $C^{r(d)}([0,1]^{d})$ with $\sup_{d}d/r(d)<\infty$ on the basis of Bakhvalov’s result. The conjecture formulated in [W03] states that multivariate integration in $C^{r(d)}([0,1]^{d})$ is intractable even if $r(d)=\infty$. That is, even when we consider infinitely differentiable functions with all partial derivatives bounded by one, $n({\varepsilon},B_{d})$ cannot be bounded by a polynomial in ${\varepsilon}^{-1}$ and $d$. Although we are not able to establish this conjecture in full generality, we shall prove that multivariate integration in $C^{\infty}([0,1]^{d})$ is not strongly tractable. This is achieved by showing that for a fixed $n$, the $n$th minimal error goes to one as $d$ approaches infinity. More precisely, we show that for any $n$ and $\eta$ there exists $d=d(n,\eta)$ such that for any linear algorithm there exists a polynomial that is a sum of univariate polynomials, which belongs to the unit ball of $C^{\infty}([0,1]^{d})$, whose integral is at least $1-\eta$, and the algorithm outputs zero. This proof technique allows us to prove the lack of strong tractability, but seems too weak to establish the lack of tractability. ## 2 Multivariate Integration in $C^{\infty}([0,1]^{d})$ We precisely define the problem of multivariate integration studied in this paper. Let $F_{d}=C^{\infty}([0,1]^{d})$ be the space of real functions defined on the unit cube $[0,1]^{d}$ that are infinitely differentiable with the norm $||f||_{d}\,=\,\sup\\{\,|D^{\alpha}f(x)|\,:\ x\in[0,1]^{d},\ \alpha-\text{any multiindex}\,\\}.$ Here $\alpha=[\alpha_{1},\alpha_{2},\ldots,\alpha_{d}]$ with non-negative integers $\alpha_{j}$, $|\alpha|=\alpha_{1}+\alpha_{2}+\ldots+\alpha_{d}$, and $D^{\alpha}f=\frac{\partial^{|\alpha|}}{\partial x_{1}^{\alpha_{1}}\partial x_{2}^{\alpha_{2}}\ldots\partial x_{d}^{\alpha_{d}}}f$ stands for the differentiation operator. Let $B_{d}$ denote the unit ball of this space. The multivariate integration problem is defined as an approximation of integrals $I_{d}(f)=\int_{[0,1]^{d}}f(t)\,dt\qquad\forall f\in B_{d},$ where the integral is taken with respect to the Lebesgue measure. It is well known that adaption and the use of non–linear algorithms do not help for the multivariate integration problem, as proven in [B71], see also [NW01]. That is why we consider only linear algorithms, $A_{n,d}(f)=\sum_{j=1}^{n}a_{j}f(x_{j})$ for some real coefficients $a_{j}$ and some sample points $x_{j}$ from $[0,1]^{d}$. Here $n$ denotes the number of function values used by the algorithm. Of course the $a_{i}$ and $x_{i}$ may depend on $n$. The (worst case) error of the algorithm $A_{n,d}$ is defined as ${\rm err}(A_{n,d})=\sup\\{|I_{d}(f)-A_{n,d}(f)|:\ f\in B_{d}\\}.$ Let ${\rm LIN}_{n,d}$ denote the class of all linear algorithms that use $n$ function values. The $n$th minimal error is defined as $e(n,B_{d})=\inf\\{{\rm err}(A_{n,d}):\ A_{n,d}\in{\rm LIN}_{n,d}\\}.$ We shall prove the following theorem: ###### Theorem 2.1 For any positive integer $n$ we have $\lim_{d{\rightarrow}\infty}e(n,C^{\infty}([0,1]^{d}))=1.$ This theorem easily implies that multivariate integration in $C^{\infty}([0,1]^{d})$ is not strongly tractable. Indeed, were it strongly tractable, we would have a polynomial bound on $n({\varepsilon},B_{d})$ independent of $d$, thus having a linear algorithm of error at most ${\varepsilon}$ and using at most $n=n({\varepsilon},B_{d})$ function values. Taking, say, ${\varepsilon}<\tfrac{1}{2}$ and $n>n(\frac{1}{2},B_{d})$ we would get $e(n,B_{d})\leq\tfrac{1}{2}$ independently of $d$, and this would contradict the theorem that $e(n,B_{d})$ goes to one when $d$ approaches infinity. ### 2.1 Proof of the Theorem We take an arbitrary positive integer $n$ and $\eta\in(0,1)$. The idea of the proof is to separate variables and, for sufficiently large $d$ and any $A_{n,d}\in{\rm LIN}_{n,d}$, to find a polynomial $f\in B_{d}$ that is a sum of univariate polynomials such that $|I_{d}(f)-A_{n,d}(f)|>1-\eta$. It will suffice to find such a polynomial for which $f(x_{j})=0$ at all points $x_{j}$ used by $A_{n,d}$. Then $A_{n,d}(f)=0$, and thus ${\rm err}(A_{n,d})\geq|I_{d}(f)|>1-\eta$. Since $A_{n,d}$ is an arbitrary linear algorithm this implies that $e(n,B_{d})\geq 1-\eta$ for sufficiently large $d$. For the zero algorithm $A_{n,d}\equiv 0$ we have we have $err(A_{n,d})=1$ and hence we have $e(n,B_{d})\leq 1$. Since $\eta$ can be arbitrarily small, this completes the proof of Theorem 2.1. Suppose for the moment that we have the following lemma, which will be proven in the next subsection. ###### Lemma 2.2 For any positive integer $n$, and any $\eta\in(0,1)$ there exists a constant $K_{\eta,n}$ such that for any choice of $y_{1},y_{2},\ldots,y_{n}\in[0,1]$ there exists a polynomial $f:[0,1]{\rightarrow}{\mathbb{R}}$ satisfying the following conditions: 1. 1. $\max_{x\in[0,1]}|f(x)|\leq 1$, 2. 2. $\max_{k=0,1\ldots}\max_{x\in[0,1]}|f^{(k)}(x)|\leq K_{\eta,n}$, 3. 3. $\int_{0}^{1}f(x)\,dx>1-\eta$, 4. 4. $f(y_{j})=0$ for $j=1,2,\ldots,n$. Having Lemma 2.2, we take any $d\geq K_{\eta,n}$ and any $A\in{\rm LIN}_{n,d}$ that uses sample points $x_{j}\,=\,[x_{j}^{1},x_{j}^{2},\ldots,x_{j}^{d}]\,\in\,[0,1]^{d}$ for $j=1,2,\ldots,n$. For $i=1,2,\ldots,d$, let $f_{i}$ be the polynomial given by Lemma 2.2 for $y_{j}=x_{j}^{i}$, with $j=1,2,\ldots,n$. Consider the multivariate polynomial $f(t_{1},t_{2},\ldots,t_{d})=\frac{1}{d}\sum_{i=1}^{d}f_{i}(t_{i})\qquad t_{i}\in[0,1].$ The values of $f$ are bounded by $1$ since they are arithmetic means of the values of $f_{i}$ from $[-1,1]$. Any mixed derivative of such a function $f$ is 0, while $\left|\frac{\partial^{k}}{\partial x_{i}^{k}}f(a)\right|\,=\,\left|\frac{1}{d}f_{i}^{(k)}(a^{i})\right|\,\leq\,\frac{K_{\eta,n}}{d}\,\leq\,1.$ Thus $f$ belongs to $B_{d}$. Additionally, $\int_{[0,1]^{d}}f(t)\,dt\,=\,\frac{1}{d}\sum_{i=1}^{d}\int_{[0,1]}f_{i}(x)\,dx\,>\,1-\eta.$ Furthermore $f(x_{j})=0$ since $f_{i}(x_{j}^{i})=0$ for all $j=1,2,\ldots,n$. Thus $f$ is a function we needed to prove Theorem 2.1. ### 2.2 Proof of the Lemma We will use the Stone-Weierstrass theorem to find a function satisfying Lemma 2.2. For $\delta={\eta}/{(7n)}$, let the function $g:[0,2+\delta]{\rightarrow}{\mathbb{R}}$ be defined as $1-\delta$ on $[0,1-\delta]\cup[1+\delta,2+\delta]$, $-2\delta$ at 1 and linear on $[1-\delta,1]$ and $[1,1+\delta]$. It is obviously a continuous function, so by the Stone-Weierstrass theorem we can approximate it by a polynomial $P$ of degree $N=N(\eta,n)$ such that $\max_{x\in[0,2+\delta]}|g(x)-P(x)|<\delta.$ The polynomial $P$ is negative at 1 and positive at $1+\delta$, so it has a root at some $y_{0}\in(1,1+\delta)$. Let $P_{i}(x)=P(x+y_{0}-y_{i})$. As $y_{0}-y_{i}\in(0,1+\delta)$ the polynomial $P_{i}$ satisfies $\max_{x\in[0,1]}|P_{i}(x)-g(x+y_{0}-y_{i})|<\delta$. Now take $f(x)=\prod_{i=1}^{n}P_{i}^{2}(x)\qquad\forall x\in[0,1].$ Note that $f(y_{j})=\prod_{i=1}^{n}P^{2}(y_{j}+y_{0}-y_{i})=0$ since the $j$th factor is $P^{2}(y_{0})=0$. The polynomial $P_{i}$ satisfies $1-2\delta\,<\,P_{i}(x)\,<\,1\qquad\forall x\in[0,1]\setminus[y_{i}-2\delta,y_{i}+\delta].$ Thus $(1-2\delta)^{2n}\,<\,f(x)\,<\,1\qquad\forall x\in[0,1]\setminus\bigcup_{i=1}^{n}[y_{i}-2\delta,y_{i}+\delta]$ and, of course, $f(x)\geq 0$ on the whole interval $[0,1]$. This allows us to approximate the integral of $f$. Indeed, $\int_{0}^{1}f(x)\,dx\,\geq\,\int_{[0,1]\setminus\bigcup_{i=1}^{n}[y_{i}-2\delta,y_{i}+\delta]}\,f(x)\,dx\,>\,(1-3\delta n)(1-2\delta)^{2n}.$ Using the Bernoulli inequality we conclude that the last expression is at least $(1-3n\delta)(1-4n\delta)\,\geq\,1-7n\delta=1-\eta.$ The function $f$ is a polynomial on $[0,1]$. Its coefficients are continuous functions of $(y_{1},y_{2},\ldots,y_{n})\in[0,1]^{n}$ since the coefficients of each $P_{i}$ are continuous functions of $y_{i}$, and $f$ is the product of $P_{i}$’s. The upper bound of the $j$th derivative of $f$ is a continuous function of $f$’s coefficients, and thus a continuous function of $(y_{1},y_{2},\ldots,y_{n})$. As a continuous function on a compact set it is bounded for each $j$, and so all derivarives up to the $2nN$th order have a common bound, say, $K_{\eta,n}$, independent of $(y_{1},y_{2},\ldots,y_{n})$. This means that the second condition of Lemma 2.2 is satisfied and the proof of Lemma 2.2 is completed. ### 2.3 Acknowledgements This work was completed with the enormous help of Prof. Henryk Woźniakowski, who suggested the problem, gave me the theoretical tools to solve it and worked with me on creating a publishable paper from the mathematical reasoning. The help and support of my father was also crucial to finish this work. ## References * [B71] N. S. Bakhvalov, On the optimality of linear methods for operator approximation in convex classes of functions, USSR Comput. Maths. Math. Phys. 11 (1971), 244-249 * [NW01] E. Novak and H. Woźniakowski, When are integration and discrepancy tractable?, FOCM Proceedings of Oxford 1999, eds. R. A. DeVore, A. Iserlis and E. Süli, Cambridge University Press, 211-266, 2001. * [W03] H. Woźniakowski, Open Problems for Tractability of Multivariate Integration, J. Complexity 19 (2003), 434-444.

arxiv-papers

2024-09-04T02:48:54.137202

0803.0477

# Minimal Niven numbers H. Fredricksen1, E. J. Ionascu2, F. Luca3, P. Stănică1 1 Department of Applied Mathematics, Naval Postgraduate School Monterey, CA 93943, USA; {HalF,pstanica}@nps.edu 2Department of Mathematics, Columbus State University Columbus, GA 31907, USA; ionascu_eugen@colstate.edu 3Instituto de Matemáticas, Universidad Nacional Autónoma de México C.P. 58089, Morelia, Michoacán, México; fluca@matmor.unam.mx (December $21^{st}$, 2007) ###### Abstract Define $a_{k}$ to be the smallest positive multiple of $k$ such that the sum of its digits in base $q$ is equal to $k$. The asymptotic behavior, lower and upper bound estimates of $a_{k}$ are investigated. A characterization of the minimality condition is also considered. †† Mathematics Subject Classification: 11L20, 11N25, 11N37 Key Words: sum of digits, Niven Numbers. Work by F. L. was started in the Spring of 2007 while he visited the Naval Postgraduate School. He would like to thank this institution for its hospitality. H. F. acknowledges support from the National Security Agency under contract RMA54. Research of P. S. was supported in part by a RIP grant from Naval Postgraduate School. ## 1 Motivation A positive integer $n$ is a Niven number (or a Harshad number) if it is divisible by the sum of its (decimal) digits. For instance, 2007 is a Niven number since 9 divides 2007. A $q$-Niven number is an integer $k$ which is divisible by the sum of its base $q$ digits, call it $s_{q}(k)$ (if $q=2$, we shall use $s(k)$ for $s_{2}(k)$). Niven numbers have been extensively studied by various authors (see Cai [3], Cooper and Kennedy [4], De Koninck and Doyon [5], De Koninck, Doyon and Katai [6], Grundman [7], Mauduit, Pomerance and Sárközy [11], Mauduit and Sárközy [12], Vardi [16], just to cite a few of the most recent works). In this paper, we define a natural sequence in relation to $q$-Niven numbers. For a fixed but arbitrary $k\in{\mathbb{N}}$ and a base $q\geq 2$, one may ask whether or not there exists a $q$-Niven number whose sum of its digits is precisely $k$. We will show later that the answer to this is affirmative. Therefore, it makes sense to define $a_{k}$ to be the smallest positive multiple of $k$ such that $s_{q}(a_{k})=k$. In other words, $a_{k}$ is the smallest Niven number whose sum of the digits is a given positive integer $k$. We denote by $c_{k}$ the companion sequence $c_{k}={a_{k}}/{k}$, $k\in{\mathbb{N}}$. Obviously, $a_{k}$, respectively, $c_{k}$, depend on $q$, but we will not make this explicit to avoid cluttering the notation. In this paper we give constructive methods in Sections 3, 4 and 7 by two different techniques for the binary and nonbinary cases, yielding sharp upper bounds for $a_{k}$. We find elementary upper bounds true for all $k$, and then better nonelementary ones true for most odd $k$. Throughout this paper, we use the Vinogradov symbols $\gg$ and $\ll$ and the Landau symbols $O$ and $o$ with their usual meanings. The constants implied by such symbols are absolute. We write $x$ for a large positive real number, and $p$ and $q$ for prime numbers. If ${{\mathcal{A}}}$ is a set of positive integers, we write ${{\mathcal{A}}}(x)={{\mathcal{A}}}\cap[1,x]$. We write $\ln x$ for the natural logarithm of $x$ and $\log x=\max\\{\ln x,1\\}$. ## 2 Easy proof for the existence of $a_{k}$ In this section we present a simple argument that shows that the above defined sequence $a_{k}$ is well defined. First we assume that $k$ satisfies $\gcd(k,q)=1$. By Euler’s theorem, we can find an integer $t$ such that $q^{t}\equiv 1\pmod{k}$, and then define $K=1+q^{t}+q^{2t}+\cdots+q^{(k-1)t}.$ Obviously, $K\equiv 0\pmod{k}$, and also $s_{q}(K)=k$. Hence, in this case, $K$ is a Niven number whose digits in base $q$ are only $0$’s and $1$’s and whose sum is $k$. If $k$ is not coprime to $q$, we write $k=ab$ where $\gcd(b,q)=1$ and $a$ divides $q^{n}$ for some $n\in{\mathbb{N}}$. As before, we can find $K\equiv 0\pmod{b}$ with $s_{q}(K)=b$. Let $u=\max\\{n,\lceil{\log_{q}K}\rceil\\}+1$, and define $K^{\prime}=(q^{u}+q^{2u}+\cdots+q^{ua})K.$ Certainly $k=ab$ is a divisor of $K^{\prime}$ and $s_{q}(K^{\prime})=ab=k$. Therefore, $a_{k}$ is well defined for every $k\in{\mathbb{N}}$. This argument gives a large upper bound, namely of size $\exp(O(k^{2}))$ for $a_{k}$. We remark that if $m$ is the minimal $q$-Niven number corresponding to $k$, then $q-1$ must divide $m-s_{q}(m)=kc_{k}-k=(c_{k}-1)k$. This observation turns out to be useful in the calculation of $c_{k}$ for small values of $k$. For instance, in base ten, the following table of values of $a_{k}$ and $c_{k}$ can be established easily by using the previous simple observation. As an example, if $k=17$ then $9$ has to divide $c_{17}-1$ and so we need only check $10,19,28$. $k$ | $10$ | $11$ | $12$ | $13$ | $14$ | $15$ | $16$ | $17$ | $18$ | $19$ | $20$ | $21$ | $22$ | $23$ ---|---|---|---|---|---|---|---|---|---|---|---|---|---|--- $c_{k}$ | $19$ | $19$ | $4$ | $19$ | $19$ | $13$ | $28$ | $28$ | $11$ | $46$ | $199$ | $19$ | $109$ | $73$ $a_{k}$ | $190$ | $209$ | $48$ | $247$ | $266$ | $195$ | $448$ | $476$ | $198$ | $874$ | $3980$ | $399$ | $2398$ | $1679$ ## 3 Elementary bounds for $a_{k}$ in the binary case For each positive integer $k$ we set $n_{k}=\lceil\log_{2}k\rceil$. Thus, $n_{k}$ is the smallest positive integer with $k\leq 2^{n_{k}}$. Assuming that $k\in\mathbb{N}$ ($k>1$) is odd, we let $t_{k}$ be the multiplicative order of 2 modulo $k$, and so, $2^{t_{k}}\equiv 1\pmod{k}$. Obviously, $t_{k}\geq n_{k}$ and $t_{k}\mid\phi(k)$, where $\phi$ is Euler’s totient function. Thus, $n_{k}\leq t_{k}\leq k-1.$ (1) ###### Lemma 1. For every odd integer $k>1$, every integer $x\in\\{0,1,\ldots,k-1\\}$ can be represented as a sum modulo $k$ of exactly $n_{k}$ distinct elements of $\displaystyle D=\\{2^{i}\,|\,i=0,\ldots,n_{k}+k-2\\}.$ ###### Proof. We find the required representation in a constructive way. Let us start with an example. If $x=0$ and $k=2^{n_{k}}-1$, then since $x\equiv k\pmod{k}$, we notice that in this case we have a representation as required by writing $k=1+2+\cdots+2^{n_{k}-1}$ (note that $n_{k}-1\leq n_{k}+k-2$ is equivalent to $k\geq 1$). Any $x\in\\{0,1,\ldots,k-1\\}$ has at most $n_{k}$ bits of which at most $n_{k}-1$ are ones. Next, let us illustrate the construction when this binary representation of $x$ contains exactly $n_{k}-1$ ones, say $x=2^{n_{k}-1}+2^{n_{k}-2}+\cdots+2+1-2^{j},~{}\text{for some}~{}j\in\\{0,1,\ldots,n_{k}-1\\}.$ First, we assume $j\leq n_{k}-2$. Using $2^{j+1}=2^{j}+2^{j}\equiv 2^{j}+2^{j+t_{k}}\pmod{k}$, we write $x\equiv 2^{j+t_{k}}+2^{n_{k}-1}+\cdots+2^{j+2}+2^{j}+2^{j-1}+\cdots+1\pmod{k},$ where both $j+t_{k}\leq n_{k}-2+k-1=n_{k}+k-3$ and $j+t_{k}>n_{k}-1$ are true according to (1). Therefore all exponents are distinct and they are contained in the required range, which gives us a representation of $x$ as a sum of exactly $n_{k}$ different elements of $D$ modulo $k$. If $j=n_{k}-1$, then $x=2^{n_{k}-1}-1$. We consider $x+k$ instead of $x$. By the definition of $n_{k}$, we must have $k\geq 2^{n_{k}-1}+1$. Hence, $x+k\geq 2^{n_{k}}$, which implies that the binary representation of $x+k$ starts with $2^{n_{k}}$ and it has at most $n_{k}$ ones. Indeed, if $s(x+k)\geq n_{k}+1$, then $x+k\geq 2^{n_{k}}+2^{n_{k}-1}+\cdots+2+1=2^{n_{k}+1}-1$, which in turn contradicts the inequality $x+k\leq k-1+k=2k-1\leq 2^{n_{k}+1}-3$ since $k$ is odd. If $s(x+k)=n_{k}$, then we are done ($k\geq 3$). If $s(x+k)=n_{k}-1$, then we proceed as before and observe that this time $j+t_{k}\leq n_{k}+k-2$ for every $j\in\\{0,1,2,\ldots,n_{k}-1\\}$ and $j+t_{k}>n_{k}$ if $j>0$ which is an assumption that we can make because in order to obtain $n_{k}-1$ ones two of the powers of $2$, out of $1,2,2^{2},\ldots,2^{n_{k}-1}$, must be missing. If $s(x+k)<n_{k}-1$, then for every zero in the representation of $x+k$, which is preceded by a one and followed by $\ell$ ($\ell\geq 0$) other zeroes, we can fill out the zeros gap in the following way. If such a zero is given by the coefficient of $2^{j}$, then we replace $2^{j+1}$ by $2^{j}+2^{j-1}+\cdots+2^{j-\ell}+2^{j-\ell+t_{k}}$. This will give $\ell+2$ ones instead of a one and $\ell+1$ zeros. We fill out all gaps this way with the exception of the gap corresponding to the smallest power of 2 and $\ell\geq 1$, where in order to insure the inequality $j^{\prime}+t_{k}>n_{k}$ ($j^{\prime}=j-\ell+1>0$) one will replace $2^{j+1}$ by $2^{j}+2^{j-1}+\cdots+2^{j-\ell+1}+2^{j-\ell+1+t_{k}}$. The result will be a representation in which all the additional powers $2^{j^{\prime}+t_{k}}$ will be distinct and the total number of powers of two is $n_{k}$. The maximum exponent of these powers is at most $j^{\prime}+t_{k}\leq n_{k}+k-2$. If the representation of $x$ starts with $2^{n_{k}-1}$, then the technique described above can be applied directly to $x$ making sure that all zero gaps are completely filled. Otherwise, we apply the previous technique to $x+k$. ∎ ###### Example 2. Let $k=11$. Then $n_{11}=4$ and $t_{11}=10$. Suppose that we want to represent $9$ as a sum of $4$ distinct terms modulo 11 from the set $D=\\{1,2,\ldots,2^{13}\\}$. Since $9=2^{3}+1$, we have $9=2^{2}+2+2+1$, so $9\equiv 2^{2}+2+2^{11}+1\pmod{11}$. If we want to represent $7=2^{2}+2^{1}+2^{0}$ then, since this representation does not contain $2^{3}$, we look at $7+11=18=2^{4}+2=2^{3}+2^{3}+2=2^{3}+2^{2}+2^{2}+2$. Thus, $7\equiv 2^{3}+2^{2}+2^{12}+2\pmod{11}$. We note that the representation given by Lemma 1 is not unique. If this construction is applied in such a way that the zero left when appropriate is always the one corresponding to the largest power of $2$, we will obtain the largest of such representations. In the previous example, we can fill out the smallest gap first and leave a zero from the gap corresponding to $2^{3}$, so $7\equiv 18\equiv 2^{4}+2=2^{3}+2^{3}+1+1\equiv 2^{3}+2^{13}+1+2^{10}\pmod{11}$. Recall that $2^{\alpha}\|m$ means that $2^{\alpha}\mid m$ but $2^{\alpha+1}\nmid m$. We write $\mu_{2}(m)$ for the exponent $\alpha$. ###### Theorem 3. For all positive integers $k$ and $\ell$, there exists a positive integer $n$ having the following properties: 1. $(a)$ $s(nk)=\ell k$, 2. $(b)$ $n\leq(2^{\ell k+n_{k}}-2^{\mu_{2}(k)})/k$. ###### Proof. It is clear that if $k$ is a power of $2$, say $k=2^{s}$, then we can take $n=2^{\ell k}-1$ and so $s(kn)=s(2^{s}+2^{s+1}+\cdots+2^{s+\ell k-1})=\ell k$. In this case, the upper bound in part $(b)$ is sharp since ${n_{k}}=s=\mu_{2}(k)$. Furthermore, if $k$ is of the form $k=2^{m}d$ for some positive integers $m,d$ with odd $d\geq 3$, then assuming that we can find an integer $n\leq(2^{2^{m}\ell d+{n_{k}}^{\prime}}-1)/d$, where ${n_{k}}^{\prime}=\lceil\log_{2}d\rceil$, such that $s(nd)=2^{m}\ell d$, then $nk$ satisfies condition $(a)$ since $s(nk)=s(2^{m}nd)=s(nd)=2^{m}\ell d=\ell k.$ We observe that condition $(b)$ is also satisfied in this case, because $(2^{2^{m}\ell d+{n_{k}}^{\prime}}-1)/d=(2^{\ell k+{n_{k}}}-2^{m})/k.$ Thus, without loss of generality, we may assume in what follows that $k\geq 3$ is odd. Consider the integer $M=2^{\ell k+n_{k}}-1=1+2^{1}+\cdots+2^{\ell k+n_{k}-1}$, and so, $s(M)=\ell k+n_{k}$. By Lemma 1, we can write $M\equiv 2^{j_{1}}+2^{j_{2}}+\cdots+2^{j_{n_{k}}}\pmod{k},$ (2) where $0\leq j_{1}<j_{2}<\cdots<j_{n}\leq k+n_{k}-2<\ell k+n_{k}-1$. Therefore, we may take $n=\dfrac{M-(2^{j_{1}}+2^{j_{2}}+\cdots+2^{j_{n_{k}}})}{k},$ which is an integer by (2) and satisfies $s(nk)=s(M-(2^{j_{1}}+2^{j_{2}}+\cdots+2^{j_{n_{k}}}))=\ell k.$ ∎ ###### Corollary 4. The sequence $(a_{k})_{k\geq 1}$ satisfies $2^{k}-1\leq a_{k}\leq 2^{k+n_{k}}-2^{\mu_{2}(k)}.$ (3) ###### Proof. The first inequality in (3) follows from the fact that if $s(a_{k})=k$, then $a_{k}\geq 1+2+\cdots+2^{k-1}=2^{k}-1$. The second inequality in (3) follows from Theorem 3 by taking $\ell=1$, and from the minimality condition in the definition of $a_{k}$. ∎ We have computed $a_{k}$ and $c_{k}$ for all $k=1,\ldots,128$, $c_{1}=1,\ c_{2}=3,\ c_{3}=7,\ldots,\ c_{20}=209715,\ldots.$ and the graph of $k\to\ln(c_{k})$ against the functions $k\to\ln(2^{k})$ and $k\to\ln(2^{k}-1)-\ln(k)$ is included in Figure 1. Figure 1: The graphs of $k\to\ln(c_{k})$ and $k\to\ln(2^{k})$, $k\to\ln(2^{k}-1)-\ln(k)$ The right hand side of inequality (3) is sharp when $k=2^{s}$, as we have already seen. For $k=2^{s}-1$, we get values of $c_{k}$ very close to $2^{k}-1$ but, in general, numerical evidence shows that $c_{k}/2^{k}$ is closer to zero more often than it is to $1$. In fact, we show in Section 6 that this is indeed the case at least for odd indices (see Corollary 11, Corollary 12 and relation (23)). ## 4 Improving binary estimates and some closed formulae In order to obtain better bounds for $a_{k}$, we introduce the following classes of odd integers. For a positive integer $m$ we define ${\cal C}_{m}=\\{k\equiv 1\pmod{2}|\ 2^{k+m}-1\equiv\sum_{i=1}^{m}2^{j_{i}}\pmod{k},\ \text{for}\ 0\leq j_{1}<j_{2}<\cdots<j_{m}\leq m+k-2\\}.$ Let us observe that ${\cal C}_{m}\subset{\cal C}_{m+1}$. Indeed, if $k$ is in ${\cal C}_{m}$, we then have $2^{k+m}-1\equiv 2^{j_{1}}+\cdots+2^{j_{m}}$ for some $0\leq j_{1}<j_{2}<\cdots<j_{m}\leq m+k-2$. Multiplying by 2 the above congruence and adding one to both sides, we get $2^{k+m+1}-1\equiv 1+2^{j_{1}+1}+\cdots+2^{j_{m}+1}$, representation which implies that $k$ belongs to ${\cal C}_{m+1}$. Note also that Lemma 1 shows that every odd integer $k\geq 3$ belongs to ${\cal C}_{u}$, where $u=\left\lceil\log k/\log 2\right\rceil$. Hence, we have $2{\mathbb{N}}+1=\bigcup_{m\in{\mathbb{N}}}{\cal C}_{m}.$ ###### Theorem 5. For every $k\in{\cal C}_{1}$, we have $2^{k}-1<a_{k}<2^{k+1}-1.$ In particular, $c_{k}/2^{k}\to 0$ as $k\to\infty$ through ${\cal C}_{1}$. Furthermore, $a_{k}=2^{k+1}-1-2^{j_{1}}$, where $j_{1}=j_{0}+st_{k}$, with $s=\lfloor(k-1-j_{0})/t_{k}\rfloor$, and $0\leq j_{0}\leq t_{k}-1$ is such that $2^{k+1}-1\equiv 2^{j_{0}}\pmod{k}$. ###### Proof. We know that $2^{k}-1\not\equiv 0\pmod{k}$ (see [13, Problem 37, p. 109]). Hence, an integer of binary length $k$ whose sum of digits is $k$ is not divisible by $k$. Therefore, $a_{k}>2^{k}-1$. Next, we assume that $a_{k}$ is an integer of binary length $k+1$ and sum of digits $k$; that is, $a_{k}=2^{k+1}-1-2^{j}$ for some $j=0,\ldots,k-1$. But $2^{k+1}-1\equiv x\pmod{k}$, and by hypothesis there exists $j_{0}$ such that $x=2^{j_{0}}$ for some $j_{0}\in\\{0,\ldots,t_{k}-1\\}$. In order to obtain $a_{k}$, we need to subtract the highest power of 2 possible because of the minimality of $a_{k}$. So, we need to take the greatest exponent $j_{1}=j_{0}+st_{k}\leq k-1$, leading to $s=\lfloor(k-1-j_{0})/t_{k}\rfloor$. Hence, $a_{k}=2^{k+1}-1-2^{j_{1}}$. ∎ Based on the above argument, we can compute, for instance, $a_{5}=55=2^{6}-1-2^{3}$, since $2^{3}-1\equiv 2^{3}$ (mod 5). Similarly, $a_{29}=2^{30}-1-2^{5}=1073741791$, since $2^{30}-1\equiv 2^{5}$ (mod 29), and $a_{25}=2^{26}-1-2^{19}=66584575$, since $2^{26}-1\equiv 2^{19}$ (mod 25), or perhaps the more interesting example $a_{253}=2^{254}-1-2^{242}$. ###### Theorem 6. If $m\in{\mathbb{N}}$, and $k\in{\cal C}_{m+1}\setminus{\cal C}_{m}$, we then have $2^{k+m-1}-1<a_{k}<2^{k+m}-1.$ Thus, $c_{k}/2^{k}\to 0$ as $k\to\infty$ in ${\cal C}_{m}$ for any fixed $m$. ###### Proof. Similar as the proof of Theorem 5. ∎ ###### Theorem 7. For all integers $k=2^{i}-1\geq 3$, we have $a_{k}\leq 2^{k+{k^{-}}}+2^{k}-2^{k-i}-1,$ (4) where ${k^{-}}$ is the least positive residue of $-k$ modulo $i$. Furthermore, the bound (4) is tight when $k=2^{i}-1$ is a Mersenne prime. In this case, we have $c_{k}/2^{k}\to 1/2$ as $k\to\infty$ through Mersenne primes, assuming that this set is infinite. ###### Proof. For the first claim, we show that the sum of binary digits of the bound of the upper bound on (4) is exactly $k$, and also that this number is a multiple of $k$. From the definition of $k^{-}$, we find that $k+k^{-}=i\alpha$ for some positive integer $\alpha$. Since $\displaystyle 2^{k+{k^{-}}}+2^{k}-2^{k-i}-1$ $\displaystyle=$ $\displaystyle 2^{k-i}(2^{i}-1)+2^{i\alpha}-1$ $\displaystyle=$ $\displaystyle(2^{i}-1)(2^{k-i}+2^{i(\alpha-1)}+2^{i(\alpha-2)}+\cdots+1),$ we get that $2^{k+{k^{-}}}+2^{k}-2^{k-i}-1$ is divisible by $k$. Further, $k^{-}\geq 1$ since $k$ is not divisible by $i$ (see the proof of Theorem 5), and $\displaystyle s\left(2^{k+{k^{-}}}+2^{k}-2^{k-i}-1\right)$ $\displaystyle=$ $\displaystyle s\left(2^{k+k^{-}-1}+\cdots+2+1+2^{k}-2^{k-i}\right)$ $\displaystyle=$ $\displaystyle s\left(2^{k+k^{-}-1}+\cdots+2^{k}+\cdots+\widehat{2^{k-i}}+\cdots+2+1+2^{k}\right)$ $\displaystyle=$ $\displaystyle s\left(2^{k+k^{-}}+2^{k-1}+\cdots+\widehat{2^{k-i}}+\cdots+2+1\right)$ $\displaystyle=$ $\displaystyle k,$ where $\hat{t}$ means that $t$ is missing in that sum. The first claim is proved. We now consider a Mersenne prime $k=2^{i}-1$. First, we show that $k\in{\cal C}_{i}\setminus{\cal C}_{i-1}$. Since $u=\lceil\log k/\log 2\rceil=i$, by Lemma 1, we know that $k\in{\cal C}_{i}$. Suppose by way of contradiction that $k\in{\cal C}_{i-1}$. Then $2^{k+i-1}-1\equiv 2^{j_{1}}+\cdots+2^{j_{i-1}}\pmod{k}$ (5) holds with some $0\leq j_{1}<j_{2}<\cdots<j_{i-1}\leq k+i-3$. Since $k$ is prime, we have that $2^{k-1}\equiv 1\pmod{k}$, and so $2^{k+i-1}-1\equiv 2^{i}-1\equiv 0\pmod{k}.$ Because $2^{i}\equiv 1\pmod{k}$, we can reduce all powers $2^{j}$ of $2$ modulo $k$ to powers with exponents less than or equal to $i-1$. We get at most $i-1$ such terms. But in this case, the sum of at least one and at most $i-1$ distinct members of the set $\\{1,2,\ldots,2^{i-1}\\}$ is positive and less than the sum of all of them, which is $k$. So, the equality (5) is impossible. To finish the proof, we need to choose the largest representation $x=2^{j_{1}}+\cdots+2^{j_{i}}$, with $0\leq j_{1}<j_{2}<\cdots<j_{i}\leq k+i-2$, such that $2^{k+i}-1\equiv x\pmod{k}$. But $2^{k+i}-1\equiv 2^{i+1}-1\equiv 1\pmod{k}$. Since the exponents $j$ are all distinct, the way to accomplish this is to take $j_{i}=k+i-2$, $j_{i-1}=k+i-3,\ldots,j_{2}=k$, and finally $j_{1}$ to be the greatest integer with the property that the resulting $x$ satisfies $x\equiv 1\pmod{k}$. Since $x=2^{j_{1}}+2^{k}(1+2+\cdots+2^{i-2})=2^{j_{1}}+2^{k}(2^{i-1}-1)\equiv 2^{j_{1}}+2^{i}-2\equiv 2^{j_{1}}-1\pmod{k}$, we need to have $2^{j_{1}}\equiv 2\pmod{k}$. Since the multiplicative order of $2$ modulo $k$ is clearly $i$, we have to take the largest $j_{1}=1+si$ such that $1+si<k$. But $i$ must be prime too and so $2^{i-1}\equiv 1\pmod{i}$. This implies $k=2^{i}-1\equiv 1\pmod{i}$. Therefore $j_{1}=k-i$. So, $a_{k}=2^{k+i}-1-x=2^{k+i}-1-2^{k-i}-2^{k+i-1}+2^{k}=2^{k+i-1}+2^{k}-2^{k-i}-1$ and the inequality given in our statement becomes an equality since $k^{-}=i-1$ in this case. Regarding the limit claim, we observe that $\frac{c_{k}}{2^{k}}=\frac{k+1}{2k}+\frac{1}{k}-\frac{1}{k2^{i}}-\frac{1}{k2^{k}}\ \longrightarrow\ \frac{1}{2},$ as $i$ (and as a result $k$) goes to infinity. ∎ Between the two extremes, Theorems 6 and 7, we find out that the first situation is more predominant (see Corollary 12). Next, we give quantitative results on the sets ${\cal C}_{m}$. However we start with a result which shows that ${\cal C}_{1}$ is of asymptotic density zero as one would less expect. ## 5 ${\cal C}_{1}$ is of density zero Here, we show that ${\cal C}_{1}$ is of asymptotic density zero. For the purpose of this section only, we omit the index and simply write ${\mathcal{C}}=\\{1\leq n:2^{n+1}-1\equiv 2^{j}\pmod{n}~{}{\text{\rm for~{}some}}~{}j=1,2,\ldots\\}.$ It is clear that ${\mathcal{C}}$ contains only odd numbers. Recall that for a positive real number $x$ and a set ${\cal A}$ we put ${\cal A}(x)={\cal A}\cap[1,x]$. We prove the following estimate. ###### Theorem 8. The estimate $\\#{\mathcal{C}}(x)\ll\frac{x}{(\log\log x)^{1/7}}$ holds for all $x>e^{e}$. ###### Proof. We let $x$ be large, and put $q$ for the smallest prime exceeding $y=\frac{1}{2}\left(\frac{\log\log x}{\log\log\log x}\right)^{1/2}.$ Clearly, for large $x$ the prime $q$ is odd and its size is $q=(1+o(1))y$ as $x\to\infty$. For an odd prime $p$ we write $t_{p}$ for the order of $2$ modulo $p$ first defined at the beginning of Section 3. Recall that this is the smallest positive integer $k$ such that $2^{k}\equiv 1\pmod{p}$. Clearly, $t_{p}\mid p-1$. We put ${\cal P}=\\{p~{}{\text{\rm prime}}:p\equiv 1\pmod{q}~{}{\text{\rm and}}~{}t_{p}\mid(p-1)/q\\}.$ (6) The effective version of Lagarias and Odlyzko of Chebotarev’s Density Theorem (see [10], or page 376 in [14]), shows that there exist absolute constants $A$ and $B$ such that the estimate $\\#{\cal P}(t)=\frac{\pi(t)}{q(q-1)}+O\left(\frac{t}{\exp\left(A{\sqrt{\log t}}/q\right)}\right)$ (7) holds for all real numbers $t$ as long as $q\leq B(\log t)^{1/8}$. In particular, we see that estimate (7) holds when $x>x_{0}$ is sufficiently large and uniformly in $t\in[z,x]$, where we take $z=\exp((\log\log x)^{100})$. We use the above estimate to compute the sum of the reciprocals of the primes $p\in{\cal P}(u)$, where we put $u=x^{1/100}$. We have $S=\sum_{p\in{\cal P}(u)}\frac{1}{p}=\sum_{\begin{subarray}{c}p\in{\cal P}\\\ p\leq z\end{subarray}}\frac{1}{p}+\sum_{\begin{subarray}{c}p\in{\cal P}\\\ z<p\leq u\end{subarray}}\frac{1}{p}=S_{1}+S_{2}.$ For $S_{1}$, we only use the fact that every prime $p\in{\cal P}$ is congruent to $1$ modulo $q$. By the Brun-Titchmarsh inequality we have $S_{1}\leq\sum_{\begin{subarray}{c}p\leq z\\\ p\equiv 1\pmod{q}\end{subarray}}\frac{1}{p}\ll\frac{\log\log z}{\phi(q)}\ll\frac{\log\log\log x}{q}=O(1).$ For $S_{2}$, we are in the range where estimate (7) applies so by Abel’s summation formula $\displaystyle S_{2}$ $\displaystyle=$ $\displaystyle\sum_{\begin{subarray}{c}p\in{\cal P}\\\ z\leq p\leq u\end{subarray}}\frac{1}{p}\ll\int_{z}^{u}\frac{d\\#{\cal P}(t)}{t}=\frac{\\#{\cal P}(t)}{t}\Big{|}_{t=z}^{t=u}$ $\displaystyle+$ $\displaystyle\int_{z}^{u}\left(\frac{\pi(t)}{q(q-1)t^{2}}+O\left(\frac{t}{\exp(A{\sqrt{\log t}}/q)}\right)\right)dt$ $\displaystyle=$ $\displaystyle\int_{z}^{u}\frac{dt}{q(q-1)t\log t}+O\left(\frac{1}{q^{2}}\right)+O\left(\int_{z}^{u}\frac{dt}{q(q-1)t(\log t)^{2}}\right)$ $\displaystyle=$ $\displaystyle\frac{\log\log u-\log\log z}{q(q-1)}+O\left(\frac{1}{q^{2}}\right)=\frac{\log\log x}{q(q-1)}+O(1).$ In the above estimates, we used the fact that $\pi(t)=\frac{t}{\log t}+O\left(\frac{t}{(\log t)^{2}}\right),$ as well as the fact that $\frac{t}{\exp(A{\sqrt{\log t}}/q)}=O\left(\frac{t}{q^{2}(\log t)^{2}}\right)$ uniformly for $t\geq z$. To summarize, we have that $S=\frac{\log\log x}{q(q-1)}+O(1)=\frac{\log\log x}{q^{2}}+O\left(\frac{\log\log x}{q^{3}}+1\right)=\frac{\log\log x}{q^{2}}+O(1).$ (8) We next eliminate a few primes from ${\cal P}$ defined in (6). Namely, we let ${\cal P}_{1}=\\{p:t_{p}<p^{1/2}/(\log p)^{10}\\},$ and ${\cal P}_{2}=\\{p:p-1~{}{\text{\rm has~{}a~{}divisor}}~{}d~{}{\text{\rm in}}~{}[p^{1/2}/(\log p)^{10},p^{1/2}(\log p)^{10}]\\}.$ A well-known elementary argument (see, for example, Lemma 4 in [2]) shows that $\\#{\cal P}_{1}(t)\ll\frac{t}{(\log t)^{2}},$ (9) therefore by the Abel summation formula one gets easily that $\sum_{p\in{\cal P}_{1}}\frac{1}{p}=O(1).$ As for ${\cal P}_{2}$, results of Indlekofer and Timofeev from [9] show that $\\#{\cal P}_{2}(t)\ll\frac{t\log\log t}{(\log t)^{1+\delta}},$ where $\delta=2-(1+\log\log 2)/\log 2=0.08\ldots$, so again by Abel’s summation formula one gets that $\sum_{p\in{\cal P}_{2}}\frac{1}{p}=O(1).$ We thus arrive at the conclusion that letting ${\cal Q}={\cal P}\backslash({\cal P}_{1}\cup{\cal P}_{2})$, we have $S^{\prime}=\sum_{p\in{\cal Q}(u)}\frac{1}{p}=S-\sum_{p\in{\cal P}_{1}(u)\cup{\cal P}_{2}(u)}\frac{1}{p}=\frac{\log\log x}{q^{2}}+O(1).$ (10) Now let us go back to the numbers $n\in{\cal C}$. Let ${\cal D}_{1}$ be the subset of ${\cal C}(x)$ consisting of the numbers free of primes in ${\cal Q}(u)$. By the Brun sieve, $\displaystyle\\#{\cal D}_{1}$ $\displaystyle\ll$ $\displaystyle x\prod_{p\in{\cal Q}(u)}\left(1-\frac{1}{p}\right)=x\exp\left(-\sum_{p\in{\cal Q}(u)}\frac{1}{p}+O\left(\sum_{p\in{\cal Q}(u)}\frac{1}{p^{2}}\right)\right)$ (11) $\displaystyle\ll$ $\displaystyle x\exp(-S^{\prime}+O(1))\ll x\exp\left(-\frac{\log\log x}{q^{2}}\right)$ $\displaystyle=$ $\displaystyle\frac{x}{(\log\log x)^{4+o(1)}}\ll\frac{x}{(\log\log x)^{3}}.$ Assume from now on that $n\in{\cal C}(x)\backslash{\cal D}_{1}$. Thus, $p\mid n$ for some prime $p\in{\cal Q}(u)$. Assume that $p^{2}\mid n$ for some $p\in{\cal Q}(u)$. Denote by ${\mathcal{D}}_{2}$ the subset of such $n\in{\mathcal{C}}(x)\backslash{\mathcal{D}}_{1}$. Keeping $p\in{\cal Q}(u)$ fixed, the number of $n\leq x$ with the property that $p^{2}\mid n$ is $\leq x/p^{2}$. Summing up now over all primes $p\equiv 1\pmod{q}$ not exceeding $x^{1/2}$, we get that the number of such $n\leq x$ is at most $\\#{\mathcal{D}}_{2}\leq\sum_{\begin{subarray}{c}p\leq x^{1/2}\\\ p\equiv 1\pmod{q}\end{subarray}}\frac{x}{p^{2}}\ll\frac{x}{q^{2}\log q}\ll\frac{x}{\log\log x}.$ (12) Let ${\mathcal{D}}_{3}={\mathcal{C}}(x)\backslash({\mathcal{D}}_{1}\cup{\mathcal{D}}_{2})$. Write $n=pm$, where $p$ does not divide $m$. We may also assume that $n\geq x/\log x$ since there are only at most $x/\log x$ positive integers failing this condition. Put $t=t_{p}$. The definition of ${\cal C}$ implies that $2^{mp+1}\equiv 2^{j}+1\pmod{p}$ for some $j=1,2,\ldots,t$, and since $2^{p}\equiv 2\pmod{p}$, we get that $2^{mp+1}\equiv 2^{m+1}\pmod{p}$. We note that $2^{m+1}\pmod{p}$ determines $m\leq x/p$ uniquely modulo $t$. We estimate the number of values that $m$ can take modulo $t$. Writing $X=\\{2^{j}\pmod{p}\\}$, we see that $\\#\\{m\pmod{p}\\}\leq I/t$, where $I$ is the number of solutions $(x_{1},x_{2},x_{2})$ to the equation $x_{1}-x_{2}-x_{3}=0,\qquad x_{1},~{}x_{2},~{}x_{3}\in X.$ (13) Indeed, to see that, note that if $m$ and $j$ are such that $2^{m+1}\equiv 1+2^{j}\pmod{p}$, then $(x_{1},x_{2},x_{3})=(2^{m+1+y},2^{y},2^{j+y})$ for $y=0,\ldots,t-1$, is also a solution of equation (13), and conversely, every solution $(x_{1},x_{2},x_{3})=(2^{y_{1}},2^{y_{2}},2^{y_{3}})$ of equation (13) arises from $2^{m+1}\equiv 1+2^{j}\pmod{p}$, where $m+1=y_{1}-y_{2}$ and $j=y_{3}-y_{2}$, by multiplying it with $2^{y_{2}}$. To estimate $I$, we use exponential sums. For a complex number $z$ put ${\bf e}(z)=\exp(2\pi iz)$. Using the fact that for $z\in\\{0,1,\ldots,p-1\\}$ the sum $\frac{1}{p}\sum_{a=0}^{p-1}{\bf e}(az/p)$ is $1$ if and only if $z=0$ and is $0$ otherwise, we get $I=\frac{1}{p}\sum_{x_{1},x_{2},x_{3}\in X}\sum_{a=0}^{p-1}{\bf e}(a(x_{1}-x_{2}-x_{3})/p).$ Separating the term for $a=0$, we get $I=\frac{(\\#X)^{3}}{p}+\frac{1}{p}\sum_{a=1}^{p-1}\sum_{x_{1},x_{2},x_{3}\in X}{\bf e}(a(x_{1}-x_{2}-x_{3})/p)=\frac{t^{3}}{p}+\frac{1}{p}\sum_{a=1}^{p-1}T_{a}T_{-a}^{2},$ where we put $T_{a}=\sum_{x_{1}\in X}{\bf e}(ax_{1}/p).$ A result of Heath- Brown and Konyagin [8], says that if $a\neq 0$, then $|T_{a}|\ll t^{3/8}p^{1/4}.$ Thus, $I=\frac{t^{3}}{p}+O(t^{9/8}p^{3/4}),$ leading to the fact that the number of values of $m$ modulo $t$ is $\\#\\{m\pmod{t}\\}\leq\frac{I}{t}\leq\frac{t^{2}}{p}+O(t^{1/8}p^{3/4}).$ Since also $m\leq x/p$, it follows that the number of acceptable values for $m$ is $\ll\frac{x}{pt}\left(\frac{t^{2}}{p}+t^{1/8}p^{3/4}\right)\ll\frac{xt}{p^{2}}+\frac{x}{t^{7/8}p^{1/4}}$ (note that $x/pt\geq 1$ because $pt<p^{2}<u^{2}<x$). Hence, $\\#{\mathcal{D}}_{3}\leq\sum_{p\in{\cal Q}(u)}\frac{xt}{p^{2}}+\sum_{p\in{\cal Q}(u)}\frac{x}{t^{7/8}p^{1/4}}=T_{1}+T_{2}.$ For the first sum $T_{1}$ above, we observe that $t\leq p/q$, therefore $t/p^{2}\leq 1/(pq)$. Thus, the first sum above is $T_{1}\ll\sum_{\begin{subarray}{c}p\in{\cal Q}(u)\end{subarray}}\frac{x}{pq}\ll\frac{xS^{\prime}}{q}\ll\frac{x\log\log x}{q^{3}}\ll x\frac{(\log\log\log x)^{3/2}}{(\log\log x)^{1/2}},$ (14) where we used again estimate (10). Finally, for the second sum $T_{2}$, we change the order of summation and thus get that $T_{2}\leq x\sum_{t\geq t_{0}}\frac{1}{t^{7/8}}\sum_{\begin{subarray}{c}p\in{\cal Q}(u)\\\ t(p)=t\end{subarray}}\frac{1}{p^{1/4}},$ (15) where $t_{0}=t_{0}(q)$ can be taken to be any lower bound on the smallest $t=t_{p}$ that can show up. We will talk about it later. For the moment, note that for a fixed $t$, $p$ is a prime factor of $2^{t}-1$. Thus, there are only $O(\log t)$ such primes. Furthermore, for each such prime we have $p>qt$. Hence, $T_{2}\ll\frac{x}{q^{1/4}}\sum_{t\geq t_{0}}\frac{\log t}{t^{9/8}}.$ Since $p\not\in{\cal P}_{1}\cup{\cal P}_{2}$, we get that $t_{p}>p^{1/2}(\log p)^{10}$. Since $p\geq 2q+1$, we get that $t\gg q^{1/2}(\log q)^{10}$. Thus, for large $x$ we may take $t_{0}=q^{1/2}(\log q)^{9}$ and get an upper bound for $T_{2}$. Hence, $\displaystyle T_{2}$ $\displaystyle\ll$ $\displaystyle\frac{x}{q^{1/4}}\sum_{t>q^{1/2}(\log q)^{9}}\frac{\log t}{t^{9/8}}\ll\frac{x}{q^{1/4}}\int_{q^{1/2}(\log q)^{9}}^{\infty}\frac{\log s}{s^{9/8}}\,ds$ (16) $\displaystyle\ll$ $\displaystyle\frac{x}{q^{1/4}}\left(-\frac{\log s}{s^{1/8}}\Big{|}_{q^{1/2}(\log q)^{9}}^{\infty}\right)\ll\frac{x}{q^{1/4+1/16}(\log q)^{1/8}}\ll\frac{x}{q^{5/16}(\log q)^{1/8}}$ $\displaystyle\ll$ $\displaystyle\frac{x(\log\log\log x)^{1/32}}{(\log\log x)^{5/32}}.$ Combining the bounds (14) and (16), we get that $\\#{\mathcal{D}}_{3}\ll\frac{x}{(\log\log x)^{1/7}},$ which together with the bounds (11) and (12) completes the proof of the theorem. ∎ Figure 2: The graph of $2\frac{\\#{\mathcal{C}}_{2}(x)}{x}$, $1\leq x\leq 63201$, $x$ odd. Although the density of ${\mathcal{C}}_{1}$ is zero, one my try to calculate the densities of ${\mathcal{C}}_{m}$ ($m>1$) hoping that they are positive and approach $1$ as $m\to\infty$. In the Figure 2 we have numerically calculated the density of ${\mathcal{C}}_{2}$ within the odd integers up to 63201. Nevertheless, we abandoned this idea having conjectured that the density of each ${\mathcal{C}}_{m}$ is still zero. However, the next section gives a way out to proving that $c_{k}/2^{k}$ goes to zero in arithmetic average over odd integers $k$. ## 6 The sets ${\mathcal{C}}_{m}$ for large $m$ In this section, we prove the following result. ###### Theorem 9. Put $m(k)=\lfloor\exp(4000(\log\log\log k)^{3})\rfloor$. The set of odd positive integers $k$ such that $k\in{\cal C}_{m(k)}$ is of asymptotic density $1/2$. In particular, most odd positive integers $k$ belong to ${\cal C}_{m(k)}$. ###### Proof. Let $x$ be large. We put $y=(\log\log x)^{3}.$ We start by discarding some of the odd positive integers $k\leq x$. We start with ${\cal A}_{1}=\\{k\leq x:q^{2}\mid k,~{}{\text{\rm or}}~{}q(q-1)\mid k,~{}{\text{\rm or}}~{}q^{2}\mid\phi(k)~{}{\text{\rm for~{}some~{}prime}}~{}q\geq y\\}.$ Clearly, if $n\in{\cal A}_{1}$, then there exists some prime $q\geq y$ such that either $q^{2}\mid n$, or $q(q-1)\mid n$, or $q^{2}\mid p-1$ for some prime factor $p$ of $n$, or $n$ is a multiple of two primes $p_{1}<p_{2}$ such that $q\mid p_{i}-1$ for both $i=1$ and $2$. The number of integers in the first category is $\displaystyle\leq\sum_{y<q\leq x^{1/2}}\left\lfloor\frac{x}{q^{2}}\right\rfloor\leq x\sum_{y<q\leq x^{1/2}}\frac{1}{q^{2}}\ll x\int_{y}^{x^{1/2}}\frac{dt}{t^{2}}\ll\frac{x}{y}=\frac{x}{(\log\log x)^{3}}=o(x)$ as $x\to\infty$. Similarly, the number of integers in the second category is $\leq\sum_{y<q<x^{1/2}+1}\left\lfloor\frac{x}{q(q-1)}\right\rfloor\ll x\sum_{y\leq q\leq x^{1/2}+1}\frac{1}{q^{2}}\ll\frac{x}{y}=\frac{x}{(\log\log x)^{3}}=o(x)$ as $x\to\infty$. The number of integers in the third category is $\displaystyle\leq\sum_{y<q\leq x^{1/2}}\sum_{\begin{subarray}{c}p\leq x\\\ p\equiv 1\pmod{q^{2}}\end{subarray}}\left\lfloor\frac{x}{p}\right\rfloor\leq x\sum_{y<q\leq x^{1/2}}\sum_{\begin{subarray}{c}p\leq x\\\ p\equiv 1\pmod{q^{2}}\end{subarray}}\frac{1}{p}$ $\displaystyle\ll x\sum_{y<q\leq x^{1/2}}\frac{\log\log x}{\phi(q^{2})}\ll x\log\log x\sum_{y<q\leq x^{1/2}}\frac{1}{q^{2}}$ $\displaystyle\ll\frac{x\log\log x}{y}=\frac{x}{(\log\log x)^{2}}=o(x)$ as $x\to\infty$, while the number of integers in the fourth and most numerous category is $\displaystyle\leq\sum_{y<q\leq x^{1/2}}\sum_{\begin{subarray}{c}p_{1}<p_{2}<x\\\ p_{i}\equiv 1\pmod{q},~{}i=1,2\end{subarray}}\left\lfloor\frac{x}{p_{1}p_{2}}\right\rfloor\leq x\sum_{y<q\leq x^{1/2}}\sum_{\begin{subarray}{c}p_{1}<p_{2}<x\\\ p_{i}\equiv 1\pmod{q},~{}i=1,2\end{subarray}}\frac{1}{p_{1}p_{2}}$ $\displaystyle\leq x\sum_{y<q\leq x^{1/2}}\frac{1}{2}\left(\sum_{\begin{subarray}{c}p\leq x\\\ p\equiv 1\pmod{q}\end{subarray}}\frac{1}{p}\right)^{2}\ll x\sum_{y<q\leq x^{1/2}}\left(\frac{\log\log x}{\phi(q)}\right)^{2}$ $\displaystyle\ll x(\log\log x)^{2}\sum_{y<q\leq x^{1/2}}\frac{1}{q^{2}}\ll\frac{x(\log\log x)^{2}}{y}=\frac{x}{\log\log x}=o(x)$ as $x\to\infty$. We now let ${\cal Q}=\\{p:t_{p}\leq p^{1/3}\\},$ and let ${\cal A}_{2}$ be the set of $k\leq x$ divisible by some $q\in{\cal Q}$ with $q>y$. To estimate $\\#{\cal A}_{2}$, we begin by estimating the counting function $\\#{\cal Q}(t)$ of ${\cal Q}$ for positive real numbers $t$. Clearly, $2^{\\#{\cal Q}(t)}\leq\prod_{q\in{\cal Q}(t)}q\leq\prod_{s\leq t^{1/3}}(2^{s}-1)<2^{\sum_{s\leq t^{1/3}}s}\leq 2^{t^{2/3}},$ so $\\#{\cal Q}(t)\leq t^{2/3}.$ (17) By Abel’s summation formula, we now get that $\\#{\cal A}_{2}\leq\sum_{\begin{subarray}{c}y\leq q\leq x\\\ q\in{\cal Q}\end{subarray}}\left\lfloor\frac{x}{q}\right\rfloor\leq x\sum_{\begin{subarray}{c}y\leq q\leq x\\\ q\in{\cal Q}\end{subarray}}\frac{1}{q}\ll x\int_{y}^{x}\frac{d\\#{\cal Q}(t)}{t}\ll\frac{x}{y^{1/3}}=\frac{x}{\log\log x}=o(x)$ as $x\to\infty$. Recall now that $P(m)$ stands for the largest prime factor of the positive integer $m$. Known results from the theory of distribution of smooth numbers show that uniformly for $3\leq s\leq t$, we have $\displaystyle\Psi(t,s)=\\#\\{m\leq t:P(m)\leq s\\}\ll t\exp(-u/2),$ (18) where $u=\log t/\log s$ (see [15, Section III.4]). Thus, putting $z=\exp\left(32(\log\log\log x)^{2}\right),$ we conclude that the estimate $\Psi(t,y)\ll\frac{t}{(\log\log x)^{5}}$ (19) holds uniformly for large $x$ once $t>z$, because in this case $u=\frac{\log t}{\log y}\geq\frac{32}{3}\log\log\log x,$ therefore $\frac{u}{2}\geq\frac{16}{3}\log\log\log x,$ so, in particular, $u/2>5\log\log\log x$ holds for all large $x$. Furthermore, if $t>Z=\exp((\log\log x)^{2})$, then $u=\frac{\log t}{\log y}=\frac{(\log\log x)^{2}}{3\log\log\log x},$ so $u/2>2\log\log x$ one $x$ is sufficiently large. Thus, in this range, inequality (19) can be improved to $\Psi(t,y)\ll\frac{x}{\exp(2\log\log x)}\ll\frac{x}{(\log x)^{2}}.$ (20) Now for a positive integer $m$, we put $d(m,y)$ for the largest divisor $d$ of $m$ which is $y$-smooth, that is, $P(d)\leq y$. Let ${\cal A}_{3}$ be the set of $k\leq x$ having a prime factor $p$ exceeding $z^{10}$ such that $d(p-1,y)>p^{1/10}$. To estimate $\\#{\cal A}_{3}$, we fix a $y$-smooth number $d$ and a prime $p$ with $z^{10}<p<d^{10}$ such that $p\equiv 1\pmod{d}$, and observe that the number of $n\leq x$ which are multiples of this prime $p$ is $\leq\lfloor x/p\rfloor$. Note also that $d>p^{1/10}>z$. Summing up over all the possibilities for $d$ and $p$, we get that $\\#{\cal A}_{3}$ does not exceed $\displaystyle\sum_{\begin{subarray}{c}z<d\\\ P(d)\leq y\end{subarray}}\sum_{\begin{subarray}{c}p\leq x\\\ p\equiv 1\pmod{d}\end{subarray}}\left\lfloor\frac{x}{p}\right\rfloor\leq x\sum_{\begin{subarray}{c}z<d\\\ P(d)\leq y\end{subarray}}\sum_{\begin{subarray}{c}p\leq x\\\ p\equiv 1\pmod{d}\end{subarray}}\frac{1}{p}\ll x\sum_{\begin{subarray}{c}z<d\\\ P(d)\leq y\end{subarray}}\frac{\log\log x}{\phi(d)}$ $\displaystyle\ll x(\log\log x)^{2}\sum_{\begin{subarray}{c}z<d\\\ P(d)\leq y\end{subarray}}\frac{1}{d}\ll x(\log\log x)^{2}\int_{z}^{x}\frac{d\Psi(t,y)}{t}$ $\displaystyle\ll x(\log\log x)^{2}\left(\frac{\Psi(t,y)}{t}\Big{|}_{z}^{x}+\int_{z}^{x}\frac{\Psi(t,y)}{t^{2}}dt\right)$ $\displaystyle\ll\frac{x}{(\log\log x)^{3}}+x(\log\log x)^{2}\int_{z}^{x}\frac{\Psi(t,y)dt}{t^{2}}.$ In the above estimates, we used aside from the Abel summation formula and inequality (19), also the minimal order of the Euler function $\phi(d)/d\gg 1/\log\log x$ valid for all $d\in[1,x]$. It remains to bound the above integral. For this, we split it at $Z$ and use estimates (19) and (20). In the smaller range, we have that $\int_{z}^{Z}\frac{\Psi(t,y)dt}{t^{2}}\ll\frac{1}{(\log\log x)^{5}}\int_{z}^{Z}\frac{dt}{t}\ll\frac{\log Z}{(\log\log x)^{5}}\ll\frac{1}{(\log\log x)^{3}}.$ In the larger range, we use estimate (20) and get $\int_{Z}^{x}\frac{\Psi(t,y)dt}{t^{2}}\ll\frac{1}{(\log x)^{2}}\int_{Z}^{x}\frac{dt}{t}\ll\frac{1}{\log x}.$ Putting these together we get that $\\#{\cal A}_{3}\ll\frac{x}{(\log\log x)^{3}}+x(\log\log x)^{2}\left(\frac{1}{(\log\log x)^{3}}+\frac{1}{\log x}\right)=o(x)$ as $x\to\infty$. Now let $\ell=d(k,z^{10})$. Put $w=\exp(1920(\log\log\log x)^{3}),$ and put ${\cal A}_{4}$ for the set of $k\leq x$ such that $\ell>w$. Note that each such $k$ has a divisor $d>w$ such that $P(d)\leq z^{10}$. Since for such $d$ we have $\frac{\log d}{\log(z^{10})}=6\log\log\log x,$ we get that in the range $t\geq w$, $u/2>3\log\log\log x,$ for large $x$, so $\Psi(t,z^{10})<\frac{t}{(\log\log x)^{3}}$ (21) uniformly for such $t$ once $x$ is large. Furthermore, if $t>Z_{1}=\exp(1280\log\log x(\log\log\log x)^{2}),$ then $u=\frac{\log t}{\log z^{10}}>4\log\log x$ therefore $u/2>2\log\log x$. In particular, $\Psi(t,z^{10})\ll\frac{x}{(\log x)^{2}}$ (22) in this range. By an argument already used previously, we have that $\\#{\cal A}_{4}$ is at most $\displaystyle\leq\sum_{\begin{subarray}{c}w<d<x\\\ P(d)\leq z^{10}\end{subarray}}\left\lfloor\frac{x}{d}\right\rfloor\leq x\sum_{\begin{subarray}{c}w<d<x\\\ P(d)\leq z^{10}\end{subarray}}\frac{1}{d}\ll x\int_{w}^{x}\frac{d\Psi(t,z^{10})}{t}$ $\displaystyle\ll x\left(\frac{\Psi(t,z^{10})}{t}\Big{|}_{t=w}^{t=x}+\int_{w}^{x}\frac{\Psi(t,z^{10})dt}{t^{2}}\right)$ $\displaystyle\ll x\left(\frac{1}{(\log\log x)^{3}}+\int_{w}^{Z_{1}}\frac{\Psi(t,z^{10})dt}{t^{2}}+\int_{Z_{1}}^{x}\frac{\Psi(t,z^{10})dt}{t^{2}}\right)$ $\displaystyle\ll x\left(\frac{1}{(\log\log x)^{3}}+\frac{\log Z_{1}}{(\log\log x)^{3}}+\frac{\log x}{(\log x)^{2}}\right)=o(x)$ as $x\to\infty$, where the above integral was estimated by splitting it at $Z_{1}$ and using estimates (21) and (22) for the lower and upper ranges respectively. Let ${\cal A}_{5}$ be the set of $k\leq x$ which are coprime to all primes $p\in[y,z^{10}]$. By the Brun method, $\\#{\cal A}_{5}\ll x\prod_{y\leq q\leq z}\left(1-\frac{1}{q}\right)\ll\frac{x\log y}{\log z}\ll\frac{x}{\log\log\log x}=o(x)$ as $x\to\infty$. We next let ${\cal A}_{6}$ be the set of $k\leq x$ such that $P(k)<w^{100}$. Clearly, $\\#{\cal A}_{6}=\Psi(x,w^{100})=x\exp\left(-c_{1}\frac{\log x}{(\log\log\log x)^{3}}\right)=o(x)$ as $x\to\infty$, where $c_{1}=1/384000$. Finally, we let ${\cal A}_{7}=\\{k\leq x:dp\mid k~{}{\text{\rm for~{}some}}~{}p\equiv 1\pmod{d}~{}{\text{\rm and}}~{}p<d^{3}\\}.$ Assume that $k\in{\cal A}_{7}$. Then there is a prime factor $p$ of $k$ and a divisor $d$ of $p-1$ of size $d>p^{1/3}$ such that $dp\mid k$. Fixing $d$ and $p$, the number of such $n\leq x$ is $\leq\lfloor x/(dp)\rfloor$. Thus, $\displaystyle\\#{\cal A}_{7}$ $\displaystyle\leq$ $\displaystyle\sum_{y\leq p\leq x}\sum_{\begin{subarray}{c}d\mid p-1\\\ d>p^{1/3}\end{subarray}}\left\lfloor\frac{x}{dp}\right\rfloor\leq x\sum_{y\leq p\leq x}\frac{1}{p}\sum_{\begin{subarray}{c}d\mid p-1\\\ d>p^{1/3}\end{subarray}}\frac{1}{d}$ $\displaystyle\ll$ $\displaystyle\sum_{y\leq p\leq x}\frac{1}{p}\left(\frac{\tau(p-1)}{p^{1/3}}\right)\ll x\sum_{y\leq p\leq x}\frac{\tau(p-1)}{p^{1+1/3}}\ll x\sum_{y\leq p\leq x}\frac{1}{p^{5/4}}$ $\displaystyle\ll$ $\displaystyle x\int_{y}^{x}\frac{dt}{t^{5/4}}\ll\frac{x}{y^{1/4}}=\frac{x}{(\log\log x)^{3/4}}=o(x)$ as $x\to\infty$. Here, we used $\tau(m)$ for the number of divisors of the positive integer $m$ and the fact that $\tau(m)\ll_{\varepsilon}m^{\varepsilon}$ holds for all $\varepsilon>0$ (with the choice of $\varepsilon=1/12$). From now on, $k\leq x$ is odd and not in $\bigcup_{1\leq i\leq 7}{\cal A}_{i}$. From what we have seen above, most odd integers below $x$ have this property. Then $\ell\leq w$ because $k\not\in{\cal A}_{4}$. Further, $k/\ell$ is square-free because $k\not\in{\cal A}_{1}$. Moreover, if $p\mid k/\ell$, then $p>z^{10}>y$, therefore $t_{p}>p^{1/3}$ because $k\not\in{\cal A}_{2}$. Since $k\not\in{\cal A}_{3}$, we get that $d(p-1,y)<p^{1/10}$, so $t_{p}^{\prime}=t_{p}/\gcd(t_{p},d(p-1,y))>p^{1/3-1/10}>p^{1/5}$ for all such $p$. Moreover, $t_{p}^{\prime}$ is divisible only by primes $>z>y$, so if $p_{1}$ and $p_{2}$ are distinct primes dividing $k/\ell$, then $t_{p_{1}}^{\prime}$ and $t_{p_{2}}^{\prime}$ are coprime because $k\not\in{\cal A}_{1}$. Finally, $\ell>y$ because $k\not\in{\cal A}_{5}$. Furthermore, for large $x$ we have that $w>y$, so $k>\ell$ and in fact $k/\ell$ is divisible by a prime $>w^{100}$ because $k\not\in{\cal A}_{6}$. We next put $n={\text{\rm lcm}}[d(\phi(k),y),\phi(\ell)]$. We let $n_{0}$ stand for the minimal positive integer such that $n_{0}\equiv-k+1\pmod{\phi(\ell)}$ and let $m=n_{0}+\ell\phi(\ell)$. Note that $m\leq 2\ell\phi(\ell)\leq 2w^{2}=2\exp(3840(\log\log\log x)^{3}).$ We may also assume that $k>x/\log x$ since there are only at most $x/\log x=o(x)$ positive integers $k$ failing this property. Since $k>x/\log x$, we get that $m<2\exp(3840(\log\log\log x)^{3})<\lfloor\exp(4000(\log\log\log k)^{3})\rfloor=m(k)$ holds for large $x$. We will now show that this value for $m$ works. First of all $m+k=n_{0}+\ell\phi(\ell)+k\equiv 1\pmod{\phi(\ell)}$ so $\begin{split}2^{m+k}-1&\equiv 1\equiv 2^{\phi(\ell)-1}+2^{\phi(\ell)-2}+\cdots+2^{\phi(\ell)-(n_{0}-1)}+2^{\phi(\ell)-(n_{0}-1)}+\\\ &+2^{x_{1}n}+\cdots+2^{x_{t}n}\pmod{\ell},\end{split}$ where $t=\ell\phi(\ell)$ and $x_{1},\ldots,x_{t}$ are any nonnegative integers. Let $U=2^{m+k}-1-2^{\phi(\ell)-1}-\cdots-2^{\phi(\ell)-(n_{0}-1)}-2^{\phi(\ell)-(n_{0}-1)}.$ Then $U\equiv\sum_{i=1}^{t}2^{x_{i}\phi(\ell)}\pmod{\ell}$ for any choice of the integers $x_{1},\ldots,x_{t}$. Let $p$ be any prime divisor of $k/\ell$. Clearly, $\gcd(t_{p},n)=d(t_{p},y),$ because $t_{p}\mid\phi(k)$ and $n\not\in{\cal A}_{1}$. In particular, $t_{p}^{\prime}=\frac{t_{p}}{\gcd(t_{p},n)}\geq\frac{t_{p}}{\gcd(d(\phi(k),y),p-1)}\geq p^{1/3-1/10}>p^{1/5}.$ Let $X=\\{2^{jn}\pmod{p}\\}$. Certainly, the order of $2^{n}$ modulo $p$ is precisely $t_{p}^{\prime}$. So, $\\#X=t_{p}^{\prime}>p^{1/5}$. A recent result of Bourgain, Glibichuk and Konyagin (see Theorem 5 in [1]), shows that there exists a constant $T$ which is absolute such that for all integers $\lambda$, the equation $\lambda\equiv 2^{x_{1}n}+\cdots+2^{x_{t}n}\pmod{p}$ has an integer solutions $0\leq x_{1},\ldots,x_{t}<t_{p}^{\prime}$ once $t>T$. In fact, for large $p$ the number of such solutions $N(t,p,\lambda)=\\#\\{(x_{1},\ldots,x_{t}):0\leq x_{1},\ldots,x_{t}\leq t_{p}\\}$ satisfies $N(t,p,\lambda)\in\left[\frac{\\#X^{t}}{2p},\frac{2\\#X^{t}}{p}\right]$ independently in the parameter $\lambda$ and uniformly in the number $t$. In particular, if we let $N_{1}(t,p,\lambda)$ be the number of such solutions with $x_{i}=x_{j}$ for some $i\neq j$, then $N_{1}(t,p,\lambda)\ll t^{2}\\#X^{t-1}/p$. Indeed, the pair $(i,j)$ with $i\neq j$ can be chosen in $O(t^{2})$ ways, and the common value of $x_{i}=x_{j}$ can be chosen in $\\#X$ ways. Once these two data are chosen, then the number of ways of choosing $x_{s}\in\\{0,1,\ldots,t_{p}^{\prime}-1\\}$ with $s\in\\{1,2,\ldots,t\\}\backslash\\{i,j\\}$ such that $\lambda-2^{x_{i}n}-2^{x_{j}n}\equiv\sum_{\begin{subarray}{c}1\leq s\leq t\\\ s\neq i,j\end{subarray}}2^{x_{s}n}\pmod{p}$ is $N(t-2,p,\lambda-2^{x_{i}n}-2^{x_{j}n})\ll\\#X^{t-2}/p$ for $t>T+2$. In conclusion, if all solutions $x_{1},\ldots,x_{t}$ have two components equal, then $p^{1/5}\ll\\#X\ll t^{2}$, so $p\ll t^{10}$. For us, $t\leq 2w^{2}$, so $p\ll w^{20}$. Since $P(k)=P(k/\ell)>w^{100}$, it follows that at least for the largest prime $p=P(k)$, we may assume that $x_{1},\ldots,x_{t}$ are all distinct modulo $p$ for a suitable value of $\lambda$. We apply the above result with $\lambda=U$, $t=\ell\phi(\ell)$ (note that since $t>y$, it follows that $t>T+2$ does indeed hold for large values of $x$), and write ${\bf x}(p)=(x_{1}(p),\ldots,x_{t}(p))$ for a solution of $U\equiv 2^{x_{1}(p)n}+\cdots+2^{x_{t}(p)n}\pmod{p},\qquad 0\leq x_{1}(p)\leq\ldots\leq x_{t}(p)<t_{p}^{\prime}.$ We also assume that for at least one prime (namely the largest one) the $x_{i}(p)$’s are distinct. Now choose integers $x_{1},\ldots,x_{t}$ such that $x_{i}\equiv x_{i}(p)\pmod{t_{p}^{\prime}}$ for all $p\mid k/\ell$. This is possible by the Chinese Remainder Lemma since the numbers $t_{p}^{\prime}$ are coprime as $p$ varies over the distinct prime factors of $k/\ell$. We assume that for each $i$, $x_{i}$ is the minimal nonnegative integer in the corresponding arithmetic progression modulo $\prod_{p\mid k/\ell}t_{p}^{\prime}$. Further, since $nx_{i}(p)$ are distinct modulo $t_{p}^{\prime}$ when $p=P(k)$, it follows that $nx_{i}$ are also distinct for $i=1,\ldots,t$. Hence, for such $x_{i}$’s we have that $U-\sum_{i=1}^{t}2^{x_{i}n}$ is a multiple of all $p\mid k/\ell$, and since $k/\ell$ is square-free, we get that $U\equiv\sum_{i=1}^{t}2^{x_{i}n}\pmod{k/\ell}.$ But the above congruence is also valid modulo $\ell$, so it is valid modulo $k={\text{\rm lcm}}[\ell,k/\ell]$, since $\ell$ and $k/\ell$ are coprime. Thus, $U\equiv\sum_{i=1}^{t}2^{x_{i}n}\pmod{k},$ or $2^{k+m-1}-1\equiv 2^{\phi(\ell)-1}+\cdots+2^{\phi(\ell)-(n_{0}-1)}+2^{\phi(\ell)-(n_{0}-1)}+\sum_{i=1}^{t}2^{x_{i}n}\pmod{k}.$ As we have said, the numbers $x_{i}n$ are distinct and they can be chosen of sizes at most $n{\text{\rm lcm}}[t_{p}^{\prime}:p\mid k\ell]\leq\phi(k)\leq k$. Finally, $nx_{i}$ are divisible by $\phi(\ell)$ whereas none of the numbers $\phi(\ell)-j$ for $j=1,\ldots,n_{0}-1$ is unless $n_{0}=1$. Thus, assuming that $n_{0}\neq 1$, we get that all the $m=t+n_{0}$ exponents are distinct except for the fact that $\phi(\ell)-(n_{0}-1)$ appears twice. Let us first justify that $n_{0}\neq 1$. Recalling the definition of $n_{0}$, we get that if this were so then $\phi(\ell)\mid k$. However, we have just said that $\ell$ has a prime factor $p>y$. If $\phi(\ell)\mid k$, then $k$ is divisible by both $p$ and $p-1$ for some $p>y$ and this is impossible since $n\not\in{\cal A}_{1}$. Finally, to deal with the repetition of the exponent $\phi(\ell)-(n_{0}-1)$, we replace this by $\phi(\ell)-(n_{0}-1)+t_{k}$, where as usual $t_{k}$ is the order of $2$ modulo $n$. We show that with this replacement, all the exponents are distinct. Indeed, this replacement will not change the value of $2^{\phi(\ell)-(n_{0}-1)+t_{k}}\pmod{k}$. Assume that after this replacement, $\phi(\ell)-(n_{0}-1)+t_{k}$ is still one of the remaining exponents. If it has become a multiple of $n$, it follows that it is in particular divisible by $t_{p}$ for all primes $p\mid\ell$. Since $t_{p}\mid t_{k}$ and $t_{p}\mid\phi(\ell)$ for all primes $p\mid\ell$, we get that $t_{p}\mid n_{0}-1$, so $t_{p}\mid k$. Since $\ell$ is divisible by some prime $p>y$ (because $k\not\in{\cal A}_{5}$), we get that $t_{p}\mid k$. Since $k\not\in{\cal A}_{2}$, we get that $t_{p}>p^{1/3}$. Thus, $k$ is divisible by a prime $p>y$ and a divisor $d$ of $p-1$ with $d>p^{1/3}$, and this is false since $n\not\in{\cal A}_{7}$. Hence, this is impossible, so it must be the case that $\phi(\ell)-(n_{0}-1)+t_{k}\in\\{\phi(\ell)-1,\ldots,\phi(\ell)-(n_{0}-1)\\}$. This shows that $t_{k}\leq n_{0}\leq\ell\phi(\ell)\leq 2^{10}w^{20}$. However, $t_{k}$ is a multiple of $t_{P(k)}\geq P(k)^{1/3}$, showing that $P(k)\leq 2^{30}w^{60}$, which is false for large $x$ since $k\not\in{\cal A}_{6}$. Thus, the new exponents are all distinct for our values of $k$. As far as their sizes go, note that since $k$ has at least two odd prime factors, it follows that $t_{k}\mid\phi(k)/2$, therefore $\phi(\ell)-(n_{0}-1)+t_{k}\leq w+\phi(k)/2<w+k/2<k$ since $k>2w$ for large $x$. Thus, we have obtained a representation of $2^{k+m}-1$ modulo $m$ of the form $2^{j_{1}}+\cdots+2^{j_{m}}\pmod{k}$ where $0\leq j_{1}<\ldots<j_{m}\leq k$, which shows that $k\in{\cal C}_{m}$. Since $m\leq m(k)$ and ${\cal C}_{m}\subset{\cal C}_{m(k)}$, the conclusion follows. ∎ ###### Remark 10. The above proof shows that in fact the number of odd $k<x$ such that $k\not\in{\cal C}_{m(k)}$ is $O(x/\log\log\log x)$. ###### Corollary 11. For large $x$, the inequality $c_{k}/2^{k}<2^{m(k)-(\log k)/(\log 2)}$ holds for all odd $k<x$ with at most $O(x/\log\log\log x)$ exceptions. ###### Proof. This follows from the fact that $c_{k}=a_{k}/k\leq 2^{k+m}/k$, where $k\in{\cal C}_{m}$ (see Theorem 6), together with above Theorem 9 and Remark 10. ∎ ###### Corollary 12. The estimate $\frac{1}{x}\sum_{\begin{subarray}{c}1\leq k\leq x\\\ k~{}{\text{\rm odd}}\end{subarray}}\frac{c_{k}}{2^{k}}=O\left(\frac{1}{\log\log\log x}\right)$ holds for all $x$. ###### Proof. If $k\leq x/\log x$ is odd, then $c_{k}/2^{k}\leq 1$, so $\sum_{\begin{subarray}{c}k\leq x/\log x\\\ k~{}{\text{\rm odd}}\end{subarray}}\frac{c_{k}}{2^{k}}\leq\frac{x}{\log x}.$ If $k\in[x/\log x,x]$ but $k\not\in{\cal C}_{m(k)}$, then still $c_{k}/2^{k}\leq 1$ and, by the Corollary 11, the number of such $k$’s is $O(x/\log\log\log x)$. Thus, $\sum_{\begin{subarray}{c}k\in[x/\log x,x]\\\ k\not\in{\cal C}_{m(k)}\\\ k~{}{\text{\rm odd}}\end{subarray}}\frac{c_{k}}{2^{k}}\ll\frac{x}{\log\log\log x}.$ For the remaining odd values of $k\leq x$, we have that $\frac{c_{k}}{2^{k}}\leq 2^{m(k)-(\log k)/(\log 2)},$ so it suffices to show that $2^{m(k)-(\log k)/(\log 2)}<\frac{1}{\log\log\log x},$ is equivalent to $(\log k)/(\log 2)-m(k)>\log\log\log\log x/\log 2,$ which in turn is implied by $\log(x/\log x)-\log\log\log\log x>(\log 2)\exp(4000(\log\log\log x)^{3}),$ and this is certainly true for large $x$. Thus, indeed, $\sum_{\begin{subarray}{c}1\leq k\leq x\\\ k~{}{\text{\rm odd}}\end{subarray}}\frac{c_{k}}{2^{k}}=O\left(\frac{x}{\log\log\log x}\right),$ which is what we wanted to prove. ∎ In particular, $\frac{1}{x}\sum_{\begin{subarray}{c}k\leq x\\\ k~{}{\text{\rm odd}}\end{subarray}}\frac{c_{k}}{2^{k}}=o(1)$ (23) as $x\to\infty$. One can adapt these techniques to obtain that the whole sequence $c_{k}/2^{k}$ is convergent to $0$ in arithmetic average. In order to do so, the sets ${\cal C}_{m}$ should be suitably modified and an analog of Theorem 9 for these new sets should be proved. We leave this for a subsequent work. ## 7 Existence and bounds for $a_{k}$ in base $q>2$ Let $q\geq 2$ be a fixed integer and let $x$ be a positive real number. Put $\displaystyle V_{k}(x)$ $\displaystyle=$ $\displaystyle\\{0\leq n<x:s_{q}(n)=k\\},$ $\displaystyle V_{k}(x;h,m)$ $\displaystyle=$ $\displaystyle\\{0\leq n<x:s_{q}(n)=k,n\equiv h\pmod{m}\\}.$ Mauduit and Sárközy proved in [12] that if $\gcd(m,q(q-1))=1$, then there exists some constant $c_{0}$ depending on $q$ such that if we put $\ell=\min\left\\{k,(q-1)\lfloor\log x/\log q\rfloor-k\right\\},$ then $V_{k}(x)$ is well distributed in residues classes modulo $m$ provided that $m<\exp(c_{0}\ell^{1/2})$. Taking $m=k$ and $h=0$, we deduce that if $k<\exp(c_{0}\ell^{1/2})$, then $V_{k}(x;0,k)=(1+o(1))V_{k}(x)/k$ as $x\to\infty$ uniformly in our range for $k$. The condition on $k$ is equivalent to $\log k\ll\ell^{1/2}$, which is implied by $k+O((\log k)^{2})\ll\log x$. Thus, we have the following result. ###### Lemma 13. Let $q\geq 2$ be fixed. There exists a constant $c_{1}$ such that if $k$ is any positive integer with $\gcd(k,q(q-1))=1$, then $V_{k}(x)$ is well distributed in arithmetic progressions of modulus $k$ whenever $x>\exp(c_{1}k)$. Corollary 2 of [12] implies that if $\Delta=\left|\frac{q-1}{2\log q}\log x-k\right|=o(\log x)\qquad{\text{\rm as}}~{}x\to\infty,$ (24) then the estimate $\\#V_{k}(x)=\frac{x}{(\log x)^{1/2}}\exp\left(-c_{3}\frac{\Delta^{2}}{\log x}+O\left(\frac{\Delta^{3}}{(\log x)^{2}}+\frac{1}{(\log x)^{1/2}}\right)\right)$ holds with some explicit constant $c_{3}$ depending on $q$. As a corollary of this result, we deduce the following result. ###### Lemma 14. If condition (24) is satisfied, then $V_{k}(x)\not=\emptyset$. In case $k$ and $q$ are coprime but $k$ and $q-1$ are not, we may apply instead Theorem B of [11] with $m=k$ and $h=0$ to arrive at a similar result. ###### Lemma 15. Assume that $q\geq 2$ is fixed. There exists a constant $c_{4}$ depending only on $q$ such that if $k$ is a positive integer with $\gcd(k,q)=1$, and $x\geq\exp(c_{4}k)$, then $V_{k}(x;0,k)\neq 0$. One can even remove the coprimality condition on $q$ and $k$. Assume that $x$ is sufficiently large such that $\Delta\leq c_{5}(\log x)^{5/8},$ (25) where $c_{5}$ is some suitable constant depending on $q$. Using Theorem C and Lemma 5 of [11] with $m=k$ and $h=0$, we obtain the following result. ###### Lemma 16. Assume that both estimates (25) and $k<2^{(\log x)^{1/4}}$ hold. Then $V_{k}(x;0,k)\neq\emptyset$. A sufficient condition on $x$ for Lemma 16 above to hold is that $x>\exp(c_{6}k)$, where $c_{6}$ is a constant is a constant that depends on $q$. Putting Lemmas 15 and 16 together we obtain the next theorem. ###### Theorem 17. For all $q\geq 2$ there exists a constant $c_{6}$ depending on $q$ such that for all $k\geq 1$ there exists $n\leq\exp(c_{6}k)$ with $s_{q}(kn)=k$. Consequently, $a_{k}=\exp(O(k))$ for all $k$, and in particular it is nonzero. The following example of Lemma 18 shows that $a_{k}=\exp(o(k))$ does not always hold as $k\to\infty$. ###### Lemma 18. If $q>2$, then $a_{q^{m}}=q^{m}\left(2q^{\frac{q^{m}-1}{q-1}}-1\right).$ If $q=2$, then $a_{2^{m}}=2^{m}(2^{2^{m}}-1)$. ###### Proof. The fact that $s_{q}(a_{q^{m}})=q^{m}$ for all $q\geq 2$ is immediate. We now show the minimality of the given $a_{q^{m}}$ with this property. Let $\alpha_{m}=(q^{m}-1)/(q-1)$. Note that every digit of $q^{\alpha_{m}}-1$ in base $q$ is maximal, so $q^{\alpha_{m}}-1$ is minimal such that $s_{q}(q^{\alpha_{m}}-1)=q^{m}-1$. Since $q^{\alpha_{m}}-1=(q-1)q^{\alpha_{m}-1}+(q-1)q^{\alpha_{m}-2}+\cdots+(q-1),$ then $a_{q^{m}}$ must contain the least term $q^{t}$, where $t>\alpha_{m}-1$ such that its sum of digits is $q^{m}$ and $q^{m}|a_{q^{m}}$. The least term is obviously $q^{\alpha_{m}}$, and it just happens that $a_{q^{m}}$ such defined satisfies the mentioned conditions. ∎ ## References * [1] J. Bourgain, A. A. Glibichuk and S. V. Konyagin, ‘Estimates for the number of sums and products and for exponential sums in finite fields of prime order’, J. London Math. Soc. 73 (2006), 380–398. * [2] W. D. Banks, M. Z. Garaev, F. Luca and I. E. Shparlinski, ‘Uniform distribution of fractional parts related to pseudoprimes’, Canadian J. Math., to appear. * [3] T. Cai, ‘On 2-Niven numbers and 3-Niven numbers’, Fibonacci Quart. 34 (1996), 118–120. * [4] C. N. Cooper and R. E. Kennedy, ‘On consecutive Niven numbers’, Fibonacci Quart. 21 (1993), 146–151. * [5] J. M. De Koninck and N. Doyon, ‘On the number of Niven numbers up to $x$’, Fibonacci Quart. 41 (2003), 431–440. * [6] J. M. De Koninck, N. Doyon, and I. Katai, ‘On the counting function for the Niven numbers’, Acta Arith. 106 (2003), 265–275. * [7] H. G. Grundman, ‘Sequences of consecutive Niven numbers’, Fibonacci Quart. 32 (1994), 174–175. * [8] D. R. Heath-Brown and S. Konyagin, ‘New bounds for Gauss sums derived from $k$th powers’, Quart. J. Math. 51 (2000), 221–235. * [9] H.-K. Indlekofer and N. M. Timofeev, ‘Divisors of shifted primes’, Publ. Math. Debrecen 60 (2002), 307–345. * [10] L. C. Lagarias and A. M. Odlyzko, ‘Effective versions of Chebotarev’s Density Theorem’, in Algebraic Number Fields (A. Frölich, ed.), Academic Press, New York, 1977, 409–464. * [11] C. Mauduit, C. Pomerance and A. Sárközy, ‘On the distribution in residue classes of integers with a fixed digit sum’, The Ramanujan J. 9 (2005), 45–62. * [12] C. Mauduit and A. Sárközy, ‘On the arithmetic structure of integers whose sum of digits is fixed’, Acta Arith. 81 (1997), 145–173. * [13] I. Niven, H.S. Zuckerman and H.L. Montgomery, ‘An introduction to the theory of numbers’, Fifth Edition, John Wiley $\&$ Sons, Inc., 1991. * [14] F. Pappalardi, ‘On Hooley’s theorem with weights’, Rend. Sem. Mat. Univ. Pol. Torino 53 (1995), 375–388. * [15] G. Tenenbaum, ‘Introduction to analytic and probabilistic number theory’, Cambridge University Press, 1995. * [16] I. Vardi, ‘Niven numbers’, §2.3 in Computational Recreations in Mathematics, Addison-Wesley, 1991, 19 and 28–31.

arxiv-papers

2024-09-04T02:48:54.141707

0803.0512

FERMILAB-Pub-08-051-T # Accumulating evidence for nonstandard leptonic decays of $D_{s}$ mesons Bogdan A. Dobrescu and Andreas S. Kronfeld Theoretical Physics Department, Fermi National Accelerator Laboratory, Batavia, Illinois, USA (March 4, 2008; revised April 28, 2008) ###### Abstract The measured rate for $D_{s}^{+}\\!\to\\!\ell^{+}\nu$ decays, where $\ell$ is a muon or tau, is larger than the standard model prediction, which relies on lattice QCD, at the 3.8$\sigma$ level. We discuss how robust the theoretical prediction is, and we show that the discrepancy with experiment may be explained by a charged Higgs boson or a leptoquark. ###### pacs: 13.20.Fc,12.60-i,14.80.-j Introduction.—The pattern of flavor and $CP$ violation of the standard model has been established by a wide range of experiments. This agreement, however, leaves room for new flavor effects to show up as calculations and measurements improve. Intriguingly, decays of the $D_{s}$ meson (the lightest $c\bar{s}$ state) could be more sensitive to new physics than any other process explored so far. It suffices that a new particle couples predominantly to leptons and up-type quarks, but not to the first generation. In this Letter we examine the leptonic decays of the $D_{s}$. Recently, the calculation of the relevant QCD matrix element has improved significantly, and more accurate measurements of the rate have been made. The average of the experimental results disagrees with the standard model by almost four standard deviations. We discuss the evidence, and propose that a nonstandard amplitude interferes with the standard $W$-mediated amplitude. We show that the tree- level exchange of a spin-0 particle with mass of order 1 TeV may account for the discrepancy. Leptonic $D_{s}$ decays.—The $D_{s}\\!\to\\!\ell\nu$ branching fraction, where $\ell$ is a charged lepton of mass $m_{\ell}$, is given in the standard model by $B(D_{s}\\!\to\\!\ell\nu)=\frac{m_{D_{s}}}{8\pi}\tau_{D_{s}}f_{D_{s}}^{2}\left|G_{F}V^{*}_{cs}m_{\ell}\right|^{2}\left(1\\!-\frac{m_{\ell}^{2}}{m_{D_{s}}^{2}}\right)^{\\!\\!2}.$ (1) Here $m_{D_{s}}$ and $\tau_{D_{s}}$ are the mass and lifetime of the $D_{s}$, $G_{F}$ is the Fermi constant, and $V_{cs}$ is a Cabibbo-Kobayashi-Maskawa (CKM) element. The decay constant $f_{D_{s}}$ is defined by $\langle 0|\,\bar{s}\gamma_{\mu}\gamma_{5}c\,|D_{s}(p)\rangle=if_{D_{s}}p_{\mu},$ (2) where $p_{\mu}$ is the 4-momentum of the $D_{s}$ meson. Although the electroweak transition proceeds at the tree level, $D_{s}^{+}\\!\to\\!W^{+}\\!\to\\!\ell^{+}\nu_{\ell}$, its rate is suppressed. The helicity of the lepton must flip, leading to the factor $m_{\ell}$ in the amplitude. For the muon, this helicity suppression $(m_{\ell}/m_{D_{s}})^{2}$ is $2.8\times 10^{-3}$. The $\tau$ mass is only 10% smaller than the $D_{s}$ mass (1.969 GeV), so there is no significant helicity suppression, but the phase space suppression [the last factor in Eq. (1)] is $3.4\times 10^{-2}$. Table 1: Experimental values of $f_{D_{s}}$. Our averages treat systematic uncertainties as uncorrelated and omit the PDG entry Yao:2006px , which is an average of earlier experiments. final state | reference | $f_{D_{s}}$ (MeV) ---|---|--- $\ell\nu$ | PDG Yao:2006px | $294\pm 27$ $\mu\nu$ | BaBar Aubert:2006sd | $283\pm 17\pm 16$ $\mu\nu$ | CLEO Pedlar:2007za | $264\pm 15\pm\;7$ $\mu\nu$ | Belle Widhalm:2007mi | $275\pm 16\pm 12$ $\tau\nu$ ($\tau\\!\to\\!\pi\nu$) | CLEO Pedlar:2007za | $310\pm 25\pm~{}8$ $\tau\nu$ ($\tau\\!\to\\!e\nu\bar{\nu}$) | CLEO Ecklund:2007zm | $273\pm 16\pm\;8$ $\mu\nu$ | our average | $273\pm 11$ $\tau\nu$ | our average | $285\pm 15$ We have collected in Table 1 all precise experimental measurements of $B(D_{s}\\!\to\\!\ell\nu)$, which are usually quoted in terms of $f_{D_{s}}$ Yao:2006px ; Rosner:2008yu . Combining the error bars in quadrature, our average of $\tau\nu$ and $\mu\nu$ final states is $\left(f_{D_{s}}\right)_{\rm expt}=277\pm 9\;\textrm{MeV}.$ (3) The most accurate calculation from lattice QCD is Follana:2007uv $\left(f_{D_{s}}\right)_{\rm QCD}=241\pm 3\;\textrm{MeV},$ (4) where statistical and systematic uncertainties are combined in the fitting methods. The only other modern lattice-QCD calculation agrees, $249\pm 3\pm 16$ MeV Aubin:2005ar , but its quoted error is five times larger and would not influence a weighted average with Eq. (4). The discrepancy between Eqs. (3) and (4) is 15% and 3.8$\sigma$. Table 1 also shows averages for each mode separately: for $\tau\nu$ ($\mu\nu$) alone, the discrepancy is 18% and 2.9$\sigma$ (13% and 2.7$\sigma$). If the BaBar result is omitted from the average, as in Ref. Rosner:2008yu , then the discrepancy is 3.4$\sigma$. On the other hand, if the earlier measurements Yao:2006px as well as the BaBar result are included, we find a 4.1$\sigma$ discrepancy. Experiments.—CLEO Pedlar:2007za ; Ecklund:2007zm produces $D_{s}$ pairs near threshold, where the multiplicity is low. Their method reconstructs one $D_{s}^{(*)}$ and then counts how often the opposite-side $D_{s}$ decays leptonically. When the charged lepton is a muon, the neutrino is “detected” by requiring the missing mass-squared to peak at zero. When the charged lepton is a $\tau$, the identification is made through the subsequent decays $\tau\\!\to\\!e\nu\bar{\nu}$ and $\tau\\!\to\\!\pi\bar{\nu}$. BaBar Aubert:2006sd observes $D_{s}$ coming from the decay $D_{s}^{*}\\!\to\\!D_{s}\gamma$, produced well above threshold. They compare the relative number of subsequent $D_{s}\\!\to\\!\mu^{+}\nu$ and $D_{s}\\!\to\\!\phi\pi$, and then use their own measurement of $B(D_{s}\\!\to\\!\phi\pi)$ to determine $B(D_{s}\\!\to\\!\ell\nu)$. Belle Widhalm:2007mi also observes $D_{s}$ via $D_{s}^{*}\\!\to\\!D_{s}\gamma$, but the whole event is reconstructed, using a Monte Carlo technique. In summary, all these measurements have central values and error bars that are straightforward to interpret, and to combine to obtain Eq. (3). The measured branching fraction and Eq. (1) yield $|V_{cs}|f_{D_{s}}$. Three- generation CKM unitarity is assumed, either taking $|V_{cs}|$ from a global fit to flavor physics Yao:2006px , or setting $|V_{cs}|=|V_{ud}|$. The difference is numerically irrelevant. Relaxing the assumption cannot lead to agreement between theory and experiment because unitarity, even for more than three generations, requires $|V_{cs}|<1$, whereas the discrepancy would require $|V_{cs}|\approx 1.1$. Radiative corrections.—The measurements are not, strictly speaking, for $D_{s}\\!\to\\!\ell\nu$ alone, because some photons are always radiated. The radiative corrections have been studied, focusing on effects that could overcome the helicity suppression Burdman:1994ip ; Hwang:2005uk . For $D_{s}\\!\to\\!\tau^{+}\nu$ there is no sizable helicity suppression. In the rest frame of the $D_{s}$, the $\tau$ acquires only 9.3 MeV of kinetic energy, so it cannot radiate much. Explicit calculation Burdman:1994ip shows that the radiative corrections are too small to account for the discrepancy Wang:2001mm . For $D_{s}\\!\to\\!\mu^{+}\nu$ radiative corrections could play a role due to processes of the form $D_{s}\\!\to\\!\gamma D_{s}^{*}\\!\to\\!\gamma\mu^{+}\nu$, where $D_{s}^{*}$ is a (virtual) vector or axial-vector meson. The transition $D_{s}^{*}\\!\to\\!\mu^{+}\nu$ is not helicity-suppressed, so the factor $\alpha$ for radiation is compensated by a relative factor $m_{D_{s}}^{2}/m_{\mu}^{2}$ for omitting helicity suppression. Using Eq. (12) of Ref. Burdman:1994ip and imposing the CLEO Pedlar:2007za cut $E_{\gamma}>300$ MeV, we find that the radiative rate is around 1% and, hence, insufficient to explain the discrepancy. Lattice QCD—There are many lattice-QCD calculations for $f_{D_{s}}$ in the literature, but only Refs. Follana:2007uv ; Aubin:2005ar include 2+1 flavors of sea quarks, which is necessary to find agreement for many “gold-plated” quantities, namely those for which errors are easiest to control Davies:2003ik . Both calculations start with lattice gauge fields generated by the MILC Collaboration Bernard:2001av , which employ “rooted staggered fermions” for the sea quarks. At finite lattice spacing this approach has small violations of unitarity and locality. Theoretical and numerical evidence suggests that these vanish in the continuum limit, such that QCD is obtained, with the undesirable features controlled with chiral perturbation theory. The strengths and weaknesses of this approach have been reviewed in detail Sharpe:2006re . Reference Follana:2007uv reports an error five times smaller than that of Ref. Aubin:2005ar for several reasons. The largest uncertainties in Ref. Aubin:2005ar come from a power-counting estimate of the discretization error for the charm quark, and from uncertainties in the chiral extrapolation. Reference Follana:2007uv employs a different discretization for the charm quark, which allows a controlled extrapolation to the continuum limit. Thus, the discretization error here is driven by the underlying numerical data. The action for the charm quark in Ref. Follana:2007uv , called HISQ Follana:2006rc , is the same as that used for the light valence quarks. As a result the statistical errors are smaller than those of the heavy-quark method used in Ref. Aubin:2005ar , and the axial current automatically has the physical normalization. The suitability of HISQ for charm is one of its design features, it has been tested via the charmonium spectrum Follana:2006rc , and the computed $D$ and $D_{s}$ masses agree with experiment. The $D^{+}$ decay constant $f_{D^{+}}$ also agrees with experiment, at $1\sigma$. Another feature of Ref. Follana:2007uv is the way the lattice-spacing and sea-quark mass dependence is fitted. Full details are not yet published, but it is noteworthy that the same analysis yields $f_{\pi}$ and $f_{K}$ in agreement with experiment Yao:2006px and earlier, equally precise, lattice- QCD calculations Aubin:2004fs . The $D_{s}$ meson is simpler than the pion or kaon for lattice QCD, because none of the valence quarks is light, so $f_{D_{s}}$ is easier to determine than $f_{\pi}$. We find that simple extrapolations lead to the same central values for both $m_{D_{s}}$ and $f_{D_{s}}$. The error bar in Eq. (4) is smaller than that in Eq. (3). Therefore, it is the combined experimental error that provides the yardstick for the deviation. To illustrate, if the lattice-QCD error bar were doubled, the discrepancy becomes 2.7$\sigma$, 2.5$\sigma$, and 3.3$\sigma$ for $\tau$, $\mu$, and combined. Hence, even if additional sources of uncertainty are uncovered, evidence for a deviation may well remain. Nonstandard effective interactions.—Although the experiments quote the final states as $\mu^{+}\nu_{\mu}$ and $\tau^{+}\nu_{\tau}$ (and their charge conjugates), the flavor of the neutrino is not detected. Nonstandard physics could lead to any neutrino flavor, even a sterile neutrino. However, given the large effect that needs to be explained, we shall restrict our attention to amplitudes that could interfere with the standard model, which fixes the neutrino flavor. Lorentz-invariant new physics may contribute to $D_{s}\\!\to\\!\ell\nu_{\ell}$ only through the following effective Lagrangian: $\frac{C_{A}^{\ell}}{M^{2}}\left(\bar{s}\gamma_{\mu}\gamma_{5}c\right)\left(\bar{\nu}_{L}\gamma^{\mu}\ell_{L}\right)+\frac{C_{P}^{\ell}}{M^{2}}\left(\bar{s}\gamma_{5}c\right)\left(\bar{\nu}_{L}\ell_{R}\right)+{\rm H.c.},$ (5) where $C_{A}^{\ell}$ and $C_{P}^{\ell}$ are complex dimensionless parameters, $M$ is the mass of some particle whose exchange induces the 4-fermion operators (5), and the $c,s,\ell$ fields are taken in the mass-eigenstate basis. The hadronic matrix element required for the decay induced by $(\bar{s}\gamma_{5}c)(\bar{\nu}_{L}\ell_{R})$ is related to the one of Eq. (2) by partial conservation of the axial current: $(m_{c}+m_{s})\langle 0|\,\bar{s}i\gamma_{5}c\,|D_{s}\rangle=f_{D_{s}}m^{2}_{D_{s}}$. The branching fraction in the presence of the operators (5) is given by Eq. (1) with $G_{F}V^{*}_{cs}m_{\ell}$ replaced by $G_{F}V^{*}_{cs}m_{\ell}+\frac{1}{\sqrt{2}M^{2}}\left(C_{A}^{\ell}m_{\ell}+\frac{C_{P}^{\ell}\,m_{D_{s}}^{2}}{m_{c}+m_{s}}\right)~{},$ (6) with no helicity suppression in the last term. The imaginary part of $V_{cs}$ is negligible (in the standard CKM parametrization Yao:2006px ), so constructive interference, which would increase $B(D_{s}\\!\to\\!\ell\nu)$, requires the real part of $C_{A}^{\ell}$ or $C_{P}^{\ell}$ to be nonzero and positive. Assuming only one nonzero coefficient, the amplitude for $\tau^{+}\nu_{\tau}$ ($\mu^{+}\nu_{\mu}$) could be increased by 12% (8.4%) only if $\displaystyle\frac{M}{(\mathop{\rm Re}C_{A}^{\ell})^{1/2}}$ $\displaystyle\lesssim$ $\displaystyle\left\\{\begin{array}[]{rl}710~{}\textrm{GeV}&\textrm{for}~{}\ell=\tau\\\\[2.84526pt] \hphantom{3}850~{}\textrm{GeV}&\textrm{for}~{}\ell=\mu\end{array}\right.\;,$ (9) $\displaystyle\frac{M}{(\mathop{\rm Re}C_{P}^{\ell})^{1/2}}$ $\displaystyle\lesssim$ $\displaystyle\left\\{\begin{array}[]{rl}920~{}\textrm{GeV}&\textrm{for}~{}\ell=\tau\\\\[2.84526pt] 4500~{}\textrm{GeV}&\textrm{for}~{}\ell=\mu\end{array}\right.\;,$ (12) thereby reducing the discrepancy to $1\sigma$ in each case. These bounds are a key new result of this Letter, because they constrain any model of new physics. The effective interaction (5) also contributes to the semileptonic decays $D\\!\to\\!K\mu^{+}\nu$. This proceeds through two amplitudes, corresponding to angular momentum $J=1$ or $0$ for the lepton pair. For $J=1$, the standard- model amplitude and that from $C_{A}^{\mu}$ are not helicity suppressed, while that from $C_{P}^{\mu}$ is. For $J=0$, the pattern of helicity suppression is as for the leptonic decay. Hence, only the $J=1$ part of the rate will be visible, and as the accuracy of the lattice-QCD calculations improves, the comparison with experiment will help decide which interactions are responsible for the effect in $D_{s}\\!\to\\!\ell\nu$. The current status favors $C_{P}^{\mu}\neq 0$ rather than $C_{A}^{\mu}\neq 0$, because the lattice-QCD prediction for $D\\!\to\\!K\mu\nu$ Aubin:2004ej agrees with experiment Widhalm:2006wz , albeit at the $\sim 7\%$ level. New particles.—There are three choices for the electric charge of a boson that can mediate the four-fermion operators (5): $+1,+2/3,-1/3$, corresponding to the three diagrams shown in Fig. 1. The exchanged boson (taken to be emitted from the vertex where $c$ is absorbed) is a color singlet if the electric charge is +1, and a color triplet if the electric charge is $+2/3$ or $-1/3$. We shall consider only the cases where the new boson has spin 0 or 1, and its interactions are renormalizable. 11 1 (20,20)(0,0)(0,40)(20,20) (90,40)(70,20)(70,20)(90,0) (20,20)(70,20)4 (-9,37)[c]$c$(-9,5)[c]$\bar{s}$(103,37)[c]$\ell^{+}$(99,5)[c]$\nu$(45,30)[c]$(+1)$(20,20)(0,0)(40,0)(20,20) (20,60)(40,80)(0,80)(20,60) (20,60)(20,20)3 (-5,70)[c]$c$(-5,10)[c]$\bar{s}$(44,70)[c]$\nu$(49,10)[c]$\ell^{+}$(44,40)[c]$(+2/3)$(20,20)(0,0)(20,20)(40,0) (40,80)(20,60)(0,80)(20,60) (20,60)(20,20)3 (-5,70)[c]$c$(-5,10)[c]$\bar{s}$(46,70)[c]$\ell^{+}$(45,10)[c]$\nu$(44,40)[c]$(-1/3)$ Figure 1: Four-fermion operators induced by boson exchange. A new vector boson, $W^{\prime}$, of electric charge $+1$ would contribute only to $C_{A}^{\ell}$. Such a boson must be associated with a new gauge symmetry, which makes it difficult to allow large couplings to left-handed leptons. One possibility is that $W$ and $W^{\prime}$ mix, but the constraint from electroweak data on mixing ($\lesssim 10^{-2}$) is too strong to allow noticeable deviations in $D_{s}$ decays. Another possibility is that some new vector-like fermions transform under the new gauge symmetry and mix with the left-handed leptons. Such mixing is also tightly constrained, especially by the nonobservation of vector-like fermions at LEP and the Tevatron. Overall, a $W^{\prime}$ is inconsistent with Eq. (9), barring perhaps some finely-tuned elaborate model (e.g., with large $W$-$W^{\prime}$ mixing whose electroweak effects are cancelled by other particles). A spin-0 particle of charge +1, $H^{+}$, appears in models with two or more Higgs doublets. Its interactions, in the mass eigenstate basis for charged fermions, include $H^{+}\left(y_{c}\bar{c}_{R}s_{L}+y_{s}\bar{c}_{L}s_{R}+y_{\ell}\bar{\nu}_{\ell}\ell\right)+{\rm H.c.},$ (13) where $y_{c},y_{s},y_{\ell}$ are complex Yukawa couplings. The exchange of $H^{+}$ induces $C_{A}^{\ell}=0$ and $C_{P}^{\ell}=\frac{1}{2}\left(y_{c}^{*}-y_{s}^{*}\right)y_{\ell}~{},$ (14) taking $M$ equal to the $H^{+}$ mass. If $H^{+}$ is the charged Higgs boson present in the Type-II two-Higgs-doublet model, then $y_{c}/y_{s}=m_{c}/(m_{s}\tan^{2}\\!\beta)$ so that $C_{P}^{\ell}$ can have either sign Hewett:1995aw , but the Yukawa couplings are too small to be compatible with Eq. (12). Other models lead to large constructive interference. For example, a two-Higgs-doublet model where one doublet gives the $c$, $u$ (but not $d$, $s$, $b$, or $t$) and lepton masses, and has a vacuum expectation value of $\sim 2$ GeV, yields $|y_{s}|\ll y_{\tau},y_{c}^{*}\sim O(1)$. Thus, $C_{P}^{\ell}>0$ and the limits (12) are satisfied for $M\lesssim 500$ GeV. Furthermore, such a model explains why the deviations in $\tau\nu$ and $\mu\nu$ are comparable. It is encouraging that this two-Higgs-doublet model does not induce tree-level flavor-changing neutral currents, and the off-diagonal couplings of $H^{+}$ are CKM suppressed. Given that this model has not been previously studied, its 1-loop contributions to flavor-changing processes (such as $b\\!\to\\!s\gamma$) need to be computed before deciding whether some fine tuning is required to evade experimental bounds. The charge $-1/3$ and $+2/3$ exchanges correspond to leptoquarks. A scalar charge $+2/3$ exchange arises for the $(3,2,+7/6)$ set of $SU(3)_{c}\\!\times\\!SU(2)_{W}\\!\times\\!U(1)_{Y}$ charges. This leptoquark appears, for example, in a new theory of quark and lepton masses Dobrescu:2008 . Let $r=(r_{u},r_{d})$ be the doublet leptoquark, where $r_{d}$ is its charge $+2/3$ component. The interaction terms relevant here, written in the same basis as (5), are $\lambda_{c\ell}r_{d}\bar{c}_{R}\nu_{L}^{\ell}+\lambda^{\prime}_{s\ell}r_{d}\bar{s}_{L}\ell_{R}$. The $r_{d}$ exchange gives $C_{A}^{\ell}=0$ and $C_{P}^{\ell}=-\lambda_{c\ell}^{*}\lambda^{\prime}_{s\ell}/4$. Since the leptoquark couplings can have any phase, the new amplitude can interfere constructively. Still, various flavor processes constrain the couplings of $r$. Even if its couplings to first-generation fermions were negligible, the lepton-flavor violating decays $\tau\\!\to\\!\mu\bar{s}s$, where $\bar{s}s$ hadronizes to $\eta$, $\eta^{\prime}$, $\phi$ or $K\bar{K}$, set a lower limit on $M^{2}\\!/|\lambda^{\prime}_{s\tau}\lambda^{\prime}_{s\mu}|$, which is hard to reconcile with Eq. (12). One way out would be a model with two $r$ leptoquarks, with one coupling to $\tau$ and the other one to $\mu$. The constraint from $\tau\\!\to\\!\mu\bar{s}s$ similarly disfavors spin-1 leptoquarks of charge $+2/3$. A scalar leptoquark of charge $-1/3$ (also discussed in Dobrescu:2008 ) arises in the case of two sets of $SU(3)_{c}\times SU(2)_{W}\times U(1)_{Y}$ charges: $(3,1,-1/3)$ or $(3,3,-1/3)$. Let us denote the former by $\tilde{d}$. Its Yukawa couplings are given by $\tilde{d}\left[\kappa_{\ell}\left(\bar{c}_{L}\ell_{L}^{c}-\bar{s}_{L}\nu_{L}^{\ell c}\right)+\kappa^{\prime}_{\ell}\,\bar{c}_{R}\ell_{R}^{c}\right]+{\rm H.c.},$ (15) where $\kappa_{\ell}$ and $\kappa^{\prime}_{\ell}$ are complex parameters. These interactions are present, for example, in $R$-parity violating supersymmetric models (their effect on $D_{s}\\!\to\\!e^{+}\nu$ has been analyzed in Ref. Akeroyd:2002pi ). The $\tilde{d}$ exchange, as in the last diagram of Fig. 1, gives (for $M$ equal to the $\tilde{d}$ mass) $C_{A}^{\ell}=\frac{1}{4}\,|\kappa_{\ell}|^{2}\;\;\;,\;\;\;C_{P}^{\ell}=\frac{1}{4}\,\kappa_{\ell}\kappa^{\prime*}_{\ell}~{}~{}.$ (16) For $|\kappa_{\ell}^{\prime}/\kappa_{\ell}|\ll m_{\ell}m_{c}/m_{D_{s}}^{2}$, the interference is automatically constructive [see Eq. (6)], and the resulting deviations in $\tau\nu$ and $\mu\nu$ are approximately equal if $|\kappa_{\mu}|\approx|\kappa_{\tau}|$. Moreover, there are no severe constraints from other processes on the couplings $\kappa_{\ell}$ and $\kappa^{\prime}_{\ell}$ with $\ell=\tau$ or $\mu$. The $\tilde{d}$ couplings to the electron can be forbidden by a symmetry, and the ones to first- generation quarks could be small. The $(3,3,-1/3)$ scalar leptoquark includes an $SU(2)_{W}$ component of charge $-4/3$ which mediates $\tau\\!\to\\!\mu\bar{s}s$. The vector leptoquark of charge $-1/3$ has the same problem. Conclusions.—We have argued that the $3.8\sigma$ discrepancy between the standard model and the combined experimental measurements of $D_{s}\\!\to\\!\ell\nu$ appears so far to be robust, and thus it is worth interpreting it in terms of new physics. The upper bounds (9) and (12) on the scale of four-fermion operators are low enough to allow exploration of the underlying physics at the LHC. A $\tilde{d}$ scalar leptoquark of charge $-1/3$ may solve the $D_{s}$ puzzle without running into conflict with any other measurements. At the LHC, the $\tilde{d}$ can be strongly produced in pairs, and the final states would be $\ell^{+}\ell^{-}jj$, where $\ell$ is a $\tau$ or a $\mu$, and $j$ is a $c$-jet. Given that there are two $\ell j$ pairs, each of them forming a resonance at the $\tilde{d}$ mass, the backgrounds can be kept under control. The current limits on the $\tilde{d}$ mass from similar searches at the Tevatron are around 200 GeV Abulencia:2005ua . An alternative explanation is provided by an $H^{+}$ exchange in a (new) model where a Higgs doublet gives masses to the charged leptons and $c$ and $u$ quarks, and a second Higgs doublet gives masses to the down-type and top quarks. Both the leptoquark and charged Higgs solutions lead naturally to comparable increases in the branching fractions for $D_{s}\\!\to\\!\tau^{+}\nu$ and $D_{s}\\!\to\\!\mu^{+}\nu$, as suggested by the data. Acknowledgments.—We thank P. Fox, E. Lunghi, S. Stone and R. Van de Water for helpful discussions. Fermilab is operated by Fermi Research Alliance, LLC, under US DoE Contract DE-AC02-07CH11359. ## References * (1) W. M. Yao et al. [Particle Data Group], J. Phys. G 33, 1 (2006) and 2007 partial update for 2008. * (2) J. L. Rosner and S. Stone, arXiv:0802.1043 [hep-ex]. * (3) B. Aubert et al. [BABAR Collaboration], Phys. Rev. Lett. 98, 141801 (2007) [arXiv:hep-ex/0607094]. * (4) T. K. Pedlar et al. [CLEO Collaboration], Phys. Rev. D 76, 072002 (2007) [arXiv:0704.0437 [hep-ex]]. * (5) K. Abe _et al._ [Belle Collaboration], arXiv:0709.1340 [hep-ex]. * (6) K. M. Ecklund et al. [CLEO Collaboration], arXiv:0712.1175 [hep-ex]. * (7) E. Follana, C. T. H. Davies, G. P. Lepage and J. Shigemitsu [HPQCD Collaboration], Phys. Rev. Lett. 100, 062002 (2008) [arXiv:0706.1726 [hep-lat]]. * (8) C. Aubin et al. [Fermilab Lattice and MILC Collaborations], Phys. Rev. Lett. 95, 122002 (2005) [arXiv:hep-lat/0506030]. * (9) G. Burdman, J. T. Goldman and D. Wyler, Phys. Rev. D 51, 111 (1995) [arXiv:hep-ph/9405425]. * (10) C. W. Hwang, Eur. Phys. J. C 46, 379 (2006) [arXiv:hep-ph/0512006]. * (11) Owing to the small energy release, a reported $-8$% correction to $D_{s}\\!\to\\!\tau\nu$ [G. L. Wang, C. H. Chang and T. F. Feng, arXiv:hep-ph/0102251] is unlikely to be correct. * (12) C. T. H. Davies et al. [HPQCD, MILC, and Fermilab Lattice Collaborations], Phys. Rev. Lett. 92, 022001 (2004) [arXiv:hep-lat/0304004]. * (13) C. Bernard et al. [MILC Collaboration], Phys. Rev. D 64, 054506 (2001) [arXiv:hep-lat/0104002]; C. Aubin et al. [MILC Collaboration], Phys. Rev. D 70, 094505 (2004) [arXiv:hep-lat/0402030]. * (14) S. R. Sharpe, PoS LAT2006, 022 (2006) [arXiv:hep-lat/0610094]; A. S. Kronfeld, arXiv:0711.0699 [hep-lat]. * (15) E. Follana et al. [HPQCD Collaboration], Phys. Rev. D 75, 054502 (2007) [arXiv:hep-lat/0610092]. * (16) C. Aubin et al. [MILC Collaboration], Phys. Rev. D 70, 114501 (2004) [arXiv:hep-lat/0407028]. * (17) C. Aubin et al. [Fermilab Lattice and MILC Collaborations], Phys. Rev. Lett. 94, 011601 (2005). * (18) L. Widhalm et al., Phys. Rev. Lett. 97, 061804 (2006) [arXiv:hep-ex/0604049]; B. Aubert et al. [BABAR Collaboration], Phys. Rev. D 76, 052005 (2007) [arXiv:hep-ex/0607077]; S. Dobbs et al. [CLEO Collaboration], arXiv:0712.1020 [hep-ex]. * (19) J. L. Hewett, arXiv:hep-ph/9505246. The results in A. G. Akeroyd, Prog. Theor. Phys. 111, 295 (2004) [arXiv:hep-ph/0308260], A. G. Akeroyd and C. H. Chen, Phys. Rev. D 75, 075004 (2007) [arXiv:hep-ph/0701078] are valid only for $\tan^{2}\\!\beta\gg m_{c}/m_{s}\approx 13$. * (20) B. A. Dobrescu and P. J. Fox, Fermilab-Pub-08-049-T. * (21) A. G. Akeroyd and S. Recksiegel, Phys. Lett. B 554, 38 (2003) [arXiv:hep-ph/0210376]. * (22) A. Abulencia et al. [CDF Collaboration], Phys. Rev. D 73, 051102 (2006); V. M. Abazov et al. [D0 Collaboration], Phys. Lett. B 636, 183 (2006).

arxiv-papers

2024-09-04T02:48:54.148167

0803.0523

††thanks: Present address # New lattice action for heavy quarks Mehmet B. Oktay Department of Physics, University of Illinois at Urbana- Champaign, Urbana, Illinois 61801, USA School of Mathematics, Trinity College, Dublin 2, Ireland Andreas S. Kronfeld Theoretical Physics Department, Fermi National Accelerator Laboratory, Batavia, Illinois, USA (March 4, 2008) ###### Abstract We extend the Fermilab method for heavy quarks to include interactions of dimensions 6 and 7 in the action. There are, in general, many new interactions, but we carry out the calculations needed to match the lattice action to continuum QCD at the tree level, finding six non-zero couplings. Using the heavy-quark theory of cutoff effects, we estimate how large the remaining discretization errors are. We find that our tree-level matching, augmented with one-loop matching of the dimension-five interactions, can bring these errors below 1%, at currently available lattice spacings. ###### pacs: 11.15.Ha, 12.38.Gc ††preprint: FERMILAB-PUB-08/054-T ## I Introduction An important application of lattice gauge theory is to calculate hadronic matrix elements relevant to experiments in flavor physics. With recent advances in lattice calculations with $n_{f}=2+1$ flavors of dynamical quarks Bernard:2002bk ; Aubin:2004ej ; Aubin:2005ar ; Allison:2004be , we now have an exciting prospect of genuine QCD calculations. To match the experimental uncertainty, available now or in the short term, it is essential to control all other sources of theoretical uncertainty as well as possible. An attractive target is to reduce the uncertainty, from any given source, to 1–2%. This target will be hard to hit if one relies on increases in computer power alone: methodological improvements are needed too. Many of the important processes are electroweak transitions of heavy charmed or $b$-flavored quarks. A particular challenge stems from heavy-quark discretization effects, because $m_{Q}a\not\ll 1$. The key to meeting the challenge is to observe that heavy quarks are non-relativistic in the rest frame of the containing hadron Eichten:1987xu ; Lepage:1987gg . The scale of the heavy-quark mass, $m_{Q}$, can (and should) be separated from the soft scales inside the hadron and treated with an effective field theory instead of computer simulation. Even so, at available lattice spacings Bernard:2002bk , many calculations of $D$-meson ($B$-meson) properties suffer from a discretization error of around 7% (5%) Aubin:2004ej ; Aubin:2005ar . Thus, it makes sense to develop a more accurate discretization. In this paper we extend the accuracy of the “Fermilab” method for heavy quarks El-Khadra:1996mp to include in the lattice action all interactions of dimension six. We also include certain interactions of dimension seven. Because heavy quarks are non-relativistic, they are commensurate with related dimension-6 terms, in the power counting of heavy-quark effective theory (HQET) for heavy-light hadrons Eichten:1987xu or non-relativistic QCD (NRQCD) for quarkonium Lepage:1987gg . The Fermilab method starts with Wilson fermions Wilson:1975hf and the clover action Sheikholeslami:1985ij . With these actions lattice spacing effects are bounded for large $m_{Q}a$, thanks to heavy-quark symmetry. They can be reduced systematically by allowing an asymmetry between spatial and temporal interactions. Asymmetry in the lattice action compensates for the non- relativistic kinematics, enabling a relativistic description through the Symanzik effective field theory Symanzik:1979ph . Alternatively, one may interpret Wilson fermions non-relativistically from the outset El- Khadra:1996mp , and set up the improvement program matching lattice gauge theory and continuum QCD to each other through HQET and NRQCD Kronfeld:2000ck ; Harada:2001fi . The Symanzik description makes it possible to design a lattice action that behaves smoothly as $m_{Q}a\to 0$, converging to the universal continuum limit. The HQET description, on the other hand, makes semiquantitative estimates of discretization errors more transparent. The new action introduced below has nineteen bilinear interactions beyond those of the asymmetric version of the clover action, as well as many four- quark interactions. Several of these couplings are redundant, and many more vanish when matching to continuum QCD at the tree level. We study semiquantitatively how many of the new operators are needed to achieve 1–2% accuracy. We find, in the end, that only _six_ new interactions are essential for such accuracy. The action is designed with some flexibility, so that one may choose the computationally least costly version of the action. This paper is organized as follows. Section II considers the description of lattice gauge theory via continuum effective field theories. Then, in some detail, we identify a full set of operators describing heavy-quark discretization effects. We then determine how many of these are redundant, and which redundant directions should be used to preserve the good high-mass behavior. We have two goals in this analysis. One is to design the new, more highly improved, action; for this step a Symanzik-like description is more helpful, and the resulting action is given in Sec. III. The other is to estimate the discretization errors of the new action; here the HQET and NRQCD descriptions are more useful. To make error estimates, and to use the new action in numerical work, we need matching calculations; they are in Sec. IV. Our error estimates are in Sec. V. Section VI concludes. Some of the material is technical and appears in appendices: Feynman rules needed for the matching calculation are in Appendix A; some details of the Compton scattering amplitude used for matching are in Appendix B; a discussion of improvement of the gauge action on anisotropic lattices (which one needs only if the heavy quarks are not quenched) is in Appendix C. Some of these results have been reported earlier Oktay:2002mj . ## II Effective Field Theory In this section we discuss how to understand and control discretization effects using effective field theories. We start with a brief overview, focusing on issues that arise for heavy quarks, those with mass $m_{Q}\gg\Lambda$. For more details, the reader may consult earlier work El- Khadra:1996mp ; Kronfeld:2000ck ; Harada:2001fi ; Aoki:2001ra ; Christ:2006us or a pedagogical review Kronfeld:2002pi . Here we catalog all interactions of dimension 6 and also certain interactions of dimension 7 that, for heavy quarks, are of comparable size when $m_{Q}a\not\ll 1$. ### II.1 Overview Cutoff effects in lattice field theories are most elegantly studied with continuum effective field theories. The idea originated with Symanzik Symanzik:1979ph and was extended to gluons and light quarks by Weisz and collaborators Weisz:1982zw ; Weisz:1983bn ; Luscher:1984xn ; Sheikholeslami:1985ij . One develops a relationship ${\cal L}_{\mathrm{lat}}\doteq{\cal L}_{\mathrm{Sym}},$ (1) where $\doteq$ means that the two Lagrangians generate the same on-shell spectrum and matrix elements. The lattice itself regulates the ultraviolet behavior of the underlying (lattice) theory ${\cal L}_{\mathrm{lat}}$. On the other hand, a continuum scheme, which does not need to be specified in detail, regulates (and renormalizes) the ultraviolet behavior of the effective theory ${\cal L}_{\mathrm{Sym}}$. In lattice QCD (with Wilson fermions), the local effective Lagrangian (LE${\cal L}$) is ${\cal L}_{\mathrm{Sym}}=\frac{1}{2g^{2}}\mathop{\mathrm{tr}}[F_{\mu\nu}F^{\mu\nu}]-\sum_{f}\bar{q}_{f}({D\kern-6.49994pt/}+m_{f})q_{f}+\sum_{i}a^{\dim{\cal L}_{i}-4}K_{i}(g^{2},ma;c_{j};\mu a){\cal L}_{i},$ (2) where $g^{2}$ and $m_{f}$ are the gauge coupling and quark mass (of flavor $f$), renormalized at scale $\mu\lesssim a^{-1}$. The (continuum) QCD Lagrangian appears as the first two terms. The sum consists of higher dimension operators ${\cal L}_{i}$, multiplied by short-distance coefficients $K_{i}$. These terms describe cutoff effects. The short-distance coefficients depend on the renormalization point and on the couplings, including couplings $c_{j}$ of improvement terms in ${\cal L}_{\mathrm{lat}}$. Equation (2) is fairly well-established to all orders in perturbation theory Luscher:1998pe ; Adams:2007gh and believed to hold non-perturbatively as well. If $a$ is small enough, the terms ${\cal L}_{i}$ may be treated as operator insertions, leading to a description of lattice gauge theory as “QCD + small corrections”. In heavy-quark physics $m_{Q}\gg\Lambda$, where $\Lambda$ is the QCD scale, so one is led to consider what happens when $m_{Q}a\not\ll 1$. The short-distance coefficients depend explicitly on the mass. Time derivatives of heavy-quark or heavy-antiquark fields in the ${\cal L}_{i}$ also generate mass dependence of observables. With field redefinitions—or, equivalently, with the equations of motion—these time derivatives can be eliminated. Focusing on a single heavy flavor $Q$, the result of these manipulations is El-Khadra:1996mp ; Aoki:2001ra ; Christ:2006us ${\cal L}_{\mathrm{Sym}}=\cdots-\bar{Q}\left(\gamma_{4}D_{4}+m_{1}+\sqrt{\frac{m_{1}}{m_{2}}}\bm{\gamma}\cdot\bm{D}\right)Q+\sum_{i}a^{\dim\bar{\cal L}_{i}-4}\bar{K}_{i}(g^{2},m_{2}a;\mu a)\bar{\cal L}_{i},$ (3) where the ellipsis denotes the unaltered LE${\cal L}$ for gluons and light quarks. By construction the $\bar{\cal L}_{i}$ do not have any time derivatives acting on quarks or antiquarks. The advantage of Eq. (3) is that all dependence on the heavy-quark mass is in the short-distance coefficients $m_{1}$, $\sqrt{m_{1}/m_{2}}$, and $\bar{K}_{i}(m_{2}a)$. Matrix elements of the $\bar{\cal L}_{i}$ generate soft scales. The heavy-quark symmetry of Wilson quarks (with either the Wilson Wilson:1975hf or Sheikholeslami-Wohlert Sheikholeslami:1985ij actions) guarantees that the coefficients $\bar{K}_{i}(m_{2}a)$ are bounded for all $m_{2}a$. This feature can be preserved by improving the lattice Lagrangian with discretizations of the $\bar{\cal L}_{i}$, thereby avoiding higher time derivatives El-Khadra:1996mp ; Kronfeld:2000ck . For such improved actions, Eq. (3) neatly isolates the potentially most serious problem of heavy quarks into the deviation of the coefficient $\sqrt{m_{1}/m_{2}}$ from $1$. Fortunately, the problem can be circumvented in two simple ways. One is a Wilson-like action with two hopping parameters El-Khadra:1996mp , tuned so that $m_{1}=m_{2}$. Then Eq. (3) once again takes the form “QCD + small corrections”. The new lattice action introduced in Sec. III has two hopping parameters for this reason. Another solution is to interpret Wilson fermions in a non-relativistic framework. One can replace the Symanzik description with one using a non- relativistic effective field theory for the quarks (and antiquarks) Kronfeld:2000ck . For the leading $\bar{Q}$-$Q$ term in Eq. (3) $\bar{Q}\left(\gamma_{4}D_{4}+m_{1}+\sqrt{\frac{m_{1}}{m_{2}}}\bm{\gamma}\cdot\bm{D}\right)Q\doteq\bar{h}^{(+)}\left(D_{4}+m_{1}-\frac{\bm{D}^{2}+z_{B}(m_{2}a,\mu a)i\bm{\Sigma}\cdot\bm{B}}{2m_{2}}\right)h^{(+)}+\cdots$ (4) where $z_{B}$ is a matching coefficient, and $h^{(+)}$ is a heavy-quark field satisfying $h^{(+)}=+\gamma_{4}h^{(+)}$. Another set of terms appears for the antiquark, with field $h^{(-)}$ satisfying $h^{(-)}=-\gamma_{4}h^{(-)}$. The non-relativistic effective theory conserves heavy quarks and heavy antiquarks separately. As a consequence, the rest mass $m_{1}$ has no effect on mass splittings and matrix elements.111A simple proof can be found in Ref. Kronfeld:2000ck . For lattice gauge theory this implies that the bare quark mass (or hopping parameter) should not be adjusted via $m_{1}$. Instead, the bare mass should be adjusted to normalize the kinetic energy $\bm{D}^{2}/2m_{2}$. One can develop the non-relativistic effective theory for the lattice artifacts $\bar{\cal L}_{i}$ by using heavy-quark fields instead of Dirac quark fields Kronfeld:2000ck . Higher-dimension operators in the heavy-quark theory receive contributions from the expansions of Eq. (4) and of the $\bar{\cal L}_{i}$. Coalescing the coefficients of like operators obtains a description of lattice gauge theory with heavy quarks ${\cal L}_{\mathrm{lat}}\doteq\cdots-\bar{h}^{(+)}(D_{4}+m_{1})h^{(+)}+\sum_{i}{\cal C}_{i}^{\mathrm{lat}}(g^{2},m_{2};m_{2}a,c_{j};\mu/m_{2}){\cal O}_{i},$ (5) where the operators ${\cal O}_{i}$ on the right-hand side are those of a (continuum) heavy-quark effective theory, of dimension 5 and higher, built out of heavy-quark fields $h^{(\pm)}$, gluons, and light quarks. (The leading ellipsis denotes term for the gluons and light quarks only.) The ${\cal C}_{i}$ are short-distance coefficients, which depend on $g^{2}$, the heavy- quark mass, the ratio of short distances $m_{2}a$, and also all couplings $c_{j}$ in the lattice action. The logic and structure is the same as the non- relativistic description of QCD, ${\cal L}_{\mathrm{QCD}}\doteq\cdots-\bar{h}^{(+)}(D_{4}+m_{Q})h^{(+)}+\sum_{i}{\cal C}_{i}^{\mathrm{cont}}(g^{2},m_{Q};\mu/m_{Q}){\cal O}_{i}.$ (6) Thus, improvement of lattice gauge theory is attained by adjusting couplings $c_{j}$ until ${\cal C}_{i}^{\mathrm{lat}}(c_{j})-{\cal C}_{i}^{\mathrm{cont}}$ vanishes (identically, or perhaps to some accuracy) for the first several ${\cal O}_{i}$. It does not matter whether one carries out the improvement program by adjusting $\bar{K}_{i}(c_{j})=0$ or ${\cal C}_{i}^{\mathrm{lat}}(c_{j})={\cal C}_{i}^{\mathrm{cont}}$ Harada:2001fi . The results for the $c_{j}$ are the same, provided one identifies $m_{Q}$ with $m_{2}$. The matching assumes that $\bm{p}a\ll 1$, but at the same time $m_{2}a\not\ll 1$. One is thus led to non-relativistic kinematics ($\bm{p}/m_{2}\ll 1$) in the matching calculation, where both descriptions—Eqs. (3) and (5)—are valid. Kinematics are encoded into the operators $\bar{\cal L}_{i}$ or ${\cal O}_{i}$ and are not transferred to the short-distance coefficients. Hence, kinematics cannot influence matching conditions on the $c_{j}$. In particular, when indeed $m_{2}a\ll 1$ (which may be impractical, but is conceivable theoretically) relativistic kinematics ($\bm{p}\sim m_{2}$) are possible, and it follows from the Symanzik effective field theory that the solution of $\bar{K}_{i}(c_{j})=0$ yields the same $c_{j}$ for both relativistic and non- relativistic kinematics. ### II.2 Quark bilinears in the LE${\cal L}$ In the rest of this section we construct the LE${\cal L}$ appropriate to heavy quarks. The two main steps are first to list all of the ${\cal L}_{i}$ that can appear, and second to decide which should be considered redundant. In part it is a generalization of the dimension-6 analysis of Ref. Sheikholeslami:1985ij to the case without axis-interchange symmetry. At dimension 6 there are quark bilinears, four-quark interactions, and interactions that contain only the gauge field. We shall start with the bilinears and turn to the others further below. In each case, we first consider complete lists of operators, and then consider which can be chosen to be redundant. Table 1 contains a list of all quark bilinears through dimension 6 that can appear in the effective Lagrangian. Table 1: Bilinear interactions that could appear in the Symanzik LE${\cal L}$ through dimension 6. Dim | With axis-interchange symmetry | Without axis-interchange symmetry | HQET $\lambda^{s}$ | NRQCD $\upsilon^{t}$ ---|---|---|---|--- 3 | $\bar{q}q$ | | $\bar{Q}Q$ | | | 4 | $\bar{q}{D\kern-6.49994pt/}q$ | | $\bar{Q}(\gamma_{4}D_{4}+m_{1})Q$ | | $1$ | $\upsilon^{2}$ | | | $\bar{Q}\bm{\gamma}\cdot\bm{D}Q$ | | $\lambda$ | $\upsilon^{2}$ 5 | $\bar{q}D^{2}q$ | $\varepsilon_{1}$ | $\bar{Q}D_{4}^{2}Q$ | $\varepsilon_{1}$ | | | | | $\bar{Q}\bm{D}^{2}Q$ | $\delta_{1}$ | $\lambda$ | $\upsilon^{2}$ | $-{\textstyle\frac{i}{2}}\bar{q}\sigma_{\mu\nu}F_{\mu\nu}q$ | | $\bar{Q}i\bm{\Sigma}\cdot\bm{B}Q$ | | $\lambda$ | $\upsilon^{4}$ | | | $\bar{Q}\bm{\alpha}\cdot\bm{E}Q$ | | $\lambda^{2}$ | $\upsilon^{4}$ 6 | $\bar{q}\gamma_{\mu}D_{\mu}^{3}q$ | | $\bar{Q}\gamma_{i}D_{i}^{3}Q$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $\bar{q}\\{{D\kern-6.49994pt/},D^{2}\\}q$ | $\varepsilon_{2}$ | $\bar{Q}\gamma_{4}D_{4}^{3}Q$ | $\varepsilon_{2}$ | | | | | $\bar{Q}\\{\gamma_{4}D_{4},\bm{D}^{2}\\}Q$ | $\delta_{2}$ | | | | | $\bar{Q}\\{D_{4}^{2},\bm{\gamma}\cdot\bm{D}\\}Q$ | $\vartheta_{2}$ | | | | | $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{D}^{2}\\}Q$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $-{\textstyle\frac{i}{2}}\bar{q}\\{{D\kern-6.49994pt/},\sigma_{\mu\nu}F_{\mu\nu}\\}q$ | $\varepsilon_{F}$ | $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}Q$ | $\varepsilon_{F}$ | $\lambda^{2}$ | $\upsilon^{4}$ | | | $\bar{Q}\\{\gamma_{4}D_{4},i\bm{\Sigma}\cdot\bm{B}\\}Q$ | $\delta_{B}$ | | | | | $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},i\bm{\Sigma}\cdot\bm{B}\\}Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ | | | $\bar{Q}[D_{4},\bm{\gamma}\cdot\bm{E}]Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{q}[D_{\mu},F_{\mu\nu}]\gamma_{\nu}q$ | | $\bar{Q}\gamma_{4}(\bm{D}\cdot\bm{E}-\bm{E}\cdot\bm{D})Q$ | | $\lambda^{2}$ | $\upsilon^{4}$ | | | $\bar{Q}\bm{\gamma}\cdot(\bm{D}\times\bm{B}+\bm{B}\times\bm{D})Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ The second column contains interactions that respect axis-interchange symmetry; the fourth column contains the extension to the case without axis- interchange symmetry. The meaning of the other columns is explained below. Covariant derivatives act on all fields to the right, $D_{\mu}FQ=(\partial_{\mu}F+[A_{\mu},F])Q+F\,D_{\mu}Q.$ (7) This notation is convenient for the interactions with commutators and anti- commutators. To arrive at the lists we exploit identities such as $\displaystyle{D\kern-6.49994pt/}^{2}$ $\displaystyle=$ $\displaystyle D^{2}-{\textstyle\frac{i}{2}}\sigma_{\mu\nu}F_{\mu\nu},$ (8) $\displaystyle 2\gamma_{4}D_{4}\bm{\gamma}\cdot\bm{D}\gamma_{4}D_{4}$ $\displaystyle=$ $\displaystyle\\{\gamma_{4}D_{4},\bm{\alpha}\cdot\bm{E}\\}-\\{D_{4}^{2},\bm{\gamma}\cdot\bm{D}\\},$ (9) $\displaystyle 2\bm{\gamma}\cdot\bm{D}\gamma_{4}D_{4}\bm{\gamma}\cdot\bm{D}$ $\displaystyle=$ $\displaystyle\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}-\\{\gamma_{4}D_{4},(\bm{\gamma}\cdot\bm{D})^{2}\\}.$ (10) Some interactions are omitted, because the underlying lattice gauge theory is invariant under cubic rotations, spatial inversion, time reflection, and charge conjugation.222Reference Sheikholeslami:1985ij included the dimension-6 interaction $\bar{q}[{D\kern-6.49994pt/},D^{2}]q$. Reference El- Khadra:1996mp included the dimension-5 interaction $\bar{Q}[\gamma_{4}D_{4},\bm{\gamma}\cdot\bm{D}]Q$. Both are odd under charge conjugation and, thus, may be omitted. The fourth column is arranged so that its entries are part of the corresponding interactions in the second column. It is easy to show that the list is complete, by writing out all independent ways to have three covariant derivatives, expressing the $\bm{E}$ and $\bm{B}$ fields as anti-commutators of covariant derivatives. One finds 11 possibilities, and then one can use identities to manipulate this list to that given in the fourth column of Table 1. The LE${\cal L}$ contains several redundant directions. The equation of motion of the leading LE${\cal L}$ plays a key role in specifying which operator insertions may be considered redundant. Let us assume, for the moment, that $m_{1}=m_{2}$, so that the equation of motion in the Symanzik LE${\cal L}$ is the Dirac equation. Below we shall use the non-relativistic effective field theory to address the case $m_{1}\neq m_{2}$. The quark fields are integration variables in a functional integral, so an equally valid description is obtained by changing variables $\displaystyle Q$ $\displaystyle\mapsto$ $\displaystyle e^{J}Q,$ (11) $\displaystyle\bar{Q}$ $\displaystyle\mapsto$ $\displaystyle\bar{Q}e^{\bar{J}},$ (12) where $\displaystyle J=a\varepsilon_{1}({D\kern-6.49994pt/}+m)$ $\displaystyle+$ $\displaystyle a\delta_{1}\bm{\gamma}\cdot\bm{D}+a^{2}\varepsilon_{2}({D\kern-6.49994pt/}+m)^{2}-a^{2}{\textstyle\frac{1}{2}}\varepsilon_{F}i\sigma_{\mu\nu}F_{\mu\nu}+a^{2}\delta_{2}(\bm{\gamma}\cdot\bm{D})^{2}$ (13) $\displaystyle+$ $\displaystyle a^{2}\delta_{B}i\bm{\Sigma}\cdot\bm{B}+a^{2}\vartheta_{2}[\gamma_{4}D_{4},\bm{\gamma}\cdot\bm{D}]$ and similarly for $\bar{J}$ with separate parameters $\bar{\varepsilon}_{i}$, $\bar{\delta}_{i}$, and $\bar{\vartheta}_{i}$. If the $\delta$ parameters (and $\vartheta_{2}$, $\bar{\vartheta}_{2}$) vanish, then $J$ and $\bar{J}$ preserve invariance under interchange of all four axes. One can propagate the change of variables to the LE${\cal L}$, and trace which coefficients of dimensions 5 and 6 are shifted by amounts proportional to the parameters in $J$ and $\bar{J}$. To avoid generating terms that violate charge conjugation one chooses $\bar{\varepsilon}_{i}=+\varepsilon_{i}$, $\bar{\delta}_{i}=+\delta_{i}$, $\bar{\vartheta}_{2}=-\vartheta_{2}$. We then see that there are two redundant directions at dimension 5, and five at dimension 6. That means that two couplings in the dimension-5 lattice action may be set by convenience, and five in the dimension-6 lattice action. The third and fifth columns show the correspondence between parameters in the change of variables and the interactions that we choose to be redundant. As expected from general arguments El-Khadra:1996mp ; Aoki:2001ra ; Christ:2006us , all interactions in which $\gamma_{4}D_{4}$ acts on $Q$ or (after integration by parts) $\bar{Q}$ are redundant. There is quite a bit of freedom here. One could choose $\varepsilon_{F}$ to eliminate $\bar{Q}[D_{4},\bm{\gamma}\cdot\bm{E}]Q=\bar{Q}\\{\gamma_{4}D_{4},\bm{\alpha}\cdot\bm{E}\\}Q$ instead of $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}Q$. But the former is suppressed, relative to the latter, in heavy-quark systems. Moreover, in HQET and NRQCD one has $\displaystyle\bar{Q}\bm{\alpha}\cdot\bm{E}Q$ $\displaystyle\doteq$ $\displaystyle\bar{h}^{(+)}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}h^{(+)}/2m_{2}+\cdots,$ (14) $\displaystyle\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}Q$ $\displaystyle\doteq$ $\displaystyle\bar{h}^{(+)}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}h^{(+)}+\cdots,$ (15) which mean that $\bar{Q}\bm{\alpha}\cdot\bm{E}Q$ and $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}Q$ generate nearly the same effects in heavy-quark systems. Thus, we prefer to take $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}Q$ to be redundant. To understand the general pattern of redundant interactions, let us introduce some notation. Let $\mathcal{B}$ ($\mathcal{E}$) be a combination of gauge fields, derivatives, and Dirac matrices that commutes (anti-commutes) with $\gamma_{4}$. An example of $\mathcal{B}$ ($\mathcal{E}$) is $i\bm{\Sigma}\cdot\bm{B}$ ($\bm{\alpha}\cdot\bm{E}$). Also, let us write $\mathcal{B}_{\pm}$ (and $\mathcal{E}_{\pm}$) when $\bar{Q}\mathcal{B}_{\pm}Q$ (or $\bar{Q}\mathcal{E}_{\pm}Q$) has charge conjugation $\pm 1$. Because we wish to eliminate time derivatives of quark and antiquark fields, we would like $\bar{Q}\\{\gamma_{4}D_{4},\mathcal{B}_{+}\\}Q$ and $\bar{Q}[\gamma_{4}D_{4},\mathcal{E}_{-}]Q$ to be redundant. That is always possible: simply add to $J$ in Eq. (13) terms of the form $\delta_{\mathcal{B}_{+}}\mathcal{B}_{+}$ and $\vartheta_{\mathcal{E}_{-}}\mathcal{E}_{-}$. As a consequence, neither $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\mathcal{B}_{+}\\}Q$ nor $\bar{Q}[\bm{\gamma}\cdot\bm{D},\mathcal{E}_{-}]Q$ is redundant. On the other hand, in $\bar{Q}[\gamma_{4}D_{4},\mathcal{B}_{-}]Q$ and $\bar{Q}\\{\gamma_{4}D_{4},\mathcal{E}_{+}\\}Q$ the time derivative acts only on gauge fields. Thus, by adding to $J$ terms of the form $\vartheta_{\mathcal{B}_{-}}\mathcal{B}_{-}$ and $\delta_{\mathcal{E}_{+}}\mathcal{E}_{+}$ it is possible to choose $\bar{Q}[\bm{\gamma}\cdot\bm{D},\mathcal{B}_{-}]Q$ and $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\mathcal{E}_{+}\\}Q$ to be redundant. Instead of $\bar{Q}[\bm{\gamma}\cdot\bm{D},\mathcal{B}_{-}]Q$ or $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\mathcal{E}_{+}\\}Q$ it may be convenient to choose an operator related through an identity. ### II.3 Power counting The small corrections of an effective field theory are small, because the product of the short-distance coefficients and the operators yield a ratio of a short-distance scale to a long-distance scale. For light quarks in the Symanzik effective field theory, the essential ratio is $a/\Lambda^{-1}=\Lambda a$, and dimensional analysis reveals the power of $\Lambda a$ to which any contribution is suppressed. In particular, $\mathcal{B}$\- and $\mathcal{E}$-type interactions of the same dimension are equally important. For heavy quarks the physics is different, because $m_{Q}^{-1}$ is a short distance. The ratio $a/m_{Q}^{-1}=m_{Q}a$ should not be taken commensurate with $\Lambda a$ El-Khadra:1996mp . Instead, interactions should be classified in a way that brings out the physics. It is natural to turn to HQET and NRQCD. Let us start with heavy-light hadrons and HQET. $\mathcal{E}$-type interactions of given dimension are $\Lambda/m_{Q}$ times smaller than $\mathcal{B}$-type interactions of the same dimension. Because $\Lambda/m_{Q}\ll 1$ and $\Lambda a\ll 1$, it makes sense to count powers of $\lambda$, where $\lambda$ is either of the small parameters Kronfeld:2000ck ; Harada:2001fi ; Christ:2006us $\lambda\sim a\Lambda,\Lambda/m_{Q}.$ (16) This power counting pertains whether $m_{Q}<a$, $m_{Q}\sim a$, or $m_{Q}>a$. Writing the corrections in the Symanzik fashion (with Dirac quark fields $Q$ and $\bar{Q}$), each $\bar{\cal L}_{i}$ is suppressed by $\lambda^{s}$, with $s=\dim\mathcal{L}-4+n_{\Gamma}.$ (17) Here $n_{\Gamma}=0$ or $1$ for interactions of the form $\bar{Q}\mathcal{B}_{+}Q$ or $\bar{Q}\mathcal{E}_{+}Q$, respectively. The sixth column of Table 1 (labelled HQET) shows the suppression of each interaction, relative to the (leading) contribution from the light degrees of freedom. In the following we call the power counting for heavy-light hadrons, based on Eq. (17), “HQET power counting.” Now let us recall how to classify interactions in quarkonium according to the power of the relative internal velocity, $\upsilon$. Because color source and sink are both non-relativistic, chromoelectric fields carry a power of $\upsilon^{3}$, and chromomagnetic fields a power of $\upsilon^{4}$ Lepage:1992tx . $\mathcal{E}$-type interactions are suppressed by a power of $p/m_{Q}=\upsilon$, analogously to their suppression in heavy-light hadrons. Thus, bilinears are suppressed by $\upsilon^{t}$, where now $t=\dim\mathcal{L}-3+n_{E}+2n_{B}+n_{\Gamma},$ (18) and $n_{E}$ ($n_{B}$) is the number of chromoelectric (chromomagnetic) fields. The seventh column of Table 1 (labelled NRQCD) shows the suppression of each interaction. In the following we call the power counting for quarkonium, based on Eq. (18), “NRQCD power counting.” Glancing down the sixth and seventh column of Table 1, one sees several terms of order $\lambda^{3}$ and $\upsilon^{6}$, from Eqs. (17) and (18) one realizes that some dimension-7 interactions are of the same order. They are listed in Table 2. Table 2: Dimension-(7,0) bilinear interactions that are commensurate, for heavy quarks, with those of order $\lambda^{3}$ (in HQET) or $\upsilon^{4}$, $\upsilon^{6}$ (in NRQCD). Dim | Without axis-interchange symmetry | HQET $\lambda^{s}$ | NRQCD $\upsilon^{t}$ ---|---|---|--- 7 | $\bar{Q}D_{i}^{4}Q$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $\sum_{i\neq j}\bar{Q}i\Sigma_{i}D_{j}B_{i}D_{j}Q$ | $\delta[\sum_{i}\gamma_{i}D_{i}^{3}]$ | $\lambda^{3}$ | $\upsilon^{6}$ | $\sum_{i\neq j}\bar{Q}\\{D_{j}^{2},i\Sigma_{i}B_{i}\\}Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}(\bm{D}^{2})^{2}Q$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $\bar{Q}\\{\bm{D}^{2},i\bm{\Sigma}\cdot\bm{B}\\}Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}\bm{\gamma}\cdot\bm{D}i\bm{\Sigma}\cdot\bm{B}\bm{\gamma}\cdot\bm{D}Q$ | $\delta[\\{\bm{\gamma}\cdot\bm{D},i\bm{\Sigma}\cdot\bm{B}\\}]$ | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}D_{i}i\bm{\Sigma}\cdot\bm{B}D_{i}Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}\bm{D}\cdot(\bm{B}\times\bm{D})Q$ | $\delta[\bm{\gamma}\cdot(\bm{D}\times\bm{B}+\bm{B}\times\bm{D})]$ | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}(i\bm{\Sigma}\cdot\bm{B})^{2}Q$ | $\delta[\\{\bm{\gamma}\cdot\bm{D},\bm{D}^{2}\\}]$ | $\lambda^{3}$ | $\upsilon^{8}$ | $\bar{Q}\bm{B}\cdot\bm{B}Q$ | | $\lambda^{3}$ | $\upsilon^{8}$ | $\bar{Q}(\bm{\alpha}\cdot\bm{E})^{2}Q$ | $\delta[[D_{4},\bm{\gamma}\cdot\bm{E}]]$ | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{Q}\bm{E}\cdot\bm{E}Q$ | | $\lambda^{3}$ | $\upsilon^{6}$ There are two interactions with four derivatives, six with the chromomagnetic field and two derivatives, and four with two $\bm{E}$ or two $\bm{B}$ fields. A third combination of four derivatives is omitted, using the identity $D_{i}\bm{D}^{2}D_{i}=(\bm{D}^{2})^{2}+\bm{D}\cdot(\bm{B}\times\bm{D})-\bm{B}^{2}$. Other dimension-7 operators carry power $\lambda^{4}$ in HQET power counting, or $\upsilon^{8}$ (or higher) in NRQCD power counting. Five combinations are redundant (as shown), and we shall see below how they and the others arise in matching calculations. The $(d,n_{\Gamma})=(7,1)$ operator $\bar{Q}\\{\bm{D}^{2},\bm{\alpha}\cdot\bm{E}\\}Q$ and several $(d,n_{\Gamma})=(8,0)$ operators, all with $n_{E}=1$ and $n_{D}+n_{\Gamma}=3$, have NRQCD power-counting $\upsilon^{6}$. Reference Lepage:1992tx includes spin-dependent ones, to obtain the next-to-leading corrections to spin- dependent mass splittings. We have not included these operators in our analysis, but a straightforward extension of the matching calculation in Sec. IV.2.1 would suffice to determine their couplings. Although this description of cutoff effects is somewhat cumbersome, it provides a valuable foundation for our new action, given in Sec. III. To obtain the new action, we simply discretize the interactions in Tables 1 and 2, except those with higher time derivatives. The discretization of $\bar{Q}\bm{\gamma}\cdot\bm{D}Q$ is needed to obtain a lattice action that behaves smoothly as $m_{Q}a\to 0$ El-Khadra:1996mp , reproducing the universal continuum limit of QCD. Similarly, discretizations of the $\mathcal{E}$-type interactions, such as $\bar{Q}\bm{\alpha}\cdot\bm{E}Q$ and $\bar{Q}\\{\bm{\gamma}\cdot\bm{D},\bm{D}^{2}\\}Q$, are needed to retain that feature here. ### II.4 Heavy-quark description For understanding the size of heavy-quark discretization effects, it is simpler to switch to a non-relativistic description. (When $m_{1}\neq m_{2}$, it is also necessary to see the connection to QCD.) The list of interactions is much shorter, because the constraint $\gamma_{4}h^{(\pm)}=\pm h^{(\pm)}$ removes the $\mathcal{E}$-type interactions. It is given in Table 3, including the dimension-7 interactions related to those in Table 2. Table 3: Bilinear interactions that could appear in the heavy-quark LE${\cal L}$ through dimension 7. Dim | Without axis-interchange symmetry | HQET $\lambda^{s}$ | NRQCD $\upsilon^{t}$ ---|---|---|--- 3 | $\bar{h}^{(\pm)}h^{(\pm)}$ | | | 4 | $\bar{h}^{(\pm)}\gamma_{4}D_{4}h^{(\pm)}$ | | | 5 | $\bar{h}^{(\pm)}D_{4}^{2}h^{(\pm)}$ | $\varepsilon_{1}$ | | | $\bar{h}^{(\pm)}\bm{D}^{2}h^{(\pm)}$ | | $\lambda$ | $\upsilon^{2}$ | $\bar{h}^{(\pm)}i\bm{\Sigma}\cdot\bm{B}h^{(\pm)}$ | | $\lambda$ | $\upsilon^{4}$ 6 | $\bar{h}^{(\pm)}\gamma_{4}D_{4}^{3}h^{(\pm)}$ | $\varepsilon_{2}$ | | | $\bar{h}^{(\pm)}\\{\gamma_{4}D_{4},\bm{D}^{2}\\}h^{(\pm)}$ | $\delta_{2}$ | | | $\bar{h}^{(\pm)}\\{\bm{\gamma}\cdot\bm{D},\bm{\alpha}\cdot\bm{E}\\}h^{(\pm)}$ | | $\lambda^{2}$ | $\upsilon^{4}$ | $\bar{h}^{(\pm)}\\{\gamma_{4}D_{4},i\bm{\Sigma}\cdot\bm{B}\\}h^{(\pm)}$ | $\delta_{B}$ | | | $\bar{h}^{(\pm)}\gamma_{4}(\bm{D}\cdot\bm{E}-\bm{E}\cdot\bm{D})h^{(\pm)}$ | | $\lambda^{2}$ | $\upsilon^{4}$ 7 | $\bar{h}^{(\pm)}D_{i}^{4}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $\sum_{i\neq j}\bar{h}^{(\pm)}\\{D_{j}^{2},i\Sigma_{i}B_{i}\\}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\sum_{i\neq j}\bar{h}^{(\pm)}i\Sigma_{i}D_{j}B_{i}D_{j}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}(\bm{D}^{2})^{2}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{4}$ | $\bar{h}^{(\pm)}\\{\bm{D}^{2},i\bm{\Sigma}\cdot\bm{B}\\}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}\bm{\gamma}\cdot\bm{D}i\bm{\Sigma}\cdot\bm{B}\bm{\gamma}\cdot\bm{D}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}D_{i}i\bm{\Sigma}\cdot\bm{B}D_{i}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}\bm{D}\cdot(\bm{B}\times\bm{D})h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}(i\bm{\Sigma}\cdot\bm{B})^{2}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{8}$ | $\bar{h}^{(\pm)}\bm{B}\cdot\bm{B}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{8}$ | $\bar{h}^{(\pm)}(\bm{\alpha}\cdot\bm{E})^{2}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ | $\bar{h}^{(\pm)}\bm{E}\cdot\bm{E}h^{(\pm)}$ | | $\lambda^{3}$ | $\upsilon^{6}$ Also, fewer changes of the field variables are possible: $\displaystyle h^{(\pm)}$ $\displaystyle\mapsto$ $\displaystyle e^{J}h,$ (19) $\displaystyle\bar{h}^{(\pm)}$ $\displaystyle\mapsto$ $\displaystyle\bar{h}e^{\bar{J}},$ (20) where now $J=a\varepsilon_{1}(\gamma_{4}D_{4}+m_{1})+a^{2}\varepsilon_{2}(\gamma_{4}D_{4}+m_{1})^{2}+a^{2}\delta_{2}\bm{D}^{2}+a^{2}\delta_{B}i\bm{\Sigma}\cdot\bm{B},$ (21) and similarly for $\bar{J}$. To avoid $C$-odd interactions, one should choose equal parameters in $J$ and $\bar{J}$. Thus, there are four redundant directions of interest—all with time derivatives of the (anti-)quark field. In the end, just as many non-redundant interactions remain as in the Symanzik description. The heavy-quark description provides a good way to estimate the size of remaining discretization effects, as in Sec. V. ### II.5 Gauge-field and four-quark interactions in the LE${\cal L}$ We now turn to interactions in the gauge sector of the LE${\cal L}$, and also to four-quark interactions. The two are connected when one considers on-shell improvement, because in quark-quark scattering short-distance gluon exchange generates the same behavior as four-quark contact interactions. Here we give a cursory sketch of the gauge action. Then we consider the four-quark interactions, including details mostly for completeness. In practice (see Sec. V), we find the four-quark corrections to be smaller than those of the bilinear interactions analyzed in the preceding subsection. The gauge sector of the LE${\cal L}$ is the same as for anisotropic lattices, where one adjusts the action so that the temporal lattice spacing $a_{t}$ differs from the spatial lattice spacing $a_{s}$. The short-distance coefficients are different; here asymmetry between spatial and temporal gauge couplings arise only from heavy-quark loops. Improved anisotropic actions have been discussed in the literature Morningstar:1996ze , but full details remain unpublished Alford:1996up . We present the details in Appendix C. We are most concerned here with effects that survive on shell, so we study here the possible changes of variables for the gauge field. With axis- interchange symmetry one has Luscher:1984xn ; Sheikholeslami:1985ij $A_{\mu}\mapsto A_{\mu}+a^{2}\varepsilon_{A}[D^{\nu},F_{\mu\nu}]+a^{2}g^{2}\sum_{f}\varepsilon_{Jf}\,t^{a}\,(\bar{q}_{f}\gamma_{\mu}t^{a}q_{f}),$ (22) with a color-adjoint vector-current term for each flavor $f$ of quark (heavy or light). The appearance of $g^{2}$ multiplying the currents is a convenient normalization convention. When one now considers giving up axis-interchange symmetry, one has $\displaystyle A_{4}$ $\displaystyle\mapsto$ $\displaystyle A_{4}+a^{2}\varepsilon_{A}(\bm{D}\cdot\bm{E}-\bm{E}\cdot\bm{D})+a^{2}g^{2}\sum_{f}\varepsilon_{Jf}\,t^{a}\,(\bar{q}_{f}\gamma_{4}t^{a}q_{f}),$ (23) $\displaystyle\bm{A}$ $\displaystyle\mapsto$ $\displaystyle\bm{A}-a^{2}(\varepsilon_{A}+\delta_{E})[D_{4},\bm{E}]+a^{2}(\varepsilon_{A}+\delta_{A})(\bm{D}\times\bm{B}+\bm{B}\times\bm{D})$ (24) $\displaystyle\hphantom{\bm{A}}+a^{2}g^{2}\sum_{f}(\varepsilon_{Jf}+\delta_{Jf})t^{a}(\bar{q}_{f}\bm{\gamma}t^{a}q_{f}),$ which reduce to Eq. (22) when the $\delta$s vanish. For a moment, let us set $\varepsilon_{Jf}=\delta_{Jf}=0$ in Eqs. (23) and (24), and focus on the gauge fields alone. As discussed in Appendix C, there are eight independent gauge-field interactions that arise at dimension six. There are three independent ways—parametrized by $\varepsilon_{A}$, $\delta_{A}$, and $\delta_{E}$—to transform the gauge field, yielding three redundant directions. Similarly, there are eight distinct classes of six-link loops, shown in Fig. 1, that can be used in an improved lattice gauge action. Figure 1: Six-link loops available for improving the gauge action on anisotropic lattices: rectangles (top row); parallelograms (middle); bent rectangles (bottom). Nomenclature from Ref. Luscher:1984xn . In Appendix C, we show that three of them—all three classes of “bent rectangles” in the bottom row of Fig. 1—may be omitted from an on-shell improved gauge action. The transformations involving the currents $\bar{q}_{f}\gamma_{\mu}t^{a}q_{f}$ are more interesting. They shift the LE${\cal L}$ [cf. Eq. (2)] by $\displaystyle{\cal L}_{\mathrm{Sym}}\mapsto{\cal L}_{\mathrm{Sym}}$ $\displaystyle-$ $\displaystyle a^{2}\sum_{f}\varepsilon_{Jf}\bar{q}_{f}\gamma_{4}(\bm{D}\cdot\bm{E}-\bm{E}\cdot\bm{D})q_{f}+a^{2}\sum_{f}(\varepsilon_{Jf}+\delta_{Jf})\bar{q}_{f}[D_{4},\bm{\gamma}\cdot\bm{E}]q_{f}$ (25) $\displaystyle-$ $\displaystyle a^{2}\sum_{f}(\varepsilon_{Jf}+\delta_{Jf})\bar{q}_{f}\bm{\gamma}\cdot(\bm{D}\times\bm{B}+\bm{B}\times\bm{D})q_{f}$ $\displaystyle-$ $\displaystyle a^{2}g^{2}\sum_{fg}\varepsilon_{Jf}(\bar{q}_{f}\gamma_{\mu}t^{a}q_{f})(\bar{q}_{g}\gamma_{\mu}t^{a}q_{g})-a^{2}g^{2}\sum_{fg,j}\delta_{Jf}(\bar{q}_{f}\gamma_{j}t^{a}q_{f})(\bar{q}_{g}\gamma_{j}t^{a}q_{g}),\quad\;\quad$ where the derivatives act only on the gauge fields. The size of these shifts—of order $g^{2}$ for four-quark operators and of order $g^{0}$ for bilinears—is commensurate with the respective terms that already appear in ${\cal L}_{\rm Sym}$. Thus, the $2n_{f}$ parameters $\varepsilon_{Jf}$ and $\delta_{Jf}$ could be used to eliminate bilinears or four-quark operators. For simulations it is preferable to remove the latter, namely $\bar{q}_{f}\gamma_{4}t^{a}q_{f}\bar{q}_{f}\gamma_{4}t^{a}q_{f}$ and $\bar{q}_{f}\bm{\gamma}t^{a}q_{f}\cdot\bar{q}_{f}\bm{\gamma}t^{a}q_{f}$. We now list the dimension-six four-quark interactions in the LE${\cal L}$. For a single flavor, the complete list is in Table 4, which also indicates that the current-current interactions are redundant. Table 4: Four-quark interactions that could appear in the LE${\cal L}$ (for a single flavor). Dim | With axis interchange | Without axis interchange ---|---|--- 6 | $(\bar{q}t^{a}q)^{2}$ | | $(\bar{Q}t^{a}Q)^{2}$ | | $(\bar{q}\gamma_{5}t^{a}q)^{2}$ | | $(\bar{Q}\gamma_{5}t^{a}Q)^{2}$ | | $(\bar{q}\gamma_{\mu}t^{a}q)^{2}$ | $\varepsilon_{J}$ | $(\bar{Q}\gamma_{4}t^{a}Q)^{2}$ | $\varepsilon_{J}$ | | | $(\bar{Q}\gamma_{i}t^{a}Q)^{2}$ | $\delta_{J}$ | $(\bar{q}\gamma_{\mu}\gamma_{5}t^{a}q)^{2}$ | | $(\bar{Q}\gamma_{4}\gamma_{5}t^{a}Q)^{2}$ | | | | $(\bar{Q}\gamma_{i}\gamma_{5}t^{a}Q)^{2}$ | | $(\bar{q}i\sigma_{\mu\nu}t^{a}q)^{2}$ | | $(\bar{Q}i\Sigma_{i}t^{a}Q)^{2}$ | | | | $(\bar{Q}\alpha_{i}t^{a}Q)^{2}$ | Interactions with the color structure $(\bar{q}\Gamma q)^{2}$ may be omitted, because they can be related to those listed through Fierz rearrangement of the fields. When considering several flavors of quark, we must keep track of flavor indices as well as color and Dirac indices. The Fierz problem becomes more intricate, and we shall find that color-singlet and color-octet structures should be maintained. Let us start with Fierz rearrangement of the Dirac indices. The four-quark terms in the LE${\cal L}$ take the form $\sum_{X}K_{X}\bar{q}_{f\alpha}\Gamma_{X}q_{g\beta}\bar{q}_{h\gamma}\Gamma_{X}q_{i\delta}=-\sum_{X,Y}K_{X}F_{XY}\bar{q}_{f\alpha}\Gamma_{Y}q_{i\delta}\bar{q}_{h\gamma}\Gamma_{Y}q_{g\beta},$ (26) where $K_{X}$ denotes short-distance coefficients, the Greek (Latin) indices label color (flavor), $F$ is the Fierz rearrangement matrix (with $F^{2}=1$), and the minus sign comes from anti-commutation of the fermion fields. Equation (26) leaves the flavor and color indices uncontracted, but to get terms in the LE${\cal L}$, the color indices must be contracted (one way or another), and the flavor labels must yield a flavor-neutral interaction. Without loss, we can choose the side of Eq. (26) such that the Dirac matrices contract quark fields of the same flavor. Then one can use Fierz identities for SU($N$) generators (${t^{a}}^{\dagger}=-t^{a}$) $\displaystyle Nt^{a}_{\alpha\beta}t^{a}_{\gamma\delta}$ $\displaystyle=$ $\displaystyle-t^{a}_{\alpha\delta}t^{a}_{\gamma\beta}-(N^{2}-1)\delta_{\alpha\delta}\delta_{\gamma\beta}/2N,$ (27) $\displaystyle\delta_{\alpha\beta}\delta_{\gamma\delta}$ $\displaystyle=$ $\displaystyle\delta_{\alpha\delta}\delta_{\gamma\beta}/N-2t^{a}_{\alpha\delta}t^{a}_{\gamma\beta},$ (28) so that the color indices are contracted across the same fields as the Dirac and flavor indices. After using Fierz rearrangement to bring quarks of the same flavor next to each other, one is left with the interactions in Table 5. Table 5: Four-quark interactions that remain when Fierz rearrangement is taken into account. A sum over Dirac matrices $\Gamma_{X}$ in each of the sets $\\{1\\}$, $\\{\gamma_{4}\\}$, $\\{\bm{\gamma}\\}$, $\\{i\bm{\Sigma}\\}$, $\\{\bm{\alpha}\\}$, $\\{\bm{\gamma}\gamma_{5}\\}$, $\\{\gamma_{4}\gamma_{5}\\}$, $\\{\gamma_{5}\\}$ is assumed. (With axis-interchange symmetry, the sets would be $\\{1\\}$, $\\{\gamma_{\mu}\\}$, $\\{i\sigma_{\mu\nu}\\}$, $\\{\gamma_{\mu}\gamma_{5}\\}$, $\\{\gamma_{5}\\}$.) Quarks | Color octet | Color singlet ---|---|--- Heavy-heavy | $\bar{Q}\Gamma_{X}t^{a}Q\,\bar{Q}\Gamma_{X}t^{a}Q$ | – Heavy-heavy | $\bar{Q}_{1}\Gamma_{X}t^{a}Q_{1}\,\bar{Q}_{2}\Gamma_{X}t^{a}Q_{2}$ | $\bar{Q}_{1}\Gamma_{X}Q_{1}\,\bar{Q}_{2}\Gamma_{X}Q_{2}$ Heavy-light | $\bar{Q}\Gamma_{X}t^{a}Q\sum_{f}\bar{q}_{f}\Gamma_{X}t^{a}q_{f}$ | $\bar{Q}\Gamma_{X}Q\sum_{f}\bar{q}_{f}\Gamma_{X}q_{f}$ Light-light | $\sum_{f}\bar{q}_{f}\Gamma_{X}t^{a}q_{f}\sum_{g}\bar{q}_{g}\Gamma_{X}t^{a}q_{g}$ | $\sum_{f}\bar{q}_{f}\Gamma_{X}q_{f}\sum_{g}\bar{q}_{g}\Gamma_{X}q_{g}$ To be concrete, we consider $n_{l}$ flavors of light quarks (with $m_{q}\lesssim\Lambda$) and two flavors of heavy quarks (charm and bottom). We neglect the dependence of the coefficients on the light quark masses, because four-quark interactions are already small corrections (of dimension six). In that case, the four-quark interactions can be arranged so that only the SU($n_{l}$) flavor singlets $\sum_{f}\bar{q}_{f}\Gamma_{X}t^{a}q_{f}$ and $\sum_{f}\bar{q}_{f}\Gamma_{X}q_{f}$ appear. The parameters $\varepsilon_{Jf}$ and $\delta_{Jf}$ may be used to eliminate color-octet current-current interactions. For each heavy flavor, one finds $(\bar{Q}\gamma_{4}t^{a}Q)^{2}$ and $\sum_{i}(\bar{Q}\gamma_{i}t^{a}Q)^{2}$ to be redundant. For light quarks, we may neglect the differences in the mass, so they have common parameters, and the flavor-singlet combination $(\sum_{f}\bar{q}_{f}\gamma_{\mu}t^{a}q_{f})^{2}$ is redundant. For the light flavors, our list of operators is a Fierz rearrangement of the list in Ref. Sheikholeslami:1985ij . The leading HQET power counting for heavy-light four-quark operators follows from dimensional analysis and Eq. (17): $\lambda^{2+n_{\Gamma}}$, just as if the light-quark part were replaced by three derivatives. Heavy-heavy four- quark operators will be suppressed, once matrix elements are taken, by a heavy-quark loop, leading to $g^{2}\lambda^{4+n_{\Gamma}}$. In quarkonium, the size of heavy-light four-quark operators follows similarly from Eq. (18): $\upsilon^{3+n_{\Gamma}}$. The valence heavy-heavy operators are more interesting. They must contain two contributions, one to improve $t$-channel gluon exchange, and another to improve $s$-channel annihilation. The former have NRQCD power counting $g^{2}\upsilon^{3+n_{\Gamma}}\sim\upsilon^{4+n_{\Gamma}}$ (since $g^{2}\sim\upsilon$ Lepage:1992tx ). The latter are $\upsilon^{2}$ times smaller, because the $s$-channel gluon is far off shell, but the Dirac-matrix suppression is now $\upsilon^{1-n_{\Gamma}}$, leading to $g^{2}\upsilon^{6-n_{\Gamma}}\sim\upsilon^{7-n_{\Gamma}}$ in all. In practice, the $s$-channel contributions are suppressed further, when treated as an insertion in a color-singlet quarkonium state. At the tree level, the only color structure that can arise is the color-octet. Its matrix elements vanish in the $\bar{Q}Q$-color-singlet Fock state of quarkonium, leaving the $\upsilon^{3}$-suppressed $\bar{Q}QA$ color octet Bodwin:1994jh . Color- singlet four-quark operators arise at one loop, with an additional factor of $g^{2}\sim\upsilon$. ## III New Lattice Action In this section we introduce a new, improved lattice action for heavy quarks, designed to yield smaller discretization errors than the action in Ref. El- Khadra:1996mp . Our design is based on several lessons from the preceding section and Refs. El-Khadra:1996mp ; Kronfeld:2000ck ; Harada:2001fi . First, it is important to preserve the natural heavy-quark symmetry of Wilson fermions, so that the coefficients $\bar{K}_{i}$ stay bounded for all $m_{Q}a$. (This feature is spoiled in the standard improvement program designed for light quarks, which introduces several new terms that grow with $m_{Q}$.) Second, the new lattice action is flexible enough to match cleanly onto both the Symanzik description and the non-relativistic description. Let us write the action as follows $S=S_{D^{2}F^{2}}+S_{0}+\sum_{d=5}^{\infty}\sum_{n_{\Gamma}=0}^{1}S_{(d,n_{\Gamma})}+S_{\bar{q}q\bar{q}q},$ (29) where $S_{D^{2}F^{2}}$ is the improved gauge action [Eq. (172)], $S_{0}$ is the basic Fermilab action, the $S_{(d,n_{\Gamma})}$ consist of the bilinear terms added to improve the quark sector, and $S_{\bar{q}q\bar{q}q}$ denotes four-quark interactions. $S_{(d,n_{\Gamma})}$ consists of (discretizations of) interactions of dimension $d$, with $n_{\Gamma}$ as in the discussion of power counting, Eqs. (16)–(18). Including the interactions in $S_{(d,1)}$ couples “upper” and “lower” components, but allows a smooth limit $a\to 0$.333Lattice NRQCD, which directly discretizes the continuum heavy-quark action, can be thought of as omitting $S_{(d,1)}$ in favor of $S_{(d+1,0)}$. Our aim is to improve the action to include all interactions of dimension six. Then the power counting requires us to include $S_{(7,0)}$ as well. Finally, $S_{\bar{q}q\bar{q}q}$ consists of discretizations of four-quark operators, at dimension six, those of Table 5. The basic Fermilab action El-Khadra:1996mp is a generalization of the Wilson action Wilson:1975hf : $\displaystyle S_{0}$ $\displaystyle=$ $\displaystyle m_{0}a^{4}\sum_{x}\bar{\psi}(x)\psi(x)+a^{4}\sum_{x}\bar{\psi}(x)\gamma_{4}{D_{4}}_{\mathrm{lat}}\psi(x)-{\textstyle\frac{1}{2}}a^{5}\sum_{x}\bar{\psi}(x){\triangle_{4}}_{\mathrm{lat}}\psi(x)$ (30) $\displaystyle+\,\zeta a^{4}\sum_{x}\bar{\psi}(x)\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}}\psi(x)-{\textstyle\frac{1}{2}}r_{s}\zeta a^{5}\sum_{x}\bar{\psi}(x)\triangle^{(3)}_{\mathrm{lat}}\psi(x).$ We denote lattice fermions fields with $\psi$ to distinguish them from the continuum quark fields in Sec. II. The dimension-five Wilson terms are included in $S_{0}$ to remove doubler states. The remaining dimension-five interactions are Sheikholeslami:1985ij ; El-Khadra:1996mp $\displaystyle S_{(5,0)}=S_{B}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{2}}c_{B}\zeta a^{5}\sum_{x}\bar{\psi}(x)i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\psi(x),$ (31) $\displaystyle S_{(5,1)}=S_{E}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{2}}c_{E}\zeta a^{5}\sum_{x}\bar{\psi}(x)\bm{\alpha}\cdot\bm{E}_{\mathrm{lat}}\psi(x),$ (32) where the notation $S_{B}$ and $S_{E}$ is from Ref. El-Khadra:1996mp , and the discretizations ${D_{\mu}}_{\mathrm{lat}}$, ${\triangle_{\mu}}_{\mathrm{lat}}$, $\triangle^{(3)}_{\mathrm{lat}}$, $\bm{B}_{\mathrm{lat}}$, $\bm{E}_{\mathrm{lat}}$ are defined below. The new interactions in Eq. (29) introduced in this paper are $\displaystyle S_{(6,0)}$ $\displaystyle=$ $\displaystyle r_{E}a^{6}\sum_{x}\bar{\psi}(x)\\{\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}},\bm{\alpha}\cdot\bm{E}_{\mathrm{lat}}\\}\psi(x)$ (33) $\displaystyle+$ $\displaystyle z_{E}a^{6}\sum_{x}\bar{\psi}(x)\gamma_{4}\left(\bm{D}_{\mathrm{lat}}\cdot\bm{E}_{\mathrm{lat}}-\bm{E}_{\mathrm{lat}}\cdot\bm{D}_{\mathrm{lat}}\right)\psi(x),$ $\displaystyle S_{(6,1)}$ $\displaystyle=$ $\displaystyle c_{1}a^{6}\sum_{x}\bar{\psi}(x)\sum_{i}\gamma_{i}{D_{i}}_{\mathrm{lat}}{\triangle_{i}}_{\mathrm{lat}}\psi(x)+c_{2}a^{6}\sum_{x}\bar{\psi}(x)\\{\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}},\triangle^{(3)}_{\mathrm{lat}}\\}\psi(x)$ (34) $\displaystyle+$ $\displaystyle c_{3}a^{6}\sum_{x}\bar{\psi}(x)\\{\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}},i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\\}\psi(x)$ $\displaystyle+$ $\displaystyle z_{3}a^{6}\sum_{x}\bar{\psi}(x)\bm{\gamma}\cdot\left(\bm{D}_{\mathrm{lat}}\times\bm{B}_{\mathrm{lat}}+\bm{B}_{\mathrm{lat}}\times\bm{D}_{\mathrm{lat}}\right)\psi(x)$ $\displaystyle+$ $\displaystyle c_{EE}a^{6}\sum_{x}\bar{\psi}(x)\\{\gamma_{4}{D_{4}}_{\mathrm{lat}},\bm{\alpha}\cdot\bm{E}_{\mathrm{lat}}\\}\psi(x),$ $\displaystyle S_{(7,0)}$ $\displaystyle=$ $\displaystyle c_{4}a^{7}\sum_{x}\bar{\psi}(x)\sum_{i}{\triangle_{i}}_{\mathrm{lat}}^{2}\psi(x)+c_{5}a^{7}\sum_{x}\bar{\psi}(x)\sum_{i}\sum_{j\neq i}\\{i\Sigma_{i}{B_{i}}_{\mathrm{lat}},{\triangle_{j}}_{\mathrm{lat}}\\}\psi(x)$ (35) $\displaystyle+$ $\displaystyle r_{5}a^{7}\sum_{x}\bar{\psi}(x)\sum_{i}\sum_{j\neq i}i\Sigma_{i}\left[D_{j}B_{i}D_{j}\right]_{\mathrm{lat}}\psi(x)$ $\displaystyle+$ $\displaystyle z_{6}a^{7}\sum_{x}\bar{\psi}(x)\left(\triangle^{(3)}_{\mathrm{lat}}\right)^{2}\psi(x)+z_{7}a^{7}\sum_{x}\bar{\psi}(x)\\{\triangle^{(3)}_{\mathrm{lat}},i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\\}\psi(x)$ $\displaystyle+$ $\displaystyle z^{\prime}_{7}a^{7}\sum_{x}\bar{\psi}(x)[D_{i}i\bm{\Sigma}\cdot\bm{B}D_{i}]_{\mathrm{lat}}\psi(x)$ $\displaystyle+$ $\displaystyle r_{7}a^{7}\sum_{x}\bar{\psi}(x)\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}}i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}}\psi(x)$ $\displaystyle+$ $\displaystyle r^{\prime}_{7}a^{7}\sum_{x}\bar{\psi}(x)[\bm{D}\cdot\left(\bm{B}\times\bm{D}\right)]_{\mathrm{lat}}\psi(x)$ $\displaystyle+$ $\displaystyle r_{BB}a^{7}\sum_{x}\bar{\psi}(x)\left(i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\right)^{2}\psi(x)+z_{BB}a^{7}\sum_{x}\bar{\psi}(x)\bm{B}_{\mathrm{lat}}\cdot\bm{B}_{\mathrm{lat}}\psi(x)$ $\displaystyle-$ $\displaystyle r_{EE}a^{7}\sum_{x}\bar{\psi}(x)\left(\bm{\alpha}\cdot\bm{E}_{\mathrm{lat}}\right)^{2}\psi(x)+z_{EE}a^{7}\sum_{x}\bar{\psi}(x)\bm{E}_{\mathrm{lat}}\cdot\bm{E}_{\mathrm{lat}}\psi(x).$ All couplings in Eqs. (30)–(35) are real; explicit factors of $i$ are fixed by reflection positivity Osterwalder:1977pc of the continuum action. Some of the improvement terms extend over more than one timeslice, so there are small violations of reflection positivity for the lattice action. We expect that the associated problems are not severe, as with the improved gauge action Luscher:1984is . Equations (33)–(35) contain 19 new couplings. The convention for couplings $c_{i}$, $r_{i}$ and $z_{i}$ is as follows. In matching calculations we find that couplings $z_{i}$ vanish at the tree level, while the couplings $c_{i}$ do not. Couplings $r_{i}$ are redundant and, for this reason, could be omitted. The analysis in Sect. II gives the _number_ of redundant interactions, rather than the specific choices of interactions themselves. The possibilities for the dimension-7 redundant directions are as follows. One of $(c_{4},c_{5},r_{5})$ is redundant; we choose $r_{5}$. Furthermore, one of $(z_{6},z_{7},r_{7},r_{BB})$, another of $(z_{7},r_{7},r_{BB})$, and another of $(z_{7},r_{7},r^{\prime}_{7},r_{BB})$ are redundant; we choose $r_{7}$, $r^{\prime}_{7}$, and $r_{BB}$. But because pragmatic considerations could motivate other choices, we keep all of them in our analysis. This strategy also provides a good way for the matching calculations to verify the formal analysis of the LE${\cal L}$. In future numerical work, we recommend choosing $r_{s}$, as usual, to solve the doubling problem (in practice $r_{s}\geq 1$). The others may be chosen to save computer time, which presumably means choosing the couplings of computationally demanding interactions to vanish. The difference operators and fields with the subscript “lat” are taken to be $\displaystyle{D_{\rho}}_{\mathrm{lat}}$ $\displaystyle=$ $\displaystyle(T_{\rho}-T_{-\rho})/2a$ (36) $\displaystyle{\triangle_{\rho}}_{\mathrm{lat}}$ $\displaystyle=$ $\displaystyle(T_{\rho}+T_{-\rho}-2)/a^{2},\quad\triangle^{(3)}_{\mathrm{lat}}=\sum_{i=1}^{3}{\triangle_{i}}_{\mathrm{lat}},$ (37) $\displaystyle{F_{\rho\sigma}}_{\mathrm{lat}}$ $\displaystyle=$ $\displaystyle\frac{1}{8a^{2}}\sum_{\bar{\rho}=\pm\rho}\sum_{\bar{\sigma}=\pm\sigma}\mathop{\mathrm{sgn}}\bar{\rho}\mathop{\mathrm{sgn}}\bar{\sigma}\left[T_{\bar{\rho}}T_{\bar{\sigma}}T_{-\bar{\rho}}T_{-\bar{\sigma}}-T_{\bar{\sigma}}T_{\bar{\rho}}T_{-\bar{\sigma}}T_{-\bar{\rho}}\right],$ (38) where the covariant translation operators $T_{\pm\rho}$ translate all fields to the right one site in the $\pm\rho$ direction, and multiply by the appropriate link matrix Kronfeld:1984zv . These discretizations are conventional for $S_{0}+S_{B}+S_{E}$. For the new interactions, we have re- used the same ingredients. For the interactions with couplings $r_{5}$ and $z^{\prime}_{7}$ one can consider $\left[D_{j}B_{i}D_{j}\right]_{\mathrm{lat}}={D_{j}}_{\mathrm{lat}}{B_{i}}_{\mathrm{lat}}{D_{j}}_{\mathrm{lat}},$ (39) or $\left[D_{j}B_{i}D_{j}\right]_{\mathrm{lat}}=\frac{1}{2a^{2}}\left[(1-T_{-j}){B_{i}}_{\mathrm{lat}}(T_{j}-1)+(T_{j}-1){B_{i}}_{\mathrm{lat}}(1-T_{-j})\right].$ (40) In tree-level matching calculation, both lead to the same dependence on $r_{5}$ and $z^{\prime}_{7}$. Equation (39) has the advantage that is re-uses elements that are already defined (in a computer program, say) for the dimension-4 and -5 action. Equation (40) is more local, however, and may have other advantages. A FermiQCD DiPierro:2003sz computer code of the new action indicates that Eq. (39) is faster Massimo:2008sz . This code also indicates that it is advantageous to choose the redundant directions so that one may set $r_{5}=r_{7}=0$. The improved gluon action $S_{D^{2}F^{2}}$ is defined in Appendix C. The four- quark action $S_{\bar{q}q\bar{q}q}$ contains the obvious discretization of the (continuum) operators explained in Sec. II.5 and listed in Tables 4 and 5: simply substitute lattice fermion fields for the continuum fields, and assign each a real coupling. When matching to continuum QCD, the couplings in $S_{\bar{q}q\bar{q}q}$ start at order $g^{2}$, making them commensurate with order-$g^{2}$ matching effects in $S_{(6,1)}+S_{(7,0)}$, such as tree-level quark-quark scattering. To incorporate the four-quark action in a Monte Carlo simulation, one would introduce auxiliary fields to recover a bilinear action. In the next section we show, however, that these operators are not necessary for the target accuracy of 1–2%, so this cumbersome set-up can be avoided for now. ## IV Matching Conditions In this section we derive improvement conditions on the new couplings at the tree level. We calculate on-shell observables for small $\bm{p}a$ without any assumption on $m_{Q}a$. We look at the energy as a function of 3-momentum, which is sensitive to $c_{1}$, $c_{2}$, $c_{4}$, and $z_{6}$. We then look at the interaction of a quark with classical background chromoelectric and chromomagnetic fields. The former is sensitive to $c_{E}$, $r_{E}$, and $z_{E}$; the latter to all but $c_{EE}$, $r_{EE}$, $z_{EE}$, $r_{BB}$, and $z_{BB}$. To ensure that these results are compatible with the improved gauge action, we next compute the amplitude for quark-quark scattering. This step also matches the four-quark interactions, which are not written out explicitly in Sec. III. Finally, we compute the amplitude for Compton scattering to match $c_{EE}$, $r_{EE}$, $z_{EE}$, $r_{BB}$, and $z_{BB}$. ### IV.1 Energy The energy of a heavy quark on the lattice is defined through the exponential fall-off in time of the propagator. For small momentum $\bm{p}$ the energy can be written $E=m_{1}+\frac{\bm{p}^{2}}{2m_{2}}-{\textstyle\frac{1}{6}}w_{4}a^{3}\sum_{i}p_{i}^{4}-\frac{\left(\bm{p}^{2}\right)^{2}}{8m_{4}^{3}}+\cdots,$ (41) where the coefficients $m_{1}$, $m_{2}$, $m_{4}$ and $w_{4}$ depend on the couplings in the action. Appendix A contains the Feynman rule for the propagator and recalls the general formula for the energy, Eq. (127). By explicit calculation we find $\displaystyle m_{1}a$ $\displaystyle=$ $\displaystyle\ln(1+m_{0}a),$ (42) $\displaystyle\frac{1}{m_{2}a}$ $\displaystyle=$ $\displaystyle\frac{2\zeta^{2}}{m_{0}a(2+m_{0}a)}+\frac{r_{s}\zeta}{1+m_{0}a},$ (43) $\displaystyle w_{4}$ $\displaystyle=$ $\displaystyle\frac{2\zeta(\zeta+6c_{1})}{m_{0}a(2+m_{0}a)}+\frac{r_{s}\zeta-24c_{4}}{4(1+m_{0}a)},$ (44) $\displaystyle\frac{1}{m_{4}^{3}a^{3}}$ $\displaystyle=$ $\displaystyle\frac{8\zeta^{4}}{[m_{0}a(2+m_{0}a)]^{3}}+\frac{4\zeta^{4}+8r_{s}\zeta^{3}(1+m_{0}a)}{[m_{0}a(2+m_{0}a)]^{2}}+\frac{r_{s}^{2}\zeta^{2}}{(1+m_{0}a)^{2}}$ (45) $\displaystyle+\frac{32\zeta c_{2}}{m_{0}a(2+m_{0}a)}-\frac{8z_{6}}{1+m_{0}a}.$ The dimension-6 and -7 couplings $(c_{1},c_{4})$ and $(c_{2},z_{6})$ modify $w_{4}$ and $m_{4}a$, but not $m_{1}a$ or $m_{2}a$. To match Eq. (41) to the continuum QCD, one requires $m_{4}=m_{2}$ and $w_{4}=0$. From $m_{4}=m_{2}$ one obtains the tuning condition $\displaystyle 16\zeta c_{2}$ $\displaystyle=$ $\displaystyle\frac{4\zeta^{4}(\zeta^{2}-1)}{[m_{0}a(2+m_{0}a)]^{2}}-\frac{\zeta^{3}[2\zeta+4r_{s}(1+m_{0}a)-6r_{s}\zeta^{2}/(1+m_{0}a)]}{m_{0}a(2+m_{0}a)}$ (46) $\displaystyle+$ $\displaystyle\frac{3r_{s}^{2}\zeta^{4}}{(1+m_{0}a)^{2}}+\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)}\left[8z_{6}+\frac{r_{s}^{3}\zeta^{3}}{(1+m_{0}a)^{2}}-\frac{r_{s}^{2}\zeta^{2}}{1+m_{0}a}\right],$ which (at fixed $m_{0}a$) prescribes a line in the $(c_{2},z_{6})$ plane. From $w_{4}=0$ one obtains the tuning condition $0=\zeta^{2}+6\zeta c_{1}+\left(r_{s}\zeta-24c_{4}\right)\frac{m_{0}a(2+m_{0}a)}{8(1+m_{0}a)},$ (47) which (at fixed $m_{0}a$) prescribes a line in the $(c_{1},c_{4})$ plane. As $m_{0}a\to 0$, both lines become vertical: the coefficients $c_{1}$ and $c_{2}$ of dimension-6 operators are fixed, whereas the coefficients of $c_{4}$ and $z_{6}$ dimension-7 operators are undetermined. At this stage it is tempting to choose $c_{4}$ and $z_{6}$ to be two of the redundant couplings, but below we shall see that there are better choices. ### IV.2 Background Field To compute the interaction of a lattice quark with a continuum background field, we have to compute vertex diagrams with one gluon attached to the quark line. The Feynman rules are given in Eqs. (146) and (147). Our Feynman rules introduce a gauge potential via $U_{\mu}(x)=\exp\left[g_{0}A_{\mu}(x+{\textstyle\frac{1}{2}}e_{\mu}a)\right],$ (48) where $e_{\mu}$ is a unit vector in the $\mu$ direction, and take the Fourier transform of the gauge field to be $A_{\mu}(x)=\int\frac{d^{4}k}{(2\pi)^{4}}e^{ik\cdot x}A_{\mu}(k).$ (49) A background field would, however, lead to parallel transporters $U_{\mu}(x)={\sf P}\exp\left[g_{0}\int_{0}^{1}A_{\mu}(x+se_{\mu}a)ds\right].$ (50) Equation (48) is a convention. If we use Eq. (50) instead, vertices, propagators, and external line factors for gluons would change, in such a way that Feynman diagrams for on-shell amplitudes end up being the same. To use the interaction with a background classical field as a matching condition, we must compute the current $J_{\mu}$ that couples to the background field $A_{\mu}$ in Eq. (50). Current conservation requires $k\cdot J(k)=0,$ (51) where $k$ is the external gluon’s momentum. The usual convention for $A_{\mu}(k)$, from Eqs. (48) and (49), yields a current $\hat{J}_{\mu}$ satisfying $\hat{k}\cdot\hat{J}(k)=0,$ (52) where $\hat{k}_{\mu}=(2/a)\sin(k_{\mu}a/2)$. One sees, therefore, that a classical gluon line with Lorentz index $\mu$ must be multiplied by $n_{\mu}(k)=\frac{\hat{k}_{\mu}}{k_{\mu}}\approx 1-\frac{k_{\mu}^{2}a^{2}}{24}.$ (53) One should think of $n_{\mu}(k)$ as a wave-function factor for the external line. Its appearance has been noted previously by Weisz Weisz:1982zw . In the rest of this subsection we match the vertex function in lattice gauge theory with our new action to that in the continuum gauge theory. The incoming quark’s momentum is $p$, the outgoing $p^{\prime}$, and the gluon’s $K=p^{\prime}-p$. The current is given by (no implied sum on $\mu$) $J_{\mu}=n_{\mu}(K)\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})\Lambda_{\mu}(p^{\prime},p)u(\xi,\bm{p})\mathcal{N}(p),$ (54) where $\Lambda_{\mu}(p^{\prime},p)$ is the vertex function derived in Appendix A. The external quarks take normalization factors $\mathcal{N}$ as well as spinor factors El-Khadra:1996mp . #### IV.2.1 Chromoelectric field: $\mu=4$ For the interaction with the chromoelectric background field, we use the time component $J_{4}$. To $O(\bm{p}^{2}/m^{2})$ the current in continuum QCD is $J_{4}=\bar{u}(\xi^{\prime},\bm{0})\left[1-\frac{\bm{K}^{2}-2i\bm{\Sigma}\cdot(\bm{K}\times\bm{P})}{8m^{2}}\right]u(\xi,\bm{0}),$ (55) where $P=(p^{\prime}+p)/2$. After a short calculation with the new lattice action we find $J_{4}=\bar{u}(\xi^{\prime},\bm{0})\left[1-\frac{\bm{K}^{2}-2i\bm{\Sigma}\cdot(\bm{K}\times\bm{P})}{8m_{E}^{2}}+\frac{z_{E}\bm{K}^{2}a^{2}}{1+m_{0}a}\right]u(\xi,\bm{0}),$ (56) where $\frac{1}{4m_{E}^{2}a^{2}}=\frac{\zeta^{2}}{[m_{0}a(2+m_{0}a)]^{2}}+\frac{\zeta^{2}c_{E}}{m_{0}a(2+m_{0}a)}+\frac{2r_{E}}{1+m_{0}a}.$ (57) The correct (tree-level) matching is achieved if one adjusts $z_{E}=0$ (58) and $(c_{E},r_{E})$ such that $m_{E}=m_{2}$: $\zeta^{2}c_{E}+r_{E}\frac{2m_{0}a(2+m_{0}a)}{1+m_{0}a}=\frac{\zeta^{2}(\zeta^{2}-1)}{m_{0}a(2+m_{0}a)}+\frac{r_{s}\zeta^{3}}{1+m_{0}a}+\frac{r_{s}^{2}\zeta^{2}m_{0}a(2+m_{0}a)}{4(1+m_{0}a)^{2}}.$ (59) At fixed $m_{0}a$ the latter prescribes a line in the $(c_{E},r_{E})$ plane. As before, this line becomes vertical at $m_{0}a=0$, fixing $c_{E}=1$ and leaving $r_{E}$ undetermined. To obtain conditions on $c_{EE}$, $r_{EE}$, and $z_{EE}$, we shall have to turn to Compton scattering in Sec. IV.4. #### IV.2.2 Chromomagnetic field: $\mu=i$ For the interaction with the chromomagnetic background field, we use the spatial components $J_{i}$. To $O(\bm{p}^{3}/m^{3})$ the current in continuum QCD is $\displaystyle J_{i}$ $\displaystyle=$ $\displaystyle-i\bar{u}(\xi^{\prime},\bm{0})\left\\{P_{i}\left(\frac{1}{m}-\frac{\bm{P}^{2}+{\textstyle\frac{1}{4}}\bm{K}^{2}}{2m^{3}}\right)-\frac{K_{i}\,\bm{P}\cdot\bm{K}}{8m^{3}}\right.$ (60) $\displaystyle\hskip 20.00003pt-\left.\varepsilon_{ijl}i\Sigma_{l}K_{j}\left(\frac{1}{2m}-\frac{\bm{P}^{2}+{\textstyle\frac{1}{4}}\bm{K}^{2}}{4m^{3}}\right)+\varepsilon_{ijl}i\Sigma_{l}P_{j}\frac{\bm{P}\cdot\bm{K}}{4m^{3}}\right\\}u(\xi,\bm{0}).$ After another short calculation we find $\displaystyle J_{i}$ $\displaystyle=$ $\displaystyle-i\bar{u}(\xi^{\prime},\bm{0})\left\\{P_{i}\left(\frac{1}{m_{2}}-\frac{\bm{P}^{2}+{\textstyle\frac{1}{4}}\bm{K}^{2}}{2m_{4}^{3}}\right)-\frac{K_{i}\,\bm{P}\cdot\bm{K}}{8m_{2}m_{E}^{2}}+\frac{z_{E}a^{2}K_{i}\,\bm{P}\cdot\bm{K}}{m_{2}(1+m_{0}a)}\right.$ (61) $\displaystyle\hskip 20.00003pt+\left.{\textstyle\frac{1}{8}}w_{B_{1}}a^{3}\left[P_{i}\bm{K}^{2}-K_{i}\,\bm{P}\cdot\bm{K}\right]-{\textstyle\frac{1}{16}}w_{B_{2}}a^{3}\varepsilon_{ijl}K_{j}i\Sigma_{l}\bm{K}^{2}\right.$ $\displaystyle\hskip 20.00003pt-\left.{\textstyle\frac{1}{4}}w_{B_{3}}a^{3}\varepsilon_{ijl}K_{j}P_{l}i\bm{\Sigma}\cdot\bm{P}+{\textstyle\frac{1}{4}}w_{X}a^{3}X_{i}\right.$ $\displaystyle\hskip 20.00003pt-\left.{\textstyle\frac{2}{3}}w_{4}a^{3}P_{i}(P_{i}^{2}+{\textstyle\frac{1}{4}}K_{i}^{2})+{\textstyle\frac{1}{12}}w^{\prime}_{B}a^{3}\varepsilon_{ijl}i\Sigma_{l}K_{j}(K_{i}^{2}+K_{j}^{2})\right.$ $\displaystyle\hskip 20.00003pt+\left.{\textstyle\frac{1}{12}}(w_{4}+w_{4}^{\prime})a^{3}\varepsilon_{ijl}i\Sigma_{l}K_{j}[(3P_{i}^{2}+{\textstyle\frac{1}{4}}K_{i}^{2})+(3P_{j}^{2}+{\textstyle\frac{1}{4}}K_{j}^{2})]\right.$ $\displaystyle\hskip 20.00003pt-\left.\varepsilon_{ijl}i\Sigma_{l}K_{j}\left(\frac{1}{2m_{B}}-\frac{\bm{P}^{2}+{\textstyle\frac{1}{4}}\bm{K}^{2}}{4m_{B^{\prime}}^{3}}\right)+\varepsilon_{ijl}i\Sigma_{l}P_{j}\frac{\bm{P}\cdot\bm{K}}{4m_{2}m_{E}^{2}}\right\\}u(\xi,\bm{0}),$ where $m_{2}$, $m_{4}^{3}$, $w_{4}$, and $m_{E}^{2}$ have been introduced already, and $\displaystyle\frac{1}{m_{B}a}$ $\displaystyle=$ $\displaystyle\frac{1}{m_{2}a}+\frac{(c_{B}-r_{s})\zeta}{1+m_{0}a},$ (62) $\displaystyle\frac{1}{m_{B^{\prime}}^{3}a^{3}}$ $\displaystyle=$ $\displaystyle\frac{1}{m_{4}^{3}a^{3}}-\frac{r_{s}(r_{s}-c_{B})\zeta^{2}}{(1+m_{0}a)^{2}}+\frac{8(z_{6}-z_{7})+4(r_{7}-z^{\prime}_{7})}{1+m_{0}a},$ (63) $\displaystyle w_{B_{3}}$ $\displaystyle=$ $\displaystyle\frac{4(r_{s}-c_{B})\zeta^{3}(1+m_{0}a)}{[m_{0}a(2+m_{0}a)]^{2}}+\frac{16(c_{2}-c_{3})\zeta}{m_{0}a(2+m_{0}a)}+\frac{8r_{7}}{1+m_{0}a},$ (64) $\displaystyle w_{B_{2}}$ $\displaystyle=$ $\displaystyle w_{B_{3}}+\frac{16z_{3}\zeta}{m_{0}a(2+m_{0}a)}-\frac{8z^{\prime}_{7}}{1+m_{0}a},$ (65) $\displaystyle w_{B_{1}}$ $\displaystyle=$ $\displaystyle w_{B_{2}}-\frac{8(r^{\prime}_{7}-z^{\prime}_{7})}{1+m_{0}a},$ (66) $\displaystyle w^{\prime}_{B}$ $\displaystyle=$ $\displaystyle\frac{c_{B}\zeta-4(c_{5}-r_{5})}{1+m_{0}a},$ (67) $\displaystyle w^{\prime}_{4}$ $\displaystyle=$ $\displaystyle-\frac{r_{s}\zeta-24c_{4}+16(2c_{5}+r_{5})}{4(1+m_{0}a)}.$ (68) The term $w_{X}a^{3}\bm{X}$ is discussed below. Comparing Eqs. (60) and (61), one sees that the first four terms match the continuum if $m_{2}=m_{4}=m_{E}=m$. The other terms do not match unless one adjusts $c_{B}=r_{s}$ El-Khadra:1996mp and $z_{E}=0$ [as in Eq. (58)] and, furthermore, demands $w_{4}=w^{\prime}_{4}=w_{B_{1}}=w_{B_{2}}=w_{B_{3}}=w^{\prime}_{B}=0$: $\displaystyle c_{3}$ $\displaystyle=$ $\displaystyle c_{2}+\frac{r_{7}}{\zeta}\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)},$ (69) $\displaystyle z_{3}$ $\displaystyle=$ $\displaystyle\frac{r^{\prime}_{7}}{\zeta}\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)},$ (70) $\displaystyle c_{4}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{24}}r_{s}\zeta+{\textstyle\frac{1}{3}}c_{B}\zeta+2r_{5},$ (71) $\displaystyle c_{5}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{4}}c_{B}\zeta+r_{5},$ (72) $\displaystyle z_{7}$ $\displaystyle=$ $\displaystyle z_{6}+{\textstyle\frac{1}{2}}(r_{7}-r^{\prime}_{7}),$ (73) $\displaystyle z^{\prime}_{7}$ $\displaystyle=$ $\displaystyle r^{\prime}_{7}.$ (74) Taken with Eqs. (46) and (47), these tuning conditions put eight constraints on the nine (non-redundant) couplings for interactions made solely out of spatial derivatives (and, hence, chromomagnetic fields). To eliminate $z_{6}$ from the right-hand side of Eq. (73), and to obtain conditions on $r_{BB}$ and $z_{BB}$, we shall have to turn to Compton scattering in Sec. IV.4. Equations (69)–(74) make concrete several abstract features of Sec. II. If one would like to take $c_{4}$ to be redundant in Eq. (47), then one cannot take $r_{5}$ to be redundant here, and similarly for $z_{6}$ and $r_{7}$ or $r^{\prime}_{7}$. Also, a mistuned $c_{5}-r_{5}$ leads to $w^{\prime}_{B}\neq 0$ and a spin-dependent contribution $[1+{\textstyle\frac{1}{6}}w^{\prime}_{B}m_{2}a(K_{i}^{2}+K_{j}^{2})a^{2}]\varepsilon_{ijl}i\Sigma_{l}K_{j}/2m_{2}$. The mismatch here is suppressed by $\lambda^{2}$ in the HQET counting—as expected from Table 2—and by $a^{3}$ in the usual Symanzik counting. The only undesired term in Eq. (61) not yet discussed is ${\textstyle\frac{1}{4}}w_{X}a^{3}X_{i}$, where $\displaystyle\bm{X}$ $\displaystyle=$ $\displaystyle(i\bm{\Sigma}\times\bm{K})\,\bm{P}^{2}-(i\bm{\Sigma}\times\bm{P})\,\bm{P}\cdot\bm{K}-\bm{P}\,[i\bm{\Sigma}\cdot(\bm{K}\times\bm{P})]+(\bm{K}\times\bm{P})\,i\bm{\Sigma}\cdot\bm{P},\quad$ (75) $\displaystyle w_{X}$ $\displaystyle=$ $\displaystyle\frac{4r_{s}\zeta^{3}(1+m_{0}a)}{[m_{0}a(2+m_{0}a)]^{2}}+\frac{16c_{2}\zeta}{m_{0}a(2+m_{0}a)}.$ (76) One cannot tune $w_{X}=0$. Fortunately, however, $\bm{X}=\bm{0}$. A simple geometric proof is as follows: if, by chance, $\bm{P}$ is parallel to $\bm{K}$, then setting $\bm{P}\propto\bm{K}$ one sees that the last two terms on the right-hand side of Eq. (75) vanish and the first two cancel. In the general case that $\bm{P}$ is not parallel to $\bm{K}$, then $\bm{K}$, $\bm{P}$, and $\bm{K}\times\bm{P}$ are three linearly independent vectors. But one easily sees that $\bm{K}\cdot\bm{X}=\bm{P}\cdot\bm{X}=(\bm{K}\times\bm{P})\cdot\bm{X}=0;$ (77) thus, $\bm{X}=\bm{0}$. Such identities are very useful in simplifying expressions for the Compton scattering amplitude. ### IV.3 Quark-quark scattering To match the four-quark action, $S_{\bar{q}q\bar{q}q}$, one must work out the quark-quark scattering amplitude. With the current $J_{\mu}$ derived in the previous subsection, this is a relatively simple task. The main new ingredient is the improved gluon propagator. For $k^{2}a^{2}\ll 1$, one finds Weisz:1982zw $D_{\mu\nu}(k)=n_{\mu}(k)D_{\mu\nu}^{\rm cont}(k)n_{\nu}(k)\left[1+xa^{2}k^{2}\right]+O(a^{4}),$ (78) where $x$ is the redundant coupling of the pure-gauge action, cf. Appendix C and Ref. Luscher:1984xn . This approximation suffices for evaluating $t$-channel gluon exchange. Once the bilinear action has been matched correctly, the lattice amplitude (using, say, Feynman gauge) is clearly merely $\mathcal{A}_{\rm lat}(12\to 12)=\mathcal{A}_{\rm cont}(12\to 12)+xa^{2}t^{a}J_{1}\cdot J_{2}t^{a},$ (79) where 1 and 2 label the scattered quark flavors, and both $t^{a}$ have uncontracted color indices. We find, therefore, that the tree-level couplings of $S_{\bar{q}q\bar{q}q}$ are, at most, proportional to $x$. They can be eliminated, at the tree level, by setting $x=0$, with the added benefit of simplifying the gauge action $S_{D^{2}F^{2}}$. Note, however, that the approximation in Eq. (78) and, thus, Eq. (79), breaks down for $s$-channel annihilation of heavy quarks. As discussed in Sec. II.5, these interactions are suppressed for other reasons, so the four-quark operators needed to correct them may be neglected. ### IV.4 Compton scattering The matching of Secs. IV.1–IV.3 leaves four non-redundant couplings of the new action undetermined: $z_{6}$, $c_{EE}$, $z_{EE}$, and $z_{BB}$. To find four more matching conditions, we turn to Compton scattering. We shall proceed with the gauge-action redundant coupling $x=0$. The amplitude is $\mathcal{A}_{\rm lat}^{ab}(qg\to qg)=\sum_{\mu\nu}\bar{\epsilon}^{\prime}_{\nu}(k^{\prime})n_{\nu}(k^{\prime})\hat{\mathcal{M}}_{\mu\nu}^{ab}\epsilon_{\mu}(k)n_{\mu}(k),$ (80) where $\bar{\epsilon}_{\nu}$ and $\epsilon_{\mu}$ are continuum polarization vectors, and $\hat{\mathcal{M}}_{\mu\nu}^{ab}$ denotes the sum of Feynman diagrams shown in Fig. 2. Figure 2: Feynman diagrams for Compton scattering in lattice gauge theory. The factors $n_{\nu}(k^{\prime})$ and $n_{\mu}(k)$ appear in Eq. (80) to account for lattice gluons. With them one can verify that $\sum_{\rm pol.}\epsilon_{\mu}(k)n_{\mu}(k)n_{\nu}(k)\bar{\epsilon}_{\nu}(k)=-D_{\mu\nu}(k),$ (81) as usual. We find it convenient to associate these factors with the diagrams and introduce $\mathcal{M}_{\mu\nu}^{ab}=n_{\nu}(k^{\prime})\hat{\mathcal{M}}_{\mu\nu}^{ab}n_{\mu}(k)$. Then $\displaystyle\mathcal{M}_{\mu\nu}^{ab}$ $\displaystyle=$ $\displaystyle t^{b}t^{a}n_{\nu}(k^{\prime})\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})\Lambda_{\nu}(p^{\prime},q)S(q)\Lambda_{\mu}(q,p)u(\xi,\bm{p})\mathcal{N}(p)n_{\mu}(k)$ $\displaystyle+$ $\displaystyle t^{a}t^{b}n_{\nu}(k^{\prime})\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})\Lambda_{\mu}(p^{\prime},q^{\prime})S(q^{\prime})\Lambda_{\mu}(q^{\prime},p)u(\xi,\bm{p})\mathcal{N}(p)n_{\mu}(k)$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}\\{t^{a},t^{b}\\}n_{\nu}(k^{\prime})\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})aX_{\mu\nu}(p,k,-k^{\prime})u(\xi,\bm{p})\mathcal{N}(p)n_{\mu}(k)$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}[t^{a},t^{b}]n_{\nu}(k^{\prime})\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})aY_{\mu\nu}(p,k,-k^{\prime})u(\xi,\bm{p})\mathcal{N}(p)n_{\mu}(k),$ $\displaystyle+$ $\displaystyle t^{c}V^{abc}_{\mu\nu\sigma}(k,-k^{\prime},-K)D_{\sigma\rho}(K)n_{\nu}(k^{\prime})\mathcal{N}(p^{\prime})\bar{u}(\xi^{\prime},\bm{p}^{\prime})\Lambda_{\rho}(p^{\prime},p)u(\xi,\bm{p})\mathcal{N}(p)n_{\mu}(k)$ where $q=p+k=p^{\prime}+k^{\prime}$, $q^{\prime}=p-k^{\prime}=p^{\prime}-k$, $K=k-k^{\prime}=p^{\prime}-p$. The propagator $S(q)$ and vertex factors $\Lambda_{\mu}$, $X_{\mu\nu}$ and $Y_{\mu\nu}$ are defined in Appendix A. The gluon propagator, to the accuracy needed, is given in Eq. (78), and to the same accuracy the triple-gluon vertex is (with $x=0$) $\displaystyle V^{abc}_{\mu\nu\sigma}(k,-k^{\prime},-K)$ $\displaystyle=$ $\displaystyle if^{abc}\left[n_{\mu}(k)n_{\nu}(k^{\prime})n_{\sigma}(K)\right]^{-1}\left\\{\vphantom{{\textstyle\frac{1}{2}}}\right.$ (83) $\displaystyle\delta_{\mu\nu}[(k+k^{\prime})_{\sigma}(1-{\textstyle\frac{1}{12}}\delta_{\mu\sigma}K^{2}a^{2})+{\textstyle\frac{1}{12}}K_{\sigma}(k^{2}_{\mu}-{k^{\prime}_{\mu}}^{2})a^{2}]$ $\displaystyle-$ $\displaystyle\delta_{\nu\sigma}[(k^{\prime}-K)_{\mu}(1-{\textstyle\frac{1}{12}}\delta_{\nu\mu}k^{2}a^{2})+{\textstyle\frac{1}{12}}k_{\mu}({k^{\prime}_{\nu}}^{2}-K_{\nu}^{2})a^{2}]$ $\displaystyle-$ $\displaystyle\delta_{\sigma\mu}[(K+k)_{\nu}(1-{\textstyle\frac{1}{12}}\delta_{\sigma\nu}{k^{\prime}}^{2}a^{2})-{\textstyle\frac{1}{12}}k^{\prime}_{\nu}(K^{2}_{\sigma}-k_{\sigma}^{2})a^{2}]\left.\vphantom{{\textstyle\frac{1}{2}}}\right\\}.$ Note that the factors $n_{\sigma}(K)$, _etc._ , arise naturally. Note also that $K\cdot J=k\cdot\epsilon=k^{\prime}\cdot\bar{\epsilon}^{\prime}=k^{2}={k^{\prime}}^{2}=0$, so most of the lattice artifacts in the vertex drop out. The remaining one is necessary to cancel a similar lattice artifact from the other diagrams, cf. Eqs. (164) and (165). We may choose the polarization vectors such that $\bar{\epsilon}^{\prime}_{4}=\epsilon_{4}=0$. Then we need only focus on $\mathcal{M}_{mn}$. We have verified that $\mathcal{M}_{44}$ is improved by (a subset of) the improvement conditions needed for $\mathcal{A}(qg\to qg)$ calculated with these polarization vectors. The present the results, let us introduce some notation. Write the momenta as $\displaystyle P$ $\displaystyle=$ $\displaystyle(p^{\prime}+p)/2,$ (84) $\displaystyle R$ $\displaystyle=$ $\displaystyle(k+k^{\prime})/2,$ (85) $\displaystyle K$ $\displaystyle=$ $\displaystyle p^{\prime}-p=k-k^{\prime},$ (86) so $q=P+R$ and $q^{\prime}=P-R$. Note that $P_{0}=-iP_{4}=2m_{1}+\cdots$ is larger than the other momenta, and $K_{0}=-iK_{4}=({\bm{p}^{\prime}}^{2}-\bm{p}^{2})/2m_{2}$ is smaller. Next separate the diagrams according to a color decomposition, $\mathcal{M}_{\mu\nu}^{ab}={\textstyle\frac{1}{2}}\\{t^{a},t^{b}\\}\mathcal{M}_{\mu\nu}+{\textstyle\frac{1}{2}}[t^{a},t^{b}]\mathcal{N}_{\mu\nu},$ (87) where the second term would be absent in an Abelian gauge theory. Finally, write $\mathcal{M}_{\mu\nu}=\sum_{n=0}^{3}\sum_{s=0}^{n}R_{0}^{n-1-2s}\mathcal{M}_{\mu\nu}^{(n,n-1-2s)},$ (88) and similarly for $\mathcal{N}_{\mu\nu}$, where the superscript $(n,r)$ denotes the power in $1/m$ and $R_{0}$. Most of these terms are well-matched with Eqs. (58), (59), (69)–(74). New matching conditions come from $\mathcal{M}_{mn}^{(3,2)}$, $\mathcal{N}_{mn}^{(3,2)}$, $\mathcal{M}_{mn}^{(3,0)}$, and $\mathcal{N}_{mn}^{(3,0)}$. The $(n,r)=(3,2)$ amplitudes are $\displaystyle\mathcal{M}_{mn}^{(3,2)}$ $\displaystyle=$ $\displaystyle\frac{\delta_{mn}}{4m_{EE}^{3}}+\frac{2a^{3}z_{EE}\delta_{mn}}{1+m_{0}a},$ (89) $\displaystyle\mathcal{N}_{mn}^{(3,2)}$ $\displaystyle=$ $\displaystyle\frac{\varepsilon_{mni}i\Sigma_{i}}{4m_{EE}^{3}},$ (90) where $\displaystyle\frac{1}{m_{EE}^{3}a^{3}}$ $\displaystyle=$ $\displaystyle\frac{8[\zeta+{\textstyle\frac{1}{2}}c_{E}\zeta m_{0}a(2+m_{0}a)]^{2}}{[m_{0}a(2+m_{0}a)]^{3}}+\frac{4\zeta^{2}}{[m_{0}a(2+m_{0}a)]^{2}}$ (91) $\displaystyle+$ $\displaystyle\frac{16c_{EE}\zeta}{m_{0}a(2+m_{0}a)(1+m_{0}a)}+\frac{8(c_{EE}\zeta+r_{EE})}{1+m_{0}a}.$ To match to continuum QCD one requires $z_{EE}=0$ (92) and the adjustment of $(c_{EE},r_{EE})$ so that $m_{EE}=m_{2}$. As with, say, $(c_{E},r_{E})$, at fixed $m_{0}a$ the latter prescribes a line in the $(c_{EE},r_{EE})$ plane, which becomes vertical at $m_{0}a=0$, fixing $c_{EE}=-{\textstyle\frac{1}{8}}$ and leaving $r_{EE}$ undetermined. The $(n,r)=(3,0)$ amplitudes are $\displaystyle\mathcal{M}_{mn}^{(3,0)}$ $\displaystyle=$ $\displaystyle\left.\mathcal{M}_{mn}^{(3,0)}\right|_{\rm matched}-\frac{2a^{3}}{e^{m_{1}a}}(z_{BB}+z_{6}+r_{7}-r_{BB}-z^{\prime}_{7})M_{mn},$ (93) $\displaystyle M_{mn}$ $\displaystyle=$ $\displaystyle\delta_{mn}(\bm{R}^{2}-{\textstyle\frac{1}{4}}\bm{K}^{2})-(R_{m}-{\textstyle\frac{1}{2}}K_{m})(R_{n}+{\textstyle\frac{1}{2}}K_{n}),$ (94) $\displaystyle\mathcal{N}_{mn}^{(3,0)}$ $\displaystyle=$ $\displaystyle\left.\mathcal{N}_{mn}^{(3,0)}\right|_{\rm matched}-\frac{2a^{3}}{e^{m_{1}a}}(z_{6}+r_{7}-r_{BB}-z^{\prime}_{7})N_{mn},$ (95) $\displaystyle N_{mn}$ $\displaystyle=$ $\displaystyle\varepsilon_{mnr}(R_{r}i\bm{\Sigma}\cdot\bm{R}-{\textstyle\frac{1}{4}}K_{r}i\bm{\Sigma}\cdot\bm{K})-{\textstyle\frac{1}{2}}(i\Sigma_{n}\varepsilon_{mrs}+i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s},$ (96) where “matched” denotes terms (spelled out in Appendix B) that already match, if the conditions derived so far are applied. Equations (93) and (95) yield the new conditions $\displaystyle z_{BB}+z_{6}-z^{\prime}_{7}$ $\displaystyle=$ $\displaystyle r_{BB}-r_{7},$ (97) $\displaystyle z_{6}-z^{\prime}_{7}$ $\displaystyle=$ $\displaystyle r_{BB}-r_{7}.$ (98) Solving these, and noting $z^{\prime}_{7}=r^{\prime}_{7}$ [Eq. (74)], we find $\displaystyle z_{BB}$ $\displaystyle=$ $\displaystyle 0,$ (99) $\displaystyle z_{6}$ $\displaystyle=$ $\displaystyle r_{BB}+r^{\prime}_{7}-r_{7},$ (100) which completes the set of conditions needed to match the new lattice action. ### IV.5 Matching Summary Equations (46), (47), (71)–(74), (99), and (100) can now be combined to yield $\displaystyle 6\zeta c_{1}$ $\displaystyle=$ $\displaystyle-\zeta^{2}+(c_{B}\zeta+6r_{5})\frac{m_{0}a(2+m_{0}a)}{1+m_{0}a},$ (101) $\displaystyle 16\zeta c_{2}$ $\displaystyle=$ $\displaystyle\frac{4\zeta^{4}(\zeta^{2}-1)}{[m_{0}a(2+m_{0}a)]^{2}}-\frac{\zeta^{3}[2\zeta+4r_{s}(1+m_{0}a)-6r_{s}\zeta^{2}/(1+m_{0}a)]}{m_{0}a(2+m_{0}a)}$ $\displaystyle+$ $\displaystyle\frac{3r_{s}^{2}\zeta^{4}}{(1+m_{0}a)^{2}}+\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)}\left[8(r_{BB}+r^{\prime}_{7}-r_{7})+\frac{r_{s}^{3}\zeta^{3}}{(1+m_{0}a)^{2}}-\frac{r_{s}^{2}\zeta^{2}}{1+m_{0}a}\right],$ $\displaystyle c_{3}$ $\displaystyle=$ $\displaystyle c_{2}+\frac{r_{7}}{\zeta}\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)}+\frac{(r_{s}-c_{B})\zeta^{2}(1+m_{0}a)}{4m_{0}a(2+m_{0}a)},$ (103) $\displaystyle c_{4}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{24}}r_{s}\zeta+{\textstyle\frac{1}{3}}c_{B}\zeta+2r_{5},$ (104) $\displaystyle c_{5}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{4}}c_{B}\zeta+r_{5},$ (105) $\displaystyle z_{3}$ $\displaystyle=$ $\displaystyle\frac{r^{\prime}_{7}}{\zeta}\frac{m_{0}a(2+m_{0}a)}{2(1+m_{0}a)},$ (106) $\displaystyle z_{6}$ $\displaystyle=$ $\displaystyle r_{BB}+r^{\prime}_{7}-r_{7},$ (107) $\displaystyle z_{7}$ $\displaystyle=$ $\displaystyle r_{BB}-{\textstyle\frac{1}{2}}(r_{7}-r^{\prime}_{7}),$ (108) $\displaystyle z^{\prime}_{7}$ $\displaystyle=$ $\displaystyle r^{\prime}_{7},$ (109) $\displaystyle z_{BB}$ $\displaystyle=$ $\displaystyle 0,$ (110) To run a numerical simulation, we would like to have as few new couplings as possible. The matching calculations verified the presence of several redundant directions. We may, therefore, take $r_{5}=r_{7}=r^{\prime}_{7}=r_{BB}=0$ (111) to all orders in perturbation theory. Hence $\displaystyle c_{B}$ $\displaystyle=$ $\displaystyle r_{s},$ (112) $\displaystyle c_{1}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{6}}\zeta+c_{B}\frac{m_{0}a(2+m_{0}a)}{6(1+m_{0}a)},$ (113) $\displaystyle c_{2}=c_{3}$ $\displaystyle=$ $\displaystyle\frac{\zeta^{3}(\zeta^{2}-1)}{[2m_{0}a(2+m_{0}a)]^{2}}-\frac{\zeta^{2}[\zeta+2r_{s}(1+m_{0}a)-3r_{s}\zeta^{2}/(1+m_{0}a)]}{8m_{0}a(2+m_{0}a)}$ (114) $\displaystyle+$ $\displaystyle\frac{3r_{s}^{2}\zeta^{3}}{16(1+m_{0}a)^{2}}+\frac{m_{0}a(2+m_{0}a)r_{s}^{2}\zeta}{32(1+m_{0}a)^{2}}\left[\frac{r_{s}\zeta}{1+m_{0}a}-1\right],$ $\displaystyle c_{4}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{24}}r_{s}\zeta+{\textstyle\frac{1}{3}}c_{B}\zeta,$ (115) $\displaystyle c_{5}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{4}}c_{B}\zeta,$ (116) and $z_{3}=z_{6}=z_{7}=z^{\prime}_{7}=z_{BB}=0.$ (117) From the chromoelectric interactions we require $m_{E}=m_{2}$ and $m_{EE}=m_{2}$, whence $c_{E}=\frac{\zeta^{2}-1}{m_{0}a(2+m_{0}a)}+\frac{r_{s}\zeta}{1+m_{0}a}+\frac{r_{s}^{2}m_{0}a(2+m_{0}a)}{4(1+m_{0}a)^{2}}-\frac{r_{E}}{\zeta^{2}}\frac{2m_{0}a(2+m_{0}a)}{1+m_{0}a},$ (118) $\displaystyle c_{EE}[2+m_{0}a(2+m_{0}a)]$ $\displaystyle=$ $\displaystyle\frac{\zeta(\zeta^{2}-1)(1+m_{0}a)}{[m_{0}a(2+m_{0}a)]^{2}}+\frac{c_{E}\zeta(\zeta^{2}-1)(1+m_{0}a)}{m_{0}a(2+m_{0}a)}$ (119) $\displaystyle+$ $\displaystyle\frac{\zeta(r_{s}\zeta-1-m_{0}a)}{2m_{0}a(2+m_{0}a)}+{\textstyle\frac{1}{2}}r_{s}c_{E}\zeta^{2}+2r_{E}\zeta-{\textstyle\frac{1}{4}}c_{E}^{2}\zeta(1+m_{0}a)$ $\displaystyle+$ $\displaystyle\frac{r_{s}r_{E}m_{0}a(2+m_{0}a)}{1+m_{0}a}-\frac{r_{EE}}{\zeta}m_{0}a(2+m_{0}a),$ and we also find $z_{E}=z_{EE}=0.$ (120) Without loss one may set the redundant $r_{E}=r_{EE}=0$ to simplify the action and Eqs. (118) and (119). In summary, of the nineteen new couplings in Eqs. (33)–(35), we find only _six_ that are non-zero at tree-level matching. Moreover, once the bilinear action has been matched, and the redundant gauge coupling $x=0$, the only non- zero four-quark interaction would correspond to (highly suppressed) $Q\bar{Q}$ annihilation. In the next section we shall examine the size of the remaining uncertainties, to justify that this level of matching suffices. ## V Errors from Truncation In this section we give a semi-quantitative analysis of heavy-quark discretization effects with the new action. Our aim is to study the accuracy needed in matching lattice gauge theory to continuum QCD. Several elements are needed. First, we need estimates of the mismatch at short distances. This is straightforward, because the calculations of Sec. IV can be applied to work out how large the mismatch is for the unimproved action. Second, we need estimates of the long-distance effects, which is possible parametrically, by counting powers of $\Lambda$ and $\upsilon$. Finally, the size of discretization effects depends on the lattice spacing (obviously) so we must note the range that is tractable today and in the near future. The error analysis is convenient using the non-relativistic description. Heavy-quark effects of operators that are related as in Eqs. (14) and (15) are lumped into one short-distance coefficient ${\cal C}_{i}^{\mathrm{lat}}$ per HQET operator in Table 3. In Sec. IV the short-distance coefficients are $1/2m_{2}$, $1/2m_{B}$, $1/4m_{E}^{2}$, $1/8m_{4}^{3}$, $w_{4}$, $w_{B_{i}}$, etc. In the corresponding continuum short-distance coefficients ${\cal C}_{i}^{\mathrm{cont}}$, these masses are replaced with a single mass $m_{Q}$. To eliminate discretization effects from the kinetic energy, one should identify $m_{Q}$ with $m_{2}$. Comparison of Eqs. (5) and (6) then says that heavy-quark discretization effects take the form $\mathtt{error}_{i}=\left({\cal C}_{i}^{\mathrm{lat}}-{\cal C}_{i}^{\mathrm{cont}}\right)\langle{\cal O}_{i}\rangle.$ (121) For example, the error from $(\bm{p}^{2})^{2}/8m_{4}^{3}$ is $\mathtt{error}_{m_{4}}=\left(\frac{1}{8m_{4}^{3}a^{3}}-\frac{1}{(2m_{2}a)^{3}}\right)a^{3}\langle(\bm{p}^{2})^{2}\rangle.$ (122) See Refs. Kronfeld:2000ck ; Harada:2001fi for further details, and Ref. Kronfeld:2003sd for the application of this technique to compare several heavy-quark formalisms. We estimate the matrix elements $\langle{\cal O}_{i}\rangle$ using the power counting of HQET and NRQCD for heavy-light hadrons and quarkonium, respectively. The power of $\lambda$ or $\upsilon$ is listed in Table 3. The coefficient mismatches are obtained from Sec. IV, where explicit expressions show how the coefficients depend on the new couplings. In particular, when the new couplings vanish, we derive the mismatch for the Wilson and clover actions. Explicit calculations of the mismatch at higher orders of perturbation theory are not yet available. (They would be tantamount to higher-loop matching.) Nevertheless, the asymptotic behavior remains constrained, when $m_{Q}a\ll 1$ because of the presence of the $\mathcal{E}$-type operators, when $m_{Q}a\not\ll 1$ by heavy-quark symmetry, and when $m_{Q}a\sim 1$ because the Wilson time derivative ensures only one pole in the propagator Kronfeld:2000ck . It turns out that the most pessimistic asymptotic behavior for $1/2m_{B}$, $1/4m_{E}^{2}$, etc., is the same at higher orders as in the tree level formulas in Sec. IV. It seems reasonable, therefore, to multiply the tree- level mismatch with $\alpha_{s}^{l}$ to estimate the $l$-loop mismatch. We use one-loop running for $\alpha_{s}(a)$ starting with $\alpha_{s}(1/11~{}{\rm fm})=1/3$. This yields the high end of the Brodsky-Lepage-Mackenzie coupling Brodsky:1982gc calculated for similar quantities Harada:2002jh . The resulting estimates for the mismatch of rotationally symmetric operators are shown in Fig. 3, as a function of the lattice spacing $a=m_{2}a/m_{Q}$, $Q\in\\{c,b\\}$. Figure 3: Relative truncation errors for the new action. The light gray or red curves stand for $c$ quarks; dark gray or blue for $b$. Dotted curves show the error when the contribution is unimproved. Dashed and solid curves show the error for tree-level and one-loop matching, respectively, of the needed operators. $\Lambda=700$ MeV, $m_{c}=1400$ MeV, $m_{b}=4200$ MeV; $\upsilon^{2}_{\bar{c}c}=0.3$, $\upsilon^{2}_{\bar{b}b}=0.1$. Vertical lines show lattice spacings available with the MILC ensembles Bernard:2001av . We show the relative error in mass splittings, which are of order $\Lambda$ in heavy-light hadrons and of order $m_{Q}\upsilon^{2}$ in quarkonium. The left set of plots uses HQET power counting, for heavy-light hadrons, while the right set of plots uses NRQCD power counting, for quarkonia. The light gray or red (dark gray or blue) curves show the estimate for hadrons containing $c$ ($b$) quarks. The dotted curves show the error when the corresponding correction term is omitted completely, i.e., the errors in the Wilson action. The dashed (solid) curves show the estimate of the error for tree-level (one- loop) matching. The vertical lines highlight $a=0.125$ fm, $0.09$ fm, $0.06$ fm and $0.045$ fm, corresponding to the ensembles of gauge fields with $n_{f}=2+1$ flavors from the MILC collaboration Bernard:2001av . To drive the each contribution to heavy-quark discretization effects below 1%, we find that one-loop matching is necessary for $c_{B}$, the coupling of the chromomagnetic clover term. Tree-level matching is sufficient for the chromoelectric clover coupling $c_{E}$, though one-loop matching would be desirable for charmonium and charmed hadrons. The lowest plots, labeled “from $1/8m_{4}^{3}$” are for the relativistic correction terms, with couplings $c_{2}$ and $z_{6}$. They also apply to $1/8m_{B^{\prime}}^{3}$ and the related chromomagnetic couplings $c_{3}$ and $z_{7}$. The one-loop mismatches of four-quark interactions are suppressed not only by a loop factor, but also by $\lambda^{2}$ or $\upsilon^{2}$, so they should fall below 1% too. Similar results for operators that break rotational symmetry are shown in Fig. 4. Figure 4: Relative truncation errors for the new action, from discretization effects that break rotational symmetry. The curves have the same meaning as in Fig. 3. To drive these contributions to heavy-quark discretization effects below 1%, we again find it sufficient to tune the couplings of the new action at the tree level. There are some other noteworthy features of Figs. 3 and 4. For $m_{Q}a\ll 1$, the discretization effects vanish as a power of $a$, as one would deduce from the Symanzik effective field theory. Because we identify $m_{2}$ with the mass in the ${\cal C}_{i}^{\mathrm{cont}}$, the powers of $a$ are balanced by $\Lambda$ or $m_{Q}\upsilon$, not $m_{Q}$. Had we identified $m_{1}$ with the physical mass, errors of order $(m_{Q}a)^{n}$ would have appeared. For $m_{Q}a\sim 1$, the tree-level curves flatten out. The error cannot grow without bound, because of the heavy-quark symmetries of the Wilson action and our improvements to it. Indeed, the curves for the $b$ quark are usually lower than those for the $c$ quark, which bodes well for calculations relevant to the Cabibbo-Kobayashi-Maskawa matrix. The underlying reason for the pattern is that the static approximation works better for $b$-flavored hadrons than for charmed hadrons. The $1/m_{b}^{n}$ contributions start out smaller, so their mismatches are also smaller. Similarly, the leading NRQCD works better for bottomonium than charmonium. The mismatches from $1/8m_{4}^{3}$ and $w_{4}/6$ deviate from the pattern, however, because NRQCD’s relative suppression $\upsilon^{2}_{\bar{b}b}/\upsilon^{2}_{\bar{c}c}$ is not as strong as HQET’s $(m_{c}/m_{b})^{3}$. Mismatches from $w_{B_{i}}/4$ and $(w_{4}+w^{\prime}_{4})/4$ are of order $\upsilon^{4}$ and again follow the pattern. In tree-level improvement, one should avoid choices where it is known that one-loop corrections from tadpole diagrams will be large Lepage:1992xa . Therefore, we envision following some sort of tadpole improvement. In the action, write each link matrix as $u_{0}[U_{\mu}/u_{0}]$ and absorb all but one pre-factor of $u_{0}$ into a tadpole-improved coupling $\tilde{c}_{i}$ and $\tilde{r}_{i}$. (In several cases, it will be necessary to expand expressions such as ${D_{i}}_{\rm lat}{\triangle_{i}}_{\rm lat}$, ${\triangle_{i}}_{\rm lat}^{2}$, and Eq. (39), to eliminate any instance of $U_{\mu}U_{\mu}^{\dagger}=1$ before inserting $u_{0}$.) Then apply the conditions of Sec. IV to $\tilde{c}_{i}$ and $\tilde{r}_{i}$ instead of $c_{i}$ and $r_{i}$, and take the $u_{0}$ factors in the denominator from the Monte Carlo simulation. ## VI Conclusions In this paper we have presented the formalism and explicit calculations needed to define a new lattice action for heavy quarks. Our aim was to obtain an action whose discretization errors would be $\lesssim 1\%$ at currently available lattice spacings. Combining our matching calculations, power counting, and the heavy-quark theory of discretization effects, we have argued that the proposed action should meet its target. Setting to zero the redundant couplings and those that vanish when matched at the tree level, our action can be written $S=S_{0}+S_{B}+S_{E}+S_{\rm new}$, where $\displaystyle S_{\rm new}$ $\displaystyle=$ $\displaystyle c_{1}a^{6}\sum_{x}\bar{\psi}(x)\sum_{i}\gamma_{i}{D_{i}}_{\mathrm{lat}}{\triangle_{i}}_{\mathrm{lat}}\psi(x)+c_{2}a^{6}\sum_{x}\bar{\psi}(x)\\{\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}},\triangle^{(3)}_{\mathrm{lat}}\\}\psi(x)$ (123) $\displaystyle+$ $\displaystyle c_{3}a^{6}\sum_{x}\bar{\psi}(x)\\{\bm{\gamma}\cdot\bm{D}_{\mathrm{lat}},i\bm{\Sigma}\cdot\bm{B}_{\mathrm{lat}}\\}\psi(x)+c_{EE}a^{6}\sum_{x}\bar{\psi}(x)\\{\gamma_{4}{D_{4}}_{\mathrm{lat}},\bm{\alpha}\cdot\bm{E}_{\mathrm{lat}}\\}\psi(x)$ $\displaystyle+$ $\displaystyle c_{4}a^{7}\sum_{x}\bar{\psi}(x)\sum_{i}{\triangle_{i}}_{\mathrm{lat}}^{2}\psi(x)+c_{5}a^{7}\sum_{x}\bar{\psi}(x)\sum_{i}\sum_{j\neq i}\\{i\Sigma_{i}{B_{i}}_{\mathrm{lat}},{\triangle_{j}}_{\mathrm{lat}}\\}\psi(x).$ The new action has six additional nonzero couplings, which depend on the couplings in $S_{0}+S_{B}+S_{E}$ according to Eqs. (113)–(116) and (119). To achieve 1% accuracy, $S_{B}$ must be, and $S_{E}$ could well be, matched at the one-loop level Nobes:2003nc . Another lattice action achieves similar accuracy for charmed quarks, namely the highly-improved staggered quark (HISQ) action Follana:2006rc . Our approach is computationally more demanding than HISQ. Its advantage, however, is the intriguing result that our discretization errors for bottom quarks are _smaller_ than for charmed quarks. That means that experience with charmed hadrons and charmonium can inform analogous calculation of properties of $b$-flavored hadrons. Finally, we note that there is tension between the most accurate calculation of the $D_{s}$ meson decay constant, $f_{D_{s}}$ Follana:2007uv , which uses HISQ, and experimental measurements Dobrescu:2008fd . Our action is a candidate for the charmed quark in a cross-check of the HISQ $f_{D_{s}}$, because its discretization errors can be expected to be small enough to strengthen or dissipate the disagreement, while possessing different systematic errors. ###### Acknowledgements. We thank Massimo Di Pierro, Aida El-Khadra, and Paul Mackenzie for helpful conversations. Colin Morningstar provided useful correspondence on unpublished details of improved anisotropic gauge actions Morningstar:1996ze ; Alford:1996up . M.B.O. was supported in part by the United States Department of Energy under Grant No. DE-FG02-91ER40677, and by Science Foundation of Ireland grants 04/BRG/P0266 and 06/RFP/PHY061. A.S.K. thanks Trinity College, Dublin, for hospitality while part of this work was being carried out. Fermilab is operated by Fermi Research Alliance, LLC, under Contract No. DE- AC02-07CH11359 with the United States Department of Energy. ## Appendix A Feynman Rules In this Appendix we present Feynman rules for the new action needed to carry out the matching calculations of Sec. IV. These are the quark and gluon propagators and three- and four-point vertices. The corresponding Feynman diagrams are shown in Fig. 5. Figure 5: Feynman rules for the action $S$ given by Eqs. (29)–(35). The quark propagator [Fig. 5(a)] is modified only through $c_{2}$, $c_{1}$, $z_{6}$, and $c_{4}$. It reads $aS^{-1}(p)=i\gamma_{4}\sin(p_{4}a)+i\bm{\gamma}\cdot\bm{K}(p)+\mu(p)-\cos(p_{4}a)$ (124) where $\displaystyle K_{i}(p)$ $\displaystyle=$ $\displaystyle\sin(p_{i}a)\left[\zeta-2c_{2}\hat{\bm{p}}^{2}a^{2}-c_{1}\hat{p}_{i}^{2}a^{2}\right]$ (125) $\displaystyle\mu(p)$ $\displaystyle=$ $\displaystyle 1+m_{0}a+\hat{\bm{p}}^{2}a^{2}\left[{\textstyle\frac{1}{2}}r_{s}\zeta+z_{6}\hat{\bm{p}}^{2}a^{2}\right]+c_{4}\sum_{i}(\hat{p}_{i}a)^{4}$ (126) The tree-level mass shell is $p_{4}=iE$, where the energy satisfies $\cosh Ea=\frac{1+\mu^{2}+\bm{K}^{2}}{2\mu(\bm{p})}.$ (127) Incoming external fermion lines receive factors $u(\xi,\bm{p})\mathcal{N}(p)$ or $v(\xi,\bm{p})\mathcal{N}(p)$, where $\displaystyle\mathcal{N}(p)$ $\displaystyle=$ $\displaystyle\left(\frac{L}{\mu(\bm{p})\sinh E}\right)^{1/2},$ (128) $\displaystyle u(\xi,\bm{p})$ $\displaystyle=$ $\displaystyle\frac{L+\sinh E-i\bm{\gamma}\cdot\bm{K}}{\sqrt{2L(L+\sinh E)}}u(\xi,\bm{0}),$ (129) $\displaystyle v(\xi,\bm{p})$ $\displaystyle=$ $\displaystyle\frac{L+\sinh E+i\bm{\gamma}\cdot\bm{K}}{\sqrt{2L(L+\sinh E)}}v(\xi,\bm{0}),$ (130) $L=\mu(\bm{p})-\cosh E$; $\gamma_{4}u(\xi,\bm{0})=u(\xi,\bm{0})$, $\gamma_{4}v(\xi,\bm{0})=-v(\xi,\bm{0})$. Outgoing external fermion lines receive factors $\mathcal{N}(p)\bar{u}(\xi,\bm{p})$ or $\mathcal{N}(p)\bar{v}(\xi,\bm{p})$, where $\bar{u}(\xi,\bm{p})=u^{\dagger}(\xi,\bm{p})\gamma_{4}$, $\bar{v}(\xi,\bm{p})=v^{\dagger}(\xi,\bm{p})\gamma_{4}$. The gluon propagator [Fig. 5(b)] is not easy to express in closed form. We refer the reader to two papers of Weisz for details Weisz:1982zw and a correction Weisz:1983bn for the propagator on isotropic lattices. The improved vertex is in Ref. Weisz:1983bn . Now let us turn to vertices with one [Fig. 5(c)–(d)] or two [Fig. 5(e)–(g)] gluons attached to a quark line. The new terms in the bilinear part of the action are all built from difference and clover operators that already appear in $S_{0}+S_{B}+S_{E}$. Consequently, the new terms in the Feynman rules for these vertices can be obtained using the chain rule. The difference operators are given in Eqs. (36)–(38). To simplify notation, let us drop the subscript “lat” in this Appendix. One-gluon vertices need $\displaystyle{D_{\rho}}^{\;a}_{,\mu}(P,k)=\frac{\partial D_{\rho}}{\partial A^{a}_{\mu}(k)}$ $\displaystyle=$ $\displaystyle g_{0}t^{a}\delta_{\rho\mu}\cos[(P+{\textstyle\frac{1}{2}}k)_{\mu}a],$ (131) $\displaystyle{\triangle_{\rho}}^{\;a}_{,\mu}(P,k)=\frac{\partial\triangle_{\rho}}{\partial A^{a}_{\mu}(k)}$ $\displaystyle=$ $\displaystyle g_{0}t^{a}\delta_{\rho\mu}(2i/a)\sin[(P+{\textstyle\frac{1}{2}}k)_{\mu}a],$ (132) $\displaystyle{F_{\rho\sigma}}^{\;a}_{,\mu}(k)=\frac{\partial{F_{\rho\sigma}}}{\partial A^{a}_{\mu}(k)}$ $\displaystyle=$ $\displaystyle g_{0}t^{a}\cos{\textstyle\frac{1}{2}}k_{\mu}a\left[\delta_{\mu\sigma}iS_{\rho}(k)-\delta_{\mu\rho}iS_{\sigma}(k)\right].$ (133) It is convenient to write out the chromomagnetic and chromoelectric cases of Eq. (133): $\displaystyle{B_{i}}^{\;a}_{,m}(k)=\frac{\partial{B_{i}}}{\partial A^{a}_{m}(k)}$ $\displaystyle=$ $\displaystyle- g_{0}t^{a}\cos({\textstyle\frac{1}{2}}k_{m}a)\varepsilon_{mri}iS_{r}(k),$ (134) $\displaystyle{E_{i}}^{\;a}_{,m}(k)=\frac{\partial{E_{i}}}{\partial A^{a}_{m}(k)}$ $\displaystyle=$ $\displaystyle g_{0}t^{a}\cos({\textstyle\frac{1}{2}}k_{m}a)\delta_{mi}iS_{4}(k),$ (135) $\displaystyle{E_{i}}^{\;a}_{,4}(k)=\frac{\partial{E_{i}}}{\partial A^{a}_{4}(k)}$ $\displaystyle=$ $\displaystyle- g_{0}t^{a}\cos({\textstyle\frac{1}{2}}k_{4}a)iS_{i}(k),$ (136) since $B_{i}={\textstyle\frac{1}{2}}\varepsilon_{ijk}F_{jk}$ and $E_{i}=F_{4i}$ appear in Eq. (29). Two-gluon vertices need $\displaystyle{D_{\rho}}^{\;ab}_{,\mu\nu}(P,k,l)=\frac{\partial^{2}D_{\rho}}{\partial A^{a}_{\mu}(k)\partial A^{b}_{\nu}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}{\textstyle\frac{1}{2}}\\{t^{a},t^{b}\\}\delta_{\mu\nu}\delta_{\rho\mu}ai\sin[(P+{\textstyle\frac{1}{2}}K)_{\mu}a],$ (137) $\displaystyle{\triangle_{\rho}}^{\;ab}_{,\mu\nu}(P,k,l)=\frac{\partial^{2}\triangle_{\rho}}{\partial A^{a}_{\mu}(k)\partial A^{b}_{\nu}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}{\textstyle\frac{1}{2}}\\{t^{a},t^{b}\\}\delta_{\mu\nu}\delta_{\rho\mu}2\cos[(P+{\textstyle\frac{1}{2}}K)_{\mu}a],$ (138) where $K=k+l$. For the clover operator it is convenient to introduce $C_{\mu\nu}(k,l)=2\cos{\textstyle\frac{1}{2}}(k+l)_{\mu}a\cos{\textstyle\frac{1}{2}}l_{\mu}a\cos{\textstyle\frac{1}{2}}(k+l)_{\nu}a\cos{\textstyle\frac{1}{2}}k_{\nu}a-\cos{\textstyle\frac{1}{2}}k_{\mu}a\cos{\textstyle\frac{1}{2}}l_{\nu}a.$ (139) Then one has ($K=k+l$) $\displaystyle{F_{\rho\sigma}}^{\;ab}_{,\mu\nu}(k,l)=\frac{\partial^{2}F_{\rho\sigma}}{\partial A^{a}_{\mu}(k)\partial A^{b}_{\nu}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}[t^{a},t^{b}]\left\\{\vphantom{\hat{K}}(\delta_{\mu\rho}\delta_{\nu\sigma}-\delta_{\mu\sigma}\delta_{\nu\rho})C_{\mu\nu}(k,l)\right.$ $\displaystyle-$ $\displaystyle\left.{\textstyle\frac{1}{4}}\delta_{\mu\nu}a^{2}\hat{K}_{\mu}\left[\delta_{\mu\rho}\left(S_{\sigma}(k)-S_{\sigma}(l)\right)-\delta_{\mu\sigma}\left(S_{\rho}(k)-S_{\rho}(l)\right)\right]\right\\},$ $\displaystyle{B_{i}}^{\;ab}_{,mn}(k,l)=\frac{\partial^{2}{B_{i}}}{\partial A^{a}_{m}(k)\partial A^{b}_{n}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}[t^{a},t^{b}]\left\\{\varepsilon_{mni}C_{mn}(k,l)-{\textstyle\frac{1}{4}}\delta_{mn}\varepsilon_{mri}a^{2}\hat{K}_{m}\left[S_{r}(k)-S_{r}(l)\right]\right\\},$ $\displaystyle{E_{i}}^{\;ab}_{,mn}(k,l)=\frac{\partial^{2}{E_{i}}}{\partial A^{a}_{m}(k)\partial A^{b}_{n}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}[t^{a},t^{b}]{\textstyle\frac{1}{4}}\delta_{mn}\delta_{mi}a^{2}\hat{K}_{m}\left[S_{4}(k)-S_{4}(l)\right],$ (142) $\displaystyle{E_{i}}^{\;ab}_{,4n}(k,l)=\frac{\partial^{2}{E_{i}}}{\partial A^{a}_{4}(k)\partial A^{b}_{n}(l)}$ $\displaystyle=$ $\displaystyle g_{0}^{2}[t^{a},t^{b}]\delta_{ni}C_{4n}(k,l),$ (143) $\displaystyle{E_{i}}^{\;ab}_{,44}(k,l)=\frac{\partial^{2}{E_{i}}}{\partial A^{a}_{4}(k)\partial A^{b}_{4}(l)}$ $\displaystyle=$ $\displaystyle- g_{0}^{2}[t^{a},t^{b}]{\textstyle\frac{1}{4}}a^{2}\hat{K}_{4}\left[S_{i}(k)-S_{i}(l)\right].$ (144) The Feynman rules for one gluon are then ${\rm Fig.~{}\ref{fig:feynman}(c,d)}=-g_{0}t^{a}_{ij}\Lambda_{\mu}(p^{\prime},p),$ (145) with $\displaystyle\Lambda_{4}(p^{\prime},p)$ $\displaystyle=$ $\displaystyle\gamma_{4}\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]-i\sin[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]+{\textstyle\frac{i}{2}}c_{E}\zeta a\bm{\alpha}\cdot\bm{S}(k)\cos({\textstyle\frac{1}{2}}k_{4}a)$ (146) $\displaystyle+$ $\displaystyle ir_{E}a^{2}\gamma_{4}\bm{\Sigma}\cdot\left\\{\bm{S}(k)\times\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\right\\}\cos({\textstyle\frac{1}{2}}k_{4}a)$ $\displaystyle-$ $\displaystyle(r_{E}-z_{E})a^{2}\gamma_{4}\bm{S}(k)\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}k_{4}a)$ $\displaystyle+$ $\displaystyle c_{EE}a^{2}\bm{\gamma}\cdot\bm{S}(k)\left[S_{4}(p^{\prime})-S_{4}(p)\right]\cos({\textstyle\frac{1}{2}}k_{4}a),$ $\displaystyle\Lambda_{m}(p^{\prime},p)$ $\displaystyle=$ $\displaystyle\zeta\gamma_{m}\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{m}a]-ir_{s}\zeta\sin[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{m}a]$ (147) $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}c_{B}\zeta a\varepsilon_{mri}\Sigma_{i}S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)-{\textstyle\frac{i}{2}}c_{E}\zeta a\alpha_{m}S_{4}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle ir_{E}a^{2}\varepsilon_{mri}\Sigma_{i}\gamma_{4}S_{4}(k)\left[S_{r}(p^{\prime})+S_{r}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(r_{E}-z_{E})a^{2}\gamma_{4}S_{4}(k)\left[S_{m}(p^{\prime})-S_{m}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle c_{2}a^{2}\left\\{\gamma_{m}\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{m}a]\left(\widehat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)+\bm{\gamma}\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\widehat{(p^{\prime}+p)}_{m}\right\\}$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}c_{1}a^{2}\gamma_{m}\left\\{\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{m}a]\left(\widehat{p_{m}^{\prime}}^{2}+\hat{p}_{m}^{2}\right)+\left[S_{m}(p^{\prime})+S_{m}(p)\right]\widehat{(p^{\prime}+p)}_{m}\right\\}$ $\displaystyle-$ $\displaystyle c_{3}a^{2}\varepsilon_{mri}\gamma_{4}\gamma_{5}S_{r}(k)\left[S_{i}(p^{\prime})+S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(c_{3}-z_{3})a^{2}\bm{\gamma}\cdot\bm{S}(k)\left[S_{m}(p^{\prime})-S_{m}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle(c_{3}-z_{3})a^{2}\gamma_{m}\bm{S}(k)\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle c_{EE}a^{2}\gamma_{m}S_{4}(k)\left[S_{4}(p^{\prime})-S_{4}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle iz_{6}a^{3}\widehat{(p^{\prime}+p)}_{m}\left(\widehat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)$ $\displaystyle-$ $\displaystyle ic_{4}a^{3}\widehat{(p^{\prime}+p)}_{m}\left(\widehat{p_{m}^{\prime}}^{2}+\hat{p}_{m}^{2}\right)$ $\displaystyle-$ $\displaystyle(z_{7}+c_{5})a^{3}\varepsilon_{mri}\Sigma_{i}S_{r}(k)\left(\hat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle c_{5}a^{3}\varepsilon_{mri}\Sigma_{i}S_{r}(k)\left(\hat{p^{\prime}}_{i}^{2}+\hat{p}_{i}^{2}\right)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle r_{5}a^{3}\varepsilon_{mri}\Sigma_{i}S_{r}(k)\left[S_{i}(p^{\prime})S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(r_{7}-z^{\prime}_{7}-r_{5})a^{3}\varepsilon_{mri}\Sigma_{i}S_{r}(k)\left[\bm{S}(p^{\prime})\cdot\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle r_{7}a^{3}\varepsilon_{mri}\left[S_{i}(p^{\prime})\bm{\Sigma}\cdot\bm{S}(p)+S_{i}(p)\bm{\Sigma}\cdot\bm{S}(p)\right]S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle i(r_{7}-r^{\prime}_{7})a^{3}\left[S_{m}(p^{\prime})\bm{S}(p)\cdot\bm{S}(k)-S_{m}(p)\bm{S}(p^{\prime})\cdot\bm{S}(k)\right]\cos({\textstyle\frac{1}{2}}k_{m}a).$ In the $r_{5}$ and $z^{\prime}_{7}$ terms, Eq. (39) has been assumed. If instead one prefers Eq. (40) then replace $\left[S_{j}(p^{\prime})S_{j}(p)\right]\to\left[\cos({\textstyle\frac{1}{2}}k_{j}a)\hat{p}^{\prime}_{j}\hat{p}_{j}\right].$ Both choices have the same effect on Eq. (61). The two-gluon rules are ${\rm Fig.~{}\ref{fig:feynman}(e,f,g)}=-{\textstyle\frac{1}{2}}g_{0}^{2}\\{t^{a},t^{b}\\}_{ij}aX_{\mu\nu}(p,k,l)-{\textstyle\frac{1}{2}}g_{0}^{2}[t^{a},t^{b}]_{ij}aY_{\mu\nu}(p,k,l),$ (148) with $\displaystyle X_{mn}(p,k,l)$ $\displaystyle=$ $\displaystyle i\zeta\delta_{mn}\gamma_{m}\sin({\textstyle\frac{1}{2}}s_{m}a)-r_{s}\zeta\delta_{mn}\cos({\textstyle\frac{1}{2}}s_{m}a)$ (149) $\displaystyle-$ $\displaystyle 2r_{E}a\varepsilon_{mni}\gamma_{4}\Sigma_{i}\left[\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)S_{4}(k)\right.$ $\displaystyle\quad-\left.\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)S_{4}(l)\right]$ $\displaystyle+$ $\displaystyle i(r_{E}-z_{E})a^{2}\gamma_{4}\delta_{mn}\sin({\textstyle\frac{1}{2}}s_{m}a)\left[S_{m}(k)S_{4}(k)+S_{m}(l)S_{4}(l)\right]$ $\displaystyle+$ $\displaystyle 4ic_{2}\gamma_{m}\left[\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\sin({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\,\left.\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right]$ $\displaystyle+$ $\displaystyle 4ic_{2}\gamma_{n}\left[\sin({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\,\left.\cos({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right]$ $\displaystyle+$ $\displaystyle 2ic_{2}a\delta_{mn}\cos({\textstyle\frac{1}{2}}s_{m}a)\,\bm{\gamma}\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]$ $\displaystyle-$ $\displaystyle ic_{2}a^{2}\delta_{mn}\gamma_{m}\sin({\textstyle\frac{1}{2}}s_{m}a)\left(\widehat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)$ $\displaystyle+$ $\displaystyle ic_{1}a\delta_{mn}\gamma_{m}\hat{s}_{m}\left[4\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)-1\right]$ $\displaystyle+$ $\displaystyle 2ic_{3}\varepsilon_{mnr}\gamma_{4}\gamma_{5}\left[\sin(l_{r}a)\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)\right.$ $\displaystyle\quad-\left.\sin(k_{r}a)\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\right]$ $\displaystyle+$ $\displaystyle 2i(c_{3}-z_{3})a\left\\{[\delta_{mn}\bm{\gamma}\cdot\bm{S}(l)-\gamma_{n}S_{m}(l)]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)\right.$ $\displaystyle\quad+\left.[\delta_{mn}\bm{\gamma}\cdot\bm{S}(k)-\gamma_{m}S_{n}(k)]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\right\\}$ $\displaystyle-$ $\displaystyle 8z_{6}\left[\sin({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\sin({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad-\left.\cos({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right]$ $\displaystyle-$ $\displaystyle 2z_{6}a^{2}\delta_{mn}\cos({\textstyle\frac{1}{2}}s_{m}a)\left(\widehat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)$ $\displaystyle-$ $\displaystyle 2c_{4}a^{2}\delta_{mn}\left\\{\cos({\textstyle\frac{1}{2}}s_{m}a)\left(\widehat{p^{\prime}}_{m}^{2}+\hat{p}_{m}^{2}\right)+\cos[{\textstyle\frac{1}{2}}(k-l)_{m}a]\hat{s}_{m}^{2}-\hat{k}_{m}\hat{l}_{m}\right\\}$ $\displaystyle+$ $\displaystyle 2i(z_{7}+c_{5})a^{2}\Sigma_{i}\left[\hat{s}_{n}\varepsilon_{mri}S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\left.\hat{s}_{m}\varepsilon_{nri}S_{r}(l)\cos({\textstyle\frac{1}{2}}l_{n}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right]$ $\displaystyle+$ $\displaystyle 2ic_{5}a^{2}\varepsilon_{mnr}\left[\hat{s}_{n}\Sigma_{n}S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad-\left.\hat{s}_{m}\Sigma_{m}S_{r}(l)\cos({\textstyle\frac{1}{2}}l_{n}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right]$ $\displaystyle+$ $\displaystyle ir_{5}a^{2}\varepsilon_{mnr}\left\\{\Sigma_{n}S_{n}(s)S_{r}(k)-\Sigma_{m}S_{m}(s)S_{r}(l)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle ir_{7}a^{2}\Sigma_{n}\varepsilon_{mri}S_{r}(k)\left\\{\left[S_{i}(p^{\prime})+S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\left.\left[S_{i}(p^{\prime})-S_{i}(p)\right]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle ir_{7}a^{2}\Sigma_{m}\varepsilon_{nri}S_{r}(l)\left\\{\left[S_{i}(p^{\prime})+S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right.$ $\displaystyle\quad+\left.\left[S_{i}(p^{\prime})-S_{i}(p)\right]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\right\\}\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle ir_{7}a^{2}\varepsilon_{mnr}S_{r}(k)\bm{\Sigma}\cdot\left\\{\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\left.\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle ir_{7}a^{2}\varepsilon_{mnr}S_{r}(l)\bm{\Sigma}\cdot\left\\{\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right.$ $\displaystyle\quad+\left.\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\right\\}\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle i(z^{\prime}_{7}+r_{5}-r_{7})a^{2}\varepsilon_{mri}S_{n}(s)\Sigma_{i}S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle i(z^{\prime}_{7}+r_{5}-r_{7})a^{2}\varepsilon_{nri}S_{m}(s)\Sigma_{i}S_{r}(l)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}S_{n}(k)\left[S_{m}(p^{\prime})-S_{m}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}S_{m}(l)\left[S_{n}(p^{\prime})-S_{n}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}S_{n}(k)\left[S_{m}(p^{\prime})+S_{m}(p)\right]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}S_{m}(l)\left[S_{n}(p^{\prime})+S_{n}(p)\right]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\delta_{mn}\bm{S}(k)\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos^{2}({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\delta_{mn}\bm{S}(l)\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos^{2}({\textstyle\frac{1}{2}}l_{m}a)$ $\displaystyle+$ $\displaystyle{\textstyle\frac{1}{4}}(r^{\prime}_{7}-r_{7})a^{4}\delta_{mn}\hat{s}_{m}\left\\{S_{m}(k)\bm{S}(k)+S_{m}(l)\bm{S}(l)\right\\}\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]$ $\displaystyle+$ $\displaystyle 2(r_{BB}-z_{BB})a^{2}\left[\delta_{mn}\bm{S}(k)\cdot\bm{S}(l)-S_{m}(l)S_{n}(k)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle 2(r_{EE}+z_{EE})a^{2}\delta_{mn}S_{4}(k)S_{4}(l)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a),$ where now $p^{\prime}=p+k+l$, and $s=p^{\prime}+p=2p+k+l$; $\displaystyle X_{44}(p,k,l)$ $\displaystyle=$ $\displaystyle i\gamma_{4}\sin[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]-\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]$ (150) $\displaystyle+$ $\displaystyle ic_{EE}a^{2}\left[\bm{\gamma}\cdot\bm{S}(k)S_{4}(k)+\bm{\gamma}\cdot\bm{S}(l)S_{4}(l)\right]\sin[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]$ $\displaystyle-$ $\displaystyle 2(r_{EE}+z_{EE})a^{2}\bm{S}(k)\cdot\bm{S}(l)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}l_{4}a),$ $\displaystyle X_{4m}(p,k,l)$ $\displaystyle=$ $\displaystyle-2r_{E}a\varepsilon_{mri}\gamma_{4}\Sigma_{i}S_{r}(k)\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ (151) $\displaystyle-$ $\displaystyle i(r_{E}-z_{E})a^{2}\gamma_{4}\hat{k}_{m}^{2}\sin({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle ic_{EE}a^{2}\gamma_{m}\hat{l}_{4}^{2}\sin[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]\cos({\textstyle\frac{1}{2}}l_{4}a)\cos({\textstyle\frac{1}{2}}l_{m}a)$ $\displaystyle+$ $\displaystyle 2(r_{EE}+z_{EE})a^{2}S_{m}(k)S_{4}(l)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}l_{m}a),$ $\displaystyle Y_{mn}(p,k,l)$ $\displaystyle=$ $\displaystyle- ic_{B}\zeta\Sigma_{i}\bar{C}_{mni}(k,l)-{\textstyle\frac{1}{4}}c_{E}\zeta a^{2}\delta_{mn}\alpha_{m}\hat{K}_{m}\left[S_{4}(k)-S_{4}(l)\right]$ (152) $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}r_{E}a^{3}\varepsilon_{mni}\gamma_{4}\Sigma_{i}\left[\hat{s}_{n}\hat{k}_{n}S_{4}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)+\hat{s}_{m}\hat{l}_{m}S_{4}(l)\cos({\textstyle\frac{1}{2}}l_{n}a)\right]$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}r_{E}a^{3}\delta_{mn}\varepsilon_{mri}\gamma_{4}\Sigma_{i}\hat{K}_{m}[S_{r}(p^{\prime})+S_{r}(p)][S_{4}(k)-S_{4}(l)]$ $\displaystyle+$ $\displaystyle 2i(r_{E}-z_{E})a\delta_{mn}\gamma_{4}\left[S_{4}(k)\cos^{2}({\textstyle\frac{1}{2}}k_{m}a)-S_{4}(l)\cos^{2}({\textstyle\frac{1}{2}}l_{m}a)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)$ $\displaystyle-$ $\displaystyle{\textstyle\frac{i}{2}}(r_{E}-z_{E})a^{3}\gamma_{4}\delta_{mn}\hat{K}_{m}[S_{m}(p^{\prime})-S_{m}(p)][S_{4}(k)-S_{4}(l)]$ $\displaystyle-$ $\displaystyle 4ic_{2}\gamma_{m}\left[\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\,\left.\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\sin({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right]$ $\displaystyle+$ $\displaystyle 4ic_{2}\gamma_{n}\left[\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\right.$ $\displaystyle\quad+\,\left.\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\sin({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right]$ $\displaystyle-$ $\displaystyle 2c_{1}\delta_{mn}i\gamma_{m}\cos[(p^{\prime}+p)_{m}a]\sin[{\textstyle\frac{1}{2}}(k-l)_{m}a]$ $\displaystyle-$ $\displaystyle 2ic_{3}a\gamma_{4}\gamma_{5}\bar{C}_{mni}(k,l)[S_{i}(p^{\prime})+S_{i}(p)]$ $\displaystyle-$ $\displaystyle{\textstyle\frac{i}{4}}c_{3}a^{3}\gamma_{4}\gamma_{5}\varepsilon_{mnr}\left[\hat{k}_{r}\hat{k}_{n}\hat{s}_{n}\cos({\textstyle\frac{1}{2}}k_{m}a)+\hat{l}_{r}\hat{l}_{m}\hat{s}_{m}\cos({\textstyle\frac{1}{2}}l_{n}a)\right]$ $\displaystyle-$ $\displaystyle 2i(c_{3}-z_{3})a\left(\vphantom{\hat{K}_{m}}C_{mn}(k,l)\left\\{\gamma_{m}[S_{n}(p^{\prime})-S_{n}(p)]-\gamma_{n}[S_{m}(p^{\prime})-S_{m}(p)]\right\\}\right.$ $\displaystyle+\left.{\textstyle\frac{1}{4}}\delta_{mn}a^{2}\hat{K}_{m}[S_{m}(p^{\prime})-S_{m}(p)]\bm{\gamma}\cdot[\bm{S}(k)-\bm{S}(l)]\right.$ $\displaystyle-\left.{\textstyle\frac{1}{4}}\delta_{mn}\gamma_{m}a^{2}\hat{K}_{m}[\bm{S}(p^{\prime})-\bm{S}(p)]\cdot[\bm{S}(k)-\bm{S}(l)]\right.$ $\displaystyle+\left.[\delta_{mn}\bm{\gamma}\cdot\bm{S}(l)-\gamma_{n}S_{m}(l)]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)\right.$ $\displaystyle-\left.[\delta_{mn}\bm{\gamma}\cdot\bm{S}(k)-\gamma_{m}S_{n}(k)]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\vphantom{\hat{K}_{m}}\right)$ $\displaystyle+$ $\displaystyle{\textstyle\frac{i}{2}}c_{EE}a^{3}\gamma_{m}\delta_{mn}\hat{K}_{m}\left[S_{4}(p^{\prime})-S_{4}(p)\right]\left[S_{4}(k)-S_{4}(l)\right]$ $\displaystyle-$ $\displaystyle 2z_{6}a^{2}\left[\cos({\textstyle\frac{1}{2}}s_{m}a)\hat{l}_{m}\hat{s}_{n}\cos({\textstyle\frac{1}{2}}k_{n}a)-\cos({\textstyle\frac{1}{2}}s_{n}a)\hat{k}_{n}\hat{s}_{m}\cos({\textstyle\frac{1}{2}}l_{m}a)\right]$ $\displaystyle+$ $\displaystyle 4c_{4}\delta_{mn}\sin[(p^{\prime}+p)_{m}a]\sin[{\textstyle\frac{1}{2}}(k-l)_{m}a]$ $\displaystyle-$ $\displaystyle 2i(z_{7}+c_{5})a^{2}\Sigma_{i}\left[\varepsilon_{mri}S_{r}(k)\hat{k}_{n}\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\right.$ $\displaystyle\quad-\left.\varepsilon_{nri}S_{r}(l)\hat{l}_{m}\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)+\bar{C}_{mni}(k,l)\left(\widehat{\bm{p}^{\prime}}^{2}+\hat{\bm{p}}^{2}\right)\right]$ $\displaystyle-$ $\displaystyle 2ic_{5}a^{2}\varepsilon_{mnr}\left[\Sigma_{n}S_{r}(k)\hat{k}_{n}\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\right.$ $\displaystyle\quad+\left.\Sigma_{m}S_{r}(l)\hat{l}_{m}\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)\right]$ $\displaystyle+$ $\displaystyle 2ic_{5}a^{2}\Sigma_{i}\left(\widehat{p^{\prime}}_{i}^{2}+\hat{p}_{i}^{2}\right)\bar{C}_{mni}(k,l)$ $\displaystyle+$ $\displaystyle ir_{5}a^{2}\varepsilon_{mnr}\Sigma_{n}\left[\hat{K}_{n}\cos({\textstyle\frac{1}{2}}l_{n}a)-\hat{l}_{n}\sin^{2}({\textstyle\frac{1}{2}}s_{n}a)\right]S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle ir_{5}a^{2}\varepsilon_{mnr}\Sigma_{m}\left[\hat{K}_{m}\cos({\textstyle\frac{1}{2}}k_{m}a)-\hat{k}_{m}\sin^{2}({\textstyle\frac{1}{2}}s_{m}a)\right]S_{r}(l)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle 2ir_{5}a^{2}\Sigma_{i}S_{i}(p^{\prime})S_{i}(p)\bar{C}_{mni}(k,l)$ $\displaystyle+$ $\displaystyle ir_{7}a^{2}\Sigma_{n}\varepsilon_{mri}S_{r}(k)\left\\{\left[S_{i}(p^{\prime})-S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\right.$ $\displaystyle\quad+\left.\left[S_{i}(p^{\prime})+S_{i}(p)\right]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle ir_{7}a^{2}\Sigma_{m}\varepsilon_{nri}S_{r}(l)\left\\{\left[S_{i}(p^{\prime})-S_{i}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\right.$ $\displaystyle\quad+\left.\left[S_{i}(p^{\prime})+S_{i}(p)\right]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\right\\}\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle ir_{7}a^{2}\varepsilon_{mnr}S_{r}(k)\left\\{\bm{\Sigma}\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}k_{m}a)\right.$ $\displaystyle\quad+\left.\bm{\Sigma}\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}k_{m}a)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle ir_{7}a^{2}\varepsilon_{mnr}S_{r}(l)\left\\{\bm{\Sigma}\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}l_{n}a)\right.$ $\displaystyle\quad+\left.\bm{\Sigma}\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}l_{n}a)\right\\}\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle 2ir_{7}a^{2}\bm{\Sigma}\cdot\left[\bm{S}(p^{\prime})S_{i}(p)+\bm{S}(p)S_{i}(p^{\prime})\right]\bar{C}_{mni}(k,l)$ $\displaystyle+$ $\displaystyle i(z^{\prime}_{7}+r_{5}-r_{7})a^{2}\varepsilon_{mri}S_{r}(k)\Sigma_{i}\left\\{\hat{K}_{n}\cos({\textstyle\frac{1}{2}}l_{n}a)-\hat{l}_{n}\sin^{2}({\textstyle\frac{1}{2}}s_{n}a)\right\\}\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle-$ $\displaystyle i(z^{\prime}_{7}+r_{5}-r_{7})a^{2}\varepsilon_{nri}S_{r}(l)\Sigma_{i}\left\\{\hat{K}_{m}\cos({\textstyle\frac{1}{2}}k_{m}a)-\hat{k}_{m}\sin^{2}({\textstyle\frac{1}{2}}s_{m}a)\right\\}\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle 2i(z^{\prime}_{7}+r_{5}-r_{7})a^{2}\Sigma_{i}\bm{S}(p^{\prime})\cdot\bm{S}(p)\bar{C}_{mni}(k,l)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\left[S_{m}(p^{\prime})+S_{m}(p)\right]S_{n}(k)\cos({\textstyle\frac{1}{2}}s_{n}a)\cos({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\left[S_{n}(p^{\prime})+S_{n}(p)\right]S_{m}(l)\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\left[S_{m}(p^{\prime})-S_{m}(p)\right]S_{n}(k)\sin({\textstyle\frac{1}{2}}s_{n}a)\sin({\textstyle\frac{1}{2}}k_{n}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\left[S_{n}(p^{\prime})-S_{n}(p)\right]S_{m}(l)\sin({\textstyle\frac{1}{2}}s_{m}a)\sin({\textstyle\frac{1}{2}}l_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\delta_{mn}\bm{S}(k)\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{n}a)\cos^{2}({\textstyle\frac{1}{2}}k_{n}a)$ $\displaystyle-$ $\displaystyle(r^{\prime}_{7}-r_{7})a^{2}\delta_{mn}\bm{S}(l)\cdot\left[\bm{S}(p^{\prime})+\bm{S}(p)\right]\cos({\textstyle\frac{1}{2}}s_{m}a)\cos^{2}({\textstyle\frac{1}{2}}l_{m}a)$ $\displaystyle+$ $\displaystyle{\textstyle\frac{1}{4}}(r^{\prime}_{7}-r_{7})a^{4}\delta_{mn}\hat{s}_{m}\left\\{S_{m}(k)\bm{S}(k)-S_{m}(l)\bm{S}(l)\right\\}\cdot\left[\bm{S}(p^{\prime})-\bm{S}(p)\right]$ $\displaystyle+$ $\displaystyle 2(r^{\prime}_{7}-r_{7})a^{2}\left[S_{m}(p^{\prime})S_{n}(p)-S_{m}(p)S_{n}(p^{\prime})\right]C_{mn}(k,l)$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}(r^{\prime}_{7}-r_{7})a^{4}\delta_{mn}\hat{K}_{m}\left[S_{m}(p^{\prime})\bm{S}(p)-S_{m}(p)\bm{S}(p^{\prime})\right]\cdot\left[\bm{S}(k)-\bm{S}(l)\right]$ $\displaystyle+$ $\displaystyle ir_{BB}a^{2}\varepsilon_{mnr}\left[S_{r}(k)\bm{\Sigma}\cdot\bm{S}(l)+S_{r}(l)\bm{\Sigma}\cdot\bm{S}(k)\right]\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle+$ $\displaystyle ir_{BB}a^{2}\left(\Sigma_{m}\varepsilon_{nri}+\Sigma_{n}\varepsilon_{mri}\right)S_{r}(k)S_{i}(l)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a)$ $\displaystyle-$ $\displaystyle 2ir_{EE}a^{2}\varepsilon_{mni}\Sigma_{i}S_{4}(k)S_{4}(l)\cos({\textstyle\frac{1}{2}}k_{m}a)\cos({\textstyle\frac{1}{2}}l_{n}a),$ where $\bar{C}_{mni}(k,l)=\varepsilon_{mni}C_{mn}(k,l)-{\textstyle\frac{1}{4}}\delta_{mn}\varepsilon_{mri}a^{2}\hat{K}_{m}[S_{r}(k)-S_{r}(l)]$; $\displaystyle Y_{44}(p,k,l)$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{2}}c_{E}\zeta a\bm{\alpha}\cdot[\bm{S}(k)-\bm{S}(l)]\sin[{\textstyle\frac{1}{2}}(k+l)_{4}a]$ (153) $\displaystyle-$ $\displaystyle r_{E}a^{2}\gamma_{4}\bm{\Sigma}\cdot\\{[\bm{S}(p^{\prime})+\bm{S}(p)]\times[\bm{S}(k)-\bm{S}(l)]\\}\sin[{\textstyle\frac{1}{2}}(k+l)_{4}a]$ $\displaystyle+$ $\displaystyle i(r_{E}-z_{E})a^{2}\gamma_{4}[\bm{S}(p^{\prime})-\bm{S}(p)]\cdot[\bm{S}(k)-\bm{S}(l)]\sin[{\textstyle\frac{1}{2}}(k+l)_{4}a]$ $\displaystyle+$ $\displaystyle 2ic_{EE}a\left[\bm{\gamma}\cdot\bm{S}(k)\cos^{2}({\textstyle\frac{1}{2}}k_{4}a)-\bm{\gamma}\cdot\bm{S}(l)\cos^{2}({\textstyle\frac{1}{2}}l_{4}a)\right]\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]$ $\displaystyle-$ $\displaystyle 2ic_{EE}a\bm{\gamma}\cdot\left[\bm{S}(k)-\bm{S}(l)\right]\sin^{2}[{\textstyle\frac{1}{2}}(k+l)_{4}a]\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]$ $\displaystyle-$ $\displaystyle 2ir_{EE}a^{2}\bm{\Sigma}\cdot\left[\bm{S}(k)\times\bm{S}(l)\right]\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}l_{4}a),$ $\displaystyle Y_{4m}(p,k,l)$ $\displaystyle=$ $\displaystyle- c_{E}\zeta\alpha_{m}C_{4m}(k,l)$ (154) $\displaystyle-$ $\displaystyle 2r_{E}a\varepsilon_{mri}\gamma_{4}\Sigma_{i}[S_{r}(p^{\prime})+S_{r}(p)]C_{4m}(k,l)$ $\displaystyle-$ $\displaystyle r_{E}a^{2}\varepsilon_{mri}\gamma_{4}\Sigma_{i}\sin({\textstyle\frac{1}{2}}s_{m}a)\hat{k}_{m}S_{r}(k)\cos({\textstyle\frac{1}{2}}k_{4}a)$ $\displaystyle-$ $\displaystyle 2i(r_{E}-z_{E})a\gamma_{4}[S_{m}(p^{\prime})-S_{m}(p)]C_{4m}(k,l)$ $\displaystyle-$ $\displaystyle 2i(r_{E}-z_{E})a\gamma_{4}S_{m}(k)\cos({\textstyle\frac{1}{2}}s_{m}a)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}k_{m}a)$ $\displaystyle+$ $\displaystyle 2ic_{EE}a\gamma_{m}[S_{4}(p^{\prime})-S_{4}(p)]C_{4m}(k,l)$ $\displaystyle+$ $\displaystyle 2ic_{EE}a\gamma_{m}S_{4}(l)\cos[{\textstyle\frac{1}{2}}(p^{\prime}+p)_{4}a]\cos({\textstyle\frac{1}{2}}l_{4}a)\cos({\textstyle\frac{1}{2}}l_{m}a)$ $\displaystyle-$ $\displaystyle 2ir_{EE}a^{2}\varepsilon_{mri}\Sigma_{i}S_{r}(k)S_{4}(l)\cos({\textstyle\frac{1}{2}}k_{4}a)\cos({\textstyle\frac{1}{2}}l_{m}a).$ ## Appendix B Details of Compton Amplitudes The parts of the Compton scattering amplitude not exhibited in Sec. IV.4 are shown here. First the color-symmetric contributions: $\displaystyle\mathcal{M}_{mn}^{(1,0)}$ $\displaystyle=$ $\displaystyle\frac{\delta_{mn}}{m_{2}},$ (155) $\displaystyle\mathcal{M}_{mn}^{(2,-1)}$ $\displaystyle=$ $\displaystyle\frac{P_{m}(R+{\textstyle\frac{1}{2}}K)_{n}+P_{n}(R-{\textstyle\frac{1}{2}}K)_{m}}{m_{2}^{2}}$ (156) $\displaystyle+$ $\displaystyle\frac{[(R-{\textstyle\frac{1}{2}}K)_{m}\varepsilon_{nri}(R-{\textstyle\frac{1}{2}}K)_{r}-(R+{\textstyle\frac{1}{2}}K)_{n}\varepsilon_{mri}(R+{\textstyle\frac{1}{2}}K)_{r}]i\Sigma_{i}}{2m_{2}m_{B}}$ $\displaystyle+$ $\displaystyle\frac{2(i\Sigma_{m}\varepsilon_{nrs}+i\Sigma_{n}\varepsilon_{mrs})R_{r}K_{s}-4\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{R}+\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{K}}{8m_{B}^{2}},$ $\displaystyle\mathcal{M}_{mn}^{(2,1)}$ $\displaystyle=$ $\displaystyle\frac{\varepsilon_{mni}i\Sigma_{i}}{2m_{E}^{2}},$ (157) $\displaystyle\mathcal{M}_{mn}^{(3,-2)}$ $\displaystyle=$ $\displaystyle\left[4P_{m}P_{n}+(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}\right]\frac{4\bm{R}^{2}-\bm{K}^{2}}{16m_{2}^{3}}$ (158) $\displaystyle+$ $\displaystyle[P_{m}(R+{\textstyle\frac{1}{2}}K)_{n}+P_{n}(R-{\textstyle\frac{1}{2}}K)_{m}]\frac{\bm{P}\cdot\bm{R}}{m_{2}^{3}}$ $\displaystyle-$ $\displaystyle\left[P_{n}\varepsilon_{mri}(R_{r}+{\textstyle\frac{1}{2}}K_{r})-P_{m}\varepsilon_{nri}(R_{r}-{\textstyle\frac{1}{2}}K_{r})\right]i\Sigma_{i}\frac{4\bm{R}^{2}-\bm{K}^{2}}{8m_{2}^{2}m_{B}}$ $\displaystyle-$ $\displaystyle\left[\varepsilon_{mri}(R+{\textstyle\frac{1}{2}}K)_{r}(R+{\textstyle\frac{1}{2}}K)_{n}-\varepsilon_{nri}(R-{\textstyle\frac{1}{2}}K)_{r}(R-{\textstyle\frac{1}{2}}K)_{m}\right]i\Sigma_{i}\frac{\bm{P}\cdot\bm{R}}{2m_{2}^{2}m_{B}}$ $\displaystyle-$ $\displaystyle\left[(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}+{\textstyle\frac{1}{2}}(i\Sigma_{n}\varepsilon_{mrs}-i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s}\right]\frac{4\bm{R}^{2}-\bm{K}^{2}}{16m_{2}m_{B}^{2}}$ $\displaystyle+$ $\displaystyle\delta_{mn}\frac{(4\bm{R}^{2}-\bm{K}^{2})^{2}}{64m_{2}m_{B}^{2}}+(i\Sigma_{n}\varepsilon_{mrs}+i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s}\frac{\bm{P}\cdot\bm{R}}{4m_{2}m_{B}^{2}}$ $\displaystyle+$ $\displaystyle\left(\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{R}-\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{K}\right)\frac{4\bm{R}^{2}-\bm{K}^{2}}{32m_{2}m_{B}^{2}}$ $\displaystyle-$ $\displaystyle\left(4\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{R}-\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{K}\right)\frac{\bm{P}\cdot\bm{R}}{8m_{2}m_{B}^{2}},$ $\displaystyle\left.\mathcal{M}_{mn}^{(3,0)}\right|_{\rm match}$ $\displaystyle=$ $\displaystyle-\frac{\delta_{mn}\bm{P}^{2}+2P_{m}P_{n}}{2m_{4}^{3}}-\left(\frac{1}{m_{4}^{3}}+\frac{1}{m_{B}m_{E}^{2}}\right)\delta_{mn}\frac{4\bm{R}^{2}+\bm{K}^{2}}{16}$ (159) $\displaystyle-$ $\displaystyle\left[\frac{1}{4m_{2}m_{E}^{2}}-\frac{1}{4m_{B}m_{E}^{2}}-\frac{2z_{E}a^{2}}{e^{m_{1}a}m_{2}}\right]\left(R_{m}R_{n}+{\textstyle\frac{1}{4}}K_{m}K_{n}\right)$ $\displaystyle+$ $\displaystyle a^{3}\left(\frac{(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{16e^{2m_{1}a}}+a^{3}{\textstyle\frac{1}{16}}w_{B_{2}}\right)\delta_{mn}(4\bm{R}^{2}-\bm{K}^{2})$ $\displaystyle-$ $\displaystyle a^{3}\left(\frac{(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{16e^{2m_{1}a}}+a^{3}{\textstyle\frac{1}{16}}w_{B_{2}}\right)(4R_{m}R_{n}-K_{m}K_{n})$ $\displaystyle+$ $\displaystyle a^{3}{\textstyle\frac{1}{8}}w_{B_{1}}(\delta_{mn}\bm{K}^{2}-K_{m}K_{n})-a^{3}w_{4}\delta_{mn}\left(2P_{m}^{2}+{\textstyle\frac{1}{3}}R_{m}^{2}+{\textstyle\frac{1}{12}}K_{m}^{2}\right)$ $\displaystyle+$ $\displaystyle\left[\frac{1}{2m_{B^{\prime}}^{3}}+\frac{1}{2m_{2}m_{E}^{2}}+a^{3}{\textstyle\frac{1}{2}}(w_{4}+w^{\prime}_{4})\right]\varepsilon_{mni}i\Sigma_{i}\bm{P}\cdot\bm{R}$ $\displaystyle-$ $\displaystyle\left(\frac{1}{4m_{2}m_{E}^{2}}-\frac{1}{4m_{B}m_{E}^{2}}+{\textstyle\frac{1}{4}}a^{3}w_{B_{3}}\right)\varepsilon_{mnr}P_{r}i\bm{\Sigma}\cdot\bm{R}$ $\displaystyle-$ $\displaystyle\left[\frac{1}{2m_{B^{\prime}}^{3}}+\frac{1}{4m_{2}m_{E}^{2}}-\frac{1}{4m_{B}m_{E}^{2}}+a^{3}{\textstyle\frac{1}{2}}(w_{4}+w^{\prime}_{4})-a^{3}{\textstyle\frac{3}{4}}w_{B_{3}}\right]\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{P}$ $\displaystyle+$ $\displaystyle a^{3}{\textstyle\frac{1}{2}}(w_{4}+w^{\prime}_{4})\varepsilon_{mnr}R_{r}(P_{m}i\Sigma_{m}+P_{n}i\Sigma_{n})$ $\displaystyle-$ $\displaystyle a^{3}\left[\frac{(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{8e^{2m_{1}a}}+a^{3}{\textstyle\frac{1}{8}}(w_{B_{2}}-w_{B_{1}})\right](R_{m}K_{n}-R_{n}K_{m})$ $\displaystyle+$ $\displaystyle\frac{1}{8m_{2}m_{E}^{2}}(K_{n}\varepsilon_{mri}+K_{m}\varepsilon_{nri})P_{r}i\Sigma_{i}$ $\displaystyle-$ $\displaystyle\left(\frac{1}{8m_{B}m_{E}^{2}}-a^{3}{\textstyle\frac{1}{8}}w_{B_{3}}\right)(i\Sigma_{n}\varepsilon_{mrs}+i\Sigma_{m}\varepsilon_{nrs})P_{r}K_{s}$ $\displaystyle+$ $\displaystyle\left[\frac{1}{4m_{B^{\prime}}^{3}}-\frac{1}{8m_{2}m_{E}^{2}}+{\textstyle\frac{1}{4}}a^{3}(w_{4}+w^{\prime}_{4})\right](P_{n}\varepsilon_{mri}+P_{m}\varepsilon_{nri})K_{r}i\Sigma_{i}$ $\displaystyle-$ $\displaystyle a^{3}{\textstyle\frac{1}{4}}(w_{4}+w^{\prime}_{4})\varepsilon_{mnr}K_{r}(P_{m}i\Sigma_{m}-P_{n}i\Sigma_{n}).$ The color-antisymmetric contributions from Fig. 2(a)-(c): $\displaystyle\mathcal{N}_{mn}^{(1,0)}$ $\displaystyle=$ $\displaystyle\frac{\varepsilon_{mni}i\Sigma_{i}}{m_{B}},$ (160) $\displaystyle\mathcal{N}_{mn}^{(2,-1)}$ $\displaystyle=$ $\displaystyle-\frac{4P_{m}P_{n}+(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}}{2m_{2}^{2}}$ (161) $\displaystyle-$ $\displaystyle\frac{[P_{m}\varepsilon_{nri}(R-{\textstyle\frac{1}{2}}K)_{r}-P_{n}\varepsilon_{mri}(R+{\textstyle\frac{1}{2}}K)_{r}]i\Sigma_{i}}{m_{2}m_{B}}$ $\displaystyle-$ $\displaystyle\frac{(i\Sigma_{m}\varepsilon_{nrs}-i\Sigma_{n}\varepsilon_{mrs})R_{r}K_{s}+\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{R}-\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{K}}{4m_{B}^{2}}$ $\displaystyle-$ $\displaystyle\frac{\delta_{mn}(\bm{R}^{2}-{\textstyle\frac{1}{4}}\bm{K}^{2})-(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}}{2m_{B}^{2}},$ $\displaystyle\mathcal{N}_{mn}^{(2,1)}$ $\displaystyle=$ $\displaystyle\frac{\delta_{mn}}{2m_{E}^{2}}-\frac{4a^{2}z_{E}\delta_{mn}}{1+m_{0}a},$ (162) $\displaystyle\mathcal{N}_{mn}^{(3,-2)}$ $\displaystyle=$ $\displaystyle-\left[4P_{m}P_{n}+(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}\right]\frac{\bm{P}\cdot\bm{R}}{2m_{2}^{3}}$ (163) $\displaystyle-$ $\displaystyle[P_{m}(R+{\textstyle\frac{1}{2}}K)_{n}+P_{n}(R-{\textstyle\frac{1}{2}}K)_{m}]\frac{4\bm{R}^{2}-\bm{K}^{2}}{8m_{2}^{3}}$ $\displaystyle+$ $\displaystyle\left[P_{n}\varepsilon_{mri}(R_{r}+{\textstyle\frac{1}{2}}K_{r})-P_{m}\varepsilon_{nri}(R_{r}-{\textstyle\frac{1}{2}}K_{r})\right]i\Sigma_{i}\frac{\bm{P}\cdot\bm{R}}{m_{2}^{2}m_{B}}$ $\displaystyle+$ $\displaystyle\left[\varepsilon_{mri}(R+{\textstyle\frac{1}{2}}K)_{r}(R+{\textstyle\frac{1}{2}}K)_{n}-\varepsilon_{nri}(R-{\textstyle\frac{1}{2}}K)_{r}(R-{\textstyle\frac{1}{2}}K)_{m}\right]i\Sigma_{i}\frac{4\bm{R}^{2}-\bm{K}^{2}}{16m_{2}^{2}m_{B}}$ $\displaystyle+$ $\displaystyle\left[(R-{\textstyle\frac{1}{2}}K)_{m}(R+{\textstyle\frac{1}{2}}K)_{n}+{\textstyle\frac{1}{2}}(i\Sigma_{n}\varepsilon_{mrs}-i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s}\right]\frac{\bm{P}\cdot\bm{R}}{2m_{2}m_{B}^{2}}$ $\displaystyle-$ $\displaystyle\delta_{mn}(4\bm{R}^{2}-\bm{K}^{2})\frac{\bm{P}\cdot\bm{R}}{8m_{2}m_{B}^{2}}-(i\Sigma_{n}\varepsilon_{mrs}+i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s}\frac{4\bm{R}^{2}-\bm{K}^{2}}{32m_{2}m_{B}^{2}}$ $\displaystyle-$ $\displaystyle\left(\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{R}-\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{K}\right)\frac{\bm{P}\cdot\bm{R}}{4m_{2}m_{B}^{2}}$ $\displaystyle+$ $\displaystyle\left(4\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{R}-\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{K}\right)\frac{4\bm{R}^{2}-\bm{K}^{2}}{64m_{2}m_{B}^{2}},$ $\displaystyle\left.\mathcal{N}_{mn}^{(3,0)}\right|_{\rm match}$ $\displaystyle=$ $\displaystyle-\left(\frac{1}{2m_{B^{\prime}}^{3}}+\frac{1}{2m_{2}m_{E}^{2}}\right)\varepsilon_{mni}i\Sigma_{i}\bm{P}^{2}$ (164) $\displaystyle+$ $\displaystyle\left(\frac{1}{2m_{2}m_{E}^{2}}+a^{3}{\textstyle\frac{1}{2}}w_{B_{3}}\right)\varepsilon_{mnr}P_{r}i\bm{\Sigma}\cdot\bm{P}$ $\displaystyle-$ $\displaystyle a^{3}{\textstyle\frac{1}{2}}(w_{4}+w^{\prime}_{4})\varepsilon_{mni}i\Sigma_{i}(P_{m}^{2}+P_{n}^{2})$ $\displaystyle-$ $\displaystyle\left[\frac{1}{4m_{B^{\prime}}^{3}}+\frac{1}{8m_{2}m_{E}^{2}}+\frac{1}{8m_{B}m_{E}^{2}}-\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt+\left.\vphantom{\frac{1}{8}}a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+w^{\prime}_{B})-a^{3}{\textstyle\frac{1}{8}}w_{B_{2}}\right]\varepsilon_{mni}i\Sigma_{i}\bm{R}^{2}$ $\displaystyle+$ $\displaystyle\left[\frac{1}{4m_{B^{\prime}}^{3}}-\frac{1}{4m_{4}^{3}}+\frac{1}{8m_{2}m_{E}^{2}}-\frac{1}{8m_{B}m_{E}^{2}}-\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt+\left.a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+w^{\prime}_{B})+a^{3}{\textstyle\frac{1}{8}}w_{B_{2}}+\frac{a^{3}(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{4e^{2m_{1}a}}\right]\varepsilon_{mnr}R_{r}i\bm{\Sigma}\cdot\bm{R}$ $\displaystyle-$ $\displaystyle a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+w^{\prime}_{B})\varepsilon_{mnr}R_{r}(i\Sigma_{m}R_{m}+i\Sigma_{n}R_{n})$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{4}}\left[\frac{3}{4m_{B^{\prime}}^{3}}-\frac{1}{8m_{2}m_{E}^{2}}-\frac{1}{8m_{B}m_{E}^{2}}+\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt+\left.\vphantom{\frac{1}{8}}a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+7w^{\prime}_{B})-a^{3}{\textstyle\frac{7}{8}}w_{B_{2}}\right]\varepsilon_{mni}i\Sigma_{i}\bm{K}^{2}$ $\displaystyle+$ $\displaystyle{\textstyle\frac{1}{4}}\left[\frac{1}{4m_{4}^{3}}+\frac{1}{4m_{B^{\prime}}^{3}}-\frac{1}{8m_{2}m_{E}^{2}}-\frac{3}{8m_{B}m_{E}^{2}}+\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt+\left.a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+7w^{\prime}_{B})-a^{3}{\textstyle\frac{5}{8}}w_{B_{2}}-\frac{a^{3}(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{4e^{2m_{1}a}}\right]\varepsilon_{mnr}K_{r}i\bm{\Sigma}\cdot\bm{K}$ $\displaystyle-$ $\displaystyle a^{3}{\textstyle\frac{1}{24}}(w_{4}+w^{\prime}_{4}+7w^{\prime}_{B})\varepsilon_{mnr}K_{r}(i\Sigma_{m}K_{m}+i\Sigma_{n}K_{n})$ $\displaystyle+$ $\displaystyle\left(\frac{1}{2m_{B}m_{E}^{2}}+a^{3}{\textstyle\frac{1}{2}}w_{B_{1}}\right)\delta_{mn}\bm{P}\cdot\bm{R}+a^{3}{\textstyle\frac{4}{3}}w_{4}\delta_{mn}P_{m}R_{m}$ $\displaystyle+$ $\displaystyle\left[\frac{1}{2m_{4}^{3}}+\frac{1}{4m_{2}m_{E}^{2}}-\frac{1}{4m_{B}m_{E}^{2}}-\frac{2a^{2}z_{E}}{m_{2}e^{m_{1}a}}-a^{3}{\textstyle\frac{1}{4}}w_{B_{1}}\right](P_{m}R_{n}+P_{n}R_{m})$ $\displaystyle+$ $\displaystyle a^{3}{\textstyle\frac{1}{2}}w^{\prime}_{B}\delta_{mn}(R_{m}K_{r}-K_{m}R_{r})\varepsilon_{mri}i\Sigma_{i}$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}\left[\frac{1}{4m_{B^{\prime}}^{3}}-\frac{1}{8m_{2}m_{E}^{2}}+\frac{1}{8m_{B}m_{E}^{2}}-\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt+\left.\vphantom{\frac{1}{8}}a^{3}{\textstyle\frac{1}{6}}(w_{4}+w^{\prime}_{4}+4w^{\prime}_{B})-a^{3}{\textstyle\frac{1}{8}}w_{B_{2}}\right](R_{n}\varepsilon_{mri}+R_{m}\varepsilon_{nri})K_{r}i\Sigma_{i}$ $\displaystyle-$ $\displaystyle{\textstyle\frac{1}{2}}\left[\frac{1}{4m_{B^{\prime}}^{3}}+\frac{1}{8m_{2}m_{E}^{2}}-\frac{1}{8m_{B}m_{E}^{2}}+\frac{a^{2}z_{E}}{m_{B}e^{m_{1}a}}\right.$ $\displaystyle\hskip 20.00003pt-\left.\vphantom{\frac{1}{8}}a^{3}{\textstyle\frac{3}{8}}w_{B_{2}}\right](K_{n}\varepsilon_{mri}+K_{m}\varepsilon_{nri})R_{r}i\Sigma_{i}$ $\displaystyle+$ $\displaystyle{\textstyle\frac{1}{2}}\left[\frac{1}{4m_{4}^{3}}-a^{3}{\textstyle\frac{1}{4}}(w_{B_{2}}+w_{B_{3}})-\frac{a^{3}(r_{s}^{2}-c_{B}^{2})\zeta^{2}}{4e^{2m_{1}a}}\right](i\Sigma_{n}\varepsilon_{mrs}+i\Sigma_{m}\varepsilon_{nrs})R_{r}K_{s}$ $\displaystyle+$ $\displaystyle a^{3}{\textstyle\frac{1}{12}}(w_{4}+w^{\prime}_{4}+4w^{\prime}_{B})\varepsilon_{mnr}K_{r}(i\Sigma_{m}R_{m}-i\Sigma_{n}R_{n})$ $\displaystyle+$ $\displaystyle\left[\frac{1}{4m_{4}^{3}}-\frac{1}{8m_{2}m_{E}^{2}}-\frac{1}{8m_{B}m_{E}^{2}}+\frac{a^{2}z_{E}}{m_{2}e^{m_{1}a}}-a^{3}{\textstyle\frac{3}{8}}w_{B_{1}}\right](P_{m}K_{n}-P_{n}K_{m})$ $\displaystyle-$ $\displaystyle a^{2}{\textstyle\frac{1}{3}}\delta_{mn}\left(\frac{P_{m}R_{m}}{m_{2}}-\frac{R_{m}\varepsilon_{mri}K_{r}i\Sigma_{i}}{2m_{B}}\right).$ The terms on the last line do not match, but we still must add to Eqs. (160)–(164) the contribution of the diagram with the three-gluon vertex [Fig. 2(d)], which is $\displaystyle\mathcal{N}_{\mu\nu}^{\ref{fig:compton}(d)}$ $\displaystyle=$ $\displaystyle-2iK^{-2}\left[2\delta_{\mu\nu}R\cdot J-(k^{\prime}-K)_{\mu}J_{\nu}-(k+K)_{\nu}J_{\mu}\right]+ia^{2}{\textstyle\frac{1}{3}}\delta_{\mu\nu}R_{\mu}J_{\mu}$ (165) $\displaystyle+$ $\displaystyle{\textstyle\frac{i}{6}}a^{2}K^{-2}\left[k_{\mu}k_{\nu}(k^{\prime}-K)_{\nu}J_{\nu}+k^{\prime}_{\nu}k^{\prime}_{\mu}(K+k)_{\mu}J_{\mu}\right]$ and no $\mathcal{M}_{\mu\nu}$ contribution. Here $J_{\mu}$ is the current of Sec. IV.2. The first lattice artifact cancels the last line of Eq. (164). The second lattice artifact vanishes upon contraction with the external-gluon polarization vectors. ## Appendix C Improved Gauge Action In this Appendix we outline how to improve the gauge action, when axis- interchange symmetry is given up. The improvement program is the same as for anisotropic lattices, which has been worked out Alford:1996up and summarized Morningstar:1996ze . Since it has not been published, we give the main details here. Table 6 lists the interactions in the Symanzik LE${\cal L}$, with and without axis-interchange symmetry. Table 6: Dimension-6 gauge-field interactions that could appear in the LE${\cal L}$. With axis-interchange | Without axis-interchange ---|--- $\sum_{\mu}\mathop{\mathrm{tr}}[(D_{\mu}F_{\mu\nu})(D_{\mu}F_{\mu\nu})]$ | | $\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(D_{4}\bm{E})]$ | | | $\sum_{i}\mathop{\mathrm{tr}}[(D_{i}E_{i})(D_{i}E_{i})]$ | | | $\sum_{j\neq k}\mathop{\mathrm{tr}}[(D_{j}B_{k})(D_{j}B_{k})]$ | $\mathop{\mathrm{tr}}[F_{\mu\nu}F_{\nu\rho}F_{\rho\mu}]$ | | $\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{E}\times\bm{E})]$ | | | $\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{B}\times\bm{B})]$ | $\mathop{\mathrm{tr}}[(D_{\mu}F_{\mu\nu})(D_{\rho}F_{\rho\nu})]$ | $\varepsilon_{A}$ | $\mathop{\mathrm{tr}}[(\bm{D}\cdot\bm{E})(\bm{D}\cdot\bm{E})]$ | $\varepsilon_{A}$ | | $\mathop{\mathrm{tr}}[(\bm{D}\times\bm{B})\cdot(\bm{D}\times\bm{B})]$ | $\delta_{A}$ | | $\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(\bm{D}\times\bm{B})]$ | $\delta_{E}$ Without axis-interchange symmetry there are eight operators. Other operators can be written as linear combinations of the operators in the table and total derivatives. For example, previous work Weisz:1982zw ; Weisz:1983bn ; Luscher:1984xn used $\mathop{\mathrm{tr}}[(D_{\mu}F_{\rho\nu})(D_{\mu}F_{\rho\nu})]$, but we find it easier to use $\mathop{\mathrm{tr}}[F_{\mu\nu}F_{\nu\rho}F_{\rho\mu}]$. With the Bianchi identity $D_{\mu}F_{\rho\nu}+D_{\rho}F_{\nu\mu}+D_{\nu}F_{\mu\rho}=0$, one can show that ${\textstyle\frac{1}{2}}\mathop{\mathrm{tr}}[(D_{\mu}F_{\rho\nu})(D_{\mu}F_{\rho\nu})]=\mathop{\mathrm{tr}}[(D_{\mu}F_{\mu\nu})(D_{\rho}F_{\rho\nu})]-2\mathop{\mathrm{tr}}[F_{\mu\nu}F_{\nu\rho}F_{\rho\mu}]+\partial,$ (166) where $\partial$ denotes the omission of total derivatives that make no contribution to the action. Thus, only two of these three operators are needed. Table 6 is laid out in a suggestive way: operators in the right column clearly descend from those in the left. It is a little harder to show that there are no more Alford:1996up . When parity and charge conjugation are taken into account there are $10$ operators with two $D$s and two $E$s and another $10$ where the two $E$s are replaced with two $B$s. Of these $2\times 6$ may be eliminated in favor of total derivatives and others, leaving $2\times 4=8$ of this type. Three of these may be eliminated with the Bianchi identities $\displaystyle\bm{D}\cdot\bm{B}$ $\displaystyle=$ $\displaystyle 0,$ (167) $\displaystyle\bm{D}\times\bm{E}$ $\displaystyle=$ $\displaystyle D_{4}\bm{B}.$ (168) One application of the second Bianchi identity is less than obvious: $\mathop{\mathrm{tr}}[(D_{4}\bm{B})\cdot(D_{4}\bm{B})]=2\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{E}\times\bm{E})]-\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(\bm{D}\times\bm{B})]+\partial.$ (169) To find Eq. (169) one uses Eq. (168) for one factor of $D_{4}\bm{B}$, and then integrates by parts. In the end, there are 5 independent operators with two $D$s and two $E$s or two $B$s. In addition, there are $6$ operators with one each of $D_{4}$, $\bm{D}$, $\bm{E}$, and $\bm{B}$; $4$ may be eliminated in favor of total derivatives, and another may be eliminated with a Bianchi identity, leaving 1. Finally, there are the two operators $\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{E}\times\bm{E})]$ and $\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{B}\times\bm{B})]$. Thus, the total is 8, and the list in Table 6 is complete. There are three redundant interactions, corresponding to the transformations in Eqs. (22)–(24) that only involve gauge fields. They change the LE${\cal L}$ by $\displaystyle{\cal L}_{\mathrm{Sym}}$ $\displaystyle\mapsto$ $\displaystyle{\cal L}_{\mathrm{Sym}}+a^{2}\frac{2}{g^{2}}\left\\{\varepsilon_{A}\mathop{\mathrm{tr}}[(\bm{D}\cdot\bm{E})(\bm{D}\cdot\bm{E})]+(\varepsilon_{A}+\delta_{A})\mathop{\mathrm{tr}}[(\bm{D}\times\bm{B})\cdot(\bm{D}\times\bm{B})]\right.$ (170) $\displaystyle-\left.(2\varepsilon_{A}+\delta_{A}+\delta_{E})\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(\bm{D}\times\bm{B})]+(\varepsilon_{A}+\delta_{E})\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(D_{4}\bm{E})]\right\\}.$ By appropriate choice of the parameters $\varepsilon_{A}$, $\delta_{A}$, and $\delta_{E}$, one can remove $\mathop{\mathrm{tr}}[(\bm{D}\cdot\bm{E})(\bm{D}\cdot\bm{E})]$ and two of the other three induced interactions from the LE${\cal L}$. Below we shall see that it is most convenient to choose the redundant directions as shown in the last three lines of Table 6. To construct an improved gauge action, it is enough to consider the eight classes of six-link loops shown in Fig. 1, as well as plaquettes. Generalizing from Ref. Luscher:1984xn , we label sets of unoriented loops as in Table 7. Table 7: Unoriented loops on the lattice, up to length 6. Set $i$ | Type of loop ---|--- $0t$ | Temporal plaquettes $0s$ | Spatial plaquettes $1t$ | Rectangles with temporal long side $1t^{\prime}$ | Rectangles with temporal short side $1s$ | Spatial rectangles $2t$ | “Parallelograms” with two temporal sides $2s$ | Spatial “parallelograms” $3t$ | Bent rectangles with temporal bend edge $3t^{\prime}$ | Bent rectangles with temporal sides, but spatial bend edge $3s$ | Spatial bent rectangles Then let $S_{i}=\sum_{{\cal C}\in{\cal S}_{i}}2\mathop{\mathrm{Re}}\mathop{\mathrm{tr}}[1-U({\cal C})],$ (171) where $U({\cal C})$ is the product of link matrices around the curve ${\cal C}$. The gauge action is $S_{D^{2}F^{2}}=\frac{1}{g_{0}^{2}}\sum_{i}c_{i}S_{i},$ (172) where the $c_{i}$ are chosen so that $S_{D^{2}F^{2}}\geq 0$ and so that classical continuum limit is correct. The classical continuum limit is needed not only to determine the normalization of the $c_{i}$, but also to deduce which terms in the lattice action correspond to the redundant operators of the LE${\cal L}$. The classical continuum limit of the $S_{i}$ is easy to find with the procedure given in Ref. Luscher:1984xn . For the plaquette terms we find $\displaystyle S_{0t}$ $\displaystyle=$ $\displaystyle-\frac{a_{t}}{a_{s}}\int_{x}\mathop{\mathrm{tr}}[\bm{E}\cdot\bm{E}]+\frac{a_{t}^{3}}{12a_{s}}\int_{x}\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(D_{4}\bm{E})]+\frac{a_{t}a_{s}}{12}\int_{x}\sum_{i}\mathop{\mathrm{tr}}[(D_{i}E_{i})(D_{i}E_{i})],\hskip 20.00003pt$ (173) $\displaystyle S_{0s}$ $\displaystyle=$ $\displaystyle-\frac{a_{s}}{a_{t}}\int_{x}\mathop{\mathrm{tr}}[\bm{B}\cdot\bm{B}]+\frac{a_{s}^{3}}{12a_{t}}\int_{x}\sum_{j\neq k}\mathop{\mathrm{tr}}[(D_{j}B_{k})(D_{j}B_{k})],$ (174) where $a_{t}$ and $a_{s}$ are temporal and spatial lattice spacings, respectively. Here $\int_{x}=a_{t}a_{s}^{3}\sum_{x}\doteq\int d^{4}x.$ (175) It is convenient to express the six-link loops through $S_{0t}$ and $S_{0s}$, plus further terms of order $a^{2}$. The rectangles yield $\displaystyle S_{1t}$ $\displaystyle=$ $\displaystyle 4S_{0t}+\frac{a_{t}^{3}}{a_{s}}\int_{x}\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(D_{4}\bm{E})],$ (176) $\displaystyle S_{1t^{\prime}}$ $\displaystyle=$ $\displaystyle 4S_{0t}+a_{t}a_{s}\int_{x}\sum_{i}\mathop{\mathrm{tr}}[(D_{i}E_{i})(D_{i}E_{i})],$ (177) $\displaystyle S_{1s}$ $\displaystyle=$ $\displaystyle 8S_{0s}+\frac{a_{s}^{3}}{a_{t}}\int_{x}\sum_{j\neq k}\mathop{\mathrm{tr}}[(D_{j}B_{k})(D_{j}B_{k})];$ (178) the “parallelograms” $\displaystyle S_{2t}=8S_{0t}+4S_{0s}$ $\displaystyle-$ $\displaystyle 4a_{t}a_{s}\int_{x}\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{E}\times\bm{E})]-2a_{t}a_{s}\int_{x}\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(\bm{D}\times\bm{B})]$ (179) $\displaystyle+$ $\displaystyle a_{t}a_{s}\int_{x}\mathop{\mathrm{tr}}[(\bm{D}\cdot\bm{E})(\bm{D}\cdot\bm{E})]-a_{t}a_{s}\int_{x}\sum_{i}\mathop{\mathrm{tr}}[(D_{i}E_{i})(D_{i}E_{i})],\hskip 20.00003pt$ $\displaystyle S_{2s}=4S_{0s}$ $\displaystyle-$ $\displaystyle\frac{4a_{s}^{3}}{3a_{t}}\int_{x}\mathop{\mathrm{tr}}[\bm{B}\cdot(\bm{B}\times\bm{B})]+\frac{a_{s}^{3}}{a_{t}}\int_{x}\mathop{\mathrm{tr}}[(\bm{D}\times\bm{B})\cdot(\bm{D}\times\bm{B})]$ (180) $\displaystyle-$ $\displaystyle\frac{a_{s}^{3}}{a_{t}}\int_{x}\sum_{j\neq k}\mathop{\mathrm{tr}}[(D_{j}B_{k})(D_{j}B_{k})];$ and the bent rectangles $\displaystyle S_{3t}$ $\displaystyle=$ $\displaystyle 8S_{0t}+a_{t}a_{s}\int_{x}\mathop{\mathrm{tr}}[(\bm{D}\cdot\bm{E})(\bm{D}\cdot\bm{E})]-a_{t}a_{s}\int_{x}\sum_{i}\mathop{\mathrm{tr}}[(D_{i}E_{i})(D_{i}E_{i})],$ (181) $\displaystyle S_{3t^{\prime}}$ $\displaystyle=$ $\displaystyle 8S_{0t}+8S_{0s}-2a_{t}a_{s}\int_{x}\sum_{i}\mathop{\mathrm{tr}}[(D_{4}\bm{E})\cdot(\bm{D}\times\bm{B})],$ (182) $\displaystyle S_{3s}$ $\displaystyle=$ $\displaystyle 8S_{0s}+\frac{a_{s}^{3}}{a_{t}}\int_{x}\mathop{\mathrm{tr}}[(\bm{D}\times\bm{B})\cdot(\bm{D}\times\bm{B})]-\frac{a_{s}^{3}}{a_{t}}\int_{x}\sum_{j\neq k}\mathop{\mathrm{tr}}[(D_{j}B_{k})(D_{j}B_{k})].$ (183) We see immediately that the bent rectangles are the only place that the redundant interactions appear, so one may set $c_{3t}$, $c_{3t^{\prime}}$, and $c_{3s}$ at will, without sacrificing on-shell improvement. Indeed, the bent rectangles may be completely omitted from the improved action. To normalize the lattice gauge action to the classical continuum limit, one must choose $\displaystyle c_{0t}+4(c_{1t}+c_{1t^{\prime}})+8c_{2t}+8(c_{3t}+c_{3t^{\prime}})$ $\displaystyle=$ $\displaystyle\xi_{0},$ (184) $\displaystyle c_{0s}+8c_{1s}+4(c_{2t}+c_{2s})+8(c_{3s}+c_{3t^{\prime}})$ $\displaystyle=$ $\displaystyle\xi_{0}^{-1},$ (185) where $\xi_{0}$ is the bare anisotropy. At the tree level $\xi_{0}=a_{s}/a_{t}$. The essence of Eqs. (184) and (185) is to trade $c_{0t}$ and $c_{0s}$ for the bare coupling $g_{0}^{2}$ and the bare anisotropy $\xi_{0}$. To derive on-shell improvement conditions (at the tree level), one must allow for the transformations in Eqs. (23) and (24). We find on-shell improvement, at the tree level, when $\displaystyle\xi_{0}^{-1}c_{0t}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{5}{3}}-12x_{t^{\prime}}-4x_{s}-4(1+\xi_{0}^{-2})x_{t},$ (186) $\displaystyle\xi_{0}c_{0s}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{5}{3}}-4x_{t}-4(4+\xi_{0}^{2})x_{s},$ (187) $\displaystyle\xi_{0}^{-1}c_{1t}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{12}}+x_{t},$ (188) $\displaystyle\xi_{0}^{-1}c_{1t^{\prime}}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{12}}+x_{t^{\prime}},$ (189) $\displaystyle\xi_{0}c_{1s}$ $\displaystyle=$ $\displaystyle-{\textstyle\frac{1}{12}}+x_{s},$ (190) $\displaystyle c_{2t}$ $\displaystyle=$ $\displaystyle c_{2s}=0,$ (191) $\displaystyle\xi_{0}^{-1}c_{3t}$ $\displaystyle=$ $\displaystyle x_{t^{\prime}},$ (192) $\displaystyle\xi_{0}^{-1}c_{3t^{\prime}}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1}{2}}(x_{s}+\xi_{0}^{-2}x_{t}),$ (193) $\displaystyle\xi_{0}c_{3s}$ $\displaystyle=$ $\displaystyle x_{s},$ (194) where $x_{t}$, $x_{t^{\prime}}$, and $x_{s}$ are free parameters. In the main text of the paper, we consider isotropic lattices, but allow for the possibility that heavy-quark vacuum polarization requires some asymmetry in the couplings, starting at the one-loop level. Thus, we consider $\xi_{0}=1$ and $x_{t}=x_{t^{\prime}}=x_{s}=x$ and recover Luscher:1984xn $\displaystyle c_{0t}$ $\displaystyle=$ $\displaystyle c_{0s}={\textstyle\frac{5}{3}}-24x,$ (195) $\displaystyle c_{1t}$ $\displaystyle=$ $\displaystyle c_{1t^{\prime}}=c_{1s}=-{\textstyle\frac{1}{12}}+x,$ (196) $\displaystyle c_{2t}$ $\displaystyle=$ $\displaystyle c_{2s}=0,$ (197) $\displaystyle c_{3t}$ $\displaystyle=$ $\displaystyle c_{3t^{\prime}}=c_{3s}=x.$ (198) Positivity of the action requires $x<5/72$ and is guaranteed if $|x|<1/16$ Luscher:1984xn . Beyond the tree level asymmetry in these couplings may indeed arise. But the full freedom of the three redundant directions remains, so one may still choose $c_{3t}=x_{t}=0$, $c_{3t^{\prime}}=x_{t^{\prime}}=0$, and $c_{3s}=x_{s}=0$. ## References * (1) C. T. H. Davies et al. [HPQCD, MILC, and Fermilab Lattice Collaborations], Phys. Rev. Lett. 92, 022001 (2004) [arXiv:hep-lat/0304004]; C. Aubin et al. [HPQCD, MILC, and UKQCD Collaborations], Phys. Rev. D 70, 031504 (2004) [arXiv:hep-lat/0405022]; C. Aubin et al. [MILC Collaboration], Phys. Rev. D 70, 114501 (2004) [arXiv:hep-lat/0407028]. * (2) C. Aubin et al. [Fermilab Lattice, MILC, and HPQCD Collaborations], Phys. Rev. Lett. 94, 011601 (2005) [arXiv:hep-ph/0408306]; M. Okamoto et al., Nucl. Phys. B Proc. Suppl. 140, 461 (2005) [arXiv:hep-lat/0409116]. * (3) C. Aubin et al. [Fermilab Lattice, MILC, and HPQCD Collaborations], Phys. Rev. Lett. 95, 122002 (2005) [arXiv:hep-lat/0506030]. * (4) I. F. Allison _et al._ [HPQCD and Fermilab Lattice Collaborations], Phys. Rev. Lett. 94, 172001 (2005) [arXiv:hep-lat/0411027]. * (5) E. Eichten, Nucl. Phys. B Proc. Suppl. 4, 170 (1988); E. Eichten and B. R. Hill, Phys. Lett. B 234, 511 (1990). * (6) G. P. Lepage and B. A. Thacker, Nucl. Phys. B Proc. Suppl. 4, 199 (1988); B. A. Thacker and G. P. Lepage, Phys. Rev. D 43, 196 (1991). * (7) A. X. El-Khadra, A. S. Kronfeld, and P. B. Mackenzie, Phys. Rev. D 55, 3933 (1997) [arXiv:hep-lat/9604004]. * (8) K. G. Wilson, in New Phenomena in Subnuclear Physics, edited by A. Zichichi (Plenum, New York, 1977). * (9) B. Sheikholeslami and R. Wohlert, Nucl. Phys. B 259, 572 (1985). * (10) K. Symanzik, in _Recent Developments in Gauge Theories_ , edited by G. ’t Hooft _et al_. (Plenum, New York, 1980); in _Mathematical Problems in Theoretical Physics_ , edited by R. Schrader _et al_. (Springer, New York, 1982); Nucl. Phys. B 226, 187, 205 (1983). * (11) A. S. Kronfeld, Phys. Rev. D 62, 014505 (2000) [arXiv:hep-lat/0002008]. * (12) J. Harada _et al._ , Phys. Rev. D 65, 094513 (2002) [arXiv:hep-lat/0112044]; 71, 019903(E) (2005); 65, 094514 (2002) [arXiv:hep-lat/0112045]. * (13) M. B. Oktay _et al._ , Nucl. Phys. B Proc. Suppl. 119, 464 (2003) [arXiv:hep-lat/0209150]; 129, 349 (2004) [arXiv:hep-lat/0310016]; A. S. Kronfeld and M. B. Oktay, PoS LAT2006, 159 (2006) [arXiv:hep-lat/0610069]. * (14) S. Aoki, Y. Kuramashi, and S. i. Tominaga, Prog. Theor. Phys. 109, 383 (2003) [arXiv:hep-lat/0107009]. * (15) N. H. Christ, M. Li, and H. W. Lin, Phys. Rev. D 76, 074505 (2007) [arXiv:hep-lat/0608006]. * (16) A. S. Kronfeld, in _At the Frontiers of Particle Physics: Handbook of QCD_ , Vol. 4, edited by M. Shifman (World Scientific, Singapore, 2002) [arXiv:hep-lat/0205021]. * (17) P. Weisz, Nucl. Phys. B 212, 1 (1983). * (18) P. Weisz and R. Wohlert, Nucl. Phys. B 236, 397 (1984); 247, 544(E) (1984). * (19) M. Lüscher and P. Weisz, Commun. Math. Phys. 97, 59 (1985); 98, 433(E) (1985). * (20) M. Lüscher, in _Fields, Strings, and Critical Phenomena_ , edited by E. Brézin and J. Zinn-Justin (Elsevier, Amsterdam, 1990); in _Probing the Standard Model of Particle Interactions_ , edited by R. Gupta, A. Morel, E. DeRafael, and F. David (Elsevier, Amsterdam, 1999) [arXiv:hep-lat/9802029]. * (21) D. H. Adams and W. Lee, Phys. Rev. D 77, 045010 (2008) [arXiv:0709.0781 [hep-lat]]. * (22) G. P. Lepage, L. Magnea, C. Nakhleh, U. Magnea, and K. Hornbostel, Phys. Rev. D 46, 4052 (1992) [arXiv:hep-lat/9205007]. * (23) C. Morningstar, Nucl. Phys. B Proc. Suppl. 53, 914 (1997) [arXiv:hep-lat/9608019]. * (24) M. Alford, T. Klassen, G. P. Lepage, C. Morningstar, M. Peardon, and H. Trottier (unpublished). * (25) G. T. Bodwin, E. Braaten, and G. P. Lepage, Phys. Rev. D 51, 1125 (1995) [arXiv:hep-ph/9407339]; 55, 5853(E) (1997). * (26) K. Osterwalder and E. Seiler, Ann. Phys. 110, 440 (1978). * (27) M. Lüscher and P. Weisz, Nucl. Phys. B 240, 349 (1984). * (28) A. S. Kronfeld and D. M. Photiadis, Phys. Rev. D 31, 2939 (1985). * (29) M. Di Pierro et al. [FermiQCD Collaboration], Nucl. Phys. B Proc. Suppl. 129, 832 (2004) [arXiv:hep-lat/0311027]. * (30) M. Di Pierro, private communication. * (31) A. S. Kronfeld, Nucl. Phys. B Proc. Suppl. 129, 46 (2004) [arXiv:hep-lat/0310063]. * (32) S. J. Brodsky, G. P. Lepage, and P. B. Mackenzie, Phys. Rev. D 28, 228 (1983). * (33) J. Harada, S. Hashimoto, A. S. Kronfeld, and T. Onogi, Phys. Rev. D 67, 014503 (2003) [arXiv:hep-lat/0208004]; A. X. El-Khadra, E. Gamiz, A. S. Kronfeld and M. A. Nobes, PoS LATTICE 2007, 242 (2007) [arXiv:0710.1437 [hep-lat]]. * (34) C. Bernard et al. [MILC Collaboration], Phys. Rev. D 64, 054506 (2001) [arXiv:hep-lat/0104002]; C. Aubin et al. [MILC Collaboration], Phys. Rev. D 70, 094505 (2004) [arXiv:hep-lat/0402030]. * (35) G. P. Lepage and P. B. Mackenzie, Phys. Rev. D 48, 2250 (1993) [arXiv:hep-lat/9209022]. * (36) M. A. Nobes and H. D. Trottier, Nucl. Phys. B Proc. Suppl. 129, 355 (2004) [arXiv:hep-lat/0309086]; S. Aoki, Y. Kayaba and Y. Kuramashi, Nucl. Phys. B 689, 127 (2004) [arXiv:hep-lat/0401030]. * (37) E. Follana et al. [HPQCD Collaboration], Phys. Rev. D 75, 054502 (2007) [arXiv:hep-lat/0610092]. * (38) E. Follana, C. T. H. Davies, G. P. Lepage and J. Shigemitsu [HPQCD Collaboration], Phys. Rev. Lett. 100, 062002 (2008) [arXiv:0706.1726 [hep-lat]]. * (39) B. A. Dobrescu and A. S. Kronfeld, Phys. Rev. Lett. 100, 241802 (2008) [arXiv:0803.0512 [hep-ph]].

arxiv-papers

2024-09-04T02:48:54.155614

0803.0658

# The defining ideals of conjugacy classes of nilpotent matrices and a conjecture of Weyman Riccardo Biagioli Institut Camille Jordan, UMR 5208 du CNRS, Université de Lyon, Université Lyon 1, biagioli@math.univ-lyon1.fr Sara Faridi Department of Mathematics, Dalhousie University, Halifax, Canada, faridi@mathstat.dal.ca (research supported by NSERC) Mercedes Rosas Departamento de Álgebra, Universidad de Sevilla, mrosas@us.es (research supported by a Ramón y Cajal grant, MEC) ###### Abstract Tanisaki introduced generating sets for the defining ideals of the schematic intersections of the closure of conjugacy classes of nilpotent matrices with the set of diagonal matrices. These ideals are naturally labeled by integer partitions. Given such a partition $\lambda$, we define several methods to produce a reduced generating set for the associated ideal ${\mathcal{I}}_{\lambda}$. For particular shapes we find nice generating sets. By comparing our sets with some generating sets of ${\mathcal{I}}_{\lambda}$ arising from a work of Weyman, we find a counterexample to a related conjecture of Weyman. ## 1 Introduction Let $X$ be the set of $n\times n$ matrices over a field $k$ of characteristic $0$. In his paper Kostant [K] showed that the ideal of polynomial functions vanishing on the set of nilpotent matrices in $X$, is given by the invariants of the action by conjugation of $GL(n)$ on $X$. Let $C_{\lambda}$ be the conjugacy class of nilpotent matrices in $X$ having Jordan block sizes $\lambda^{\prime}_{1},\ldots,\lambda^{\prime}_{h}$, with $\lambda$ a partition of $n$ and $\lambda^{\prime}$ its transpose. Let $\overline{C}_{\lambda}$ be the nilpotent orbit variety defined as the Zariski closure of $C_{\lambda}$. De Concini and Procesi [DP] asked for a description of the ideal ${\mathcal{J}}_{\lambda}$ of polynomial functions vanishing on $C_{\lambda}$, for a general partition $\lambda$. They were interested in a refinement of Kostant’s result, which corresponds to the case $\lambda=(1^{n})$. De Concini and Procesi described a set of elements of ${\mathcal{J}}_{\lambda}$ that they conjectured to be a generating set. Later, Tanisaki [T] conjectured a simpler generating set, and Eisenbud and Saltman [ES] generalized Tanisaki’s conjecture to rank varieties. Finally, in 1989 Weyman [W1] used geometric methods to show that the three conjectures hold, and conjectured a minimal generating set $\mathcal{W}_{\lambda}$ for these ideals. In the present paper we focus on a related family of ideals that we denote by ${\mathcal{I}}_{\lambda}$ and call _De Concini-Procesi ideals_. These are the ideals of the scheme-theoretic intersection of nilpotent orbit varieties $\overline{C}_{\lambda}$ with the set of diagonal matrices. De Concini and Procesi [DP] produced a set of generators for these ideals that was later simplified by Tanisaki [T]. In both cases, the sets of generators are highly nonminimal. In the case $\lambda=(1^{n})$, Kostant’s theorem implies that the elementary symmetric functions of the eigenvalues of the matrices give a minimal set of generators for ${\mathcal{I}}_{(1^{n})}$. Our work in this paper is motivated by the search for a minimal generating set for De Concini–Procesi ideals. To this end, we simplify the generating set described by Tanisaki using elementary facts of the theory of symmetric functions. We provide several reduction methods. The obtained sets are minimal in special cases, and are generally much smaller. The main tool we use is a special filling of the Young diagram of the partition $\lambda$ which we call the _regular filling_. Clearly, by adding the defining ideal of the diagonal matrices to any generating set for the ideal ${\mathcal{J}}_{\lambda}$, we obtain a generating set for ${\mathcal{I}}_{\lambda}$. The following question is natural: Is it true that, after adding these generators to Weyman’s conjectured minimal generating set for ${\mathcal{J}}_{\lambda}$, a minimal generating set for ${\mathcal{I}}_{\lambda}$ is obtained ? We give a negative answer to this question and provide some infinite families of counterexamples. With the help of Macaulay 2 we verify that one of these counterexamples is also a counterexample to the original conjecture of Weyman on a minimal generating set of ${\mathcal{J}}_{\lambda}$. This has been a well studied problem that has been open for the past seventeen years. We hope that our methods together with those of Weyman will eventually lead to a complete solution of the problem of finding a minimal generating set for both ideals ${\mathcal{I}}_{\lambda}$ and ${\mathcal{J}}_{\lambda}$. Our paper is organized as follows. In Section 2 we introduce some basic tools from the theory of symmetric functions. In Section 3, we introduced Tanisaki’s generating set for the De Concini-Procesi ideal, and derive a simple combinatorial description for it. This leads to a simple rule to read a set of generators of the ideal directly from a special filling of the Young diagram of the partition that call the _regular filling_. In Section 4 we show that only generators read from the top entries of the regular filling are necessary in order to construct a generating set for ${\mathcal{I}}_{\lambda}$. The resulting generating set is in a one-to-one correspondence with a generating set that arises from the work of Weyman [W1]. In the case where the partition $\lambda$ is a hook, our result coincides with the minimal generating set we introduced in [BFR]. For a general shape though, this generating set could be far from minimal. In Section 5 we reduce the number of generators coming from each column of the Young diagram. Finally in Section 6, we provide many examples and counterexamples to the modified version of Weyman’s conjecture, and discuss classes where our reductions work best. Inside those families we are able to find a counterexample to the original conjecture of Weyman on a minimal generating set for the ideal ${\mathcal{J}}_{\lambda}$. Throughout the paper, we raise new questions whose answers could help illuminate the problem of finding minimal generating sets for ${\mathcal{I}}_{\lambda}$ and ${\mathcal{J}}_{\lambda}$. ## 2 Basic Tools We will be working in the polynomial ring $R=k[x_{1},\ldots,x_{n}]$, where $k$ may be an arbitrary field of characteristic $0$. We define a partition of $n\in\mathbb{N}$ to be a finite sequence $\lambda=(\lambda_{1},\ldots,\lambda_{k})\in\mathbb{N}^{k}$, such that $\sum_{i=1}^{k}\lambda_{i}=n$ and $\lambda_{1}\geq\ldots\geq\lambda_{k}$. If $\lambda$ is a partition of $n$ we write $\lambda\vdash n$. The nonzero terms $\lambda_{i}$ are called parts of $\lambda$. The number of parts of $\lambda$ is called the length of $\lambda$, denoted by $\ell(\lambda)$, so $\lambda_{i}=0$ if $i>\ell(\lambda)$. Let $\lambda=(\lambda_{1},\ldots,\lambda_{k})$ be a partition of $n$. The Young diagram of a partition $\lambda$ is the left-justified array with $\lambda_{i}$ squares in the $i$-th row, from bottom to top. We use the symbol $\lambda$ for both a partition and its associated Young diagram. For example, the diagram of $\lambda=(4,4,2,1)$ is illustrated in Figure 1 on the left. For a partition $\lambda=(\lambda_{1},\ldots,\lambda_{k})$ we define its conjugate partition as $\lambda^{\prime}=(\lambda_{1}^{\prime},\ldots,\lambda_{h}^{\prime})$, where for each $i\geq 1$, $\lambda_{i}^{\prime}$ is the number of parts of $\lambda$ that are bigger than or equal to $i$. The diagram of $\lambda^{\prime}$ is obtained by flipping the diagram of $\lambda$ across the diagonal. ${{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&}\kern-0.3pt\cr}}\end{matrix}\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;$ ${{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&}\kern-0.3pt\cr}}\end{matrix}$ Figure 1: The partition $\lambda=(4,4,2,1)$ and its conjugate $\lambda^{\prime}=(4,3,2,2)$. We shall need some basic definitions from the theory of symmetric functions. First, we introduce the generating series for the elementary and the complete symmetric polynomials (denoted respectively by $E(S,z)$ and $H(S,z)$). These series are defined as: $\displaystyle E(S,z)=\sum_{i\geq 0}z^{i}e_{i}(S)=\prod_{a\in S}(1+za),$ and $\displaystyle H(S,z)=\sum_{i\geq 0}z^{i}h_{i}(S)=\prod_{a\in S}\frac{1}{1-za},$ (1) where $S$ is a set of variables, and $z$ is a formal variable. Therefore, the elementary symmetric polynomial $e_{r}(S)$ is the sum of all square free monomials of degree $r$ in the variables of $S$, and the complete symmetric polynomial $h_{r}(S)$ is the sum of all monomials of degree $r$ in the variables of $S$. In order to introduce the monomial symmetric polynomials $m_{\lambda}(S)$, we say that a monomial ${\bf x}^{s}=x_{1}^{s_{1}}x_{2}^{s_{2}}\cdots x_{n}^{s_{n}}$ has type $\lambda$, if the partition $\lambda$ is obtained by rearranging the sequence $(s_{1},s_{2},\ldots,s_{n})$ in weakly descending order. Given a partition $\lambda$, the monomial symmetric polynomial $m_{\lambda}=m_{\lambda}(S)$ is defined as $m_{\lambda}(S)=\sum\mathbf{x}^{s}$ where the sum is taken over all different monomials $\mathbf{x}^{s}$ of type $\lambda$ and with all variables in $S$. If $f\in k[x_{1},\ldots,x_{n}]$ is a symmetric polynomial, and $S\subseteq\\{x_{1},\ldots,x_{n}\\}$, we define $f(S)$ as the evaluation of $f$ at the set $S$, by setting all variables $x\in\\{x_{1},\ldots,x_{n}\\}\setminus S$ to be equal to $0$ in $f$. For instance, $e_{2}(x_{1},x_{3})=x_{1}x_{3}$. The polynomial $f(S)$ is called a partially symmetric polynomial. In general, it is no longer invariant under the action of the symmetric group on $n$ letters. For simplicity, given a symmetric polynomial $f\in k[x_{1},\ldots,x_{n}]$, for all $1\leq k\leq n$, we will denote by $f(k)$ the following set of partially symmetric polynomials, $f(k)=\\{f(S)\mid S\subseteq\\{x_{1},\ldots,x_{n}\\},\ |S|=k\\}.$ For example, let $n=4$, then $e_{2}(3)=\\{x_{1}x_{2}+x_{1}x_{3}+x_{2}x_{3},\ x_{1}x_{2}+x_{1}x_{4}+x_{2}x_{4},\ x_{1}x_{3}+x_{1}x_{4}+x_{3}x_{4},\ x_{2}x_{3}+x_{2}x_{4}+x_{3}x_{4}\\}.$ Note that if $r>k$ we have $e_{r}(k)=\emptyset$. ###### Notation. Let $S\subseteq\\{x_{1},\ldots,x_{n}\\}$. For $x\in S$, and $I=\\{x_{i_{1}},\ldots,x_{i_{k}}\\}\subseteq S$, we let $\displaystyle S_{x}$ $\displaystyle=S\setminus\\{x\\}\ \ {\rm and}\ \ S_{i_{1},\ldots,i_{k}}=S\setminus I.$ We shall be using the following elementary lemma later in the paper. ###### Lemma 2.1 (Basic Lemma). Let $S\subseteq\\{x_{1},\ldots,x_{n}\\}$, $|S|=s$, and let $j\leq s$. Then 1. 1. $e_{j}(S)=e_{j}(S_{x})+xe_{j-1}(S_{x})$ for all $x\in S$; 2. 2. $\displaystyle\sum_{x\in S}e_{j}(S_{x})=(s-j)e_{j}(S)$; 3. 3. $\displaystyle\sum_{x\in S}xe_{j-1}(S_{x})=je_{j}(S)$. ###### Proof. 1. 1. Clear. 2. 2. Fix a square-free monomial $M$ of degree $j$ appearing in $e_{j}(S)$. Without loss of generality, assume $M=x_{1}\cdots x_{j}$ and $S=\\{x_{1},\ldots,x_{s}\\}$. Then each $e_{j}(S_{x_{t}})$ contains exactly one copy of $M$, for $t=j+1,\ldots,s$. There are exactly $s-j$ such indices $t$, so $M$ appears $s-j$ times in the left-hand sum. 3. 3. We use the equation in Part 1, and sum over all elements of $S$ : $\sum_{x\in S}e_{j}(S)=\sum_{x\in S}e_{j}(S_{x})+\sum_{x\in S}xe_{j-1}(S_{x})$ so by Part 2 we have $se_{j}(S)=(s-j)e_{j}(S)+\sum_{x\in S}xe_{j-1}(S_{x})$ and hence $je_{j}(S)=\sum_{x\in S}xe_{j-1}(S_{x}).$ ∎ ###### Proposition 2.2 (Another presentation of the partially symmetric polynomials). Let $S=\\{x_{1},\ldots,x_{n}\\}$, $i\leq n$, and define the ideal ${\mathcal{E}}_{i}(S)=(e_{1}(S),\ldots,e_{i}(S))$ in the polynomial ring $k[x_{1},\ldots,x_{n}]$. Let $U\subseteq S$ be a subset of cardinality $u$. Then for $i\leq n-u$ we have $e_{i}(S\setminus U)=(-1)^{i}h_{i}(U)\mbox{ mod }{\mathcal{E}}_{i}(S).$ (2) ###### Proof. This result follows from a formal manipulation of the generating functions in (1). We have $E(S\setminus U,z)=\prod_{\begin{subarray}{c}a\in S\\\ a\not\in U\end{subarray}}(1+za)=\frac{\prod_{a\in S}(1+za)}{\prod_{a\in U}(1+za)}=E(S,z)H(U,-z).$ Therefore, extracting the coefficient of $z^{i}$ from both sides of the resulting equation $E(S\setminus U,z)=E(S,z)H(U,-z)$ we obtain $e_{i}(S\setminus U)=\sum_{j=0}^{i}e_{j}(S)(-1)^{i-j}h_{i-j}(U).$ By hypothesis $e_{j}(S)$ is in the ideal for $j=1,\ldots,i$. Since $e_{0}(S)=1$, the result follows. ∎ ## 3 A new combinatorial description of Tanisaki’s generating set for ${\mathcal{I}}_{\lambda}$ In this section, we define a family of ideals ${\mathcal{I}}_{\lambda}$ in the polynomial ring $R=k[x_{1},\ldots,x_{n}]$ indexed by partitions $\lambda$ of $n$. The ideal ${\mathcal{I}}_{\lambda}$ was first introduced by De Concini and Procesi [DP] in order to describe the coordinate ring of the schematic intersection of the Zariski closure of the conjugacy class of nilpotent matrices of shape $\lambda$, with the set of diagonal matrices. In order to manipulate De Concini-Procesi ideals, we use a generating set defined by Tanisaki [T]. A nice feature of Tanisaki’s generating set is that its elements are elementary partially symmetric polynomials. Furthermore, Tanisaki’s proof of the correctness of his generating set is both elegant and elementary, and it is based on standard linear algebra facts. Finally, Tanisaki’s generating set has proven to be very fruitful in algebraic combinatorics, see for example [AB, BG, GP]. Let $\lambda=(\lambda_{1},\ldots,\lambda_{k})$ be a partition of $n$. For the purpose of the next formula, we add enough zeroes to the end of $\lambda$ so that it has $n$ terms: $\lambda=(\lambda_{1},\ldots,\lambda_{n})$. For any $1\leq k\leq n$, we define $\delta_{k}(\lambda)=\lambda^{\prime}_{n}+\lambda^{\prime}_{n-1}+\ldots+\lambda^{\prime}_{n-k+1}.$ (3) It is clear that $\delta_{n}(\lambda)\geq\delta_{n-1}(\lambda)\geq\ldots\geq\delta_{1}(\lambda)$, and that $\delta_{n}(\lambda)=n$. ###### Theorem 3.1 (Tanisaki’s generating set [T]). The ideal $\mathcal{I}_{\lambda}$ is generated by the following collection of elementary partially symmetric polynomials $\mathcal{I}_{\lambda}=\big{(}e_{r}(k)\mid k=1,\ldots,n,\;{\rm and}\ \ k\geq r>k-\delta_{k}(\lambda)\big{)}.$ (4) ###### Definition 3.2 (De Concini-Procesi ideal). We call the ideal $\mathcal{I}_{\lambda}$ defined in Theorem 3.1 the De Concini-Procesi ideal of the partition $\lambda$. Since for any partition $\lambda$ of $n$, $\delta_{n}(\lambda)=n$, when we set $k=n$ in (4) we conclude that $\mathcal{I}_{\lambda}$ contains all the elementary symmetric polynomials in all the variables $x_{1},\ldots,x_{n}$. ###### Example 3.3. Let $\lambda=(4,4,2,1,0,0,0,0,0,0,0)\vdash 11$ be the partition appearing in Figure 1. Then $(\delta_{1}(\lambda),\ldots,\delta_{11}(\lambda))=(0,0,0,0,0,0,0,2,4,7,11)$. Hence $(1-\delta_{1}(\lambda),\ldots,11-\delta_{11}(\lambda))=(1,2,3,4,5,6,7,6,5,3,0).$ Here $n=11$. For $k=1,\ldots,7$ there is no admissible $e_{r}(k)$ in the generating set described in (4). So the generating set of $\mathcal{I}_{(4421)}$ consists of the following elements We now give a simple combinatorial description of the set of generators for ${\mathcal{I}}_{\lambda}$ described in Theorem 3.1, and then demonstrate how to shorten it so that one can read a reduced generating set for ${\mathcal{I}}_{\lambda}$ directly from the diagram of the partition $\lambda$. In order to do so we introduce the notion of regular filling. ###### Definition 3.4 (The regular filling of a partition). Let $\lambda$ be a partition of $n$. Draw its Young diagram and then fill its cells with the numbers $1,2,\ldots,n$ from top to bottom and from left to right, skipping the cells in the bottom row, which should be filled at the end from right to left. This is called the regular filling of $\lambda$, denoted rf. ${{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}$ Figure 2: The regular filling of $(4,4,2,1)$. ###### Definition 3.5 (The reading process). We associate to any filling $f$ of the Young diagram of $\lambda$ a set of partial symmetric polynomials, denoted by $\mathcal{G}_{f}(\lambda)$. We read the elements of this set from the filling as follows. For a given column of $\lambda$ we add to $\mathcal{G}_{f}(\lambda)$ all the elements of the sets $e_{r}(k)$, where $k$ is the entry in the bottom cell of the column, and the degrees $r$’s are given by all the entries in that column. ###### Notation. From now on, we enumerate columns and rows of a Young diagram from left to right by starting from zero. So the “first” column will be the $0$-th column; similarly for rows. ###### Example 3.6. For the partition $\lambda=(4,4,2,1)$, the regular filling rf is illustrated in Figure 2. The reading process of this filling gives the set $\mathcal{G}_{\it rf}(\lambda)$ consisting of: the elementary symmetric polynomials $e_{1}(x_{1},\ldots,x_{11})$, $e_{2}(x_{1},\ldots,x_{11})$, $e_{3}(x_{1},\ldots,x_{11})$, $e_{11}(x_{1},\ldots,x_{11})$, coming from the $0$-th column; the partially symmetric polynomials of the sets $e_{4}(10),e_{5}(10),e_{10}(10)$ read from the first column, $e_{6}(9),e_{9}(9)$ from the second column, and $e_{7}(8),e_{8}(8)$ from the last column. By using this reading process, we are going to read Tanisaki’s generators from a special filling. ###### Definition 3.7 (The antidiagonal filling). Let $\lambda$ be a partition of $n$. Compute the partition $\delta(\lambda)$ $\delta(\lambda)=\delta_{n}(\lambda)\geq\delta_{n-1}(\lambda)\geq\ldots\geq\delta_{1}(\lambda),$ where $\delta_{k}(\lambda)$ is defined as in (3), and draw the Young diagram of its conjugate $\delta^{\prime}(\lambda)$. Now fill the $0$-th column of $\delta^{\prime}(\lambda)$ by $1,2,\ldots,n$ from top to bottom, and then fill the remainder of the diagram so that the filling is constant following each antidiagonal. We call this the antidiagonal filling of $\delta^{\prime}(\lambda)$ and denote it by af. For our running example $\lambda=(4,4,2,1,0^{7})$, we have $\delta(\lambda)=(11,7,4,2,0^{7})$; the antidiagonal filling of $\delta^{\prime}(\lambda)$ is given in Figure 3. Note that the bottom entry of the $k$-th column of $\delta^{\prime}(\lambda)$ is $n-k$. ${{{{{{{{{{{{{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}$ Figure 3: The antidiagonal filling of $\delta^{\prime}(\lambda)$. Let $\lambda$ be a partition of $n$. Compute the set $\mathcal{G}_{\it af}(\delta^{\prime}(\lambda))$ by applying the reading process to the antidiagonal filling af of $\delta^{\prime}(\lambda)$. We have the following lemma. ###### Lemma 3.8. Let $\lambda$ be a partition of $n$. Then Tanisaki’s set of generators is $\mathcal{G}_{\it af}(\delta^{\prime}(\lambda))$. In particular, ${\mathcal{I}}_{\lambda}=(\mathcal{G}_{\it af}(\delta^{\prime}(\lambda))).$ ###### Proof. Let $\lambda=(\lambda_{1},\ldots,\lambda_{n})$. Compute $\delta^{\prime}(\lambda)$ and fill its diagram with the antidiagonal filling. According to Theorem 3.1, to compute Tanisaki’s generating set, we need to find for which $k$ the interval $[k-\delta_{k}(\lambda)+1,\ldots,k-1,k]$ is nonempty; clearly this happens when $\delta_{k}(\lambda)>0$. From the definition of $\delta_{k}(\lambda)$, the only times $\delta_{k}(\lambda)>0$ is when $k=n-\lambda_{1}+1,\ldots,n$. So we are considering values $e_{r}(S)$ for sets $S$ such that $n-\lambda_{1}+1\leq|S|\leq n$. This is an interval of length $\lambda_{1}$, and the numbers $k=|S|$ we are considering are exactly the entries in the first row of $\delta^{\prime}(\lambda)$. Now, fix a column $t$ that has entry $n-t$ in its bottom cell. The generating set described in Theorem 3.1 has $e_{r}(S)$, where $|S|=n-t$ and $r=n-t-\delta_{n-t}(\lambda)+1,\ldots,n-t$. Note that there exactly $\delta_{n-t}(\lambda)$ values that $r$ takes, and that is exactly the size of the $t$-th column of $\delta^{\prime}(\lambda)$. The mentioned values of $r$ are exactly the entries of the $t$-th column of the antidiagonal filling of $\delta^{\prime}(\lambda)$. ∎ One can easily check that this procedure applied to the antidiagonal filling in Figure 3 produces the generators given in the table of Example 3.3. We are now able to show the main result of this section, namely, that ${\mathcal{I}}_{\lambda}$ is the sum of three simpler ideals. In order to do so we will use the regular filling. ###### Theorem 3.9. Let $\lambda$ be a partition of $n$. Fill the diagram of $\lambda$ with the regular filling, and compute the set $\mathcal{G}_{\it rf}(\lambda)$ by using the reading process described in Definition 3.5. Then $\mathcal{I}_{\lambda}=(\mathcal{G}_{\it rf}(\lambda)).$ ###### Proof. Compute the partition $\delta^{\prime}(\lambda)$, fill its diagram with the antidiagonal filling and read off all of Tanisaki’s generators. By Part 2 of Lemma 2.1, if $e_{r}(x_{1},\ldots,x_{j})\neq 0$ belongs to the ideal, so does $e_{r}(x_{1},\ldots,x_{J})$ for any $J>j$. Therefore, for each entry $r=1,\ldots,n$, we only need to keep the generators coming from the rightmost occurrence of that $r$ in the antidiagonal filling of $\delta^{\prime}(\lambda)$. So we delete all other occurrences of $r$ in that filling, and the corresponding cell. We obtain a filling that contains exactly one occurrence of each of the numbers from $1$ to $n$. Now observe that the differences of heights between adjacent columns of $\delta^{\prime}(\lambda)$ are given by the sequence $\lambda^{\prime}_{1},\ldots,\lambda^{\prime}_{\lambda_{1}}$. So after the deletion process, explained above, the remaining diagram will have columns of height $\lambda^{\prime}_{1},\ldots,\lambda^{\prime}_{\lambda_{1}}$. Hence it is the diagram of our partition $\lambda$. Moreover the resulting is the regular filling, and we are done. The case of the partition $\lambda=(4,4,2,1)$ is displayed in Figure 4. ∎ ${{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{\begin{array}[]{ccc}\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle*$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}&\downarrow&\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr}\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr}\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\; }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to10.0pt{\hrule height=0.3pt\vss\hbox to10.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}\end{array}$ Figure 4: From the antidiagonal to the regular filling. ###### Remark 3.10. Observe that $e_{j}(S)$ for $S$ of cardinality $j$ is a square free monomial of degree $j$. So once we have all square-free monomials of degree $n-\lambda_{1}+1$ in our ideal, then we have the ones of higher degree. These monomials are obtained when we read the generators coming from the rightmost entry of the bottom row. The following statement follows easily from the previous remark and Theorem 3.9. ###### Corollary 3.11 (First reduction of Tanisaki’s generating set for ${\mathcal{I}}_{\lambda}$). Let $\lambda$ be a partition of $n$. Then ${\mathcal{I}}_{\lambda}$ can be described as the sum of the following three ideals: ${\mathcal{I}}_{\lambda}={\mathcal{M}}_{\lambda}+{\mathcal{E}}_{\lambda}+{\mathcal{K}}_{\lambda},$ where * • ${\mathcal{M}}_{\lambda}$ is generated by all square-free monomials of degree $n-\lambda_{1}+1$; * • ${\mathcal{E}}_{\lambda}$ is generated by the elementary symmetric polynomials $e_{1}(x_{1},\ldots,x_{n}),\ldots,e_{\ell(\lambda)-1}(x_{1},\ldots,x_{n})$; * • ${\mathcal{K}}_{\lambda}$ is generated by the partially symmetric polynomials in $e_{r}(k)$, where $n-1\geq k\geq n-\lambda_{1}+1$, and $r$ in an entry of the regular filling of $\lambda$, in the same column as $k$, and strictly above it. In the particular case where the indexing partition $\lambda$ is a hook, we recover the minimal generating set for $\mathcal{I}_{\lambda}$ described in [BFR, Proposition 3.4]. ## 4 Second reduction of the generating set for ${\mathcal{I}}_{\lambda}$ Our goal in the rest of the paper is to shave off as many redundant generators as possible from the generating set given in Corollary 3.11 . It turns out that only partially symmetric polynomials coming from the top value of each column are required in the generating set. This finding already gives a large reduction in the number of generator needed in the generating set of Tanisaki. Several other reductions will be obtained in the following sections. Suppose we have a partition $\lambda$ of an integer $n$, and fill the diagram of $\lambda$ with the regular filling defined in Definition 3.4. For $k\geq 1$ we label the value in the top cell of the $k$-th column with $b_{k}$, as long as the height of the $k$-th column is $\geq 2$. If the right-most column of $\lambda$ has height 1, then we label its entry $b_{s}$. This is reflected in the diagram in Figure 5. Note that with this notation we have $b_{1}=\lambda^{\prime}_{1},\ b_{2}=\lambda^{\prime}_{1}+\lambda^{\prime}_{2}-1,\ \ldots,\ b_{k}=\lambda^{\prime}_{1}+\ldots+\lambda^{\prime}_{k}-k+1\mbox{ for }k\leq t,\ b_{s}=n-s,$ where we set $t=\lambda_{2}-1,\ \mbox{ and }\ s=\lambda_{1}-1.$ (6) Clearly if $\lambda_{1}=\lambda_{2}$, then $t=s$ and $b_{s}$ does not exist. ${{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{\tiny{\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{1}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{2}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{t}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{1}-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{2}-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle b_{3}-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle n-s-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle n$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle n-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle n-2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle n-t$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to19.0pt{\hrule height=0.3pt\vss\hbox to19.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{b_{s}=}}}{{\scriptscriptstyle{n-s}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}}$ Figure 5: Diagram of a partition $\lambda$ of $n$ with the regular filling. By Corollary 3.11) the reduced form of Tanisaki’s generating set for ${\mathcal{I}}_{\lambda}$ is the union of the following sets: (13) Our goal here is to show that it is enough to pick only one set of generators in each column, other than the $0$-th column; namely, the ones coming from the top values in each column. ###### Theorem 4.1 (Principal reduction of the generating set for ${\mathcal{I}}_{\lambda}$). Let $\lambda$ be a partition of $n$, and suppose that the diagram of $\lambda$ has been filled as in Figure 5. Then a generating set for ${\mathcal{I}}_{\lambda}$ is (20) If $\lambda=(1^{n})$ is the one-column partition, then we also need to add the element $e_{n}(n)=x_{1}\cdots x_{n}$ to this generating set. If $\lambda=(n)$ is the one-row partition, we only need generators from the last column, in other words ${\mathcal{I}}_{(n)}=(x_{1},\ldots,x_{n})$. ###### Proof. We need to show that having in the ideal all generators read from the top index of each column implies that the other partially symmetric functions coming from the larger indices in that column also belong to the ideal. We go column by column, and build a new ideal $I_{\lambda}$ by adding generators described in (20) for each column of $\lambda$. We show, each time, that $I_{\lambda}$ contains all the other generators described in (13) (coming from the same column), and therefore $I_{\lambda}={\mathcal{I}}_{\lambda}$. * _Col. 0._ There is nothing to prove here, as we are keeping all the generators $e_{1}(n),\ldots,e_{b_{1}-1}(n)$. * _Col. 1._ Assume that we have $e_{b_{1}}(S)\in I_{\lambda}$ for all $S$ with $|S|=n-1$. By Part 2 of Lemma 2.1, setting $j=b_{1}$, we see that we have $e_{b_{1}}(n)\in I_{\lambda}$. For each $i>b_{1}$, we can assume by induction on $i$ that $e_{1}(n),\ldots,e_{i-1}(n)\in I_{\lambda}\mbox{ and }\hfill e_{b_{1}}(n-1),\ldots,e_{i-1}(n-1)\in I_{\lambda}.$ Apply Part 3 of Lemma 2.1 with $j=i$, to see that $e_{i}(n)\in I_{\lambda}$. Fix a set $S$ with $|S|=n-1$ and $x\notin S$. Let $S^{x}=S\cup\\{x\\}$. Part 1 of Lemma 2.1 implies that $e_{i}(S)=e_{i}(S^{x})-xe_{i-1}(S)$ which demonstrates that $e_{i}(S)\in I_{\lambda}$. Hence $e_{i}(n-1)\in I_{\lambda}$. The fact that the generators $e_{b_{1}}(n-1)$ can be replaced by the powers $x_{1}^{b_{1}},\ldots,x_{n}^{b_{1}}$ follows directly from Proposition 2.2. Note that, in particular, we have $e_{i}(n-1)\in I_{\lambda}$, for all $i\geq b_{1}$. * _Col. j._ Suppose $I_{\lambda}$ contains all generators from the previous columns $0,\ldots,j-1$ as described in (20). Let $|S|=n-j$, and suppose $x\notin S$, so that $|S^{x}|=n-j+1$, ($S^{x}=S\cup\\{x\\}$). We know by induction that $I_{\lambda}$ contains $e_{h}(S^{x})$ for all $h\geq b_{j-1}$. Therefore, since $b_{j}>b_{j-1}$, for $i\geq b_{j}$ we have by Part 1 of Lemma 2.1 $\begin{array}[]{lll}e_{i}(S)&=e_{i}(S^{x})-xe_{i-1}(S)&=-xe_{i-1}(S)\\\ &=-x(e_{i-1}(S^{x})-xe_{i-2}(S))&=x^{2}e_{i-2}(S)\\\ &=x^{2}(e_{i-2}(S^{x})-xe_{i-3}(S))&=-x^{3}e_{i-3}(S)\\\ &\hskip 14.45377pt\vdots&\\\ &=(-1)^{i-b_{j}}x^{i-b_{j}}e_{b_{j}}(S)&(mod\ Col.\ j-1)\end{array}$ This means that once we include $e_{b_{j}}(S)$ in $I_{\lambda}$, we will have all $e_{i}(S)\in I_{\lambda}$ for $i\geq b_{j}$. ∎ In the case where $\lambda$ is a hook, the generating set described in Theorem 4.1 coincides with the minimal generating set for ${\mathcal{I}}_{\lambda}$ introduced in our earlier work [BFR]. ###### Example 4.2. Let $\lambda=(5,4,4,3)$. Then, the regular filling of $\lambda$ is ${{{{{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 16$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 15$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 14$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 13$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 12$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}}\end{matrix}$ So the generators of ${\mathcal{I}}_{\lambda}$ are Later in Example 6.4 we shall further reduce the generating set of this particular partition. ### 4.1 Remarks on a related work and conjecture of Weyman We end this section by showing some relations between the generating set of Theorem 4.1 and two generating sets for ${\mathcal{I}}_{\lambda}$ arising in the work of Weyman [W1]. In [W1] Weyman uses the representation theory of the general linear group to construct and study generating sets for the ideal ${\mathcal{J}}_{\lambda}$ of polynomial functions vanishing on the conjugacy class $\mathcal{C}_{\lambda}$. The generators in the first family, denoted by $V_{\lambda}$, are expressed as sums of minors, and come from reducible representations of $GL(n)$. The second set of generators $U_{\lambda}$, on the other hand, arises from the irreducible representations of $GL(n)$. The set $U_{\lambda}$ is smaller than $V_{\lambda}$, but how to compute its elements is not explicit in the paper. The set $V_{\lambda}$ (respectively $U_{\lambda}$) is given by the disjoint union of sets $V_{i,p}$ (respectively $U_{i,p}$), where the family of indices $(i,p)$ can be read off from a special diagram introduced by Weyman; see [W1, Example (4.5)]. We call this diagram the Weyman diagram of $\lambda$. It is possible to construct the Weyman diagram of a partition starting from the antidiagonal filling (see Definition 3.7) as follows. First, consider the antidiagonal filling of $\delta^{\prime}(\lambda)$, and justify its columns in such a way that equal entries are now in same rows. Then, replace any entry of this diagram by an $X$. The resulting picture is the Weyman diagram. In Figure 6 we illustrate the Weyman diagram corresponding to the partition $\lambda=(4,4,2,1)$. Compare this diagram to the one in Figure 3. Note that if the top $X$ in the $i$-th column of Weyman diagram of $\lambda$ has coordinates $(i,p)$, then the top cell of the $i$-th column of the regular filling of $\lambda$ is filled by $p$. $\displaystyle\begin{matrix}p=1&\underline{X}&&&&\\\ p=2&\underline{X}&&&&\\\ p=3&\underline{X}&&&&\\\ p=4&\underline{X}&\underline{X}&&&\\\ p=5&\underline{X}&X&&&\\\ p=6&\underline{X}&X&\underline{X}&&\\\ p=7&\underline{X}&X&X&\underline{X}&\\\ p=8&\underline{X}&X&X&X&\\\ p=9&\underline{X}&X&X&&\\\ p=10&\underline{X}&X&&&\\\ p=11&\underline{X}&&&&\\\ i=&0&1&2&3\\\ \end{matrix}$ Figure 6: Weyman diagram for $\lambda=(4,4,2,1)$. We would like to remark that Weyman follows a convention opposite to ours when labelling the ideals ${\mathcal{I}}_{\lambda}$ and ${\mathcal{J}}_{\lambda}$: he labels $\mathcal{J}_{\lambda}$ the ideal of polynomial functions vanishing on all nilpotent matrices with Jordan blocks $\lambda_{1},\ldots,\lambda_{n}$, while we use the transpose. On the other hand, he associates to a partition $\lambda$ what in our setting would be the Weyman diagram of $\lambda^{\prime}$. These two facts cancel out, and we do not need to take any transpose when reading statements involving his diagrams. ###### Definition 4.3 (Weyman’s generating set for $\mathcal{J}_{\lambda}$). In [W1, Theorem (4.6)] Weyman shows that the ideal ${\mathcal{J}}_{\lambda}$ is generated by the $U_{i,p}$, where the $(i,p)$’s are the coordinates of the top cells of the columns ($i\geq 1$) of the Weyman diagram of $\lambda$, together with the invariants $U_{0,p}$ with $1\leq p\leq n$. This result implies that the ideal ${\mathcal{J}}_{\lambda}$ is also generated by the $V_{i,p}$ coming from the same set of indices $(i,p)$. ###### Example 4.4. For the partition $\lambda=(4,4,2,1)$, whose Weyman diagram is in Figure 6, Weyman’s set $U_{\lambda}$ consists of $U_{0,p}$, with $1\leq p\leq 11$, $U_{1,4}$, $U_{2,6}$, and $U_{3,7}$ (and similarly for the set $V_{\lambda}$). The cells $X$ whose coordinates label this generating set are underlined. After adding the generators for the ideal defining the diagonal matrices to the two sets $V_{\lambda}$ and $U_{\lambda}$, one gets two generating sets for ${\mathcal{I}}_{\lambda}$; we denote these two generating sets by $\tilde{V_{\lambda}}$ and $\tilde{U_{\lambda}}$. Instead of going into the definitions of $V_{\lambda}$ and $U_{\lambda}$ that can be found in [W1, Section 4], we explicitly state the cardinalities of their components in order to compare them with our generating set. We emphasize the fact that Tanisaki’s generators (the ones we use) are easier to handle than Weyman’s generators. We have that $|V_{i,p}|={n\choose i}^{2}\ \text{and}\ \ |\tilde{V}_{i,p}|={n\choose i},$ and $|U_{i,p}|={n\choose i}^{2}-{n\choose i-1}^{2}\ \text{and}\ \ |\tilde{U}_{i,p}|={n\choose i}-{n\choose i-1}.$ It turns out that the cardinalities of the generating set for ${\mathcal{I}}_{\lambda}$ given by the $\tilde{V}_{i,p}$’s and the generating set given in Theorem 4.1 are the same. Moreover, it is not difficult to describe a one-to-one correspondence between the two generating sets. Under this correspondence Weyman’s $V_{i,p}$ generators correspond to our generators read from the top cell of the $i$-th column of the regular filling, as described in Theorem 4.1. Weyman conjectured that a special subset of $U_{\lambda}$ gives a minimal generating set of ${\mathcal{J}}_{\lambda}$; see Conjecture 5.1 and Remark 5.3 of [W1]. ###### Conjecture 4.5 (Weyman’s original conjecture). Let $\lambda$ be a partition. The set consisting of $U_{0,p}$ for $1\leq p\leq\ell(\lambda)$, and $U_{i,p}$, where $(i,p)$ labels a top cell of the $i$-th row (in the Weyman diagram of $\lambda$), such that there are no $X$’s to the right of or on the line segment joining $(i,p)$ with $(0,1)$, is a minimal set of generators $\mathcal{W}_{\lambda}$ of ${\mathcal{J}}_{\lambda}$. A very interesting question is the following. ###### Question 4.6 (Diagonal version of Weyman’s conjecture). Is the generating set $\tilde{\mathcal{W}}_{\lambda}$ for ${\mathcal{I}}_{\lambda}$ arising from Weyman’s conjecture minimal ? In the following sections we show that the the answer to this question is negative. Indeed, we provide some infinite families of counterexamples. These observations, together with the help of Macaulay 2 led us to the discovery that even the original conjecture of Weyman (Conjecture 4.5) fails already for one of the smallest elements in these families. ## 5 Reducing generators of ${\mathcal{I}}_{\lambda}$ of a fixed degree The aim of this section is to consider the generating set of ${\mathcal{I}}_{\lambda}$ described in Theorem 4.1, and eliminate as many redundant generators as possible from each column. ###### Proposition 5.1 (Columns of height $>1$). Let $\lambda$ be a partition whose diagram is represented in Figure 5. For $k\geq 2$, if the height of the $(k-1)$-st column is $>1$, then we can eliminate ${n-1\choose k-1}+1$ generators of ${\mathcal{I}}_{\lambda}$ (as described in (20)) that come from the $k$-th column. Indeed, if $S$ denotes the set of variables $x_{1},\ldots,x_{n}$, we can eliminate the elements in the set $\\{e_{b_{k}}(S_{1,i_{2},\ldots,i_{k}})\mid 1<i_{2}<\ldots<i_{k}\leq n\\}$ and $e_{b_{k}}(S_{2,3,\ldots,k+1})$. ###### Proof. Let $k>1$, by using Part 2 of Lemma 2.1 we write $\displaystyle\sum_{j\notin\\{i_{1},\ldots,i_{k-1}\\}}e_{b_{k}}(S_{i_{1},\ldots,i_{k-1},j})=(n-b_{k}-k+1)e_{b_{k}}(S_{i_{1},\ldots,i_{k-1}})\equiv 0\ ({\rm mod}\ \mathcal{I}_{k-1})$ (22) where $\mathcal{I}_{k-1}$ is the ideal of generators coming from columns $0$ to $k-1$. So we have a system of $n\choose k-1$ linear homogeneous equations, in $n\choose k$ variables. In fact we have one equation for each choice of a $(k-1)$-subset $\\{i_{1},\ldots,i_{k-1}\\}$, and one variable $e_{b_{k}}(S_{i_{1},\ldots,i_{k-1},j})$ for each $k$-subset $\\{i_{1},\ldots,i_{k-1},j\\}$. The matrix associated to this system has columns $J$ indexed by the $k$-subsets of $\\{1,2,\ldots,n\\}$, and rows $I$ indexed by $k-1$-subsets of $\\{1,2,\ldots,n\\}$. Equation (22), says that at position $(I,J)$ the entry will be $1$ if $I\subseteq J$ and $0$ if $I\not\subseteq J$. We claim that we can drop from the generating set of Theorem 4.1 $e_{b_{k}}(S_{J})$, for all $J$ of cardinality $k$ containing $1$, and $e_{b_{k}}(S_{2,\ldots,k+1}).$ To prove this it suffices to show that the submatrix corresponding to these columns has full rank ${n-1\choose k-1}+1$. We order the columns of this submatrix in this way: we put first the the columns indexed by a $J$ containing $1$ in alphabetical order, and then column indexed by $\\{2,\ldots,k+1\\}$. Similarly, we order the rows starting with those indexed by subsets $I$ that do not contain $1$, in alphabetical order, and then the row indexed by $\\{1,\ldots,k-1\\}$, and then the other rows in any order. In Figure 7 two examples are displayed. The square submatrix given by the first ${n-1\choose k-1}+1$ rows consists of two blocks. An identity ${n-1\choose k-1}$-matrix together with an additional row: $(1,\ldots,1,0,\ldots,0)$, with $n-k+1$ ones. In fact, this last row is indexed by $\\{1,\ldots,k-1\\}$, and the entries are $1$ at columns indexed by $\\{1,2,\ldots,k-1,j\\}$ for $j>k$, and zero otherwise. By Gauss elimination, it is easy to see that this submatrix has full rank. ∎ $\begin{array}[]{c|cccccc}&12&13&14&23&24&34\\\ \hline\cr 2&{\bf 1}&{\bf 0}&{\bf 0}&{\bf 1}&1&0\\\ 3&{\bf 0}&{\bf 1}&{\bf 0}&{\bf 1}&0&1\\\ 4&{\bf 0}&{\bf 0}&{\bf 1}&{\bf 0}&1&1\\\ 1&{\bf 1}&{\bf 1}&{\bf 1}&{\bf 0}&0&0\\\ \end{array}\hskip 56.9055pt\begin{array}[]{c|cccccc}&123&124&134&234\\\ \hline\cr 23&{\bf 1}&{\bf 0}&{\bf 0}&{\bf 1}\\\ 24&{\bf 0}&{\bf 1}&{\bf 0}&{\bf 1}\\\ 34&{\bf 0}&{\bf 0}&{\bf 1}&{\bf 1}\\\ 12&{\bf 1}&{\bf 1}&{\bf 0}&{\bf 0}\\\ \hline\cr 13&{\bf 1}&{\bf 0}&{\bf 1}&{\bf 0}\\\ 14&{\bf 0}&{\bf 1}&{\bf 1}&{\bf 0}\\\ \end{array}$ Figure 7: The non-singular submatrices for $n=4$, $k=2$, and $n=4$, $k=3$. ###### Remark 5.2. The system (22) has ${n\choose k-1}$ linear equations and ${n\choose k}$ variables. If all the equations are independent, then ${n\choose k-1}$ variables are redundant. Hence only ${n\choose k}-{n\choose k-1}$ of them are necessary. Then using Gauss elimination we would obtain an explicit generating set of the same size as Weyman’s $\tilde{U}_{k,p}$. We note that there is no explicit construction for the generators in $U_{\lambda}$ in Weyman’s paper [W1]. ###### Remark 5.3. Let $\lambda$ be a partition of $n$ different than $(n)$. As a consequence of Proposition 5.1, the number of generators coming from the top cell of column $k$ in our generating set for ${\mathcal{I}}_{\lambda}$ is ${n\choose k}-{n-1\choose k-1}-1$. On the other hand, and as discussed in Section 4.1 the corresponding $\tilde{U}_{k,p}$ in Weyman’s generating set consists of ${n\choose k}-{n\choose k-1}$ elements. Since for all partitions other than $(n)$, we have that $n>k$, we conclude that the difference between the two sets is ${n-1\choose k-2}-1$, for each $k>2$. For columns $0$, $1$, and $2$ their cardinalities coincide. We now focus on eliminating generators from a column of height 1. ###### Proposition 5.4 (Columns of height 1). Let $\lambda$ be a diagram represented in Figure 5. If $s>t\geq 1$, then we can eliminate ${n-s+t\choose t}$ square-free monomial generators of ${\mathcal{I}}_{\lambda}$ coming from the last column. ###### Proof. Note that as $n-s>b_{t}$ (see Figure 5), from the proof of Theorem 4.1 we know that $e_{n-s}(n-t)\in{\mathcal{I}}_{\lambda}$. We now claim that we can drop monomial generators of the form $\begin{array}[]{ll}e_{n-s}(S_{1,2,\ldots,s-t,i_{1},\ldots,i_{t}}),&\ s-t<i_{1}<i_{2}<\ldots<i_{t}\leq n\end{array}$ from the generating set for ${\mathcal{I}}_{\lambda}$. Since there are ${n-s+t\choose t}$ such choices for sets $\\{i_{1},\ldots,i_{t}\\}$, this will settle the statement of the proposition. But this follows from the trivial identity $e_{k}(A)=\sum_{\begin{subarray}{c}J\subseteq A\\\ |J|=k\end{subarray}}e_{k}(J),$ which implies $e_{n-s}(S_{1,2,\ldots,s-t,i_{1},\ldots,i_{t}})=e_{n-s}(S_{i_{1},\ldots,i_{t}})-\sum_{\stackrel{{\scriptstyle\scriptstyle{\mbox{$\scriptstyle\\{j_{1},\ldots,j_{s-t}\\}\neq\\{1,\ldots,s-t\\}$}}}}{{\scriptstyle{\mbox{$\scriptstyle\\{j_{1},\ldots,j_{s-t}\\}\cap\\{i_{1},\ldots,i_{t}\\}=\emptyset$}}}}}e_{n-s}(S_{j_{1},\ldots,j_{s-t},i_{1},\ldots,i_{t}})\in{\mathcal{I}}_{\lambda}.$ ∎ Therefore using Propositions 5.1 and 5.4, we have reduced our generating set to that in the table in Figure 8, using the Vandermonde identity ${n\choose k}={n-1\choose k-1}+{n-1\choose k}$. $\begin{array}[]{l|l|l}{\bf Column}&{\bf Generators}&{\bf Number}\\\ \hline\cr&&\\\ 0&e_{1}(n),\ldots,e_{b_{1}-1}(n)&b_{1}-1=\lambda^{\prime}_{1}-1\\\ 1&x_{1}^{b_{1}},\ldots,x_{n}^{b_{1}}&{n\choose 1}\hfill={n-1\choose 1}+1\\\ 2&e_{b_{2}}(n-2)&{n\choose 2}-{n-1\choose 1}-1\hfill={n-1\choose 2}-1\\\ \vdots&\hskip 18.06749pt\vdots&\hskip 25.29494pt\vdots\\\ t&e_{b_{t}}(n-t)&{n\choose t}-{n-1\choose t-1}-1\hfill={n-1\choose t}-1\\\ s\ (\mbox{if }s>t)&e_{n-s}(n-s)&{n\choose s}-{n-s+t\choose t}\end{array}$ Figure 8: Number of generators in each degree in the reduced generating set for ${\mathcal{I}}_{\lambda}$ ###### Example 5.5. Consider the partition $\lambda=(4,4,2,1)$ in Figure 2. Our formula gives 177 generators, but in fact, Macaulay2 verifies that 168 generators are enough. The extra generators are in degree 7 (see table in Figure 8): Degrees | Number of generators from Table 8 | Actual number of generators required ---|---|--- 1, 2, 3 | 1 in each degree | 1 in each degree 4 | 11 | 11 6 | 44 | 44 7 | 119 | 110 While in many examples such as the previous one, the predictions of the diagonal version of Weyman’s conjecture are correct, this is not always the case. ###### Example 5.6. Consider the partition $\lambda=(5,4,1)$. Figure 9: The partition $\lambda=(5,4,1)$ We denote by $\mathcal{I}_{01}=(e_{1}(10),e_{2}(10),x_{1}^{3},\ldots,x_{10}^{3})$ the ideal generated by the elements of the $0$-th and $1$-st column. Now consider $e_{4}(8)$ coming from the second column. Let $A\subseteq\\{1,\ldots,n\\}$ be a subset of of cardinality $8$, and let $B$ be its complement ($|B|=2$). By Proposition 2.2, we have mod ${\mathcal{E}}_{3}(10)$ $\displaystyle e_{4}(A)\equiv h_{4}(B)=m_{(4)}(B)+m_{(3,1)}(B)+m_{(2,2)}(B).$ (24) Among the monomial symmetric polynomials appearing in (24), $m_{(4)}$, and $m_{(3,1)}$ are already in the ${\mathcal{I}}_{01}$, since it contains $x_{1}^{3},\ldots,x_{n}^{3}.$ So from the second column we only need to add the set $m_{(2,2)}(2)$ to the generators of $\mathcal{I}_{01}$ to obtain a bigger ideal denoted $\mathcal{I}_{012}$ included in ${\mathcal{I}}_{\lambda}$. That is, we need to add all generators of the form $(x_{i}x_{j})^{2}$ for $i<j$. Now let us consider $e_{5}(A)$, where $|A|=7$ and $B$ is its complement. From the third column $-e_{5}(A)\equiv h_{5}(B)=m_{(5)}(B)+m_{(3,2)}(B)+m_{(4,1)}(B)+m_{(3,1,1)}(B)+m_{(2,2,1)}(B).$ (25) It is clear that each one of these monomial symmetric polynomials is already in the ideal ${\mathcal{I}}_{012}$. In fact, every monomial in the first four summands in (25) contains a power $x_{i}^{3}$, and each element in $m_{(2,2,1)}(B)$ can be obtained as a combination of elements in $m_{(2,2)}(2)$. Hence the third column will not contribute any new generator. The same happens for the last column. Let $|A|=6$ and $B$ be its complement, $|B|=4$. Then $\displaystyle e_{6}(A)=h_{6}(B)$ $\displaystyle=$ $\displaystyle m_{(6)}(B)+m_{(5,1)}(B)+m_{(4,2)}(B)+m_{(3,3)}(B)$ $\displaystyle+$ $\displaystyle m_{(4,1,1)}(B)+m_{(3,2,1)}(B)+m_{(2,2,2)}(B)$ $\displaystyle+$ $\displaystyle m_{(3,1,1,1)}(B)+m_{(2,2,1,1)}(B),$ and all monomials in this sum are already in the ideal, since they contain either a power $x_{i}^{3}$, or a monomial $(x_{i}x_{j})^{2}$. So we have ${\mathcal{I}}_{\lambda}={\mathcal{I}}_{012}$. ###### Counterexample 5.7 (Counterexample to the diagonal version of Weyman’s conjecture). Example 5.6 proves that the generating set $\tilde{\mathcal{W}}_{\lambda}$ for ${\mathcal{I}}_{\lambda}$ coming from the minimal generating set for ${\mathcal{J}}_{\lambda}$ conjectured by Weyman is not in general minimal (see Question 4.6). More precisely, according to his diagram in Figure 10, some generators of degree $5$ and $6$ should be needed, while they are not, as we just showed. In Figure 10 the coordinates of the underlined $X$’s label the generators of ${\mathcal{I}}_{\lambda}$ arising from the diagonal version of Weyman’s conjecture. The generators coming from the shaded $X$’s are not needed. This is the convention that we shall use later as well. $\displaystyle\begin{matrix}p=1&\underline{X}&&&&\\\ p=2&\underline{X}&&&&\\\ p=3&\underline{X}&\underline{X}&&&\\\ p=4&X&X&\underline{X}&&\\\ p=5&X&X&X&\hbox{\pagecolor{shade}$\underline{X}$}&\\\ p=6&X&X&X&X&\hbox{\pagecolor{shade}$\underline{X}$}\\\ p=7&X&X&X&X&\\\ p=8&X&X&X&&\\\ p=9&X&X&&&\\\ p=10&X&&&&\\\ i=&0&1&2&3&4\\\ \end{matrix}$ Figure 10: Weyman diagram for $\lambda=(5,4,1)$. It might be possible to generalize the reasoning used in Example 5.6 with an algorithm, as explained below. ###### Algorithm 5.8. Consider the Young diagram of $\lambda$ filled with the regular filling. Let $b_{1},\ldots,b_{s}$ be the top-cell entries of $\lambda$ as in Figure 5. Set $\mathcal{G}_{0}=\\{e_{1}(n),\ldots,e_{b_{1}-1}(n)\\}$, and create a list of partitions $L_{0}=\emptyset.$ For all $k\geq 1$, define $U_{k}=\\{\mu\vdash b_{k}\,|\,\ell(\mu)\leq k\ \text{ and}\ \nu\not\subseteq\mu,\ \mbox{for any}\ \nu\in L_{k-1}\\},$ where $\nu\subseteq\mu$ means that the Young diagram of $\nu$ is contained in that of $\mu$. * 1) If $|U_{k}|=1$, say $U_{k}=\\{\theta\\}$, then $L_{k}=L_{k-1}\cup\\{\theta\\}$ and $\mathcal{G}_{k}=\mathcal{G}_{k-1}\cup m_{\theta}(k)$. * 2) If $|U_{k}|=0$, then $\mathcal{G}_{k}=\mathcal{G}_{k-1}$ and $L_{k}=L_{k-1}$. * 3) If $|U_{k}|>1$, then $\mathcal{G}_{k}=\mathcal{G}_{k-1}\bigcup\big{(}\bigcup_{l\geq k}h_{b_{l}}(l)\big{)}$, and stop. Denote by $\mathcal{G}$ the set produced by the algorithm at the last step. ###### Question 5.9. Is the set $\mathcal{G}$ a generating set for ${\mathcal{I}}_{\lambda}$? Clearly this algorithm produces a subset of the generating set given by the Theorem 4.1. All generators coming from cells labeled $b_{k}$ satisfying condition $2)$ in the above algorithm would become redundant. We used this algorithm to produce generating sets for all families of examples and counterexamples considered in the next section. Then, we proceeded to prove their correctness on a one by one basis. A proof of the correctness of the algorithm would be greatly welcomed. ## 6 Families of examples and a counterexample to Weyman’s conjecture We conclude the paper by producing simple generating sets for some particular families of shapes. In particular, this allows us to construct two infinite families of counterexamples to the diagonal version of Weyman’s conjecture (Question 4.6), as well as a counterexample to the original conjecture of Weyman for a minimal generating set of the ideal $\mathcal{J}_{\lambda}$ (see Conjecture 4.5). ###### Example 6.1 (The case of two-column partitions). As mentioned above a partition of $n$ of the form $\lambda=(2^{a},1^{c})$, where $a+c=\ell=\ell(\lambda)$ the length of the partition, ${\mathcal{I}}_{\lambda}$ is generated by $e_{1}(n),\ldots,e_{\ell-1}(n),$ $x_{1}^{\ell},\ldots,x_{n}^{\ell}$. ###### Theorem 6.2 (The case of partially-rectangular partitions). Let $\lambda$ be a partition of $n$, and let $k>2$ be any integer. If columns $0,1,\ldots,k-1$ of the Young diagram have the same height, then in the generating set for the ideal ${\mathcal{I}}_{\lambda}$ described in Theorem 4.1 generators coming from columns $2,\ldots,k$ are redundant. ###### Proof. The regular filling of the partition $\lambda$ has the following form. ${{{{{{{{{{{{{{{{{{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle g+1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 2g+1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle g+2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 2g+2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle kg+1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 2g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle 3g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle n$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to20.0pt{\hrule height=0.3pt\vss\hbox to20.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}\kern-0.3pt\cr}}\end{matrix}$ By Theorem 4.1 and Proposition 2.2, modulo the previous columns, the generators coming from Column $k$ are of the form $h_{kg+1}=\sum_{a_{1}+\ldots+a_{k}=kg+1}x_{j_{1}}^{a_{1}}\ldots x_{j_{k}}^{a_{k}}$ where $1\leq j_{1}\leq\ldots\leq j_{k}\leq n$. Consider a term $x_{j_{1}}^{a_{1}}\ldots x_{j_{k}}^{a_{k}}$ in the sum above. We claim that for at least one power $a_{i}$, $a_{i}\geq g+1$, making this monomial redundant in the presence of the second column generators, which are the $(g+1)$-st powers of the variables. To see this, suppose $a_{1}\leq g,\ldots,a_{k}\leq g$. Then we should have that $kg+1=a_{1}+\ldots+a_{k}\leq kg$ which is a contradiction. ∎ ###### Remark 6.3. Drawing the Weyman diagram associated to partially rectangular partitions considered in Theorem 6.2, one can see that the points $(0,1),$ $(1,g+1),$ $(2,2g+1),\ldots,(k,kg+1)$ are collinear because they can successively obtained by adding the vector $(1,g)$. Therefore, the diagonal version of Weyman’s conjecture predicts that the generators coming from cells $(2,2g+1),\ldots,(k,kg+1)$ are redundant. This is true: in fact these are precisely the redundant cells according to Theorem 6.2. ###### Example 6.4. Let $\lambda=(5,4,4,3)$ be the partition in Example 4.2. Theorem 6.2 implies that the generating set for ${\mathcal{I}}_{\lambda}$ consists of the elements in the second column in the table below (compare with Example 4.2), and the reduced number from the table in Figure 8 is in the third column. No $7$ and $10$-degree generators are needed in the generating set. In this case the prediction of the diagonal version of Weyman’s conjecture was correct: cells $(2,7)$ and $(3,10)$ are redundant; see Figure 11. Column | Generators | Numbers from Figure 8 ---|---|--- 0 | $e_{1}(16),e_{2}(16),e_{3}(16)$ | 3 1 | $x_{1}^{4},\ldots,x_{16}^{4}$ | 16 2 | redundant | – 3 | redundant | – 4 | $e_{12}(12)$ | 1365 | Total | 1384 $\displaystyle\begin{matrix}1&\underline{X}&&&&\\\ 2&\underline{X}&&&&\\\ 3&\underline{X}&&&&\\\ 4&\underline{X}&\underline{X}&&&\\\ 5&X&X&&&\\\ 6&X&X&&&\\\ 7&X&X&\hbox{\pagecolor{shade}$\underline{X}$}&&\\\ 8&X&X&X&&\\\ 9&X&X&X&&\\\ 10&X&X&X&\hbox{\pagecolor{shade}$\underline{X}$}&\\\ 11&X&X&X&X&\\\ 12&X&X&X&X&\underline{X}\\\ 13&X&X&X&X&\\\ 14&X&X&X&&\\\ 15&X&X&&&\\\ 16&X&&&&\\\ i=&0&1&2&3&4\end{matrix}$ Figure 11: An example of a partially–rectangular partition $\lambda=(5,4,4,3)$. ###### Corollary 6.5 (The case of rectangular partitions). For a rectangular partition of $n$ of the form $\lambda=(u^{\ell})$, the generating set of ${\mathcal{I}}_{\lambda}$ will simply be $e_{1}(n),\ldots,e_{\ell-1}(n),x_{1}^{\ell},\ldots,x_{n}^{\ell}$, where $n=u\,\ell.$ ###### Corollary 6.6 (The case of two-row partitions). For a two-row partition of $n$ of the form $\lambda=(u,v)$, a generating set is given by $e_{1}(n)$, $x_{1}^{2},\ldots,x_{n}^{2}$, and $e_{u}(u)$. ###### Theorem 6.7. Let $\lambda$ be a partition of $n$. 1. 1. If $\lambda=(u^{a},(u-1)^{c})$ with $g=a+c$, then a generating set of ${\mathcal{I}}_{\lambda}$ is given by $e_{1}(n),\ldots,e_{g-1}(n),x_{1}^{g},\ldots,x_{n}^{g}.$ 2. 2. If $\lambda=(u^{a},(u-1)^{c},1)$ with $u\geq 3$ and $g=a+c>1$, then ${\mathcal{I}}_{\lambda}$ is generated by $e_{1}(n),\ldots,e_{g}(n),x_{1}^{g+1},\ldots,x_{n}^{g+1},(x_{1}x_{2})^{g},(x_{1}x_{3})^{g},\ldots,(x_{n-1}x_{n})^{g}.$ 3. 3. If $\lambda=(u^{a},(u-1)^{c},1,1)$ with $u\geq 4$ and $g=a+c+1>2$, then ${\mathcal{I}}_{\lambda}$ is generated by $\displaystyle e_{1}(n),\ldots,e_{g}(n),x_{1}^{g+1},\ldots,x_{n}^{g+1},(x_{i}+x_{j})(x_{i}x_{j})^{g-1}\text{ for all $i\neq j$},\text{and }(x_{i}x_{j}x_{k})^{g-1}\text{ for all $i<j<k$}.$ ###### Proof. 1. 1. This is an easy consequence of Theorem 6.2. 2. 2. The regular filling of $(u^{a},(u-1)^{c},1)$ will be of the form: ${{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle g+1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 2g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 3g-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{lg-l}}}{{\scriptscriptstyle{+2}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{(u-1)g}}}{{\scriptscriptstyle{-u+3}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n-2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n-3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n-l$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{n-u}}}{{\scriptscriptstyle{+1}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}\kern-0.3pt\cr}}\end{matrix}$ Columns $0$ and $1$ clearly provide the generators $e_{1}(n),\ldots,e_{g}(n),x_{1}^{g+1},\ldots x_{n}^{g+1}$. By Proposition 2.2, Column $2$ provides generators of the form $h_{2g}=\sum_{a+b=2g}x_{i}^{a}x_{j}^{b}$ for $1\leq i<j\leq n$. Since we already have $x_{i}^{g+1}$ and $x_{j}^{g+1}$ in the ideal, this sum reduces to the monomial $x_{i}^{g}x_{j}^{g}$. Hence the third column provides the remaining generators $(x_{1}x_{2})^{g},(x_{1}x_{3})^{g},\ldots,(x_{n-1}x_{n})^{g}$. It remains to show that the generators coming from Columns $3,\ldots,u-1$ are redundant. Let $l$ be any integer such that $3\leq l\leq u-1$. The generators from Column $l$, by Proposition 2.2 and the fact that we have all $(g+1)$-st powers of the variables in the ideal, are of the form $h_{lg-l+2}=\sum_{\stackrel{{\scriptstyle\scriptstyle{a_{1}+\cdots+a_{l}=lg-l+2}}}{{\scriptstyle{a_{1},\ldots,a_{l}\leq g}}}}x_{i_{1}}^{a_{1}}\ldots x_{i_{l}}^{a_{l}}$ where $1\leq i_{1}<i_{2}<\ldots<i_{l}\leq n$, and in each monomial $x_{i_{1}}^{a_{1}}\ldots x_{i_{l}}^{a_{l}}$ at most one of the powers $a_{u}$ is equal to $g$. For such a monomial in the sum, we therefore have $a_{1}+\cdots+a_{l}\leq(l-1)(g-1)+g=lg-l+1\Longrightarrow lg-l+2\leq lg-l+1$ which is a contradiction. So there is no generator from Column $l$ if $l\geq 3$. 3. 3. The regular filling of $(u^{a},(u-1)^{c},1,1)$ will be of the following form. ${{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle g+1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 2g-1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle 3g-3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{lg-2l}}}{{\scriptscriptstyle{\scriptscriptstyle{+3}}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle g$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\stackrel{{\scriptstyle\scriptscriptstyle{(u-1)g}}}{{\scriptscriptstyle{-2u+5}}}$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle n$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt}\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle\cdots$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to21.0pt{\hrule height=0.3pt\vss\hbox to21.0pt{\hss$\scriptstyle$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}\kern-0.3pt\cr}}\end{matrix}$ Again Columns $0$ and $1$ provide the generators $e_{1}(n),\ldots,e_{g}(n),x_{1}^{g+1},\ldots x_{n}^{g+1}$. By Proposition 2.2, Column $2$ provides generators of the form $h_{2g-1}=\sum_{a+b=2g-1}x_{i}^{a}x_{j}^{b}$ for $1\leq i<j\leq n$. Since we already have $x_{i}^{g+1}$ and $x_{j}^{g+1}$ in the ideal, we can additionally assume that $a,b\leq g$ for each monomial $x_{i}^{a}x_{j}^{b}$ in the sum, and so at least one of $a$ or $b$ would have to be $g-1$ and the other $g$. This produces a generator of the form $x_{i}^{g}x_{j}^{g-1}+x_{i}^{g-1}x_{j}^{g}=(x_{i}+x_{j})(x_{i}x_{j})^{g-1}$. Similarly, Column $3$ will produce generators of the form $h_{3g-3}=\sum_{a+b+c=3g-3}x_{i}^{a}x_{j}^{b}x_{k}^{c}$ for $1\leq i<j<k\leq n$. Once more, we can assume that $a,b,c\leq g$, which reduces the sum above to $\begin{array}[]{l}x_{i}^{g-1}x_{j}^{g-1}x_{k}^{g-1}+x_{i}^{g-2}(x_{j}^{g}x_{k}^{g-1}+x_{j}^{g-1}x_{k}^{g})+x_{j}^{g-2}(x_{i}^{g}x_{k}^{g-1}+x_{i}^{g-1}x_{k}^{g})+x_{k}^{g-2}(x_{i}^{g}x_{j}^{g-1}+x_{i}^{g-1}x_{j}^{g})\\\ \\\ =x_{i}^{g-1}x_{j}^{g-1}x_{k}^{g-1}+x_{i}^{g-2}x_{j}^{g-1}x_{k}^{g-1}(x_{j}+x_{k})+x_{j}^{g-2}x_{i}^{g-1}x_{k}^{g-1}(x_{i}+x_{k})+x_{k}^{g-2}x_{i}^{g-1}x_{j}^{g-1}(x_{i}+x_{j}).\end{array}$ The last three summands are in the ideal already (coming from Column $2$), so the generators from Column $3$ can all be written as $x_{i}^{g-1}x_{j}^{g-1}x_{k}^{g-1}$ for $1\leq i<j<k\leq n$. We now need to show that generators coming from Column $l$, where $4\leq l\leq u-1$ are redundant. The generators from Column $l$, by Proposition 2.2 and the fact that we have all $(g+1)$-st powers of the variables in the ideal, are of the form $h_{lg-2l+3}=\sum_{\stackrel{{\scriptstyle\scriptstyle{a_{1}+\cdots+a_{l}=lg-2l+3}}}{{\scriptstyle{a_{1},\ldots,a_{l}\leq g}}}}x_{i_{1}}^{a_{1}}\ldots x_{i_{l}}^{a_{l}}$ where $1\leq i_{1}<i_{2}<\ldots<i_{l}\leq n$. Suppose that $M=x_{i_{1}}^{a_{1}}\ldots x_{i_{l}}^{a_{l}}$ is a monomial in this sum. If one of the powers, say $a_{1}$, is equal to $g$, then we must have another power among $a_{2},\ldots,a_{l}$ that is $g$ or $g-1$. If not, all of $a_{2},\ldots,a_{l}$ are $\leq g-2$, and we have $lg-2l+3=a_{1}+\cdots+a_{l}\leq g+(l-1)(g-2)=lg-2l+2$ which is a contradiction. So there is at least another power, say $a_{2}$, such that $a_{2}\geq g-1$. * • $a_{1}=a_{2}=g$. In this case, we can write $\begin{array}[]{ll}x_{i_{1}}^{g}x_{i_{2}}^{g}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}&=(x_{i_{1}}+x_{i_{2}})(x_{i_{1}}x_{i_{2}})^{g-1}[1/2x_{i_{2}}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}+1/2x_{i_{1}}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}]\\\ &-1/2x_{i_{1}}^{g+1}x_{i_{2}}^{g-1}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}-1/2x_{i_{1}}^{g-1}x_{i_{2}}^{g+1}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}\end{array}$ All the terms on the right-hand side are already in the ideal, and hence so is $x_{i_{1}}^{g}x_{i_{2}}^{g}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}$. * • $a_{1}=g$ and $a_{2}=g-1$. In this case, there is another monomial $M^{\prime}=x_{i_{1}}^{g-1}x_{i_{2}}^{g}x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}$ in the sum as well, and there is exactly one copy of $M$ and one copy of $M^{\prime}$ in the sum. Now we have $M+M^{\prime}=(x_{i_{1}}+x_{i_{2}})x_{i_{1}}^{g-1}x_{i_{2}}^{g-1}(x_{i_{3}}^{a_{3}}x_{i_{4}}^{a_{4}}...x_{i_{l}}^{a_{l}}).$ So each such monomial $M$ is paired with a unique monomial $M^{\prime}$ in the sum, and their sum is already in the ideal. Now assume that all the powers $a_{1},\ldots,a_{l}$ are $\leq g-1$. If $l-2$ of the powers $a_{1},\ldots,a_{l}$ are $\leq g-2$, then we have $lg-2l+3=a_{1}+\cdots+a_{l}\leq(l-2)(g-2)+2(g-1)=lg-2l+2$ which is a contradiction. So there are at least 3 powers among $a_{1},\ldots,a_{l}$ that are equal to $g-1$. But then the monomial $x_{i_{1}}^{a_{1}}\ldots x_{i_{l}}^{a_{l}}$ is already in ${\mathcal{I}}_{\lambda}$, because it is a multiple of a generator coming from Column $3$. ∎ ###### Corollary 6.8. Suppose that the first $l+1$ columns of a partition $\lambda$ belong to one of the three families of shapes described in Theorem 6.7. Then * a) In cases 1 and 2, the generators coming from Columns $3,\ldots,l$ are redundant. For Columns $0,1,2$ we can use the generators described in Theorem 6.7. * b) In Case 3, the generators coming from columns $4,\ldots,l$ are redundant. For Columns $0,1,2,3$ we can use the generators described in Theorem 6.7. ###### Counterexample 6.9 (Counterexamples to the diagonal version of Weyman’s conjecture). The two infinite families of partitions described in parts 2 and 3 of Theorem 6.7 are counterexamples to the diagonal version of Weyman’s conjecture. Indeed, according to it, all generators coming from each of the top cells of their diagrams should be necessary because for $k>0$, the top cells are collinear (for the first family we can move from one top cell to the next one by adding the vector $(1,g-1)$, and for the second family, by adding the vector $(1,g-2)$). But the line containing those points does not pass through $(0,1)$. Instead it passes through $(0,2)$ for the first family, and through $(0,3)$ for the second family. Let $\lambda$ be a partition such that its first $l$ columns belong to one of the two families of shapes described above, with $l>2$ for the first family and $l>3$ for the second one. The preceding corollary shows that the generators coming from Column $k$, with $3<k\leq l$ are redundant. We conclude that each such $\lambda$ is a counterexample to the diagonal version of Weyman’s conjecture. A first counterexample was shown in Counterexample 5.7. ###### Example 6.10. Consider the partition $(5,5,1,1)$ that fits inside one of the families in Theorem 6.7. As proved in that theorem, the cell containing 7 is redundant. Translated into the Weyman diagram, this means that the ${\underline{X}}$ in position $(4,7)$ is redundant (see Figure 12). ${{{{{{{{{{{{{\displaystyle\begin{matrix}\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\vbox{#}\kern-\Thickness\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 1$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 2$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 3$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 4$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 5$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 6$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 7$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 12$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 11$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 10$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 9$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt&\hbox{\vrule width=0.3pt\vbox to12.0pt{\hrule height=0.3pt\vss\hbox to12.0pt{\hss$\scriptstyle 8$\hss} \vss\hrule height=0.3pt} \vrule width=0.3pt} \kern-0.3pt }\kern-0.3pt\cr}\kern-0.3pt\cr}}\end{matrix}\hskip 28.45274pt\begin{matrix}1&\underline{X}&&&&&\\\ 2&\underline{X}&&&&&\\\ 3&\framebox{$\underline{X}$}&&&&&\\\ 4&\underline{X}&\framebox{$\underline{X}$}&&&&\\\ 5&X&X&\framebox{$\underline{X}$}&&&\\\ 6&X&X&X&\framebox{$\underline{X}$}&&\\\ 7&X&X&X&X&{\framebox{\hbox{\pagecolor{shade}$\underline{X}$}}}&\\\ 8&X&X&X&X&X&\\\ 9&X&X&X&X&&\\\ 10&X&X&X&&&\\\ 11&X&X&&&&\\\ 12&X&&&&&\\\ i=&0&1&2&3&4\end{matrix}$ Figure 12: The regular filling and the Weyman diagram of $\lambda=(5,5,1,1)$. The following table, computed with Macaulay2, confirms our prediction that the $275$ degree $7$ generators that should be in the generating set according to the diagonal version of the conjecture, are not needed. Degrees | Minimal number of generators ---|--- 1, 2, 3 | 1 in each degree 4 | 12 5 | 54 6 | 154 7 | redundant Theorems 6.2 and 6.7 can be reformulated in a suggestive geometrical way as special instances of the following statement. ###### Question 6.11. Let $\lambda$ be a partition and draw the Weyman diagram of $\lambda$. If the $X^{\prime}s$ at the top of columns $1,2,\ldots,r$ are collinear, and the line containing them passes through the point $(0,k)$, then are the generators coming from columns $k+1,\ldots,r$ redundant ? We have evidence that suggests that this statement is true: it was proven to be true when $k=1$ in Theorem 6.2, for $k=2$ in Theorem 6.7 Part 1, and for $k=3$ in Theorem 6.7 Part 2, (see Figure 12: the collinear $X$’s have been surrounded). For $k=4$, we used Macaulay2 to verify whether the statement is still true for the smallest possible member of this family, the partition (6,5,1,1,1) (see Figure 13). As predicted, all degree $9$ generators are redundant. $\displaystyle\begin{matrix}1&\underline{X}&&&&&\\\ 2&\underline{X}&&&&&\\\ 3&\underline{X}&&&&&\\\ 4&\framebox{$\underline{X}$}&&&&&\\\ 5&\underline{X}&\framebox{$\underline{X}$}&&&&\\\ 6&X&X&\framebox{$\underline{X}$}&&&\\\ 7&X&X&X&\framebox{$\underline{X}$}&&\\\ 8&X&X&X&X&\framebox{$\underline{X}$}&\\\ 9&X&X&X&X&X&{\framebox{\hbox{\pagecolor{shade}$\underline{X}$}}}\\\ 10&X&X&X&X&X&\\\ 11&X&X&X&X&&\\\ 12&X&X&X&&&\\\ 13&X&X&&&&\\\ 14&X&&&&&\\\ i=&0&1&2&3&4&5\end{matrix}$ Figure 13: An evidence regarding the statement in Question 6.11 for $\lambda=(6,5,1,1,1)$ Degrees | Minimal number of generators ---|--- 1, 2, 3, 4 | 1 in each degree 5 | 14 6 | 77 7 | 273 8 | 637 9 | redundant ### 6.1 Weyman’s original conjecture To finish our work, we focus our attention at the original conjecture of Weyman. It seems plausible that those partitions that give counterexamples to the diagonal version of Weyman’s conjecture are also counterexamples to Weyman’s original conjecture. We used Macaulay2 to verify if this was the case for the smallest shape in the families described in Counterexample 6.9. ###### Counterexample 6.12 (Counterexample to Weyman’s original conjecture). Consider the partition $(4,3,1)$ whose Weyman diagram is represented in Figure 14. The points $(1,3),(2,4)$ and $(3,5)$ are collinear, but the line that contains them does not pass through $(0,1)$. So according to Weyman’s conjecture, all these cells contribute generators to a minimal generating set of $\mathcal{J}_{(4,3,1)}$. However, Theorem 6.7 suggests that the generators coming from cell $(3,5)$ may be redundant. $\displaystyle\begin{matrix}1&\underline{X}&&&&&\\\ 2&\framebox{$\underline{X}$}&&&&&\\\ 3&\underline{X}&\framebox{$\underline{X}$}&&&&\\\ 4&X&X&\framebox{$\underline{X}$}&&&\\\ 5&X&X&X&{\framebox{\hbox{\pagecolor{shade}$\underline{X}$}}}&&\\\ 6&X&X&X&&&\\\ 7&X&X&&&&\\\ 8&X&&&&&\\\ i=&0&1&2&3\end{matrix}$ Figure 14: A counterexample to Weyman’s original conjecture: $(4,3,1)$. Using Macaulay 2, we computed the minimal generating set for $\mathcal{J}_{(4,3,1)}$ and verified that this is indeed the case. We conclude that $(4,3,1)$ is a counterexample to Weyman’s original conjecture. Degrees | Weyman’s conjecture | Minimal number of generators ---|---|--- 1 | 1 | 1 2 | 1 | 1 3 | 64 | 64 4 | 720 | 720 5 | 2352 | redundant Total | 3138 | 786 To summarize, in this particular case, Weyman’s conjecture predicts that we need 3138 generators, but only $786$ of them are really necessary. Unfortunately, even large servers were not able to handle slightly larger examples, so at this point we do not know if other partitions in the families described earlier are counterexamples to Weyman’s original conjecture. We end the paper with a natural question. ###### Question 6.13. Does the statement of Question 6.11 hold for $\mathcal{J}_{\lambda}$ ? ## Acknowledgments We wish to thank Jerzy Weyman for many interesting conversations and suggestions, as well as his interest in our project. At the same time that we were working on this project, he showed independently that $(4,3,1)$ is indeed a counterexample to his Conjecture 5.1 using geometric reasoning [W3]. Moreover, he has shown that Conjecture 5.1 holds for all the other partitions of $n\leq 9$ except for the three partitions of $9$ belonging to the shapes described in Counterexample 6.9. We also wish to Mark Shimozono pointing out some references and Emmanuel Briand for his help during this project. ## References * [AB] J.-C. Aval and N. Bergeron. _Vanishing ideals of lattice diagram determinants_ , J. Combin. Theory Series A 99 (2002), 244–260. * [BG] N. Bergeron and A. Garsia. _On certain spaces of harmonic polynomials_ , Contemp. Math., 138 (1992), 51–86. * [BFR] R. Biagioli, S. Faridi, and M. Rosas, _De Concini-Procesi ideals indexed by hooks_ , Communications in Algebra, 35 (2007), 3875–3891. * [DP] C. De Concini and C. Procesi, _Symmetric functions, conjugacy classes and the flag variety_ , Invent. Math. 64 (1981), 203–230. * [ES] D. Eisenbud and D. Saltman, _Rank varieties of matrices._ Commutative algebra (Berkeley, CA, 1987), 173–212, Math. Sci. Res. Inst. Publ., 15, Springer, New York, 1989. * [GS] D.R. Grayson and M.E. Stillman, _Macaulay 2, a software system for research in algebraic geometry_ , available at _http://www.math.uiuc.edu/Macaulay2/_. * [GP] A. Garsia and C. Procesi. _On certain graded $S_{n}$-modules and the $q$-Kotska polynomials_, Adv. Math. 94 (1992), 82–138. * [K] B. Kostant, _Lie group representations on polynomial rings._ Amer. J. Math. 85 (1963), 327–404. * [T] T. Tanisaki, _Defining ideals of the closure of conjugacy classes and representations of the Weyl groups_ , Tohoku J. Math. 34 (1982), 575–585. * [W1] J. Weyman, _The equations of conjugacy classes of nilpotent matrices._ Invent. Math. 98 (1989), no. 2, 229–245. * [W2] J. Weyman, _Two results on equations of nilpotent orbits._ J. Algebraic Geom. 11 (2002), no. 4, 791–800. * [W3] J. Weyman, _Private communication_ , August 17, 2007.

arxiv-papers

2024-09-04T02:48:54.168084

0803.0732

# Five-Year Wilkinson Microwave Anisotropy Probe (WMAP11affiliation: WMAP is the result of a partnership between Princeton University and NASA’s Goddard Space Flight Center. Scientific guidance is provided by the WMAP Science Team. ) Observations: Data Processing, Sky Maps, & Basic Results G. Hinshaw 22affiliation: Code 665, NASA/Goddard Space Flight Center, Greenbelt, MD 20771 , J. L. Weiland 33affiliation: Adnet Systems, Inc., 7515 Mission Dr., Suite A100, Lanham, Maryland 20706 , R. S. Hill 33affiliation: Adnet Systems, Inc., 7515 Mission Dr., Suite A100, Lanham, Maryland 20706 , N. Odegard 33affiliation: Adnet Systems, Inc., 7515 Mission Dr., Suite A100, Lanham, Maryland 20706 , D. Larson 44affiliation: Dept. of Physics & Astronomy, The Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218-2686 , C. L. Bennett 44affiliation: Dept. of Physics & Astronomy, The Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218-2686 , J. Dunkley 55affiliation: Dept. of Physics, Jadwin Hall, Princeton University, Princeton, NJ 08544-0708 66affiliation: Dept. of Astrophysical Sciences, Peyton Hall, Princeton University, Princeton, NJ 08544-1001 77affiliation: Astrophysics, University of Oxford, Keble Road, Oxford, OX1 3RH, UK , B. Gold 44affiliation: Dept. of Physics & Astronomy, The Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218-2686 , M. R. Greason 33affiliation: Adnet Systems, Inc., 7515 Mission Dr., Suite A100, Lanham, Maryland 20706 , N. Jarosik 55affiliation: Dept. of Physics, Jadwin Hall, Princeton University, Princeton, NJ 08544-0708 , E. Komatsu 88affiliation: Univ. of Texas, Austin, Dept. of Astronomy, 2511 Speedway, RLM 15.306, Austin, TX 78712 , M. R. Nolta 99affiliation: Canadian Institute for Theoretical Astrophysics, 60 St. George St, University of Toronto, Toronto, ON Canada M5S 3H8 , L. Page 55affiliation: Dept. of Physics, Jadwin Hall, Princeton University, Princeton, NJ 08544-0708 , D. N. Spergel 66affiliation: Dept. of Astrophysical Sciences, Peyton Hall, Princeton University, Princeton, NJ 08544-1001 1010affiliation: Princeton Center for Theoretical Physics, Princeton University, Princeton, NJ 08544 , E. Wollack 22affiliation: Code 665, NASA/Goddard Space Flight Center, Greenbelt, MD 20771 , M. Halpern 1111affiliation: Dept. of Physics and Astronomy, University of British Columbia, Vancouver, BC Canada V6T 1Z1 , A. Kogut 22affiliation: Code 665, NASA/Goddard Space Flight Center, Greenbelt, MD 20771 , M. Limon 1212affiliation: Columbia Astrophysics Laboratory, 550 W. 120th St., Mail Code 5247, New York, NY 10027-6902 , S. S. Meyer 1313affiliation: Depts. of Astrophysics and Physics, KICP and EFI, University of Chicago, Chicago, IL 60637 , G. S. Tucker 1414affiliation: Dept. of Physics, Brown University, 182 Hope St., Providence, RI 02912-1843 , E. L. Wright 1515affiliation: UCLA Physics & Astronomy, PO Box 951547, Los Angeles, CA 90095-1547 Gary.F.Hinshaw@nasa.gov ###### Abstract We present new full-sky temperature and polarization maps in five frequency bands from 23 to 94 GHz, based on data from the first five years of the WMAP sky survey. The new maps are consistent with previous maps and are more sensitive. The five-year maps incorporate several improvements in data processing made possible by the additional years of data and by a more complete analysis of the instrument calibration and in-flight beam response. We present several new tests for systematic errors in the polarization data and conclude that W band polarization data is not yet suitable for cosmological studies, but we suggest directions for further study. We do find that Ka band data is suitable for use; in conjunction with the additional years of data, the addition of Ka band to the previously used Q and V band channels significantly reduces the uncertainty in the optical depth parameter, $\tau$. Further scientific results from the five year data analysis are presented in six companion papers and are summarized in §7 of this paper. With the 5 year WMAP data, we detect no convincing deviations from the minimal 6-parameter $\Lambda$CDM model: a flat universe dominated by a cosmological constant, with adiabatic and nearly scale-invariant Gaussian fluctuations. Using WMAP data combined with measurements of Type Ia supernovae (SN) and Baryon Acoustic Oscillations (BAO) in the galaxy distribution, we find (68% CL uncertainties): $\Omega_{b}h^{2}=0.02267^{+0.00058}_{-0.00059}$, $\Omega_{c}h^{2}=0.1131\pm 0.0034$, $\Omega_{\Lambda}=0.726\pm 0.015$, $n_{s}=0.960\pm 0.013$, $\tau=0.084\pm 0.016$, and $\Delta_{\cal R}^{2}=(2.445\pm 0.096)\times 10^{-9}$ at $k=0.002~{}{\rm Mpc^{-1}}$. From these we derive: $\sigma_{8}=0.812\pm 0.026$, $H_{0}=70.5\pm 1.3$ ${\rm km~{}s^{-1}~{}Mpc^{-1}}$, $\Omega_{b}=0.0456\pm 0.0015$, $\Omega_{c}=0.228\pm 0.013$, $\Omega_{m}h^{2}=0.1358^{+0.0037}_{-0.0036}$, $z_{\rm reion}=10.9\pm 1.4$, and $t_{0}=13.72\pm 0.12\ \mbox{Gyr}$. The new limit on the tensor-to- scalar ratio is $r<0.22\ \mbox{(95\% CL)}$, while the evidence for a running spectral index is insignificant, $dn_{s}/d\ln{k}=-0.028\pm 0.020$ (68% CL). We obtain tight, simultaneous limits on the (constant) dark energy equation of state and the spatial curvature of the universe: $-0.14<1+w<0.12\ \mbox{(95\% CL)}$ and $-0.0179<\Omega_{k}<0.0081\ \mbox{(95\% CL)}$. The number of relativistic degrees of freedom, expressed in units of the effective number of neutrino species, is found to be $N_{\rm eff}=4.4\pm 1.5$ (68% CL), consistent with the standard value of 3.04. Models with $N_{\rm eff}=0$ are disfavored at $>$99.5% confidence. Finally, new limits on physically motivated primordial non-Gaussianity parameters are $-9<f_{NL}^{\rm local}<111$ (95% CL) and $-151<f_{NL}^{\rm equil}<253$ (95% CL) for the local and equilateral models, respectively. cosmic microwave background, cosmology: observations, early universe, dark matter, space vehicles, space vehicles: instruments, instrumentation: detectors, telescopes ††slugcomment: Astrophysical Journal Supplement Series, in press ## 1 INTRODUCTION The Wilkinson Microwave Anisotropy Probe (WMAP) is a Medium-Class Explorer (MIDEX) satellite aimed at elucidating cosmology through full-sky observations of the cosmic microwave background (CMB). The WMAP full-sky maps of the temperature and polarization anisotropy in five frequency bands provide our most accurate view to date of conditions in the early universe. The multi- frequency data facilitate the separation of the CMB signal from foreground emission arising both from our Galaxy and from extragalactic sources. The CMB angular power spectrum derived from these maps exhibits a highly coherent acoustic peak structure which makes it possible to extract a wealth of information about the composition and history of the universe, as well as the processes that seeded the fluctuations. WMAP data (Bennett et al., 2003; Spergel et al., 2003; Hinshaw et al., 2007; Spergel et al., 2007), along with a host of pioneering CMB experiments (Miller et al., 1999; Lee et al., 2001; Netterfield et al., 2002; Halverson et al., 2002; Pearson et al., 2003; Scott et al., 2003; Benoît et al., 2003), and other cosmological measurements (Percival et al., 2001; Tegmark et al., 2004; Cole et al., 2005; Tegmark et al., 2006; Eisenstein et al., 2005; Percival et al., 2007; Astier et al., 2006; Riess et al., 2007; Wood-Vasey et al., 2007) have established $\Lambda$CDM as the standard model of cosmology: a flat universe dominated by dark energy, supplemented by dark matter and atoms with density fluctuations seeded by a Gaussian, adiabatic, nearly scale invariant process. The basic properties of this universe are determined by five numbers: the density of matter, the density of atoms, the age of the universe (or equivalently, the Hubble constant today), the amplitude of the initial fluctuations, and their scale dependence. By accurately measuring the first few peaks in the angular power spectrum and the large-scale polarization anisotropy, WMAP data have enabled the following inferences: * • A precise (3%) determination of the density of atoms in the universe. The agreement between the atomic density derived from WMAP and the density inferred from the deuterium abundance is an important test of the standard big bang model. * • A precise (3%) determination of the dark matter density. (With five years of data and a better determination of our beam response, this measurement has improved significantly.) Previous CMB measurements have shown that the dark matter must be non-baryonic and interact only weakly with atoms and radiation. The WMAP measurement of the density puts important constraints on supersymmetric dark matter models and on the properties of other dark matter candidates. * • A definitive determination of the acoustic scale at redshift $z=1090$. Similarly, the recent measurement of baryon acoustic oscillations (BAO) in the galaxy power spectrum (Eisenstein et al., 2005) has determined the acoustic scale at redshift $z\sim 0.35$. When combined, these standard rulers accurately measure the geometry of the universe and the properties of the dark energy. These data require a nearly flat universe dominated by dark energy consistent with a cosmological constant. * • A precise determination of the Hubble Constant, in conjunction with BAO observations. Even when allowing curvature ($\Omega_{0}\neq 1$) and a free dark energy equation of state ($w\neq-1$), the acoustic data determine the Hubble constant to within 3%. The measured value is in excellent agreement with independent results from the Hubble Key Project (Freedman et al., 2001), providing yet another important consistency test for the standard model. * • Significant constraint of the basic properties of the primordial fluctuations. The anti-correlation seen in the temperature/polarization (TE) correlation spectrum on 4∘ scales implies that the fluctuations are primarily adiabatic and rule out defect models and isocurvature models as the primary source of fluctuations (Peiris et al., 2003). Further, the WMAP measurement of the primordial power spectrum of matter fluctuations constrains the physics of inflation, our best model for the origin of these fluctuations. Specifically, the 5 year data provide the best measurement to date of the scalar spectrum’s amplitude and slope, and place the most stringent limits to date on the amplitude of tensor fluctuations. However, it should be noted that these constraints assume a smooth function of scale, $k$. Certain models with localized structure in $P(k)$, and hence additional parameters, are not ruled out, but neither are they required by the data; see e.g. Shafieloo & Souradeep (2007); Hunt & Sarkar (2007). The statistical properties of the CMB fluctuations measured by WMAP are close to Gaussian; however, there are several hints of possible deviations from Gaussianity, e.g. Eriksen et al. (2007a); Copi et al. (2007); Land & Magueijo (2007); Yadav & Wandelt (2008). Significant deviations would be a very important signature of new physics in the early universe. Large-angular-scale polarization measurements currently provide our best window into the universe at $z\sim 10$. The WMAP data imply that the universe was reionized long before the epoch of the oldest known quasars. By accurately constraining the optical depth of the universe, WMAP not only constrains the age of the first stars but also determines the amplitude of primordial fluctuations to better than 3%. This result is important for constraining the growth rate of structure. This paper summarizes results compiled from 5 years of WMAP data that are fully presented in a suite of 7 papers (including this one). The new results improve upon previous results in many ways: additional data reduces the random noise, which is especially important for studying the temperature signal on small angular scales and the polarization signal on large angular scales; five independent years of data enable comparisons and null tests that were not previously possible; the instrument calibration and beam response have been much better characterized, due in part to improved analyses and to additional years of data; and, other cosmological data have become available. In addition to summarizing the other papers, this paper reports on changes in the WMAP data processing pipeline, presents the 5 year temperature and polarization maps, and gives new results on instrument calibration and on potential systematic errors in the polarization data. Hill et al. (2008) discuss the program to derive an improved physical optics model of the WMAP telescope, and use the results to better determine the WMAP beam response. Gold et al. (2008) present a new analysis of diffuse foreground emission in the WMAP data and update previous analyses using 5 year data. Wright et al. (2008) analyze extragalactic point sources and provide an updated source catalog, with new results on source variability. Nolta et al. (2008) derive the angular power spectra from the maps, including the TT, TE, TB, EE, EB, and BB spectra. Dunkley et al. (2008) produce an updated likelihood function and present cosmological parameter results based on 5 year WMAP data. They also develop an independent analysis of polarized foregrounds and use those results to test the reliability of the optical depth inference to foreground removal errors. Komatsu et al. (2008) infer cosmological parameters by combining 5 year WMAP data with a host of other cosmological data and discuss the implications of the results. Concurrent with the submission of these papers, all 5 year WMAP data are made available to the research community via NASA’s Legacy Archive for Microwave Background Data Analysis (LAMBDA). The data products are described in detail in the WMAP Explanatory Supplement (Limon et al., 2008), also available on LAMBDA. The WMAP instrument is composed of 10 differencing assemblies (DAs) spanning 5 frequencies from 23 to 94 GHz (Bennett et al., 2003): 1 DA each at 23 GHz (K1) and 33 GHz (Ka1), 2 each at 41 GHz (Q1,Q2) and 61 GHz (V1,V2), and 4 at 94 GHz (W1-W4). Each DA is formed from two differential radiometers which are sensitive to orthogonal linear polarization modes; the radiometers are designated 1 or 2 (e.g., V11 or W12) depending on polarization mode. In this paper we follow the notation convention that flux density is $S\sim\nu^{\alpha}$ and antenna temperature is $T\sim\nu^{\beta}$, where the spectral indices are related by $\beta=\alpha-2$. In general, the CMB is expressed in terms of thermodynamic temperature, while Galactic and extragalactic foregrounds are expressed in antenna temperature. Thermodynamic temperature differences are given by $\Delta T=\Delta T_{A}[(e^{x}-1)^{2}/x^{2}e^{x}]$, where $x=h\nu/kT_{0}$, $h$ is the Planck constant, $\nu$ is the frequency, $k$ is the Boltzmann constant, and $T_{0}=2.725$ K is the CMB temperature (Mather et al., 1999). A WMAP band-by- band tabulation of the conversion factors between thermodynamic and antenna temperature is given in Table 1. ## 2 CHANGES IN THE 5 YEAR DATA ANALYSIS The 1 year and 3 year data analyses were described in detail in previous papers. In large part, the 5 year analysis employs the same methods, so we do not repeat a detailed processing description here. However, we have made several improvements that are summarized here and described in more detail later in this paper and in a series of companion papers, as noted. We list the changes in the order they appear in the processing pipeline: * • There is a $\sim 1^{\prime}$ temperature-dependent pointing offset between the star tracker coordinate system (which defines spacecraft coordinates) and the instrument boresights. In the 3 year analysis we introduced a correction to account for the elevation change of the instrument boresights in spacecraft coordinates. With additional years of data, we have been able to refine our thermal model of the pointing offset, so we now include a small ($\mbox{$<$}1^{\prime}$) correction to account for the azimuth change of the instrument boresights. Details of the new correction are given in the 5 year Explanatory Supplement (Limon et al., 2008). * • We have critically re-examined the relative and absolute intensity calibration procedures, paying special attention to the absolute gain recovery obtainable from the modulation of the CMB dipole due to WMAP’s motion. We describe the revised procedure in §4 and note that the sky map calibration uncertainty has decreased from 0.5% to 0.2%. * • The WMAP beam response has now been measured in 10 independent “seasons” of Jupiter observations. In the highest resolution W band channels, these measurements now probe the beam response $\sim$44 dB down from the beam peak. However, there is still non-negligible beam solid angle below this level ($\sim$0.5%) that needs to be measured to enable accurate cosmological inference. In the 3 year analysis we produced a physical optics model of the A-side beam response starting with a pre-flight model and fitting in-flight mirror distortions to the flight Jupiter data. In the 5 year analysis we have extended the model to the B-side optics and, for both sides, we have extended the fit to include distortion modes a factor of 2 smaller in linear scale (4 times as many modes). The model is used to augment the flight beam maps below a given threshold. The details of this work are given in Hill et al. (2008). * • The far sidelobe response of the beam was determined from a combination of ground measurements and in-flight lunar data taken early in the mission (Barnes et al., 2003). For the current analysis, we have replaced a small fraction of the far sidelobe data with the physical optics model described above. We have also made the following changes in our handling of the far sidelobe pickup (Hill et al., 2008): 1) We have enlarged the “transition radius” that defines the boundary between the main beam and the far sidelobe response. This places a larger fraction of the total beam solid angle in the main beam where uncertainties are easier to quantify and propagate into the angular power spectra. 2) We have moved the far sidelobe deconvolution into the combined calibration and sky map solver (§4). This produces a self- consistent estimate of the intensity calibration and the deconvolved sky map. The calibrated time-ordered data archive has had an estimate of the far sidelobe response subtracted from each datum (as it had in the 3 year processing). * • We have updated the optimal filters used in the final step of map-making. The functional form of the filter is unchanged (Jarosik et al., 2007), but the fits have been updated to cover years 4 and 5 of the flight data. * • Each WMAP differencing assembly consists of two radiometers that are sensitive to orthogonal linear polarization states. The sum and difference of the two radiometer channels split the signal into intensity and polarization components, respectively. However, the noise levels in the two radiometers are not equal, in general, so more optimal sky map estimation is possible in theory, at the cost of mixing intensity and polarization components in the process. For the current analysis, we investigated one such weighted algorithm and found that the polarization maps were subject to unacceptable contamination by the intensity signal in cases where the beam response was non-circular and the gradient of the intensity signal was large, e.g., in K band. As a result, we reverted to the unweighted (and unbiased) estimator used in previous work. * • We have improved the sky masks used to reject foreground contamination. In previous work, we defined masks based on contours of the K band data. In the 5 year analysis we produce masks based jointly on K band and Q band contours. For a given sky cut fraction, the new masks exclude flat spectrum (e.g. free- free) emission more effectively. The new masks are described in detail in Gold et al. (2008) and are provided with the 5 year data release. In addition, we have modified the “processing” mask used to exclude very bright sources during sky map estimation. The new mask is defined in terms of low-resolution (r4) HEALPix sky pixels (Gorski et al., 2005) to facilitate a cleaner definition of the pixel-pixel inverse covariance matrices, $N^{-1}$. One side effect of this change is to introduce a few r4-sized holes around the brightest radio sources in the analysis mask, which incorporates the processing mask as a subset. * • We have amended our foreground analysis in the following ways: 1) Gold et al. (2008) perform a pixel-by-pixel analysis of the joint temperature and polarization data to study the breakdown of the Galactic emission into physical components. 2) We have updated some aspects of the Maximum Entropy (MEM) based analysis, as described in Gold et al. (2008). 3) Dunkley et al. (2008) develop a new analysis of polarized foreground emission using a Gibbs sampling approach that yields a cleaned CMB polarization map and an associated covariance matrix. 4) Wright et al. (2008) update the WMAP point source catalog and present some results on variable sources in the 5 year data. However, the basic cosmological results are still based on maps that were cleaned with the same template-based procedure that was used in the 3 year analysis. * • We have improved the final temperature power spectrum, $C_{l}^{TT}$, by using a Gibbs-based maximum likelihood estimate for $l\leq 32$ (Dunkley et al., 2008) and a pseudo-$C_{l}$ estimate for higher $l$ (Nolta et al., 2008). As with the 3 year analysis, the pseudo-$C_{l}$ estimate uses only V- and W-band data. With 5 individual years of data and six V- and W-band differencing assemblies, we can now form individual cross-power spectra from 15 DA pairs within each of 5 years and from 36 DA pairs across 10 year pairs, for a total of 435 independent cross-power spectra. * • In the 3 year analysis we developed a pseudo-$C_{l}$ method for evaluating polarization power spectra in the presence of correlated noise. In the present analysis we additionally estimate the TE, TB, EE, EB, & BB spectra and their errors using an extension of the maximum likelihood method in Page et al. (2007). However, as in the 3 year analysis, the likelihood of a given model is still evaluated directly from the polarization maps using a pixel-based likelihood. * • We have improved the form of the likelihood function used to infer cosmological parameters from the Monte Carlo Markov Chains (Dunkley et al., 2008). We use an exact maximum likelihood form for the $l\leq 32$ TT data (Eriksen et al., 2007c). We have investigated theoretically optimal methods for incorporating window function uncertainties into the likelihood, but in tests with simulated data we have found them to be biased. In the end, we adopt the form used in the 3 year analysis (Hinshaw et al., 2007), but we incorporate the smaller 5 year window function uncertainties (Hill et al., 2008) as inputs. We now routinely account for gravitational lensing when assessing parameters, and we have added an option to use low-$l$ TB and EB data for testing non-standard cosmological models. * • For testing nongaussianity, we employ an improved estimator for $f_{NL}$ (Creminelli et al., 2006; Yadav et al., 2007). The results of this analysis are described in Komatsu et al. (2008). ## 3 OBSERVATIONS AND MAPS The 5 year WMAP data encompass the period from 00:00:00 UT, 10 August 2001 (day number 222) to 00:00:00 UT, 9 August 2006 (day number 222). The observing efficiency during this time is roughly 99%; Table 2 lists the fraction of data that was lost or rejected as unusable. The Table also gives the fraction of data that is flagged due to potential contamination by thermal emission from Mars, Jupiter, Saturn, Uranus, and Neptune. These data are not used in map- making, but are useful for in-flight beam mapping (Hill et al., 2008; Limon et al., 2008). After performing an end-to-end analysis of the instrument calibration, single- year sky maps are created from the time-ordered data using the procedure described by Jarosik et al. (2007). Figure 1 shows the 5 year temperature maps at each of the five WMAP observing frequencies: 23, 33, 41, 61, and 94 GHz. The number of independent observations per pixel, $N_{\rm obs}$, is qualitatively the same as Figure 2 of Hinshaw et al. (2007) and is not reproduced here. The noise per pixel, $p$, is given by $\sigma(p)=\sigma_{0}N_{\rm obs}^{-1/2}(p)$, where $\sigma_{0}$ is the noise per observation, given in Table 1. To a very good approximation, the noise per pixel in the 5 year maps is a factor of $\sqrt{5}$ times lower than in the single-year maps. Figures 2 and 3 show the 5 year polarization maps in the form of the Stokes parameters Q and U, respectively. Maps of the relative polarization sensitivity, the Q and U analogs of $N_{\rm obs}$, are shown in Figure 13 of Jarosik et al. (2007) and are not updated here. A description of the low-resolution pixel-pixel inverse covariance matrices used in the polarization analysis is also given in Jarosik et al. (2007), and is not repeated here. The polarization maps are dominated by foreground emission, primarily synchrotron emission from the Milky Way. Figure 4 shows the polarization maps in a form in which the color scale represents polarized intensity, $P=\sqrt{Q^{2}+U^{2}}$, and the line segments indicate polarization direction for pixels with a signal-to-noise ratio greater than 1. As with the temperature maps, the noise per pixel in the 5 year polarization maps is $\sqrt{5}$ times lower than in the single-year maps. Figure 5 shows the difference between the 5 year temperature maps and the corresponding 3 year maps. All maps have been smoothed to 2∘ resolution to minimize the noise difference between them (due to the additional years of data). The left column shows the difference without any further processing, save for the subtraction of a relative offset between the maps. Table 3 gives the value of the relative offset in each band. Recall that WMAP is insensitive to absolute temperature, so we adopt a convention that sets the zero level in each map based on a model of the foreground emission at the galactic poles. While we have not changed conventions, our 3 year estimate was erroneous due to the use of a preliminary CMB signal map at the time the estimate was made. This error did not affect any cosmological results, but it probably explains the offset differences noted by Eriksen et al. (2007b) in their recent analysis of the 3 year data. The dominant structure in the left column of Figure 5 consists of a residual dipole and galactic plane emission. This reflects the updated 5 year calibration which has produced changes in the gain of order 0.3% compared to the 3 year gain estimate (see §4 for a more detailed discussion of the calibration). Table 3 gives the dipole amplitude difference in each band, along with the much smaller quadrupole and octupole power difference. (For comparison, we estimate the CMB power at $l=2,3$ to be $l(l+1)C_{l}/2\pi=211,1041$ $\mu$K2, respectively.) The right column of Figure 5 shows the corresponding sky map differences after the 3 year map has been rescaled by a single factor (in each band) to account for the mean gain change between the 3 and 5 year calibration determinations. The residual galactic plane structure in these maps is less than 0.2% of the nominal signal in Q band, and less than 0.1% in all the other bands. The large scale structure in the band-averaged temperature maps is quite robust. ### 3.1 CMB Dipole The dipole anisotropy stands apart from the rest of the CMB signal due to its large amplitude and to the understanding that it arises from our peculiar motion with respect to the CMB rest frame. In this section we present CMB dipole results based on a new analysis of the 5 year sky maps. Aside from an absolute calibration uncertainty of 0.2% (see §4), the dominint source of uncertainty in the dipole estimate arises from uncertainties in Galactic foreground subtraction. Here we present results for two different removal methods: template-based cleaning and an internal linear combination (ILC) of the WMAP multifrequency data (Gold et al., 2008). Our final results are based on a combination of these methods with uncertainties that encompass both approaches. With template-based foreground removal, we can form cleaned maps for each of the 8 high frequency DA’s, Q1-W4, while the ILC method produces one cleaned map from a linear combination of all the WMAP frequency bands. We analyze the residual dipole moment in each of these maps (a nominal dipole based on the 3 year data is subtracted from the time-ordered data prior to map-making) using a Gibbs sampling technique which generates an ensemble of full-sky CMB realizations that are consistent with the data, as detailed below. We evaluate the dipole moment of each full-sky realization and compute uncertainties from the scatter of the realizations. We prepared the data for the Gibbs analysis as follows. The $N_{\rm side}=512$, template-cleaned maps were zeroed within the KQ85 mask, smoothed with a $10^{\circ}$ FWHM Gaussian kernel, and degraded to $N_{\rm side}=16$. Zeroing the masked region prior to smoothing prevents residual cleaning errors within the mask from contaminating the unmasked data. We add random white noise (12 $\mu$K rms per pixel) to each map to regularize the pixel-pixel covariance matrix. The $N_{\rm side}=512$ ILC map was also smoothed with a $10^{\circ}$ FWHM Gaussian kernel and degraded to $N_{\rm side}=16$, but the data within the sky mask were not zeroed prior to smoothing. We add white noise of 6 $\mu$K per pixel to the smoothed ILC map to regularize its covariance matrix. Note that smoothing the data with a $10^{\circ}$ kernel reduces the residual dipole in the maps by $\sim$0.5%. We ignore this effect since the residual dipole is only $\sim$0.3% of the full dipole amplitude to start with. The Gibbs sampler was run for 10,000 steps for each of the 8 template-cleaned maps (Q1-W4) and for each of 6 independent noise realizations added to the ILC map. In both cases we applied the KQ85 mask to the analysis and truncated the CMB power at $l_{\rm max}=32$. The resulting ensembles of 80,000 and 60,000 dipole samples were analyzed independently and jointly. The results of this analysis are given in Table 4. The first row combines the results from the template-cleaned DA maps; the scatter among the 8 DA’s was well within the noise scatter for each DA, so the Gibbs samples for all 8 DA’s were combined for this analysis. The results for the ILC map are shown in the second row. The two methods give reasonably consistent results, however, the Galactic longitude of the two dipole axis estimates differ from each other by about 2$\sigma$. Since we cannot reliably identify one cleaning method to be superior to the other, we have merged the Gibbs samples from both methods to produce the conservative estimate shown in the bottom row. This approach, which enlarges the uncertainty to emcompass both estimates, gives $(d,l,b)=(3.355\pm 0.008\;{\rm mK},263.99^{\circ}\pm 0.14^{\circ},48.26^{\circ}\pm 0.03^{\circ}),$ (1) where the amplitude estimate includes the 0.2% absolute calibration uncertainty. Given the CMB monopole temperature of 2.725 K (Mather et al., 1999), this amplitude implies a Solar System peculiar velocity of $369.0\pm 0.9$ km s-1 with respect to the CMB rest frame. ## 4 CALIBRATION IMPROVEMENTS With the 5 year processing we have refined our procedure for evaluating the instrument calibration, and have improved our estimates for the calibration uncertainty. The fundamental calibration source is still the dipole anisotropy induced by WMAP’s motion with respect to the CMB rest frame (Hinshaw et al., 2003; Jarosik et al., 2007), but several details of the calibration fitting have been modified. The new calibration solution is consistent with previous results in the overlapping time range. We estimate the uncertainty in the absolute calibration is now 0.2% per differencing assembly. The basic calibration procedure posits that a single channel of time-ordered data, $d_{i}$, may be modeled as $d_{i}=g_{i}\left[\Delta T_{vi}+\Delta T_{ai}\right]+b_{i},$ (2) where $i$ is a time index, $g_{i}$ and $b_{i}$ are the instrument gain and baseline, at time step $i$, $\Delta T_{vi}$ is the differential dipole anisotropy induced by WMAP’s motion, and $\Delta T_{ai}$ is the differential sky anisotropy. We assume that $\Delta T_{vi}$ is known exactly and has the form $\Delta T_{vi}=\frac{T_{0}}{c}{\bf v}_{i}\cdot[(1+x_{\rm im}){\bf n}_{A,i}-(1-x_{\rm im}){\bf n}_{B,i}],$ (3) where $T_{0}=2.725$ K is the CMB temperature (Mather et al., 1999), $c$ is the speed of light, ${\bf v}_{i}$ is WMAP’s velocity with respect to the CMB rest frame at time step $i$, $x_{\rm im}$ is the loss imbalance parameter (Jarosik et al., 2007), and ${\bf n}_{A,i}$, and ${\bf n}_{B,i}$ are the unit vectors of the A- and B-side lines of sight at time step $i$ (in the same frame as the velocity vector). The velocity may be decomposed as ${\bf v}_{i}={\bf v}_{\rm WMAP-SSB,i}+{\bf v}_{\rm SSB-CMB},$ (4) where the first term is WMAP’s velocity with respect to the solar system barycenter, and the second is the barycenter velocity with respect to the CMB. The former is well determined from ephemeris data, while the latter has been measured by COBE-DMR with an uncertainty of 0.7% (Kogut et al., 1996). Since the latter velocity is constant over WMAP’s life span, any error in our assumed value of ${\bf v}_{\rm SSB-CMB}$ will, in theory, be absorbed into a dipole contribution to the anisotropy map, $T_{a}$. We test this hypothesis below. The differential sky signal has the form $\Delta T_{ai}=(1+x_{\rm im})[I_{a}(p_{A,i})+P_{a}(p_{A,i},\gamma_{A,i})]-(1-x_{\rm im})[I_{a}(p_{B,i})+P_{a}(p_{B,i},\gamma_{B,i})],$ (5) where $p_{A,i}$ is the pixel observed by the A-side at time step $i$ (and similarly for B), $I_{a}(p)$ is the temperature anisotropy in pixel $p$ (the intensity Stokes parameter, $I$), and $P_{a}(p,\gamma)$ is the polarization anisotropy in pixel $p$ at polarization angle $\gamma$ (Hinshaw et al., 2003) which is related to the linear Stokes parameters $Q$ and $U$ by $P_{a}(p,\gamma)=Q(p)\cos 2\gamma+U(p)\sin 2\gamma.$ (6) We further note that, in general, $I_{a}$ and $P_{a}$ depend on frequency owing to Galactic emission. A main goal of the data processing is to simultaneously fit for the calibration and sky signal. Unfortunately, since the data model is nonlinear and the number of parameters is large, the general problem is intractable. In practice, we proceed iteratively as follows. Initially we assume the gain and baseline are constant for a given time interval, typically between 1 and 24 hours, $\displaystyle g_{i}=G_{k}$ $\displaystyle\tau_{k}<t_{i}<\tau_{k+1}$ (7) $\displaystyle b_{i}=B_{k}$ $\displaystyle\tau_{k}<t_{i}<\tau_{k+1},$ (8) where $t_{i}$ is the time of the $i$th individual observation, and $\tau_{k}$ is the start time of the $k$th calibration interval. Throughout the fit we fix the velocity-induced signal, equation (3), using ${\bf v}_{\rm SSB- CMB}=[-26.29,-244.96,+275.93]$ km s-1 (in Galactic coordinates), and, for the first iteration, we assume no anisotropy signal, $\Delta T_{a}=0$. Then, for each calibration interval $k$ we perform a linear fit for $G_{k}$ and $B_{k}$ with fixed $\Delta T_{v}+\Delta T_{a}$. As we proceed through the intervals, we apply this calibration to the raw data and accumulate a new estimate of the anisotropy map as per equation 19 of Hinshaw et al. (2003). The procedure is repeated with each updated estimate of $\Delta T_{a}$. Once the calibration solution has converged, we fit the gain data, $G_{k}$, to a model that is parameterized by the instrument detector voltage and the temperatures of the receiver’s warm and cold stages, equation 2 of Jarosik et al. (2007). This parametrization still provides a good fit to the $G_{k}$ data, so we have not updated its form for the 5 year analysis. The updated best-fit parameters are given in the 5 year Explanatory Supplement (Limon et al., 2008). Note that for each radiometer, the relative gain vs. time over 5 years is determined by just two parameters. For the 5 year processing we have focused on the veracity of the “raw” calibration, $G_{k}$ and $B_{k}$. Specifically, we have improved and/or critically reexamined several aspects of the iterative fitting procedure: * • We have incorporated the effect of far sidelobe pickup directly into the iterative calibration procedure, rather than as a fixed correction (Jarosik et al., 2007). We do this by segregating the differential signal into a main beam contribution and a sidelobe contribution, $\Delta T_{i}=\Delta T_{\rm main,i}+\Delta T_{\rm side,i}.$ (9) (Hill et al. 2008 discuss how this segregation is defined in the 5 year processing.) After each iteration of the calibration and sky map estimation, we (re)compute a database of $\Delta T_{\rm side}$ on a grid of pointings using the new estimate of $I_{a}$. We then interpolate the database to estimate $\Delta T_{\rm side,i}$ for each time step $i$. Note that $\Delta T_{\rm side}$ includes contributions from both the velocity-induced signal and the intrinsic anisotropy. Ignoring sidelobe pickup can induce gain errors of up to 1.5% in K band, 0.4% in Ka band, and $\sim$0.25% in Q-W bands. * • In general, the different channels within a DA have different center frequencies (Jarosik et al., 2003); hence the different channels measure a slightly different anisotropy signal due to differences in the Galactic signal. We assess the importance of accounting for this in the calibration procedure. * • A single DA channel is only sensitive to a single linear polarization state. (WMAP measures polarization by differencing orthogonal polarization channels.) Thus we cannot reliably solve for both $P_{a}$ and for $I_{a}$ at each channel’s center frequency. We assess the relative importance of accounting for one or the other on both the gain and baseline solutions. * • We examine the sensitivity of the calibration solution to the choice of ${\bf v}_{\rm SSB-CMB}$ and to assumptions of time-dependence in the gain. ### 4.1 Calibration Tests We use a variety of end-to-end simulations to assess and control the systematic effects noted above. We summarize a number of the key tests in the remainder of this section. The first case we consider is a noiseless simulation in which we generate time-ordered data from an input anisotropy map which includes CMB and Galactic foreground signal (one map per channel, evaluated at the center frequency of each channel) and a known dipole amplitude. The input gain for each channel is fixed to be constant in time. We run the iterative calibration and sky map solver allowing for an independent sky map solution at each channel (but no polarization signal). When fitting for the calibration, we assume that ${\bf v}_{\rm SSB-CMB}$ differs from the input value by 1% to see if the known, modulated velocity term, ${\bf v}_{\rm WMAP-SSB}$, properly “anchors” the absolute gain solution. The results are shown in the top panel of Figure 6 where it is shown that the absolute gain recovery is robust to errors in ${\bf v}_{\rm SSB-CMB}$. We recover the input gain to better than 0.1% in this instance. The second case we consider is again a noiseless simulation that now includes only dipole signal (with Earth-velocity modulation), but here we vary the input gain using the flight-derived gain model (Jarosik et al., 2007). The iterative solver was run on the K band data for 1400 iterations, again starting with an initial guess that was in error by 1%. The results are shown in the bottom panel of Figure 6, which indicate systematic convergence errors of $>$0.3% in the fitted amplitude of the recovered gain model. Since the input sky signal in this case does not have any Galactic foreground or polarization components, we cannot ascribe the recovery errors to the improper handling of those effects in the iterative solver. We have also run numerous other simulations that included various combinations of instrument noise, CMB anisotropy, Galactic foreground signal (with or without individual center frequencies per channel), polarization signal, and input gain variations. The combination of runs are too numerous to report on in detail, and the results are not especially enlightening. The most pertinent trend we can identify is that when the input value of ${\bf v}_{\rm SSB-CMB}$ is assumed in the iterative solver, the recovered gain is in good agreement with the input, but when the initial guess is in error by 1%, the recovered gain will have comparable errors. We believe the lack of convergence is due to a weak degeneracy between gain variations and the sky map solution. Such a degeneracy is difficult to diagnose in the context of this iterative solver, especially given the computational demands of the system, so we are assessing the system more directly with a low-resolution parameterization of the gain and sky signal, as outlined in Appendix A. Since the latter effort is still underway, we have adopted a more pragmatic approach to evaluating the absolute gain and its uncertainty for the 5 year data release. We proceed as follows: after 50 iterations of the calibration and sky map solver, the dominant errors in the gain and sky map solution are 1) a dipole in the sky map, and 2) a characteristic wave form that reflects a relative error between ${\bf v}_{\rm SSB-CMB}$ and ${\bf v}_{\rm WMAP-SSB}$. At this point we can calibrate the amplitude of the gain error wave form to the magnitude of the velocity error in ${\bf v}_{\rm SSB-CMB}$. We can then fit the gain solution to a linear combination of the gain model of Jarosik et al. (2007) and the velocity error wave form. See Appendix B for details on this fitting procedure. In practice this fit is performed simultaneously on both channels of a radiometer since those channels share one gain model parameter. We have tested this procedure on a complete flight-like simulations that includes every important effect known, including input gain variations. The results of the gain recovery are shown in Figure 7, and based on this we conservatively assign an absolute calibration uncertainty of 0.2% per channel for the 5 year WMAP archive. ### 4.2 Summary The series of steps taken to arrive at the final 5 year calibration are as follows: * • Run the iterative calibration and sky map solver over the full 5 year data set for 50 iterations, using 24 hour calibration intervals. This run starts with $I_{a}=P_{a}=0$ and updates $I_{a}$ for each individual channel of data. $P_{a}$ is assumed to be 0 throughout this run. We keep the gain solution, $G_{k}$, from this run and discard the baseline solution. * • Run the iterative calibration and sky map solver over the full 5 year data set for 50 iterations, using 1 hour calibration intervals. This run starts with $I_{a}=P_{a}=0$ and updates both using the intensity and polarization data in the two radiometers per DA, as per Appendix D of Hinshaw et al. (2003). We keep the baseline solution, $B_{k}$, from this run and discard the gain solution. Both of these runs incorporate the sidelobe correction as noted above. * • Fit the gain solution, $G_{k}$ simultaneously for the gain model and for an error in the velocity, ${\bf\Delta v}_{\rm SSB-CMB}$, as described in Appendix B. This fit is performed on two channels per radiometer with the gain model parameter $T_{0}$ common to both channels. * • We average the best-fit velocity error over all channels within a frequency band under the assumption that the dipole is the same in each of these channels. We then fix the velocity error to a single value per frequency band and re-fit the gain model parameters for each pair of radiometer channels. Based on end-to-end simulations with flight-like noise, we estimate the absolute gain error per radiometer to be 0.2%. We believe the limiting factor in this estimate is a weak degeneracy between thermal variations in the instrument gain, which are annually modulated, and annual variations induced by errors in ${\bf v}_{\rm SSB-CMB}$. Since there is a small monotonic increase in the spacecraft temperature, additional years of data should allow improvements in our ability to separate these effects. Once we have finalized the gain model, we form a calibrated time-ordered data archive using the gain model and the 1 hour baseline estimates to calibrate the data. This archive also has a final estimate of the far sidelobe pickup subtracted from each time-ordered data point. However, we opt not to subtract a dipole estimate from the archive at this stage in the processing. ## 5 BEAM IMPROVEMENTS In addition to reassessing the calibration, the other major effort undertaken to improve the 5 year data processing was to extend the physical optics model of the WMAP telescope based on flight measurements of Jupiter. This work is described in detail in Hill et al. (2008) so we only summarize the key results with an emphasis on their scientific implications. The basic aim of the work is to use the flight beam maps from all 10 DA’s to determine the in-flight distortion of the mirrors. This program was begun for the A-side mirror during the 3 year analysis; for the 5 year analysis we have quadrupled the number of distortion modes we fit (probing distortion scales that are half the previous size), and we have developed a completely new and independent model of the B-side distortions, rather than assuming that they mirror the A-side distortions. We have also placed limits on smaller scale distortions by comparing the predicted beam response at large angles to sidelobe data collected during WMAP’s early observations of the Moon. Given the best-fit mirror model, we compute the model beam response for each DA and use it in conjunction with the flight data to constrain the faint tails of the beams, beyond $\sim 1^{\circ}$ from the beam peak. These tails are difficult to constrain with flight data alone because the Jupiter signal to noise ratio is low, but, due to their large areal extent they contain a non- negligible fraction (up to 1%) of the total beam solid angle. An accurate determination of the beam tail is required to properly measure the ratio of sub-degree-scale power to larger-scale power in the diffuse CMB emission (and to accurately assign point source flux). Figure 14 in Hill et al. (2008) compares the beam radial profiles used in the 3 year and 5 year analyses, while Figure 13 compares the $l$-space transfer functions derived from the Legendre transform of the radial profile. The important changes to note are the following. 1. 1. In both analyses we split the beam response into main beam and far sidelobe contributions. In the 5 year analysis we have enlarged the radius at which this transition is made (Hill et al., 2008). In both cases, we correct the time-ordered data for far sidelobe pickup prior to making sky maps, while the main beam contribution is only accounted for in the analysis of sky maps, e.g., in power spectrum deconvolution. As a result, the sky maps have a slightly different effective resolution which is most apparent in K band, as in Figure 5. However, in each analysis, the derived transfer functions are appropriate for the corresponding sky maps. 2. 2. In the 3 year analysis, the main beam profile was described by a Hermite polynomial expansion fit to the observations of Jupiter in the time-ordered data. This approach was numerically problematic in the 5 year analysis due to the larger transition radius; as a result, we now simply co-add the time- ordered data into radial bins to obtain the profiles. In both cases, the underlying time-ordered data is a hybrid archive consisting of flight data for points where the beam model predicts a value above a given contour, and model values for points below the contour (Hill et al., 2008). With the improved beam models and a new error analysis, we have adjusted these hybrid contours down slightly, with the result that we use proportionately more flight data (per year) in the new analysis. The radius at which the 5 year profile becomes model dominated ($>$50% of the points in a bin) is indicated by dotted lines in Figure 14 of Hill et al. (2008). 3. 3. The right column of Figure 14 in Hill et al. (2008) shows the fractional change in solid angle due to the updated profiles. The main point to note is the $\sim$1% increase in the V2 and W band channels, primarily arising in the bin from 1 to 2 degrees off the beam peak. As can be seen in Figure 3 of Hill et al. (2008), this is the angular range in which the new beam models produced the most change, owing to the incorporation of smaller distortion modes in the mirror model. The 3 year analysis made use of the model in this angular range which, in hindsight, was suppressing up to $\sim$1% of the solid angle in the V and W band beams. (The longer wavelength channels are less sensitive to distortions in this range, so the change in solid angle is smaller for K-Q bands.) In the 5 year analysis, we use relatively more flight data in this regime, so we are less sensitive to any remaining model uncertainties. Hill et al. (2008) place limits on residual model errors and propagate those errors into the overall beam uncertainty. 4. 4. Figure 13 in Hill et al. (2008) compares the beam transfer functions, $b_{l}$, derived by transforming the 3 year and 5 year radial profiles. (To factor out the effect of changing the transition radius, the 3 year profiles were extended to the 5 year radius using the far sidelobe data, for this comparison.) Since the transfer functions are normalized to 1 at $l=1$, the change is restricted to high $l$. In V and W bands, $b_{l}$ has decreased by $\sim$0.5 - 1% due largely to the additional solid angle picked up in the 1-2 degree range. This amounts to a $\sim$1 $\sigma$ change in the functions, as indicated by the red curves in the Figure. The calibrated angular power spectrum is proportional to $1/g^{2}b_{l}^{2}$, where $g$ is the mean gain and $b_{l}$ is the beam transfer function, thus the net effect of the change in gain and beam determinations is to increase the power spectrum by $\sim$0.5% at $l\lesssim 100$, and by $\sim$2.5% at high $l$. Nolta et al. (2008) give a detailed evaluation of the power spectrum while Dunkley et al. (2008) and Komatsu et al. (2008) discuss the implications for cosmology. ## 6 LOW-$l$ POLARIZATION TESTS The 3 year data release included the first measurement of microwave polarization over the full sky, in the form of Stokes Q and U maps in each of 5 bands. The analysis of WMAP polarization data is complicated by the fact that the instrument was not designed to be a true polarimeter, thus a number of systematic effects had to be understood prior to assigning reliable error estimates to the data. Page et al. (2007) presented the 3 year polarization data in great detail. In this section we extend that analysis by considering some additional tests that were not covered in the 3 year analysis. We note that all of the tests described in this section have been performed on the template-cleaned reduced-foreground maps except for the final test of the Ka band data, described at the end of the section, which tests an alternative cleaning method. ### 6.1 Year-to-Year Consistency Tests With 5 years of data it is now possible to subject the data to more stringent consistency tests than was previously possible. In general, the number of independent cross-power spectra we can form within a band with $N_{d}$ differencing assemblies is $N_{d}(N_{d}-1)/2\times N_{y}+N_{d}^{2}\times N_{y}(N_{y}-1)/2$. With 5 years of data, this gives 10 independent estimates each in K and Ka band, 45 each in Q and V band, and 190 in W band. For cross power spectra of distinct band pairs, with $N_{d1}$ and $N_{d2}$ DA’s in each band, the number is $N_{d1}N_{d2}\times N_{y}^{2}$. This gives 50 each in KaQ and KaV, 100 each in KaW and QV, and 200 each in QW and VW. (For comparison, the corresponding numbers are 3, 15, & 66, and 18, 36, & 72 with 3 years of data.) We have evaluated these individual spectra from the 5 year data and have assigned noise uncertainties to each estimate using the Fisher formalism described in Page et al. (2007). We subject the ensemble to an internal consistency test by computing the reduced $\chi^{2}$ of the data at each multipole $l$ within each band or band pair, under the hypothesis that the data at each multipole and band measures the same number from DA to DA and year to year. The results of this test are given in Table 5 for the foreground-cleaned EE, EB, and BB spectra from $l=2-10$ for all band pairs from KaKa to WW. There are several points to note in these results. 1. 1. For $l\geq 6$, the most significant deviation from 1 in reduced $\chi^{2}$, in any spectrum or band, is 1.594 in the $l=7$ BB spectrum for KaQ. With 50 degrees of freedom, this is a 3 $\sigma$ deviation, but given that we have 150 $l\geq 6$ samples in the table, we expect of order 1 such value. Thus we conclude that the Fisher-based errors provide a good description of the DA-to- DA and year-to-year scatter in the $l\geq 6$ polarization data. If anything, there is a slight tendency to overestimate the uncertainties at higher $l$. 2. 2. For $l\leq 5$, we find 37 out of 120 points where the reduced $\chi^{2}$ deviates from 1 at more than 4 $\sigma$ significance, indicating excessive internal scatter in the data relative to the Fisher errors. However, all but 5 of these occur in cross-power spectra in which one or both of the bands contain W band data. If we exclude combinations with W band, the remaining 72 points have a mode in the reduced $\chi^{2}$ distribution of 1 with a slight positive skewness due to the 5 points noted above, which all contain Q band data. This may be a sign of slight foreground residuals contributing additional noise to the Q band data, though we do not see similar evidence in the Ka band spectra which would be more foreground contaminated prior to cleaning. For Ka-V bands, we believe that the Fisher errors provide an adequate description of the scatter in this $l\leq 5$ polarization data, but we subject polarization sensitive cosmological parameter estimates, e.g., the optical depth, to additional scrutiny in §6.3. 3. 3. Of special note is $l=3$ BB which, as noted in Page et al. (2007), is the power spectrum mode that is least modulated in the WMAP time-ordered data. This mode is therefore quite sensitive to how the instrument baseline is estimated and removed and, in turn, to how the 1/f noise is modeled. In the accounting above, the $l=3$ BB data have the highest internal scatter of any low-$l$ polarization mode. In particular, every combination that includes W band data is significantly discrepant; and the two most discrepant non-W band points are also estimates of $l=3$ BB. We comment on the W band data further below, but note here that the final co-added BB spectrum (based on Ka, Q, and V band data) does not lead to a significant detection of tensor modes. However, we caution that any surprising scientific conclusions which rely heavily on the WMAP $l=3$ BB data should be treated with caution. Based on the analysis presented above, we find the W band polarization data is still too unstable at low-$l$ to be reliably used for cosmological studies. We cite more specific phenomenology and consider some possible explanations in the remainder of this section. The 5 year co-added W band EE spectrum is shown in Figures 8, in the form of likelihood profiles from $l=2-7$. At each multipole we show two curves: an estimate based on evaluating the likelihood multipole by multipole, and an estimate based on the pseudo-$C_{l}$ method (Page et al., 2007). The best-fit model EE spectrum, based on the combined Ka, Q, and V band data is indicated by the dashed lines in each panel. Both spectrum estimates show excess power relative to the model spectrum, with the most puzzling multipole being $l=7$ which, as shown in Table 5, has an internal reduced $\chi^{2}$ of 1.015, for 190 degrees of freedom. This data has the hallmark of a sky signal, but that hypothesis is implausible for a variety of reasons (Page et al., 2007). It is more likely due to a systematic effect that is common to a majority of the W band channels over a majority of the 5 years of data. We explore and rule out one previously neglected effect in §6.2. It is worth recalling that $l=7$ EE, like $l=3$ BB, is a mode that is relatively poorly measured by WMAP, as discussed in Page et al. (2007); see especially Figure 16 and its related discussion. The W band BB data also exhibit unusual behavior at $l=2,3$. In this case, these two multipoles have internal reduced $\chi^{2}$ greater than 6, and the co-added $l=2$ point is nearly 10 $\sigma$ from zero. However, with 190 points in each 5 year co-added estimate it is now possible to look for trends within the data that were relatively obscure with only 3 years of data. In particular, we note that in the $l=2$ estimate, there are 28 points that are individually more than 5 $\sigma$ from zero and that all of them contain W1 data in one or both of the DA pairs in the cross power spectrum. Similarly for $l=3$, there are 14 points greater than 5 $\sigma$ and all of those points contain W4 data in one or both of the DA pairs. We have yet to pinpoint the significance of this result, but we plan to study the noise properties of these DA’s beyond what has been reported to date, and to sharpen the phenomenology with additional years of data. ### 6.2 Emissivity Tests In this section we consider time dependent emission from the WMAP optics as a candidate for explaining the excess W band “signal” seen in the EE spectrum, mostly at $l=7$. In the end, the effect proved not to be significant, but it provides a useful illustration of a common-mode effect that we believe is still present in the W band polarization data. From a number of lines of reasoning, we know that the microwave emissivity of the mirrors is a few percent in W band, and that it scales with frequency roughly like $\nu^{1.5}$ across the WMAP frequency range, as expected for a classical metal (Born & Wolf, 1980). Hence this mechanism has the potential to explain a common-mode effect that is primarily seen in W band. Further, Figure 1 in Jarosik et al. (2007) shows that the physical temperature of the primary mirrors are modulated at the spin period by $\sim$200 $\mu$K, with a dependence on solar azimuth angle that is highly repeatable from year to year. We believe this modulation is driven by solar radiation diffracting around the WMAP sun shield reaching the tops of the primary mirrors, which are only a few degrees within the geometric shadow of the sun shield. In contrast, the secondary mirrors and feed horns are in deep shadow and show no measurable variation at the spin period, so that any emission they produce only contributes to an overall radiometer offset, and will not be further considered here. As a rough estimate, the spin modulated emission from the primary mirrors could produce as much as $\sim 0.02\times 200=4$ $\mu$K of radiometric response in W band, but the actual signal depends on the relative phase of the A and B-side mirror variations and the polarization state of the emission. In more detail, the differential signal, $d(t)$, measured by a radiometer with lossy elements is $d(t)=(1-\epsilon_{A})\,T_{A}(t)-(1-\epsilon_{B})\,T_{B}(t)+\epsilon^{\rm p}_{A}\,T^{\rm p}_{A}(t)-\epsilon^{\rm p}_{B}\,T^{\rm p}_{B}(t)$ (10) where $\epsilon_{A}\ =\epsilon^{\rm p}_{A}+\epsilon^{\rm s}_{A}+\epsilon^{\rm f}_{A}$ is the combined loss in the A-side optics: (p)rimary plus (s)econdary mirrors, plus the (f)eed horn, and likewise for the B-side. $T_{A,B}$ is the sky temperature in the direction of the A or B-side line-of-sight; and $T^{\rm p}_{A,B}$ is the physical temperature of the A or B-side primary mirror. The first two terms are the sky signal attenuated by the overall loss in the A and B side optics, respectively. The effects of loss imbalance, which arise when $\epsilon_{A}\neq\epsilon_{B}$, have been studied extensively (Jarosik et al., 2003, 2007). We account for loss imbalance in the data processing and we marginalize over residual uncertainties in the imbalance coefficients when we form the pixel-pixel inverse covariance matrices (Jarosik et al., 2007). Updated estimates of the loss imbalance coefficients based on fits to the 5 year data are reported in Table 6. In the remainder of this section we focus on the last two emissive terms in Equation 10. Recall that a WMAP differencing assembly consists of two radiometers, 1 and 2, that are sensitive to orthogonal linear polarization modes. The temperature and polarization signals are extracted by forming the sum and difference of the two radiometer outputs; thus, the emission terms we need to evaluate are $d^{\rm p}_{1}(t)\pm d^{\rm p}_{2}(t)=\left(\frac{\epsilon^{\rm p}_{A1}\pm\epsilon^{\rm p}_{A2}}{1-\epsilon}\right)\,T^{\rm p}_{A}-\left(\frac{\epsilon^{\rm p}_{B1}\pm\epsilon^{\rm p}_{B2}}{1-\epsilon}\right)\,T^{\rm p}_{B}$ (11) where $\epsilon^{\rm p}_{A1}$ is the A-side primary mirror emissivity measured by radiometer 1, and so forth. The factor of $1-\epsilon$ in the denominator applies a small correction for the mean loss, $\epsilon\equiv(\epsilon_{A}+\epsilon_{B})/2$, and arises from the process of calibrating the data to a known sky brightness temperature (§4). Note that we only pick up a polarized response if $\epsilon_{1}\neq\epsilon_{2}$. We have simulated this signal in the time-ordered data using the measured primary mirror temperatures as template inputs. The emissivity coefficients were initially chosen to be consistent with the loss imbalance constraints. However, in order to produce a measurable polarization signal, we had to boost the emissivity differences to the point where they became unphysical, that is $|\epsilon_{1}-\epsilon_{2}|>|\epsilon_{1}+\epsilon_{2}|$. Nonetheless, it was instructive to analyze this simulation by binning the resulting data (which also includes sky signal and noise) as a function of solar azimuth. The results are shown in the top panel of Figure 9 which shows 3 years of co-added W band polarization data, the $d_{1}-d_{2}$ channel; the input emissive signal is shown in red for comparison. We are clearly able to detect such a signal with this manner of binning. We also computed the low-$l$ polarization spectra and found that, despite the large spin modulated input signal, the signal induced in the power spectrum was less than 2 $\mu$K2 in $l(l+1)C^{EE}_{l}/2\pi$, which is insufficient to explain the $l=7$ feature in the W band EE spectrum. In parallel with the simulation analysis, we have binned the flight radiometer data by solar azimuth angle to search for spin modulated features in the polarization data. The results for W band are shown in the bottom panel of Figure 9 for the 5 year data. While the $\chi^{2}$ per degree of freedom relative to zero is slightly high, there is no compelling evidence for a coherent spin modulated signal at the $\sim$2 $\mu$K level. In contrast, the simulation yielded spin modulated signals of 5-10 $\mu$K and still failed to produce a significant effect in the EE spectrum. Hence we conclude that thermal emission from the WMAP optics cannot explain the excess W band EE signal. In any event, we continue to monitor the spin modulated data for the emergence of a coherent signal. ### 6.3 Ka Band Tests The analysis presented in §6.1 shows that the Ka band polarization data is comparable to the Q and V band data in its internal consistency. That analysis was performed on data that had been foreground cleaned using the template method discussed in Page et al. (2007) and updated in Gold et al. (2008). In order to assess whether or not this cleaned Ka band data is suitable for use in cosmological parameter estimation we subject it to two further tests: 1) a null test in which Ka band data is compared to the combined Q and V band data, and 2) a parameter estimation based solely on Ka band data. For the null test, we form polarization maps by taking differences, $\frac{1}{2}S_{\rm Ka}-\frac{1}{2}S_{\rm QV}$, where S = Q,U are the polarization Stokes parameters, $S_{\rm Ka}$ are the maps formed from the Ka band data, and $S_{\rm QV}$ are the maps formed from the optimal combination of the Q and V band data. We evaluate the EE power spectrum from these null maps by evaluating the likelihood mode by mode while holding the other multipoles fixed at zero. The results are shown in Figure 10, along with the best-fit model spectrum based on the final 5 year $\Lambda$CDM analysis. The spectrum is clearly consistent with zero, but to get a better sense of the power of this test, we have also used these null maps to estimate the optical depth parameter, $\tau$. The result of that analysis is shown as the dashed curve in Figure 11, where we find that the null likelihood peaks at $\tau=0$ and excludes the most-likely cosmological value with $\sim$95% confidence. As a separate test, we evaluate the $\tau$ likelihood using only the template- cleaned Ka band signal maps. The result of that test is shown as the blue curve in Figure 11. While the uncertainty in the Ka band estimate is considerably larger than the combined QV estimate (shown in red), the estimates are highly consistent. The result of combining Ka, Q, and V band data is shown in the black curve. Dunkley et al. (2008) present a complementary method of foreground cleaning that makes use of Ka band data, in conjunction with K, Q, and V band data. Using a full 6 parameter likelihood evalutaion, they compare the optical depth inferred from the two cleaning methods while using the full combined data sets in both cases: see Figure 9 of Dunkley et al. (2008) for details. Based on these tests, we conclude that the Ka band data is sufficiently free of systematic errors and residual foreground signals that it is suitable for cosmological studies. The use of this band significantly enhances the overall polarization sensitivity of WMAP. ## 7 SUMMARY OF 5-YEAR SCIENCE RESULTS Detailed presentations of the scientific results from the 5 year data are given by Gold et al. (2008), Wright et al. (2008), Nolta et al. (2008), Dunkley et al. (2008), and Komatsu et al. (2008). Starting with the 5 year temperature and polarization maps, with their improved calibration, Gold et al. (2008) give a new Markov Chain Monte Carlo-based analysis of foreground emission in the data. Their results are broadly consistent with previous analyses by the WMAP team and others (Eriksen et al., 2007c), while providing some new results on the microwave spectra of bright sources in the Galactic plane that aren’t well fit by simple power-law foreground models. Figure 12 shows the 5 year CMB map based on the internal linear combination (ILC) method of foreground removal. Wright et al. (2008) give a comprehensive analysis of the extragalactic sources in the 5 year data, including a new analysis of variability made possible by the multi-year coverage. The 5 year WMAP source catalog now contains 390 objects and is reasonably complete to a flux of 1 Jy away from the Galactic plane. The new analysis of the WMAP beam response (Hill et al., 2008) has led to more precise estimates of the point source flux scale for all 5 WMAP frequency bands. This information is incorporated in the new source catalog (Wright et al., 2008), and is also used to provide new brightness estimates of Mars, Jupiter, and Saturn (Hill et al., 2008). We find significant (and expected) variability in Mars and Saturn over the course of 5 years and use that information to provide a preliminary recalibration of a Mars brightness model (Wright, 2007), and to fit a simple model of Saturn’s brightness as a function of ring inclination. The temperature and polarization power spectra are presented in Nolta et al. (2008). The spectra are all consistent with the 3 year results with improvements in sensitivity commensurate with the additional integration time. Further improvements in our understanding of the absolute calibration and beam response have allowed us to place tighter uncertainties on the power spectra, over and above the reductions from additional data. These changes are all reflected in the new version of the WMAP likelihood code. The most notable improvements arise in the third acoustic peak of the TT spectrum, and in all of the polarization spectra; for example, we now see unambiguous evidence for a 2nd dip in the high-$l$ TE spectrum, which further constrains deviations from the standard $\Lambda$CDM model. The 5 year TT and TE spectra are shown in Figure 13. We have also generated new maximum likelihood estimates of the low-$l$ polarization spectra: TE, TB, EE, EB, and BB to complement our earlier estimates based on pseudo-$C_{l}$ methods (Nolta et al., 2008). The TB, EB, and BB spectra remain consistent with zero. The cosmological implications of the 5 year WMAP data are discussed in detail in Dunkley et al. (2008) and Komatsu et al. (2008). The now-standard cosmological model: a flat universe dominated by vacuum energy and dark matter, seeded by nearly scale-invariant, adiabatic, Gaussian random-phase fluctuations, continues to fit the 5 year data. WMAP has now determined the key parameters of this model to high precision; a summary of the 5 year parameter results is given in Table 7. The most notable improvements are the measurements of the dark matter density, $\Omega_{c}h^{2}$, and the amplitude of matter fluctuations today, $\sigma_{8}$. The former is determined with 6% uncertainty using WMAP data only (Dunkley et al., 2008), and with 3% uncertainty when WMAP data is combined with BAO and SNe constraints (Komatsu et al., 2008). The latter is measured to 5% with WMAP data, and to 3% when combined with other data. The redshift of reionization is $z_{\rm reion}$ = $11.0\pm 1.4$, if the universe were reionized instantaneously. The 2 $\sigma$ lower limit is $z_{\rm reion}\mbox{$>$}8.2$, and instantaneous reionization at $z_{\rm reion}=6$ is rejected at 3.5 $\sigma$. The WMAP data continues to favor models with a tilted primordial spectrum, $n_{s}$ = $0.963^{+0.014}_{-0.015}$. Dunkley et al. (2008) discuss how the $\Lambda$CDM model continues to fit a host of other astronomical data as well. Moving beyond the standard $\Lambda$CDM model, when WMAP data is combined with BAO and SNe observations (Komatsu et al., 2008), we find no evidence for running in the spectral index of scalar fluctuations, $dn_{s}/d\ln{k}=-0.028\pm 0.020$ (68% CL). The new limit on the tensor-to- scalar ratio is $r<0.22\ \mbox{(95\% CL)}$, and we obtain tight, simultaneous limits on the (constant) dark energy equation of state and the spatial curvature of the universe: $-0.14<1+w<0.12\ \mbox{(95\% CL)}$ and $-0.0179<\Omega_{k}<0.0081\ \mbox{(95\% CL)}$. The angular power spectrum now exhibits the signature of the cosmic neutrino background: the number of relativistic degrees of freedom, expressed in units of the effective number of neutrino species, is found to be $N_{\rm eff}=4.4\pm 1.5$ (68% CL), consistent with the standard value of 3.04. Models with $N_{\rm eff}=0$ are disfavored at $>$99.5% confidence. A summary of the key cosmological parameter values is given in Table 7, where we provide estimates using WMAP data alone and WMAP data combined with BAO and SNe observations. A complete tabulation of all parameter values for each model and dataset combination we studied is available on LAMBDA. The new data also place more stringent limits on deviations from Gaussianity, parity violations, and the amplitude of isocurvature fluctuations (Komatsu et al., 2008). For example, new limits on physically motivated primordial non- Gaussianity parameters are $-9<f_{NL}^{\rm local}<111$ (95% CL) and $-151<f_{NL}^{\rm equil}<253$ (95% CL) for the local and equilateral models, respectively. ## 8 CONCLUSIONS We have presented an overview of the 5 year WMAP data and have highlighted the improvements we have made to the data processing and analysis since the 3 year results were presented. The most substantive improvements to the processing include a new method for establishing the absolute gain calibration (with reduced uncertainty), and a more complete analysis of the WMAP beam response made possible by additional data and a higher fidelity physical optics model. Numerous other processing changes are outlined in §2. The 5 year sky maps are consistent with the 3 year maps and have noise levels that are $\sqrt{5}$ times less than the single year maps. The new maps are compared to the 3 year maps in §3. The main changes to the angular power spectrum are as follows: at low multipoles ($l\lesssim 100$) the spectrum is $\sim$0.5% higher than the 3 year spectrum (in power units) due to the new absolute gain determination. At higher multipoles it is increased by $\sim$2.5%, due to the new beam response profiles, as explained in §5 and in Hill et al. (2008). These changes are consistent with the 3 year uncertainties when one accounts for both the 0.5% gain uncertainty (in temperature units) and the 3 year beam uncertainties, which were incorporated into the likelihood code. We have applied a number of new tests to the polarization data to check internal consistency and to look for new systematic effects in the W band data (§6). As a result of these tests, and of new analyses of polarized foreground emission (Dunkley et al., 2008), we have concluded that Ka band data can be used along with Q and V band data for cosmological analyses. However, we still find a number of features in the W band polarization data that preclude its use, except in the Galactic plane where the signal to noise is relatively high. We continue to investigate the causes of this and have identified new clues to follow up on in future studies (§6.1). Scientific results gathered from the suite of 5 year papers are summarized in §7. The highlights include smaller uncertainties in the optical depth, $\tau$, due to a combination of additional years of data and to the inclusion of Ka band polarization data: instantaneous reionization at $z_{\rm reion}=6$ is now rejected at 3.5 $\sigma$. New evidence favoring a non-zero neutrino abundance at the epoch of last scattering, made possible by improved measurements of the third acoustic peak; and new limits on the nongaussian parameter $f_{NL}$, based on additional data and the application of a new, more optimal bispectrum estimator. The 5 year data continue to favor a tilted primordial fluctuation spectrum, in the range $n_{s}\sim 0.96$, but a purely scale invariant spectrum cannot be ruled out at $>$3 $\sigma$ confidence. The WMAP observatory continues to operate at L2 as designed, and the addition of two years of flight data has allowed us to make significant advances in characterizing the instrument. Additional data beyond 5 years will give us a better understanding of the instrument, especially with regards to the W band polarization data since the number of jackknife combinations scales like the square of the number of years of operation. If W band data can be incorporated into the EE power spectrum estimate, it would become possible to constrain a second reionization parameter and thereby further probe this important epoch in cosmology. The WMAP data continues to uphold the standard $\Lambda$CDM model but more data may reveal new surprises. ## 9 DATA PRODUCTS All of the WMAP data is released to the research community for further analysis through the Legacy Archive for Microwave Background Data Analysis (LAMBDA) at http://lambda.gsfc.nasa.gov. The products include the complete 5 year time-ordered data archive (both raw and calibrated); the calibrated sky maps in a variety of processing stages (single year by DA, multi-year by band, high resolution and low resolution, smoothed, foreground-subtracted, and so forth); the angular power spectra and cosmological model likelihood code; a full table of model parameter values for a variety of model and data sets (including the best-fit model spectra and Markov chains); and a host of ancillary data to support further analysis. The WMAP Explanatory Supplement provides detailed information about the WMAP in-flight operations and data products (Limon et al., 2008). The WMAP mission is made possible by the support of the Science Mission Directorate Office at NASA Headquarters. This research was additionally supported by NASA grants NNG05GE76G, NNX07AL75G S01, LTSA03-000-0090, ATPNNG04GK55G, and ADP03-0000-092. EK acknowledges support from an Alfred P. Sloan Research Fellowship. This research has made use of NASA’s Astrophysics Data System Bibliographic Services. We acknowledge use of the HEALPix, CAMB, CMBFAST, and CosmoMC packages. ## References * Astier et al. (2006) Astier, P., et al. 2006, A&A, 447, 31 * Barnes et al. (2003) Barnes, C., et al. 2003, ApJS, 148, 51 * Bennett et al. (2003) Bennett, C. L., et al. 2003, ApJ, 583, 1 * Benoît et al. (2003) Benoît, A., et al. 2003, A&A, 399, L25 * Born & Wolf (1980) Born, M. & Wolf, E. 1980, Principles of Optics, sixth edn. (Pergamon Press) * Cole et al. (2005) Cole, S., et al. 2005, MNRAS, 362, 505 * Copi et al. (2007) Copi, C. J., Huterer, D., Schwarz, D. J., & Starkman, G. D. 2007, Phys. Rev. D, 75, 023507 * Creminelli et al. (2006) Creminelli, P., Nicolis, A., Senatore, L., Tegmark, M., & Zaldarriaga, M. 2006, JCAP, 0605, 004 * Dunkley et al. (2008) Dunkley, J., et al. 2008, ArXiv e-prints, 803 * Eisenstein et al. (2005) Eisenstein, D. J., et al. 2005, ApJ, 633, 560 * Eriksen et al. (2007a) Eriksen, H. K., Banday, A. J., Górski, K. M., Hansen, F. K., & Lilje, P. B. 2007a, ApJ, 660, L81 * Eriksen et al. (2007b) Eriksen, H. K., Jewell, J. B., Dickinson, C., Banday, A. J., Gorski, K. M., & Lawrence, C. R. 2007b, ArXiv e-prints, 709 * Eriksen et al. (2007c) Eriksen, H. K., et al. 2007c, ApJ, 656, 641 * Freedman et al. (2001) Freedman, W. L., et al. 2001, ApJ, 553, 47 * Gold et al. (2008) Gold, B. et al. 2008, ApJS * Gorski et al. (2005) Gorski, K. M., Hivon, E., Banday, A. J., Wandelt, B. D., Hansen, F. K., Reinecke, M., & Bartlemann, M. 2005, ApJ, 622, 759 * Halverson et al. (2002) Halverson, N. W., et al. 2002, ApJ, 568, 38 * Hill et al. (2008) Hill, R. et al. 2008, ApJS * Hinshaw et al. (2003) Hinshaw, G., et al. 2003, ApJS, 148, 63 * Hinshaw et al. (2007) —. 2007, ApJS, 170, 288 * Hunt & Sarkar (2007) Hunt, P. & Sarkar, S. 2007, Phys. Rev. D, 76, 123504 * Jarosik et al. (2003) Jarosik, N., et al. 2003, ApJS, 145, 413 * Jarosik et al. (2007) —. 2007, ApJS, 170, 263 * Kogut et al. (1996) Kogut, A., et al. 1996, ApJ, 470, 653 * Komatsu et al. (2008) Komatsu, E., et al. 2008, ArXiv e-prints, 803 * Land & Magueijo (2007) Land, K. & Magueijo, J. 2007, MNRAS, 378, 153 * Lee et al. (2001) Lee, A. T. et al. 2001, ApJ, 561, L1 * Limon et al. (2008) Limon, M., et al. 2008, Wilkinson Microwave Anisotropy Probe (WMAP): Explanatory Supplement, http://lambda.gsfc.nasa.gov/data/map/doc/MAP_supplement.pdf * Mather et al. (1999) Mather, J. C., Fixsen, D. J., Shafer, R. A., Mosier, C., & Wilkinson, D. T. 1999, ApJ, 512, 511 * Miller et al. (1999) Miller, A. D. et al. 1999, ApJ, 524, L1 * Netterfield et al. (2002) Netterfield, C. B. et al. 2002, ApJ, 571, 604 * Nolta et al. (2008) Nolta, M. R. et al. 2008, ApJS * Page et al. (2007) Page, L., et al. 2007, ApJS, 170, 335 * Pearson et al. (2003) Pearson, T. J., et al. 2003, ApJ, 591, 556 * Peiris et al. (2003) Peiris, H. V., et al. 2003, ApJS, 148, 213 * Percival et al. (2007) Percival, W. J., Cole, S., Eisenstein, D. J., Nichol, R. C., Peacock, J. A., Pope, A. C., & Szalay, A. S. 2007, MNRAS, 381, 1053 * Percival et al. (2001) Percival, W. J., et al. 2001, MNRAS, 327, 1297 * Riess et al. (2007) Riess, A. G., et al. 2007, ApJ, 659, 98 * Scott et al. (2003) Scott, P. F., et al. 2003, MNRAS, 341, 1076 * Shafieloo & Souradeep (2007) Shafieloo, A. & Souradeep, T. 2007, ArXiv e-prints, 709 * Spergel et al. (2003) Spergel, D. N., et al. 2003, ApJS, 148, 175 * Spergel et al. (2007) —. 2007, ApJS, 170, 377 * Tegmark et al. (2004) Tegmark, M., et al. 2004, Phys. Rev. D, 69, 103501 * Tegmark et al. (2006) —. 2006, Phys. Rev. D, 74, 123507 * Wood-Vasey et al. (2007) Wood-Vasey, W. M., et al. 2007, ApJ, 666, 694 * Wright (2007) Wright, E. L. 2007, ArXiv Astrophysics e-prints * Wright et al. (2008) Wright, E. L. et al. 2008, ApJS * Yadav et al. (2007) Yadav, A. P. S., Komatsu, E., Wandelt, B. D., Liguori, M., Hansen, F. K., & Matarrese, S. 2007, ArXiv e-prints, 711 * Yadav & Wandelt (2008) Yadav, A. P. S. & Wandelt, B. D. 2008, Physical Review Letters, 100, 181301 ## Appendix A FISHER MATRIX ANALYSIS OF CALIBRATION AND SKY MAP FITS ### A.1 Least Squares Calibration and Sky Model Fitting Let $i$ be a time index in the time ordered data. Let $g^{j}$ be parameters for the gain, $a_{lm}$ be parameters for the temperature anisotropy and $b^{k}$ be parameters for the baseline offset. The model of the time-ordered data (TOD) is $m_{i}=g_{i}\left[\Delta T_{vi}+\Delta T_{ai}\right]+b_{i},$ (A1) where $i$ is a time index, $\Delta T_{vi}$ is the differential dipole signal at time step $i$, including the CMB dipole, and $\Delta T_{ai}$ is the differential anisotropy signal at time step $i$. The parameters of the model are the hourly gain and baseline values, and the sky map pixel temperatures (which goes into forming $\Delta T_{a}$. We fit for them by minimizing $\chi^{2}=\sum_{i}\frac{(c_{i}-m_{i})^{2}}{\sigma_{i}^{2}},$ (A2) where $c_{i}$ is the raw data, in counts, and $\sigma_{i}$ is the rms of the $i$th observation, in counts. The Fisher matrix requires taking the second derivative of $\chi^{2}$ with respect to all parameters being fit. In order to reduce the dimensionality of the problem to something manageable, we expand the calibration and sky signal in terms of a small number of parameters. We can write $\displaystyle g_{i}$ $\displaystyle=$ $\displaystyle\sum_{j}g^{j}G_{ji},$ (A3) $\displaystyle b_{i}$ $\displaystyle=$ $\displaystyle\sum_{k}b^{k}B_{ki},$ (A4) $\displaystyle\Delta T_{ai}$ $\displaystyle=$ $\displaystyle\sum_{lm}a_{lm}\left[Y_{lm}(\hat{n}_{Ai})-Y_{lm}(\hat{n}_{Bi})\right],$ (A5) where $G$ and $B$ are function of time (defined below), $a_{lm}$ are the harmonic coefficients of the map, and $\hat{n}_{Ai}$ is the unit vector of the $A$-side feed at time step $i$, and likewise for $B$. A reasonable set of basis functions for the gain and baseline allow for an annual modulation and a small number of higher harmonics. Note that this does not include power at the spin or precession period, which might be an important extension to consider. For now we consider the trial set $G_{ji}=\left\\{\begin{array}[]{ll}1&j=0\\\ \cos j\theta_{i}&j=1,\ldots,j_{\rm max}\\\ \sin(j-j_{\rm max})\theta_{i}&j=j_{\rm max}+1,\ldots,2j_{\rm max}\end{array}\right.,$ (A6) and $B_{ki}=\left\\{\begin{array}[]{ll}1&k=0\\\ \cos k\theta_{i}&k=1,\ldots,k_{\rm max}\\\ \sin(k-k_{\rm max})\theta_{i}&k=k_{\rm max}+1,\ldots,2k_{\rm max}\end{array}\right.,$ (A7) where $\theta=\tan^{-1}(\hat{n}_{y}/\hat{n}_{x})$. Here $\hat{n}$ is the unit vector from WMAP to the Sun, and the components are evaluated in ecliptic coordinates. ### A.2 Evaluation of the Fisher Matrix We wish to evaluate the 2nd derivative $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial p_{i}\partial p_{j}}$ (A8) where $p_{i}$ and $p_{j}$ are the parameters we are trying to fit. The needed first derivatives are $\frac{1}{2}\,\frac{\partial\chi^{2}}{\partial g^{j^{\prime}}}=-\sum_{i}\frac{(c_{i}-m_{i})G_{j^{\prime}i}\left[\Delta T_{vi}+\Delta T_{ai}\right]}{\sigma_{i}^{2}},$ (A9) $\frac{1}{2}\,\frac{\partial\chi^{2}}{\partial b^{k^{\prime}}}=-\sum_{i}\frac{(c_{i}-m_{i})B_{k^{\prime}i}}{\sigma_{i}^{2}},$ (A10) $\frac{1}{2}\,\frac{\partial\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}}=-\sum_{i}\frac{(c_{i}-m_{i})g_{i}\left[Y_{l^{\prime}m^{\prime}}(\hat{n}_{Ai})-Y_{l^{\prime}m^{\prime}}(\hat{n}_{Bi})\right]}{\sigma_{i}^{2}}.$ (A11) Then $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial g^{j^{\prime\prime}}}=\sum_{i}\frac{G_{j^{\prime}i}\left[\Delta T_{vi}+\Delta T_{ai}\right]\,G_{j^{\prime\prime}i}\left[\Delta T_{vi}+\Delta T_{ai}\right]}{\sigma_{i}^{2}}$ (A12) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial a_{l^{\prime}m^{\prime}}}=\sum_{i}\frac{g_{i}\left[Y_{l^{\prime}m^{\prime}}(\hat{n}_{Ai})-Y_{l^{\prime}m^{\prime}}(\hat{n}_{Bi})\right]G_{j^{\prime}i}\left[\Delta T_{vi}+\Delta T_{ai}\right]}{\sigma_{i}^{2}}+{\cal O}\sum_{i}(c_{i}-m_{i})$ (A13) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial b^{k^{\prime}}}=\sum_{i}\frac{B_{k^{\prime}i}G_{j^{\prime}i}\left[\Delta T_{vi}+\Delta T_{ai}\right]}{\sigma_{i}^{2}}$ (A14) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}\partial a_{l^{\prime\prime}m^{\prime\prime}}}=\sum_{i}\frac{g_{i}\left[Y_{l^{\prime}m^{\prime}}(\hat{n}_{Ai})-Y_{l^{\prime}m^{\prime}}(\hat{n}_{Bi})\right]\,g_{i}\left[Y_{l^{\prime\prime}m^{\prime\prime}}(\hat{n}_{Ai})-Y_{l^{\prime\prime}m^{\prime\prime}}(\hat{n}_{Bi})\right]}{\sigma_{i}^{2}}$ (A15) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}\partial b^{k^{\prime}}}=\sum_{i}\frac{g_{i}B_{k^{\prime}i}\left[Y_{l^{\prime}m^{\prime}}(\hat{n}_{Ai})-Y_{l^{\prime}m^{\prime}}(\hat{n}_{Bi})\right]}{\sigma_{i}^{2}}$ (A16) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial b^{k^{\prime}}\partial b^{k^{\prime\prime}}}=\sum_{i}\frac{B_{k^{\prime\prime}i}B_{k^{\prime}i}}{\sigma_{i}^{2}}$ (A17) From this we can form the inverse covariance matrix $C^{-1}=\left(\begin{array}[]{ccc}\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial g^{j^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial a_{l^{\prime\prime}m^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial g^{j^{\prime}}\partial b^{k^{\prime\prime}}}\vspace{3mm}\\\ \frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}\partial g^{j^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}\partial a_{l^{\prime\prime}m^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial a_{l^{\prime}m^{\prime}}\partial b^{k^{\prime\prime}}}\vspace{3mm}\\\ \frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial b^{k^{\prime}}\partial g^{j^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial b^{k^{\prime}}\partial a_{l^{\prime\prime}m^{\prime\prime}}}&\frac{1}{2}\frac{\partial^{2}\chi^{2}}{\partial b^{k^{\prime}}\partial b^{k^{\prime\prime}}}\end{array}\right),$ (A18) where the gain and baseline blocks are $(2j_{\rm max}+1)\times(2j_{\rm max}+1)$, and the sky map block is $(l_{\rm max}+1)^{2}\times(l_{\rm max}+1)^{2}$. If we decompose $C^{-1}$ using SVD the parameter covariance matrix can be inverted to have the form $C=\sum_{i}\frac{1}{w_{i}}V_{(i)}\otimes V_{(i)}$ (A19) where the $w_{i}$ are the singular values, and the $V_{(i)}$ are the columns of the orthogonal matrix $V$. In this form, the uncertainty in the linear combination of parameters defined by $V_{(i)}$ is $1/w_{i}$. ## Appendix B CALIBRATION MODEL FITTING WITH GAIN ERROR TEMPLATES ### B.1 Gain Error From Calibration Dipole Error Consider a simple model where the input sky consists of only a pure fixed (CMB) dipole, described by the vector ${\bf d}_{c}$, and a dipole modulated by the motion of WMAP with respect to the Sun, described by the time-dependent vector ${\bf d}_{v}(t)$. The raw data produced by an experiment observing this signal is $c(t_{i})=g(t_{i})[\Delta t_{c}(t_{i})+\Delta t_{v}(t_{i})]$ (B1) where $c(t_{i})$ is the TOD signal in counts, $g(t_{i})$ is the true gain of the instrument and $\Delta t_{m}(t_{i})$ is the differential signal produced by each dipole component ($m=c,v$) at time $t_{i}$ given the instrument pointing at that time. Note that we have suppressed the explicit baseline and noise terms here for simplicity. Now suppose we calibrate the instrument using an erroneous CMB dipole, ${\bf d}^{\prime}_{c}=r{\bf d}_{c}=(1+\Delta r){\bf d}_{c}$, where $r$ is a number of order one (and $\Delta r\ll 1$ so we can ignore terms of order $\Delta r^{2}$). The fit gain, $g_{f}(t)$, will then roughly have the form $g_{f}(t)=\frac{c(t)}{|{\bf d}^{\prime}_{c}+{\bf d}_{v}(t)|}=g(t)\frac{|{\bf d}_{c}+{\bf d}_{v}(t)|}{|r{\bf d}_{c}+{\bf d}_{v}(t)|},$ (B2) where the vertical bars indicate vector magnitude. Now define ${\bf d}\equiv{\bf d}_{c}+{\bf d}_{v}$ and expand to 1st order in $\Delta r$ to get $g_{f}(t)=g(t)\left[1-\Delta r\frac{{\bf d}(t)\cdot{\bf d}_{c}}{{\bf d}(t)\cdot{\bf d}(t)}\right].$ (B3) Note that the term $({\bf d}\cdot{\bf d}_{c})/({\bf d}\cdot{\bf d})$ is dominated by a constant component of order $d_{c}^{2}/(d_{c}^{2}+d_{v}^{2})\sim 0.99$, followed by an annually modulated term that is suppressed by a factor of order $d_{v}/d_{c}$. Thus an erroneous calibration dipole induces a specific error in the fit gain that can be identified and corrected for, assuming the time dependence of the true gain is orthogonal to this form. ### B.2 Gain Model Fitting In theory, the way to do this is as follows. We have a set of data in the form of the fit gains, $g_{f,i}$ for each calibration sequence $i$, and we have a gain model, $G(t;p_{n})$, which is a function of time and a set of model parameters $p_{n}$. Ideally we would like to fit the model to the true gain, $g(t)$, but since we don’t know the true gain, the next best thing is to modify the gain model to have the same modulation form as the dipole gains have and to fit for this modulation simultaneously with the other gain model parameters. Thus $\chi^{2}$ takes the form $\chi^{2}=\sum_{i}\frac{\left[g_{i}-G_{i}(p_{n})\right]^{2}}{\sigma_{i}^{2}}=\sum_{i}\frac{\left[g_{f,i}-G_{i}(p_{n})(1-\Delta rf_{d,i})\right]^{2}}{\sigma_{i}^{2}},$ (B4) where $f_{d,i}\equiv({\bf d}\cdot{\bf d}_{c})/({\bf d}\cdot{\bf d})$ evaluated at time $t_{i}$, or is a function generated from simulations. Since the system is nonlinear, it must be minimized using a suitable nonlinear least squares routine. However, we can analyze the parameter covariance matrix directly by explicitly evaluating the 2nd derivative of $\chi^{2}$ with respect to the model parameters $C^{-1}=\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial p_{j}\partial p_{k}}.$ (B5) First compile the necessary 1st derivatives $\frac{1}{2}\,\frac{\partial\chi^{2}}{\partial\Delta r}=\sum_{i}\frac{\left[g_{f,i}-G_{i}(p_{n})(1-\Delta rf_{d,i})\right]\,(G_{i}\,f_{d,i})}{\sigma_{i}^{2}}$ (B6) $\frac{1}{2}\,\frac{\partial\chi^{2}}{\partial p_{m}}=\sum_{i}\frac{\left[g_{f,i}-G_{i}(p_{n})(1-\Delta rf_{d,i})\right]\,(-\partial G_{i}/\partial p_{m})(1-\Delta rf_{d,i})}{\sigma_{i}^{2}}$ (B7) (We evaluate the individual $\partial G/\partial p_{m}$ terms below.) Next the various 2nd derivatives are $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial\Delta r\partial\Delta r}=\sum_{i}\frac{(G_{i}\,f_{d,i})(G_{i}\,f_{d,i})}{\sigma_{i}^{2}},$ (B8) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial\Delta r\partial p_{m}}=\sum_{i}\frac{(G_{i}\,f_{d,i})(-\partial G_{i}/\partial p_{m})(1-\Delta rf_{d,i})}{\sigma_{i}^{2}}+{\cal O}\sum_{i}(g_{i}-G_{i}),$ (B9) $\frac{1}{2}\,\frac{\partial^{2}\chi^{2}}{\partial p_{m}\partial p_{n}}=\sum_{i}\frac{(\partial G_{i}/\partial p_{m})(1-\Delta rf_{d,i})(\partial G_{i}/\partial p_{n})(1-\Delta rf_{d,i})}{\sigma_{i}^{2}}+{\cal O}\sum_{i}(g_{i}-G_{i}).$ (B10) In the last two expressions, we neglect the term proportional to $\partial^{2}G/\partial p_{m}\partial p_{n}$ because the prefactor of $(g_{i}-G_{i})$ is statistically zero for the least squares solution. Finally, we evaluate the $\partial G/\partial p_{m}$ terms. The gain model has the form (Jarosik et al., 2007) $G_{i}=\alpha\frac{\bar{V}(t_{i})-V_{0}-\beta(T_{\rm RXB}(t_{i})-T^{0}_{\rm RXB})}{T_{\rm FPA}(t_{i})-T^{0}_{\rm FPA}},$ (B11) where $T^{0}_{\rm RXB}\equiv 290$ K, and $\alpha$, $V_{0}$, and $T^{0}_{\rm FPA}$ are parameters to be fit. The necessary 1st derivatives are $\partial G_{i}/\partial\alpha=\frac{\bar{V}(t_{i})-V_{0}-\beta(T_{\rm RXB}(t_{i})-T^{0}_{\rm RXB})}{T_{\rm FPA}(t_{i})-T^{0}_{\rm FPA}},$ (B12) $\partial G_{i}/\partial V_{0}=\frac{-\alpha}{T_{\rm FPA}(t_{i})-T^{0}_{\rm FPA}},$ (B13) $\partial G_{i}/\partial\beta=\frac{-\alpha(T_{\rm RXB}(t_{i})-T^{0}_{\rm RXB})}{T_{\rm FPA}(t_{i})-T^{0}_{\rm FPA}},$ (B14) $\partial G_{i}/\partial T^{0}_{\rm FPA}=\alpha\frac{\bar{V}(t_{i})-V_{0}-\beta(T_{\rm RXB}(t_{i})-T^{0}_{\rm RXB})}{(T_{\rm FPA}(t_{i})-T^{0}_{\rm FPA})^{2}}.$ (B15) Figure 1: Five-year temperature sky maps in Galactic coordinates smoothed with a $0.2\mbox{${}^{\circ}$}$ Gaussian beam, shown in Mollweide projection. top: K band (23 GHz), middle-left: Ka band (33 GHz), bottom-left: Q band (41 GHz), middle-right: V band (61 GHz), bottom-right: W band (94 GHz). Figure 2: Five-year Stokes Q polarization sky maps in Galactic coordinates smoothed to an effective Gaussian beam of $2.0\mbox{${}^{\circ}$}$, shown in Mollweide projection. top: K band (23 GHz), middle-left: Ka band (33 GHz), bottom-left: Q band (41 GHz), middle-right: V band (61 GHz), bottom-right: W band (94 GHz). Figure 3: Five-year Stokes U polarization sky maps in Galactic coordinates smoothed to an effective Gaussian beam of $2.0\mbox{${}^{\circ}$}$, shown in Mollweide projection. top: K band (23 GHz), middle-left: Ka band (33 GHz), bottom-left: Q band (41 GHz), middle-right: V band (61 GHz), bottom-right: W band (94 GHz). Figure 4: Five-year polarization sky maps in Galactic coordinates smoothed to an effective Gaussian beam of $2.0\mbox{${}^{\circ}$}$, shown in Mollweide projection. The color scale indicates polarized intensity, $P=\sqrt{Q^{2}+U^{2}}$, and the line segments indicate polarization direction in pixels whose signal-to-noise exceeds 1. top: K band (23 GHz), middle-left: Ka band (33 GHz), bottom-left: Q band (41 GHz), middle-right: V band (61 GHz), bottom-right: W band (94 GHz). Figure 5: Difference between the 5 year and 3 year temperature maps. left column: the difference in the maps, as delivered, save for the subtraction of a relative offset (Table 3), right column: the difference after correcting the 3 year maps by a scale factor that accounts for the mean gain change, $\sim 0.3$%, between the 3 year and 5 year estimates. top to bottom: K, Ka, Q, V, W band. The differences before recalibration are dominated by galactic plane emission and a dipole residual: see Table 3, which also gives the changes for $l=2,3$. Figure 6: Gain convergence tests using the iterative sky map & calibration solver run on a pair of simulations with known, but different, inputs. Both panels show the recovered gain as a function of iteration number for a 4-channel K band simulation. The initial calibration guess was chosen to be in error by 1% to test convergence; the output solutions, extrapolated with an exponential fit, are printed in each panel. top: Results for a noiseless simulation that includes a dipole signal (with Earth-velocity modulation) plus CMB and foreground anisotropy (the former is evaluated at the center frequency of each channel). The input gain was set to be constant in time. The extrapolated solutions agree with the input values to much better than 0.1%. bottom: Results for a noiseless simulation that includes only dipole signal (with Earth-velocity modulation) but no CMB or foreground signal. In this case the input gain was set up to have flight-like thermal variations. The extrapolated absolute gain recovery was in error by $>$0.3%, indicating a small residual degeneracy between the sky model and the time-dependent calibration. Figure 7: Gain error recovery test from a flight-like simulation that includes every effect known to be important. Using the daily dipole gains recovered from the iterative sky map & calibration solver as input, the gain convergence error, shown here, is fit simultaneously with the gain model parameters, not shown, following the procedure outlined in Appendix B. The red trace indicates the true gain error for each WMAP channel, based on the known input gain and the gain solution achieved by the iterative solver on its final iteration. The black trace shows the gain error recovered by the fit, averaged by frequency band. The channel-to-channel scatter within a band is $<$0.1%, though the mean of Ka band error is of order 0.1%. Figure 8: W band EE power spectrum likelihood from $l=2-7$ using two separate estimation methods: black: maximum likelihood and red: pseudo-$C_{l}$. The vertical dashed lines indicate the best-fit model power spectrum based on fitting the combined Ka, Q, and V band data. The two spectrum estimates are consistent with each other, except at $l=3$. The maximum likelihood estimates are wider because they include cosmic variance whereas the pseudo-$C_{l}$ estimates account for noise only. Both estimates show excess power in the W band data relative to the best-fit model, and to the combined KaQV band spectrum, shown in Figure 6 of Nolta et al. (2008). The extreme excess in the $l=7$ pseudo-$C_{l}$ estimate is not so severe in the maximum likelihood, but both methods are still inconsistent with the best-fit model. Figure 9: top: Simulated W band data with a large polarized thermal emission signal injected, binned by solar azimuth angle. The red trace shows the input waveform based on the flight mirror temperature profile and a model of the polarized emissivity. The black profile is the binned co-added data which follows the input signal very well. The thickness of the points represents the 1 $\sigma$ uncertainty due to white noise. bottom: Same as the top panel but for the 5 year flight data. The reduced $\chi^{2}$ of the binned data with respect to zero is 2.1 for 36 degrees of freedom, but this does not account for 1/f noise, so the significance of this result requires further investigation. However, the much larger signal in the simulation did not produce an EE spectrum with features present in the flight W band EE spectrum, so the feature in the binned flight data cannot account for the excess $l=7$ emission. Figure 10: The EE power spectrum computed from the null sky maps, $\frac{1}{2}S_{\rm Ka}-\frac{1}{2}S_{\rm QV}$, where S = Q,U are the polarization Stokes parameters, and $S_{\rm QV}$ is the optimal combination of the Q and V band data. The pink curve is the best-fit theoretical spectrum from Dunkley et al. (2008). The spectrum derived from the null maps is consistent with zero. Figure 11: Estimates of the optical depth from a variety of data combinations. The dashed curve labeled Null uses the same null sky maps used in Figure 10. The optical depth obtained from Ka band data alone (blue) is consistent with independent estimates from the combined Q and V band data (red). The final 5 year analysis uses Ka, Q, and V band data combined (black). These estimates all use a 1-parameter likelihood estimation, holding other parameters fixed except for the fluctuation amplitude, which is adjusted to fit the first acoustic peak in the TT spectrum (Page et al., 2007). The degeneracy between $\tau$ and other $\Lambda$CDM parameters is small: see Figure 7 of Dunkley et al. (2008). Figure 12: The foreground-reduced Internal Linear Combination (ILC) map based on the 5 year WMAP data. Figure 13: The temperature (TT) and temperature-polarization correlation (TE) power spectra based on the 5 year WMAP data. The addition of 2 years of data provide more sensitive measurements of the third peak in TT and the high-$l$ TE spectrum, especially the second trough. Table 1: Differencing Assembly (DA) Properties DA | $\lambda$aaEffective wavelength and frequency for a thermodynamic spectrum. | $\nu$aaEffective wavelength and frequency for a thermodynamic spectrum. | $g(\nu)$bbConversion from antenna temperature to thermodynamic temperature, $\Delta T=g(\nu)\Delta T_{A}$. | $\theta_{\rm FWHM}$ccFull-width-at-half-maximum from radial profile of A- and B-side average beams. Note: beams are not Gaussian. | $\sigma_{0}$(I)ddNoise per observation for resolution 9 and 10 $I$, $Q$, & $U$ maps, to $\sim$0.1% uncertainty. $\sigma(p)=\sigma_{0}N_{\rm obs}^{-1/2}(p)$. | $\sigma_{0}$(Q,U)ddNoise per observation for resolution 9 and 10 $I$, $Q$, & $U$ maps, to $\sim$0.1% uncertainty. $\sigma(p)=\sigma_{0}N_{\rm obs}^{-1/2}(p)$. | $\nu_{\rm s}$eeEffective frequency for synchrotron (s), free-free (ff), and dust (d) emission, assuming spectral indices of $\beta=-2.9,-2.1,+2.0$, respectively, in antenna temperature units. | $\nu_{\rm ff}$eeEffective frequency for synchrotron (s), free-free (ff), and dust (d) emission, assuming spectral indices of $\beta=-2.9,-2.1,+2.0$, respectively, in antenna temperature units. | $\nu_{\rm d}$eeEffective frequency for synchrotron (s), free-free (ff), and dust (d) emission, assuming spectral indices of $\beta=-2.9,-2.1,+2.0$, respectively, in antenna temperature units. ---|---|---|---|---|---|---|---|---|--- | (mm) | (GHz) | | (∘) | (mK) | (mK) | (GHz) | (GHz) | (GHz) K1 | 13.17 | 22.77 | 1.0135 | 0.807 | 1.436 | 1.453 | 22.47 | 22.52 | 22.78 Ka1 | 9.079 | 33.02 | 1.0285 | 0.624 | 1.470 | 1.488 | 32.71 | 32.76 | 33.02 Q1 | 7.342 | 40.83 | 1.0440 | 0.480 | 2.254 | 2.278 | 40.47 | 40.53 | 40.85 Q2 | 7.382 | 40.61 | 1.0435 | 0.475 | 2.141 | 2.163 | 40.27 | 40.32 | 40.62 V1 | 4.974 | 60.27 | 1.0980 | 0.324 | 3.314 | 3.341 | 59.65 | 59.74 | 60.29 V2 | 4.895 | 61.24 | 1.1010 | 0.328 | 2.953 | 2.975 | 60.60 | 60.70 | 61.27 W1 | 3.207 | 93.49 | 1.2480 | 0.213 | 5.899 | 5.929 | 92.68 | 92.82 | 93.59 W2 | 3.191 | 93.96 | 1.2505 | 0.196 | 6.565 | 6.602 | 93.34 | 93.44 | 94.03 W3 | 3.226 | 92.92 | 1.2445 | 0.196 | 6.926 | 6.964 | 92.34 | 92.44 | 92.98 W4 | 3.197 | 93.76 | 1.2495 | 0.210 | 6.761 | 6.800 | 93.04 | 93.17 | 93.84 Table 2: Lost and Rejected Data Category | K-band | Ka-band | Q-band | V-band | W-band ---|---|---|---|---|--- Lost or incomplete telemetry(%) | 0.12 | 0.12 | 0.12 | 0.12 | 0.12 Spacecraft anomalies(%) | 0.44 | 0.46 | 0.52 | 0.44 | 0.48 Planned stationkeeping maneuvers(%) | 0.39 | 0.39 | 0.39 | 0.39 | 0.39 Planet in beam (%) | 0.11 | 0.11 | 0.11 | 0.11 | 0.11 | —— | —— | —— | —— | —— Total lost or rejected (%) | 1.06 | 1.08 | 1.14 | 1.06 | 1.10 Table 3: Change in low-$l$ Power from 3 year Data Band | $l=0$aa$l=0,1$ \- Amplitude in the difference map, outside the processing cut, in $\mu$K. | $l=1$aa$l=0,1$ \- Amplitude in the difference map, outside the processing cut, in $\mu$K. | $l=2$bb$l=2,3$ \- Power in the difference map, outside the processing cut, $l(l+1)\,C_{l}/2\pi$, in $\mu$K2. | $l=3$bb$l=2,3$ \- Power in the difference map, outside the processing cut, $l(l+1)\,C_{l}/2\pi$, in $\mu$K2. ---|---|---|---|--- | ($\mu$K) | ($\mu$K) | ($\mu$K2) | ($\mu$K2) K | 9.3 | 5.1 | 4.1 | 0.7 Ka | 18.9 | 2.1 | 2.8 | 0.2 Q | 18.3 | 0.4 | 2.5 | 0.5 V | 14.4 | 7.3 | 1.2 | 0.0 W | 16.4 | 3.5 | 1.0 | 0.0 Table 4: WMAP 5 year CMB Dipole AnisotropyaaThe CMB dipole components for two different galactic cleaning methods are given in the first two rows. The Gibbs samples from each set are combined in the last row to produce an estimate with conservative uncertainties that encompasses both cases. Cleaning | $d_{x}$bbThe cartesian dipole components are given in Galactic coordinates. The quoted uncertainties reflect the effects of noise and sky cut, for illustration. An absolute calibration uncertainty of 0.2% should be added in quadrature. | $d_{y}$ | $d_{z}$ | $d$ccThe spherical components of the dipole are given in Galactic coordinates. In this case the quoted uncertainty in the magnitude, $d$, includes the absolute calibration uncertainty. | $l$ | $b$ ---|---|---|---|---|---|--- method | (mK) | (mK) | (mK) | (mK) | (∘) | (∘) Templates | $-0.229\pm 0.003$ | $-2.225\pm 0.003$ | $2.506\pm 0.003$ | $3.359\pm 0.008$ | $264.11\pm 0.08$ | $48.25\pm 0.03$ ILC | $-0.238\pm 0.003$ | $-2.218\pm 0.002$ | $2.501\pm 0.001$ | $3.352\pm 0.007$ | $263.87\pm 0.07$ | $48.26\pm 0.02$ Combined | $-0.233\pm 0.005$ | $-2.222\pm 0.004$ | $2.504\pm 0.003$ | $3.355\pm 0.008$ | $263.99\pm 0.14$ | $48.26\pm 0.03$ Table 5: Polarization $\chi^{2}$ Consistency TestsaaTable gives $\chi^{2}$ per degree of freedom of the independent spectrum estimates per multipole per band or band-pair, estimated from the template-cleaned maps. See text for details. Multipole | KaKa | KaQ | KaV | KaW | QQ | QV | QW | VV | VW | WW ---|---|---|---|---|---|---|---|---|---|--- | (10)bbSecond header row indicates the number of degrees of freedom in the reduced $\chi^{2}$ for that spectrum. See text for details. | (50) | (50) | (100) | (45) | (100) | (200) | (45) | (200) | (190) EE 2 | 0.727 | 1.059 | 1.019 | 1.301 | 1.586 | 0.690 | 1.179 | 0.894 | 1.078 | 1.152 3 | 1.373 | 0.994 | 1.683 | 1.355 | 1.092 | 1.614 | 1.325 | 1.005 | 1.386 | 1.519 4 | 1.561 | 1.816 | 1.341 | 2.033 | 0.993 | 1.126 | 1.581 | 1.195 | 1.596 | 1.724 5 | 0.914 | 1.313 | 1.062 | 1.275 | 1.631 | 1.052 | 1.155 | 0.589 | 0.881 | 1.252 6 | 1.003 | 0.847 | 0.688 | 1.124 | 0.740 | 0.856 | 1.049 | 1.384 | 1.168 | 1.142 7 | 0.600 | 0.671 | 0.689 | 0.936 | 0.936 | 0.780 | 0.864 | 0.900 | 1.064 | 1.015 8 | 1.578 | 1.262 | 1.337 | 1.212 | 1.080 | 0.763 | 0.608 | 1.025 | 0.871 | 0.749 9 | 0.760 | 0.710 | 0.891 | 0.820 | 0.582 | 0.726 | 0.651 | 0.791 | 0.821 | 0.795 10 | 0.494 | 0.821 | 0.996 | 0.914 | 0.656 | 0.763 | 0.806 | 0.676 | 0.891 | 0.943 EB 2 | 0.900 | 1.297 | 1.179 | 2.074 | 1.006 | 0.915 | 2.126 | 1.242 | 2.085 | 2.309 3 | 0.719 | 1.599 | 0.651 | 2.182 | 1.295 | 0.986 | 2.739 | 1.095 | 3.276 | 3.157 4 | 0.746 | 1.702 | 1.378 | 1.777 | 1.926 | 1.110 | 1.435 | 1.028 | 1.279 | 1.861 5 | 1.161 | 0.948 | 0.945 | 1.003 | 1.149 | 1.232 | 1.468 | 0.699 | 1.122 | 1.516 6 | 0.475 | 1.183 | 0.651 | 0.687 | 0.829 | 1.023 | 0.814 | 1.201 | 1.136 | 0.960 7 | 1.014 | 1.007 | 0.829 | 0.700 | 0.817 | 0.759 | 1.112 | 0.616 | 0.802 | 1.233 8 | 0.849 | 0.897 | 1.279 | 0.861 | 0.681 | 0.689 | 0.955 | 1.021 | 0.954 | 0.996 9 | 0.743 | 0.734 | 1.007 | 1.112 | 0.820 | 0.798 | 0.686 | 0.882 | 0.808 | 0.824 10 | 0.413 | 1.003 | 1.316 | 0.859 | 0.722 | 0.900 | 0.693 | 1.124 | 0.836 | 0.852 BB 2 | 2.038 | 1.570 | 1.244 | 2.497 | 1.340 | 1.219 | 2.529 | 0.694 | 1.631 | 9.195 3 | 0.756 | 0.868 | 0.808 | 1.817 | 3.027 | 1.717 | 3.496 | 0.601 | 2.545 | 5.997 4 | 1.058 | 1.455 | 1.522 | 2.144 | 1.007 | 0.905 | 1.786 | 0.752 | 1.403 | 1.984 5 | 1.221 | 1.659 | 1.742 | 2.036 | 0.889 | 1.057 | 1.271 | 1.078 | 1.660 | 1.255 6 | 0.379 | 0.805 | 0.483 | 0.812 | 1.009 | 0.861 | 1.238 | 0.800 | 0.767 | 0.955 7 | 1.925 | 1.594 | 0.967 | 1.332 | 1.074 | 0.817 | 0.928 | 0.772 | 0.994 | 1.024 8 | 0.804 | 1.005 | 0.999 | 0.912 | 1.069 | 0.782 | 0.831 | 0.997 | 0.879 | 0.943 9 | 0.320 | 0.489 | 0.502 | 0.450 | 0.884 | 0.491 | 0.729 | 0.748 | 0.664 | 0.959 10 | 1.181 | 1.162 | 1.028 | 0.980 | 1.218 | 1.165 | 0.951 | 1.079 | 0.621 | 0.791 Table 6: Loss Imbalance CoefficientsaaLoss imbalance is defined as $x_{\rm im}=(\epsilon_{A}-\epsilon_{B})/(\epsilon_{A}+\epsilon_{B})$. See §6.2 and Jarosik et al. (2007) for details. DA | $x_{\rm im,1}$ | $x_{\rm im,2}$ ---|---|--- | (%) | (%) K1 | 0.012 | 0.589 Ka1 | 0.359 | 0.148 Q1 | -0.031 | 0.412 Q2 | 0.691 | 1.048 V1 | 0.041 | 0.226 V2 | 0.404 | 0.409 W1 | 0.939 | 0.128 W2 | 0.601 | 1.140 W3 | -0.009 | 0.497 W4 | 2.615 | 1.946 Table 7: Cosmological Parameter Summary Description | Symbol | WMAP-only | WMAP+BAO+SN ---|---|---|--- Parameters for Standard $\Lambda$CDM Model aaThe parameters reported in the first section assume the 6 parameter $\Lambda$CDM model, first using WMAP data only (Dunkley et al., 2008), then using WMAP+BAO+SN data (Komatsu et al., 2008). Age of universe | $t_{0}$ | $13.69\pm 0.13\ \mbox{Gyr}$ | $13.72\pm 0.12\ \mbox{Gyr}$ Hubble constant | $H_{0}$ | $71.9^{+2.6}_{-2.7}\ \mbox{km/s/Mpc}$ | $70.5\pm 1.3\ \mbox{km/s/Mpc}$ Baryon density | $\Omega_{b}$ | $0.0441\pm 0.0030$ | $0.0456\pm 0.0015$ Physical baryon density | $\Omega_{b}h^{2}$ | $0.02273\pm 0.00062$ | $0.02267^{+0.00058}_{-0.00059}$ Dark matter density | $\Omega_{c}$ | $0.214\pm 0.027$ | $0.228\pm 0.013$ Physical dark matter density | $\Omega_{c}h^{2}$ | $0.1099\pm 0.0062$ | $0.1131\pm 0.0034$ Dark energy density | $\Omega_{\Lambda}$ | $0.742\pm 0.030$ | $0.726\pm 0.015$ Curvature fluctuation amplitude, $k_{0}=0.002$ Mpc-1 bb$k=0.002$ Mpc-1 $\longleftrightarrow$ $l_{\rm eff}\approx 30$. | $\Delta_{\cal R}^{2}$ | $(2.41\pm 0.11)\times 10^{-9}$ | $(2.445\pm 0.096)\times 10^{-9}$ Fluctuation amplitude at $8h^{-1}$ Mpc | $\sigma_{8}$ | $0.796\pm 0.036$ | $0.812\pm 0.026$ $l(l+1)C^{TT}_{220}/2\pi$ | $C_{220}$ | $5756\pm 42$ $\mu$K2 | $5751^{+42}_{-43}$ $\mu$K2 Scalar spectral index | $n_{s}$ | $0.963^{+0.014}_{-0.015}$ | $0.960\pm 0.013$ Redshift of matter-radiation equality | $z_{\rm eq}$ | $3176^{+151}_{-150}$ | $3253^{+89}_{-87}$ Angular diameter distance to matter-radiation eq.ccComoving angular diameter distance. | $d_{A}(z_{\rm eq})$ | $14279^{+186}_{-189}\ \mbox{Mpc}$ | $14200^{+137}_{-140}\ \mbox{Mpc}$ Redshift of decoupling | $z_{*}$ | $1090.51\pm 0.95$ | $1090.88\pm 0.72$ Age at decoupling | $t_{*}$ | $380081^{+5843}_{-5841}\ \mbox{yr}$ | $376971^{+3162}_{-3167}\ \mbox{yr}$ Angular diameter distance to decoupling c,dc,dfootnotemark: | $d_{A}(z_{*})$ | $14115^{+188}_{-191}\ \mbox{Mpc}$ | $14034^{+138}_{-142}\ \mbox{Mpc}$ Sound horizon at decoupling dd$l_{A}(z_{*})\equiv\pi\,d_{A}(z_{*})\,r_{s}(z_{*})^{-1}$. | $r_{s}(z_{*})$ | $146.8\pm 1.8\ \mbox{Mpc}$ | $145.9^{+1.1}_{-1.2}\ \mbox{Mpc}$ Acoustic scale at decoupling dd$l_{A}(z_{*})\equiv\pi\,d_{A}(z_{*})\,r_{s}(z_{*})^{-1}$. | $l_{A}(z_{*})$ | $302.08^{+0.83}_{-0.84}$ | $302.13\pm 0.84$ Reionization optical depth | $\tau$ | $0.087\pm 0.017$ | $0.084\pm 0.016$ Redshift of reionization | $z_{\rm reion}$ | $11.0\pm 1.4$ | $10.9\pm 1.4$ Age at reionization | $t_{\rm reion}$ | $427^{+88}_{-65}$ Myr | $432^{+90}_{-67}$ Myr Parameters for Extended Models eeThe parameters reported in the second section place limits on deviations from the $\Lambda$CDM model, first using WMAP data only (Dunkley et al., 2008), then using WMAP+BAO+SN data (Komatsu et al., 2008). A complete listing of all parameter values and uncertainties for each of the extended models studied is available on LAMBDA. Total density ffAllows non-zero curvature, $\Omega_{k}\neq 0$. | $\Omega_{\rm tot}$ | $1.099^{+0.100}_{-0.085}$ | $1.0050^{+0.0060}_{-0.0061}$ Equation of state ggAllows $w\neq-1$, but assumes $w$ is constant. | $w$ | $-1.06^{+0.41}_{-0.42}$ | $-0.992^{+0.061}_{-0.062}$ Tensor to scalar ratio, $k_{0}=0.002$ Mpc-1 b,hb,hfootnotemark: | $r$ | $<0.43\ \mbox{(95\% CL)}$ | $<0.22\ \mbox{(95\% CL)}$ Running of spectral index, $k_{0}=0.002$ Mpc-1 b,ib,ifootnotemark: | $dn_{s}/d\ln{k}$ | $-0.037\pm 0.028$ | $-0.028\pm 0.020$ Neutrino density jjAllows a massive neutrino component, $\Omega_{\nu}\neq 0$. | $\Omega_{\nu}h^{2}$ | $<0.014\ \mbox{(95\% CL)}$ | $<0.0071\ \mbox{(95\% CL)}$ Neutrino mass jjAllows a massive neutrino component, $\Omega_{\nu}\neq 0$. | $\sum m_{\nu}$ | $<1.3\ \mbox{eV}\ \mbox{(95\% CL)}$ | $<0.67\ \mbox{eV}\ \mbox{(95\% CL)}$ Number of light neutrino families kkAllows $N_{\rm eff}$ number of relativistic species. The last column adds the HST prior to the other data sets. | $N_{\rm eff}$ | $>2.3\ \mbox{(95\% CL)}$ | $4.4\pm 1.5$ hhfootnotetext: Allows tensors modes but no running in scalar spectral index. iifootnotetext: Allows running in scalar spectral index but no tensor modes.

arxiv-papers

2024-09-04T02:48:54.177657

0803.0929

# Graph Sparsification by Effective Resistances††thanks: This material is based upon work supported by the National Science Foundation under Grants No. CCF-0707522 and CCF-0634957. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. Daniel A. Spielman Program in Applied Mathematics and Department of Computer Science Yale University Nikhil Srivastava Department of Computer Science Yale University ###### Abstract We present a nearly-linear time algorithm that produces high-quality spectral sparsifiers of weighted graphs. Given as input a weighted graph $G=(V,E,w)$ and a parameter $\epsilon>0$, we produce a weighted subgraph $H=(V,\tilde{E},\tilde{w})$ of $G$ such that $|\tilde{E}|=O(n\log n/\epsilon^{2})$ and for all vectors $x\in\mathbb{R}^{V}$ $(1-\epsilon)\sum_{uv\in E}(x(u)-x(v))^{2}w_{uv}\leq\sum_{uv\in\tilde{E}}(x(u)-x(v))^{2}\tilde{w}_{uv}\leq(1+\epsilon)\sum_{uv\in E}(x(u)-x(v))^{2}w_{uv}.$ (1) This improves upon the spectral sparsifiers constructed by Spielman and Teng, which had $O(n\log^{c}n)$ edges for some large constant $c$, and upon the cut sparsifiers of Benczúr and Karger, which only satisfied (1) for $x\in\\{0,1\\}^{V}$. A key ingredient in our algorithm is a subroutine of independent interest: a nearly-linear time algorithm that builds a data structure from which we can query the approximate effective resistance between any two vertices in a graph in $O(\log n)$ time. ## 1 Introduction The goal of sparsification is to approximate a given graph $G$ by a sparse graph $H$ on the same set of vertices. If $H$ is close to $G$ in some appropriate metric, then $H$ can be used as a proxy for $G$ in computations without introducing too much error. At the same time, since $H$ has very few edges, computation with and storage of $H$ should be cheaper. We study the notion of spectral sparsification introduced by Spielman and Teng [25]. Spectral sparsification was inspired by the notion of cut sparisification introduced by Benczúr and Karger [5] to accelerate cut algorithms whose running time depends on the number of edges. They gave a nearly-linear time procedure which takes a graph $G$ on $n$ vertices with $m$ edges and a parameter $\epsilon>0$, and outputs a weighted subgraph $H$ with $O(n\log n/\epsilon^{2})$ edges such that the weight of every cut in $H$ is within a factor of $(1\pm\epsilon)$ of its weight in $G$. This was used to turn Goldberg and Tarjan’s $\widetilde{O}(mn)$ max-flow algorithm [16] into an $\widetilde{O}(n^{2})$ algorithm for approximate $st$-mincut, and appeared more recently as the first step of an $\widetilde{O}(n^{3/2}+m)$-time $O(\log^{2}n)$ approximation algorithm for sparsest cut [19]. The cut-preserving guarantee of [5] is equivalent to satisfying (1) for all $x\in\\{0,1\\}^{n}$, which are the characteristic vectors of cuts. Spielman and Teng [23, 25] devised stronger sparsifiers which extend (1) to all $x\in\mathbb{R}^{n}$, but have $O(n\log^{c}n)$ edges for some large constant $c$. They used these sparsifiers to construct preconditioners for symmetric diagonally-dominant matrices, which led to the first nearly-linear time solvers for such systems of equations. In this work, we construct sparsifiers that achieve the same guarantee as Spielman and Teng’s but with $O(n\log n/\epsilon^{2})$ edges, thus improving on both [5] and [23]. Our sparsifiers are subgraphs of the original graph and can be computed in $\widetilde{O}(m)$ time by random sampling, where the sampling probabilities are given by the effective resistances of the edges. While this is conceptually much simpler than the recursive partitioning approach of [23], we need to solve $O(\log n)$ linear systems to compute the effective resistances quickly, and we do this using Spielman and Teng’s linear equation solver. ### 1.1 Our Results Our main idea is to include each edge of $G$ in the sparsifier $H$ with probability proportional to its effective resistance. The effective resistance of an edge is known to be equal to the probability that the edge appears in a random spanning tree of $G$ (see, e.g., [9] or [6]), and was proven in [7] to be proportional to the commute time between the endpoints of the edge. We show how to approximate the effective resistances of edges in $G$ quickly and prove that sampling according to these approximate values yields a good sparsifier. To define effective resistance, identify $G=(V,E,w)$ with an electrical network on $n$ nodes in which each edge $e$ corresponds to a link of conductance $w_{e}$ (i.e., a resistor of resistance $1/w_{e}$). Then the effective resistance $R_{e}$ across an edge $e$ is the potential difference induced across it when a unit current is injected at one end of $e$ and extracted at the other end of $e$. Our algorithm can now be stated as follows. $H=\textbf{Sparsify}(G,q)$ --- Choose a random edge $e$ of $G$ with probability $p_{e}$ proportional to $w_{e}R_{e}$, and add $e$ to $H$ with weight $w_{e}/qp_{e}$. Take $q$ samples independently with replacement, summing weights if an edge is chosen more than once. Recall that the Laplacian of a weighted graph is given by $L=D-A$ where $A$ is the weighted adjacency matrix $(a_{ij})=w_{ij}$ and $D$ is the diagonal matrix $(d_{ii})=\sum_{j\neq i}w_{ij}$ of weighted degrees. Notice that the quadratic form associated with $L$ is just $x^{T}Lx=\sum_{uv\in E}(x(u)-x(v))^{2}w_{uv}$. Let $L$ be the Laplacian of $G$ and let $\tilde{L}$ be the Laplacian of $H$. Our main theorem is that if $q$ is sufficiently large, then the quadratic forms of $L$ and $\tilde{L}$ are close. ###### Theorem 1. Suppose $G$ and $H=\mathbf{Sparsify}(G,q)$ have Laplacians $L$ and $\tilde{L}$ respectively, and $1/\sqrt{n}<\epsilon\leq 1$. If $q=9C^{2}n\log n/\epsilon^{2}$, where $C$ is the constant in Lemma 5 and if $n$ is sufficiently large, then with probability at least $1/2$ $\forall x\in\mathbb{R}^{n}\quad(1-\epsilon)x^{T}Lx\leq x^{T}\tilde{L}x\leq(1+\epsilon)x^{T}Lx.$ (2) Sparsifiers that satisfy this condition preserve many properties of the graph. The Courant-Fischer Theorem tells us that $\lambda_{i}=\max_{S:\dim(S)=k}\min_{x\in S}\frac{x^{T}Lx}{x^{T}x}.$ Thus, if $\lambda_{1},\dots,\lambda_{n}$ are the eigenvalues of $L$ and $\tilde{\lambda}_{1},\dots,\tilde{\lambda}_{n}$ are the eigenvalues of $\tilde{L}$, then we have $(1-\epsilon)\lambda_{i}\leq\tilde{\lambda}_{i}\leq(1+\epsilon)\lambda_{i},$ and the eigenspaces spanned by corresponding eigenvalues are related. As the eigenvalues of the normalized Laplacian are given by $\lambda_{i}=\max_{S:\dim(S)=k}\min_{x\in S}\frac{x^{T}D^{-1/2}LD^{-1/2}x}{x^{T}x},$ and are the same as the eigenvalues of the walk matrix $D^{-1}L$, we obtain the same relationship between the eigenvalues of the walk matrix of the original graph and its sparsifier. Many properties of graphs and random walks are known to be revealed by their spectra (see for example [6, 8, 15]). The existence of sparse subgraphs which retain these properties is interesting its own right; indeed, expander graphs can be viewed as constant degree sparsifiers for the complete graph. We remark that the condition (2) also implies $\forall x\in\mathbb{R}^{n}\quad\frac{1}{1+\epsilon}x^{T}L^{+}x\leq x^{T}\tilde{L}^{+}x\leq\frac{1}{1-\epsilon}x^{T}L^{+}x,$ where $L^{+}$ is the pseudoinverse of $L$. Thus sparsifiers also approximately preserve the effective resistances between vertices, since for vertices $u$ and $v$, the effective resistance between them is given by the formula $(\chi_{u}-\chi_{v})^{T}L^{+}(\chi_{u}-\chi_{v})$, where $\chi_{u}$ is the elementary unit vector with a coordinate 1 in position $u$. We prove Theorem 1 in Section 3. At the end of Section 3, we prove that the spectral guarantee (2) of Theorem 1 is not harmed too much if use approximate effective resistances for sampling instead of exact ones(Corollary 6). In Section 4, we show how to compute approximate effective resistances in nearly-linear time, which is essentially optimal. The tools we use to do this are Spielman and Teng’s nearly-linear time solver [23, 24] and the Johnson- Lindenstrauss Lemma [18, 1]. Specifically, we prove the following theorem, in which $R_{uv}$ denotes the effective resistance between vertices $u$ and $v$. ###### Theorem 2. There is an $\widetilde{O}(m(\log r)/\epsilon^{2})$ time algorithm which on input $\epsilon>0$ and $G=(V,E,w)$ with $r=w_{max}/w_{min}$ computes a $(24\log n/\epsilon^{2})\times n$ matrix $\widetilde{Z}$ such that with probability at least $1-1/n$ $(1-\epsilon)R_{uv}\leq\|\widetilde{Z}(\chi_{u}-\chi_{v})\|^{2}\leq(1+\epsilon)R_{uv}$ for every pair of vertices $u,v\in V$. Since $\widetilde{Z}(\chi_{u}-\chi_{v})$ is simply the difference of the corresponding two columns of $\widetilde{Z}$, we can query the approximate effective resistance between any pair of vertices $(u,v)$ in time $O(\log n/\epsilon^{2})$, and for all the edges in time $O(m\log n/\epsilon^{2})$. By Corollary 6, this yields an $\widetilde{O}(m(\log r)/\epsilon^{2})$ time for sparsifying graphs, as advertised. In Section 5, we show that $H$ can be made close to $G$ in some additional ways which make it more useful for preconditioning systems of linear equations. ### 1.2 Related Work Batson, Spielman, and Srivastava [4] have given a deterministic algorithm that constructs sparsifiers of size $O(n/\epsilon^{2})$ in $O(mn^{3}/\epsilon^{2})$ time. While this is too slow to be useful in applications, it is optimal in terms of the tradeoff between sparsity and quality of approximation and can be viewed as generalizing expander graphs. Their construction parallels ours in that it reduces the task of spectral sparsification to approximating the matrix $\Pi$ defined in Section 3; however, their method for selecting edges is iterative and more delicate than the random sampling described in this paper. In addition to the graph sparsifiers of [5, 4, 23], there is a large body of work on sparse [3, 2] and low-rank [14, 2, 22, 10, 11] approximations for general matrices. The algorithms in this literature provide guarantees of the form $\|A-\tilde{A}\|_{2}\leq\epsilon$, where $A$ is the original matrix and $\tilde{A}$ is obtained by entrywise or columnwise sampling of $A$. This is analogous to satisfying (1) only for vectors $x$ in the span of the dominant eigenvectors of $A$; thus, if we were to use these sparsifiers on graphs, they would only preserve the large cuts. Interestingly, our proof uses some of the same machinery as the low-rank approximation result of Rudelson and Vershynin [22] — the sampling of edges in our algorithm corresponds to picking $q=O(n\log n)$ columns at random from a certain rank $(n-1)$ matrix of dimension $m\times m$ (this is the matrix $\Pi$ introduced in Section 3). The use of effective resistance as a distance in graphs has recently gained attention as it is often more useful than the ordinary geodesic distance in a graph. For example, in small-world graphs, all vertices will be close to one another, but those with a smaller effective resistance distance are connected by more short paths. See, for instance [13, 12], which use effective resistance/commute time as a distance measure in social network graphs. ## 2 Preliminaries ### 2.1 The Incidence Matrix and the Laplacian Let $G=(V,E,w)$ be a connected weighted undirected graph with $n$ vertices and $m$ edges and edge weights $w_{e}>0$. If we orient the edges of $G$ arbitrarily, we can write its Laplacian as $L=B^{T}WB$, where $B_{m\times n}$ is the signed edge-vertex incidence matrix, given by $B(e,v)=\left\\{\begin{array}[]{ll}1&\textrm{if $v$ is $e$'s head}\\\ -1&\textrm{if $v$ is $e$'s tail}\\\ 0&\textrm{otherwise}\end{array}\right.$ and $W_{m\times m}$ is the diagonal matrix with $W(e,e)=w_{e}$. Denote the row vectors of $B$ by $\\{b_{e}\\}_{e\in E}$ and the span of its columns by $\mathbb{B}=\mathrm{im}(B)\subseteq\mathbb{R}^{m}$ (also called the cut space of $G$ [15]). Note that $b_{(u,v)}^{T}=(\chi_{v}-\chi_{u})$. It is immediate that $L$ is positive semidefinite since $x^{T}Lx=x^{T}B^{T}WBx=\|W^{1/2}Bx\|_{2}^{2}\geq 0\quad\textrm{ for every $x\in\mathbb{R}^{n}$.}$ We also have $\ker(L)=\ker(W^{1/2}B)=\textrm{span}(\mathbf{1})$, since $\displaystyle x^{T}Lx=0$ $\displaystyle\iff\|W^{1/2}Bx\|_{2}^{2}=0$ $\displaystyle\iff\sum_{uv\in E}w_{uv}(x(u)-x(v))^{2}=0$ $\displaystyle\iff x(u)-x(v)=0\quad\textrm{for all edges $(u,v)$}$ $\displaystyle\iff\textrm{ $x$ is constant, since $G$ is connected.}$ ### 2.2 The Pseudoinverse Since $L$ is symmetric we can diagonalize it and write $L=\sum_{i=1}^{n-1}\lambda_{i}u_{i}u_{i}^{T}$ where $\lambda_{1},\ldots,\lambda_{n-1}$ are the nonzero eigenvalues of $L$ and $u_{1},\ldots,u_{n-1}$ are a corresponding set of orthonormal eigenvectors. The Moore-Penrose Pseudoinverse of $L$ is then defined as $L^{+}=\sum_{i=1}^{n-1}\frac{1}{\lambda_{i}}u_{i}u_{i}^{T}.$ Notice that $\ker(L)=\ker(L^{+})$ and that $LL^{+}=L^{+}L=\sum_{i=1}^{n-1}u_{i}u_{i}^{T},$ which is simply the projection onto the span of the nonzero eigenvectors of $L$ (which are also the eigenvectors of $L^{+}$). Thus, $LL^{+}=L^{+}L$ is the identity on $\mathrm{im}(L)=\ker(L)^{\perp}=\mathrm{span}(\mathbf{1})^{\perp}$. We will rely on this fact heavily in the proof of Theorem 1. ### 2.3 Electrical Flows Begin by arbitrarily orienting the edges of $G$ as in Section 2.1. We will use the same notation as [17] to describe electrical flows on graphs: for a vector $\mathbf{i_{\textrm{ext}}}(u)$ of currents injected at the vertices, let $\mathbf{i}(e)$ be the currents induced in the edges (in the direction of orientation) and $\mathbf{v}(u)$ the potentials induced at the vertices. By Kirchoff’s current law, the sum of the currents entering a vertex is equal to the amount injected at the vertex: $B^{T}\mathbf{i}=\mathbf{i_{\textrm{ext}}}.$ By Ohm’s law, the current flow in an edge is equal to the potential difference across its ends times its conductance: $\mathbf{i}=WB\mathbf{v}.$ Combining these two facts, we obtain $\mathbf{i_{\textrm{ext}}}=B^{T}(WB\mathbf{v})=L\mathbf{v}.$ If $\mathbf{i_{\textrm{ext}}}\perp\mathrm{span}(\mathbf{1})=\ker(L)$ — i.e., if the total amount of current injected is equal to the total amount extracted — then we can write $\mathbf{v}=L^{+}\mathbf{i_{\textrm{ext}}}$ by the definition of $L^{+}$ in Section 2.2. Recall that the effective resistance between two vertices $u$ and $v$ is defined as the potential difference induced between them when a unit current is injected at one and extracted at the other. We will derive an algebraic expression for the effective resistance in terms of $L^{+}$. To inject and extract a unit current across the endpoints of an edge $e=(u,v)$, we set $\mathbf{i_{\textrm{ext}}}=b_{e}^{T}=(\chi_{v}-\chi_{u})$, which is clearly orthogonal to $\mathbf{1}$. The potentials induced by $\mathbf{i_{\textrm{ext}}}$ at the vertices are given by $\mathbf{v}=L^{+}b_{e}^{T}$; to measure the potential difference across $e=(u,v)$, we simply multiply by $b_{e}$ on the left: $\mathbf{v}(v)-\mathbf{v}(u)=(\chi_{v}-\chi_{u})^{T}\mathbf{v}=b_{e}L^{+}b_{e}^{T}.$ It follows that the effective resistance across $e$ is given by $b_{e}L^{+}b_{e}^{T}$ and that the matrix $BL^{+}B^{T}$ has as its diagonal entries $BL^{+}B^{T}(e,e)=R_{e}$. ## 3 The Main Result We will prove Theorem 1. Consider the matrix $\Pi=W^{1/2}BL^{+}B^{T}W^{1/2}$. Since we know $BL^{+}B^{T}(e,e)=R_{e}$, the diagonal entries of $\Pi$ are $\Pi(e,e)=\sqrt{W(e,e)}R_{e}\sqrt{W(e,e)}=w_{e}R_{e}$. $\Pi$ has some notable properties. ###### Lemma 3 (Projection Matrix). (i) $\Pi$ is a projection matrix. (ii) $\mathrm{im}(\Pi)=\mathrm{im}(W^{1/2}B)=W^{1/2}\mathbb{B}$. (iii) The eigenvalues of $\Pi$ are $1$ with multiplicity $n-1$ and $0$ with multiplicity $m-n+1$. (iv) $\Pi(e,e)=\|\Pi(\cdot,e)\|^{2}$. ###### Proof. To see (i), observe that $\displaystyle\Pi^{2}$ $\displaystyle=(W^{1/2}BL^{+}B^{T}W^{1/2})(W^{1/2}BL^{+}B^{T}W^{1/2})$ $\displaystyle=W^{1/2}BL^{+}(B^{T}WB)L^{+}B^{T}W^{1/2}$ $\displaystyle=W^{1/2}BL^{+}LL^{+}B^{T}W^{1/2}\quad\textrm{ since $L=B^{T}WB$}$ $\displaystyle=W^{1/2}BL^{+}B^{T}W^{1/2}$ since $L^{+}L$ is the identity on $\mathrm{im}(L^{+}$) $\displaystyle=\Pi.$ For (ii), we have $\mathrm{im}(\Pi)=\mathrm{im}(W^{1/2}BL^{+}B^{T}W^{1/2})\subseteq\mathrm{im}(W^{1/2}B).$ To see the other inclusion, assume $y\in\mathrm{im}(W^{1/2}B)$. Then we can choose $x\perp\ker(W^{1/2}B)=\ker(L)$ such that $W^{1/2}Bx=y$. But now $\displaystyle\Pi y$ $\displaystyle=W^{1/2}BL^{+}B^{T}W^{1/2}W^{1/2}Bx$ $\displaystyle=W^{1/2}BL^{+}Lx\quad\textrm{since $B^{T}WB=L$}$ $\displaystyle=W^{1/2}Bx\quad\textrm{since $L^{+}Lx=x$ for $x\perp\ker(L)$}$ $\displaystyle=y.$ Thus $y\in\mathrm{im}(\Pi)$, as desired. For (iii), recall from Section 2.1 that $\dim(\ker(W^{1/2}B))=1$. Consequently, $\dim(\mathrm{im}(\Pi))=\dim(\mathrm{im}(W^{1/2}B))=n-1$. But since $\Pi^{2}=\Pi$, the eigenvalues of $\Pi$ are all $0$ or $1$, and as $\Pi$ projects onto a space of dimension $n-1$, it must have exactly $n-1$ nonzero eigenvalues. (iv) follows from $\Pi^{2}(e,e)=\Pi(\cdot,e)^{T}\Pi(\cdot,e)$, since $\Pi$ is symmetric. ∎ To show that $H=(V,\tilde{E},\tilde{w})$ is a good sparsifier for $G$, we need to show that the quadratic forms $x^{T}Lx$ and $x^{T}\tilde{L}x$ are close. We start by reducing the problem of preserving $x^{T}Lx$ to that of preserving $y^{T}\Pi y$. This will be much nicer since the eigenvalues of $\Pi$ are all $0$ or $1$, so that any matrix $\tilde{\Pi}$ which approximates $\Pi$ in the spectral norm (i.e., makes $\|\tilde{\Pi}-\Pi\|_{2}$ small) also preserves its quadratic form. We may describe the outcome of $H=\mathbf{Sparsify}(G,q)$ by the following random matrix: $S(e,e)=\frac{\tilde{w_{e}}}{w_{e}}=\frac{\textrm{(\\# of times $e$ is sampled)}}{qp_{e}}.$ (3) $S_{m\times m}$ is a nonnegative diagonal matrix and the random entry $S(e,e)$ specifies the ‘amount’ of edge $e$ included in $H$ by $\mathbf{Sparsify}$. For example $S(e,e)=1/qp_{e}$ if $e$ is sampled once, $2/qp_{e}$ if it is sampled twice, and zero if it is not sampled at all. The weight of $e$ in $H$ is now given by $\tilde{w_{e}}=S(e,e)w_{e}$, and we can write the Laplacian of $H$ as: $\tilde{L}=B^{T}\tilde{W}B=B^{T}W^{1/2}SW^{1/2}B$ since $\tilde{W}=WS=W^{1/2}SW^{1/2}$. The scaling of weights by $1/qp_{e}$ in $\mathbf{Sparsify}$ implies that $\mathbb{E}\tilde{w_{e}}=w_{e}$ (since $q$ independent samples are taken, each with probability $p_{e}$), and thus $\mathbb{E}S=I$ and $\mathbb{E}\tilde{L}=L$. We can now prove the following lemma, which says that if $S$ does not distort $y^{T}\Pi y$ too much then $x^{T}Lx$ and $x^{T}\tilde{L}x$ are close. ###### Lemma 4. Suppose $S$ is a nonnegative diagonal matrix such that $\|\Pi S\Pi-\Pi\Pi\|_{2}\leq\epsilon.$ Then $\forall x\in\mathbb{R}^{n}\quad(1-\epsilon)x^{T}Lx\leq x^{T}\tilde{L}x\leq(1+\epsilon)x^{T}Lx,$ where $L=B^{T}WB$ and $\tilde{L}=B^{T}W^{1/2}SW^{1/2}B$. ###### Proof. The assumption is equivalent to $\sup_{y\in\mathbb{R}^{m},y\neq 0}\frac{|y^{T}\Pi(S-I)\Pi y|}{y^{T}y}\leq\epsilon$ since $\|A\|_{2}=\sup_{y\neq 0}|y^{T}Ay|/y^{T}y$ for symmetric $A$. Restricting our attention to vectors in $\mathrm{im}(W^{1/2}B)$, we have $\sup_{y\in\mathrm{im}(W^{1/2}B),y\neq 0}\frac{|y^{T}\Pi(S-I)\Pi y|}{y^{T}y}\leq\epsilon.$ But by Lemma 3.(ii), $\Pi$ is the identity on $\mathrm{im}(W^{1/2}B)$ so $\Pi y=y$ for all $y\in\mathrm{im}(W^{1/2}B)$. Also, every such $y$ can be written as $y=W^{1/2}Bx$ for $x\in\mathbb{R}^{n}$. Substituting this into the above expression we obtain: $\displaystyle\sup_{y\in\mathrm{im}(W^{1/2}B),y\neq 0}\frac{|y^{T}\Pi(S-I)\Pi y|}{y^{T}y}$ $\displaystyle=\sup_{y\in\mathrm{im}(W^{1/2}B),y\neq 0}\frac{|y^{T}(S-I)y|}{y^{T}y}$ $\displaystyle=\sup_{x\in\mathbb{R}^{n},W^{1/2}Bx\neq 0}\frac{|x^{T}B^{T}W^{1/2}SW^{1/2}Bx-x^{T}B^{T}WBx|}{x^{T}B^{T}WBx}$ $\displaystyle=\sup_{x\in\mathbb{R}^{n},W^{1/2}Bx\neq 0}\frac{|x^{T}\tilde{L}x-x^{T}Lx|}{x^{T}Lx}\leq\epsilon.$ Rearranging yields the desired conclusion for all $x\notin\ker(W^{1/2}B)$. When $x\in\ker(W^{1/2}B)$ then $x^{T}Lx=x^{T}\tilde{L}x=0$ and the claim holds trivially.∎ To show that $\|\Pi S\Pi-\Pi\Pi\|_{2}$ is likely to be small we use the following concentration result, which is a sort of law of large numbers for symmetric rank 1 matrices. It was first proven by Rudelson in [21], but the version we state here appears in the more recent paper [22] by Rudelson and Vershynin. ###### Lemma 5 (Rudelson & Vershynin, [22] Thm. 3.1). Let $\mathbf{p}$ be a probability distribution over $\Omega\subseteq\mathbb{R}^{d}$ such that $\sup_{y\in\Omega}\|y\|_{2}\leq M$ and $\|\mathbb{E}_{\mathbf{p}}yy^{T}\|_{2}\leq 1$. Let $y_{1}\ldots y_{q}$ be independent samples drawn from $\mathbf{p}$. Then $\mathbb{E}\left\|\frac{1}{q}\sum_{i=1}^{q}y_{i}y_{i}^{T}-\mathbb{E}yy^{T}\right\|_{2}\leq\min\left(CM\sqrt{\frac{\log q}{q}},1\right)$ where $C$ is an absolute constant. We can now finish the proof of Theorem 1. ###### Proof of Theorem 1. $\mathbf{Sparsify}$ samples edges from $G$ independently with replacement, with probabilities $p_{e}$ proportional to $w_{e}R_{e}$. Since $\sum_{e}w_{e}R_{e}=\textrm{Tr}(\Pi)=n-1$ by Lemma 3.(iii), the actual probability distribution over $E$ is given by $p_{e}=\frac{w_{e}R_{e}}{n-1}$. Sampling $q$ edges from $G$ corresponds to sampling $q$ columns from $\Pi$, so we can write $\displaystyle\Pi S\Pi$ $\displaystyle=\sum_{e}S(e,e)\Pi(\cdot,e)\Pi(\cdot,e)^{T}$ $\displaystyle=\sum_{e}\frac{(\\#\textrm{ of times $e$ is sampled})}{qp_{e}}\Pi(\cdot,e)\Pi(\cdot,e)^{T}\quad\textrm{by (\ref{defS})}$ $\displaystyle=\frac{1}{q}\sum_{e}(\\#\textrm{ of times $e$ is sampled})\frac{\Pi(\cdot,e)}{\sqrt{p_{e}}}\frac{\Pi(\cdot,e)^{T}}{\sqrt{p_{e}}}$ $\displaystyle=\frac{1}{q}\sum_{i=1}^{q}y_{i}y_{i}^{T}$ for vectors $y_{1},\ldots,y_{q}$ drawn independently with replacement from the distribution $y=\frac{1}{\sqrt{p_{e}}}\Pi(\cdot,e)\quad\textrm{with probability }p_{e}.$ We can now apply Lemma 5. The expectation of $yy^{T}$ is given by $\mathbb{E}yy^{T}=\sum_{e}p_{e}\frac{1}{p_{e}}\Pi(\cdot,e)\Pi(\cdot,e)^{T}=\Pi\Pi=\Pi,$ so $\|\mathbb{E}yy^{T}\|_{2}=\|\Pi\|_{2}=1$. We also have a bound on the norm of $y$: $\frac{1}{\sqrt{p_{e}}}\|\Pi(\cdot,e)\|_{2}=\frac{1}{\sqrt{p_{e}}}\sqrt{\Pi(e,e)}\\\ =\sqrt{\frac{n-1}{R_{e}w_{e}}}\sqrt{R_{e}w_{e}}=\sqrt{n-1}.$ Taking $q=9C^{2}n\log n/\epsilon^{2}$ gives: $\mathbb{E}\left\|\Pi S\Pi-\Pi\Pi\right\|_{2}=\mathbb{E}\left\|\frac{1}{q}\sum_{i=1}^{q}y_{i}y_{i}^{T}-\mathbb{E}yy^{T}\right\|_{2}\leq C\sqrt{\epsilon^{2}\frac{\log(9C^{2}n\log n/\epsilon^{2})(n-1)}{9C^{2}n\log n}}\leq\epsilon/2,$ for $n$ sufficiently large, as $\epsilon$ is assumed to be at least $1/\sqrt{n}$. By Markov’s inequality, we have $\|\Pi S\Pi-\Pi\|_{2}\leq\epsilon$ with probability at least $1/2$. By Lemma 4, this completes the proof of the theorem. ∎ We now show that using approximate resistances for sampling does not damage the sparsifier very much. ###### Corollary 6. Suppose $Z_{e}$ are numbers satisfying $Z_{e}\geq R_{e}/\alpha$ and $\sum_{e}w_{e}Z_{e}\leq\alpha\sum_{e}w_{e}R_{e}$ for some $\alpha\geq 1$. If we sample as in $\mathbf{Sparsify}$ but take each edge with probability $p_{e}^{\prime}=\frac{w_{e}Z_{e}}{\sum_{e}w_{e}Z_{e}}$ instead of $p_{e}=\frac{w_{e}R_{e}}{\sum_{e}w_{e}R_{e}}$, then $H$ satisfies: $(1-\epsilon\alpha)x^{T}\tilde{L}x\leq x^{T}Lx\leq(1+\epsilon\alpha)x^{T}\tilde{L}x\quad\forall x\in\mathbb{R}^{n},$ with probability at least $1/2$. ###### Proof. We note that $p_{e}^{\prime}=\frac{w_{e}S_{e}}{\sum_{e}w_{e}S_{e}}\geq\frac{w_{e}(R_{e}/\alpha)}{\alpha\sum_{e}w_{e}R_{e}}=\frac{p_{e}}{\alpha^{2}}$ and proceed as in the proof of Theorem 1. The norm of the random vector $y$ is now bounded by: $\frac{1}{\sqrt{p_{e}^{\prime}}}\|\Pi(e,\cdot)\|_{2}\leq\frac{\alpha}{\sqrt{p_{e}}}\sqrt{\Pi(e,e)}\\\ =\alpha\sqrt{n-1}$ which introduces a factor of $\alpha$ into the final bound on the expectation, but changes nothing else.∎ ## 4 Computing Approximate Resistances Quickly It is not clear how to compute all the effective resistances $\\{R_{e}\\}$ exactly and efficiently. In this section, we show that one can compute constant factor approximations to all the $R_{e}$ in time $\widetilde{O}(m\log r)$. In fact, we do something stronger: we build a $O(\log n)\times n$ matrix $\widetilde{Z}$ from which the effective resistance between any two vertices (including vertices not connected by an edge) can be computed in $O(\log n)$ time. ###### Proof of Theorem 2. If $u$ and $v$ are vertices in $G$, then the effective resistance between $u$ and $v$ can be written as: $\displaystyle R_{uv}$ $\displaystyle=(\chi_{u}-\chi_{v})^{T}L^{+}(\chi_{u}-\chi_{v})$ $\displaystyle=(\chi_{u}-\chi_{v})^{T}L^{+}LL^{+}(\chi_{u}-\chi_{v})$ $\displaystyle=((\chi_{u}-\chi_{v})^{T}L^{+}B^{T}W^{1/2})(W^{1/2}BL^{+}(\chi_{u}-\chi_{v}))$ $\displaystyle=\|W^{1/2}BL^{+}(\chi_{u}-\chi_{v})^{2}\|_{2}^{2}.$ Thus effective resistances are just pairwise distances between vectors in $\\{W^{1/2}BL^{+}\chi_{v}\\}_{v\in V}$. By the Johnson-Lindenstrauss Lemma, these distances are preserved if we project the vectors onto a subspace spanned by $O(\log n)$ random vectors. For concreteness, we use the following version of the Johnson-Lindenstrauss Lemma due to Achlioptas [1]. ###### Lemma 7. Given fixed vectors $v_{1}\ldots v_{n}\in\mathbb{R}^{d}$ and $\epsilon>0$, let $Q_{k\times d}$ be a random $\pm 1/\sqrt{k}$ matrix (i.e., independent Bernoulli entries) with $k\geq 24\log n/\epsilon^{2}$. Then with probability at least $1-1/n$ $(1-\epsilon)\|v_{i}-v_{j}\|_{2}^{2}\leq\|Qv_{i}-Qv_{j}\|_{2}^{2}\leq(1+\epsilon)\|v_{i}-v_{j}\|_{2}^{2}$ for all pairs $i,j\leq n$. Our goal is now to compute the projections $\\{QW^{1/2}BL^{+}\chi_{v}\\}$. We will exploit the linear system solver of Spielman and Teng [23, 24], which we recall satisfies: ###### Theorem 8 (Spielman-Teng). There is an algorithm $x=\mathtt{STSolve}(L,y,\delta)$ which takes a Laplacian matrix $L$, a column vector $y$, and an error parameter $\delta>0$, and returns a column vector $x$ satisfying $\|x-L^{+}y\|_{L}\leq\epsilon\|L^{+}y\|_{L},$ where $\left\|y\right\|_{L}=\sqrt{y^{T}Ly}$. The algorithm runs in expected time $\widetilde{O}\left(m\log(1/\delta)\right)$, where $m$ is the number of non-zero entries in $L$. Let $Z=QW^{1/2}BL^{+}$. We will compute an approximation $\widetilde{Z}$ by using STSolve to approximately compute the rows of $Z$. Let the column vectors $z_{i}$ and $\tilde{z_{i}}$ denote the $i$th rows of $Z$ and $\tilde{Z}$, respectively (so that $z_{i}$ is the $i$th column of $Z^{T}$). Now we can construct the matrix $\widetilde{Z}$ in the following three steps. 1. 1. Let $Q$ be a random $\pm 1/\sqrt{k}$ matrix of dimension $k\times n$ where $k=24\log n/\epsilon^{2}$. 2. 2. Compute $Y=QW^{1/2}B$. Note that this takes $2m\times 24\log n/\epsilon^{2}+m=\widetilde{O}(m/\epsilon^{2})$ time since $B$ has $2m$ entries and $W^{1/2}$ is diagonal. 3. 3. Let $y_{i}$, for $1\leq i\leq k$, denote the rows of $Y$, and compute $\tilde{z}_{i}=\mathtt{STSolve}(L,y_{i},\delta)$ for each $i$. We now prove that, for our purposes, it suffices to call STSolve with $\delta=\frac{\epsilon}{3}\sqrt{\frac{2(1-\epsilon)w_{min}}{(1+\epsilon)n^{3}w_{{max}}}}.$ ###### Lemma 9. Suppose $(1-\epsilon)R_{uv}\leq\left\|Z(\chi_{u}-\chi_{v})\right\|^{2}\leq(1+\epsilon)R_{uv},$ for every pair $u,v\in V$. If for all $i$, $\|z_{i}-\tilde{z}_{i}\|_{L}\leq\delta\|z_{i}\|_{L},$ (4) where $\delta\leq\frac{\epsilon}{3}\sqrt{\frac{2(1-\epsilon)w_{min}}{(1+\epsilon)n^{3}w_{{max}}}}$ (5) then $(1-\epsilon)^{2}R_{uv}\leq\|\widetilde{Z}(\chi_{u}-\chi_{v})\|^{2}\leq(1+\epsilon)^{2}R_{uv},$ for every $uv$. ###### Proof. Consider an arbitrary pair of vertices $u$, $v$. It suffices to show that $\left|\left\|Z(\chi_{u}-\chi_{v})\right\|-\|\tilde{Z}(\chi_{u}-\chi_{v})\|\right|\leq\frac{\epsilon}{3}\left\|Z(\chi_{u}-\chi_{v})\right\|$ (6) since this will imply $\displaystyle\left|\left\|Z(\chi_{u}-\chi_{v})\right\|^{2}-\|{\tilde{Z}(\chi_{u}-\chi_{v})}\|^{2}\right|$ $\displaystyle=\left|\left\|Z(\chi_{u}-\chi_{v})\right\|-\|\tilde{Z}(\chi_{u}-\chi_{v})\|\right|\cdot\left|\left\|Z(\chi_{u}-\chi_{v})\right\|+\|\tilde{Z}(\chi_{u}-\chi_{v})\|\right|$ $\displaystyle\leq\frac{\epsilon}{3}\cdot\left(2+\frac{\epsilon}{3}\right)\left\|Z(\chi_{u}-\chi_{v})\right\|^{2}.$ As $G$ is connected, there is a simple path $P$ connecting $u$ to $v$. Applying the triangle inequality twice, we obtain $\displaystyle\left|\left\|Z(\chi_{u}-\chi_{v})\right\|-\left\|\widetilde{Z}(\chi_{u}-\chi_{v})\right\|\right|$ $\displaystyle\leq\left\|(Z-\widetilde{Z})(\chi_{u}-\chi_{v})\right\|$ $\displaystyle\leq\sum_{ab\in P}\left\|(Z-\widetilde{Z})(\chi_{a}-\chi_{b})\right\|.$ We will upper bound this later term by considering its square: $\displaystyle\left(\sum_{ab\in P}\left\|(Z-\widetilde{Z})(\chi_{a}-\chi_{b})\right\|\right)^{2}$ $\displaystyle\leq n\sum_{ab\in P}\left\|(Z-\widetilde{Z})(\chi_{a}-\chi_{b})\right\|^{2}\qquad\text{by Cauchy-Schwarz}$ $\displaystyle\leq n\sum_{ab\in E}\left\|(Z-\widetilde{Z})(\chi_{a}-\chi_{b})\right\|^{2}$ $\displaystyle=n\left\|(Z-\widetilde{Z})B^{T}\right\|_{F}^{2}\qquad\text{writing this as a Frobenius norm}$ $\displaystyle=n\left\|B(Z-\widetilde{Z})^{T}\right\|_{F}^{2}$ $\displaystyle\leq\frac{n}{w_{min}}\left\|W^{1/2}B(Z-\widetilde{Z})^{T}\right\|_{F}^{2}\qquad\text{since $\|W^{-1/2}\|_{2}\leq 1/\sqrt{w_{min}}$}$ $\displaystyle\leq\delta^{2}\frac{n}{w_{min}}\left\|W^{1/2}BZ^{T}\right\|_{F}^{2}$ since $\|W^{1/2}B(z_{i}-\tilde{z}_{i})\|^{2}\leq\delta^{2}\|W^{1/2}Bz_{i}\|^{2}$ by (4) $\displaystyle=\delta^{2}\frac{n}{w_{min}}\sum_{ab\in E}w_{ab}\left\|Z(\chi_{a}-\chi_{b})\right\|^{2}$ $\displaystyle\leq\delta^{2}\frac{n}{w_{min}}\sum_{ab\in E}w_{ab}(1+\epsilon)R_{ab}$ $\displaystyle\leq\delta^{2}\frac{n(1+\epsilon)}{w_{min}}(n-1)\qquad\text{ by Lemma~{}\ref{lempi}.(iii).}$ On the other hand, $\left\|Z(\chi_{u}-\chi_{v})\right\|^{2}\geq(1-\epsilon)R_{uv}\geq\frac{2(1-\epsilon)}{nw_{max}},$ by Proposition 10. Combining these bounds, we have $\displaystyle\frac{\left|\left\|Z(\chi_{u}-\chi_{v})\right\|-\left\|\widetilde{Z}(\chi_{u}-\chi_{v})\right\|\right|}{\left\|Z(\chi_{u}-\chi_{v})\right\|}$ $\displaystyle\leq\delta\left({\frac{n(1+\epsilon)}{w_{min}}(n-1)}\right)^{1/2}\cdot\left(\frac{nw_{max}}{2(1-\epsilon)}\right)^{1/2}$ $\displaystyle\leq\frac{\epsilon}{3}\qquad\textrm{by (\ref{eqn:stprecision}),}$ as desired. ∎ ###### Proposition 10. If $G=(V,E,w)$ is a connected graph, then for all $u,v\in V$, $R_{uv}\geq\frac{2}{nw_{max}}.$ ###### Proof. By Rayleigh’s monotonicity law (see [6]), each resistance $R_{uv}$ in $G$ is at least the corresponding resistance $R_{uv}^{\prime}$ in $G^{\prime}=w_{{max}}\times K_{n}$ (the complete graph with all edge weights $w_{{max}}$) since $G^{\prime}$ is obtained by increasing weights (i.e., conductances) of edges in $G$. But by symmetry each resistance $R_{uv}^{\prime}$ in $G^{\prime}$ is exactly $\frac{\sum_{uv}R_{uv}^{\prime}}{\binom{n}{2}}=\frac{(n-1)/w_{{max}}}{n(n-1)/2}=\frac{2}{nw_{{max}}}.$ Thus $R_{uv}\geq\frac{2}{nw_{{max}}}$ for all $u,v\in V$. ∎ Thus the construction of $\widetilde{Z}$ takes $\widetilde{O}(m\log(1/\delta)/\epsilon^{2})=\widetilde{O}(m\log r/\epsilon^{2})$ time. We can then find the approximate resistance $\|\widetilde{Z}(\chi_{u}-\chi_{v})\|^{2}\approx R_{uv}$ for any $u,v\in V$ in $O(\log n/\epsilon^{2})$ time simply by subtracting two columns of $\widetilde{Z}$ and computing the norm of their difference. ∎ Using the above procedure, we can compute arbitrarily good approximations to the effective resistances $\\{R_{e}\\}$ which we need for sampling in nearly- linear time. By Corollary 6, any constant factor approximation yields a sparsifier, so we are done. ## 5 An Additional Property Corollary 6 suggests that $\mathbf{Sparsify}$ is quite robust with respect to changes in the sampling probabilities $p_{e}$, and that we may be able to prove additional guarantees on $H$ by tweaking them. In this section, we prove one such claim. The following property is desirable for using $H$ to solve linear systems (specifically, for the construction of ultrasparsifiers [23, 24], which we will not define here): $\textrm{For every vertex $v\in V,$}\quad\sum_{e\ni v}\frac{\tilde{w}_{e}}{w_{e}}\leq 2\deg(v).$ (7) This says, roughly, that not too many of the edges incident to any given vertex get blown up too much by sampling and rescaling. We show how to incorporate this property into our sparsifiers. ###### Lemma 11. Suppose we sample $q>4n\log n/\beta$ edges of $G$ as in $\mathbf{Sparsify}$ with probabilities that satisfy $p_{(u,v)}\geq\frac{\beta}{n\min(\deg(u),\deg(v))}$ for some constant $0<\beta<1$. Then with probability at least $1-1/n$, $\sum_{e\ni v}\frac{\tilde{w}_{e}}{w_{e}}\leq 2\deg(v)\quad\textrm{for all $v\in V$.}$ ###### Proof. For a vertex $v$, define i.i.d. random variables $X_{1},\ldots,X_{q}$ by: $X_{i}=\left\\{\begin{array}[]{ll}\frac{1}{p_{e}}&\textrm{if $e\ni v$ is the $i$th edge chosen}\\\ 0&\textrm{otherwise}\end{array}\right.$ so that $X_{i}$ is set to $1/p_{e}$ with probability $p_{e}$ for each edge $e$ attached to $v$. Let $D_{v}=\sum_{e\ni v}\frac{\tilde{w_{e}}}{w_{e}}=\sum_{e\ni v}\frac{\textrm{(\\# of times $e$ is sampled)}}{qp_{e}}=\frac{1}{q}\sum_{i=1}^{q}X_{i}.$ We want to show that with high probability, $D_{v}\leq 2\deg(v)$ for all vertices $v$. We begin by bounding the expectation and variance of each $X_{i}$: $\displaystyle\mathbb{E}X_{i}$ $\displaystyle=\sum_{e\ni v}p_{e}\frac{1}{p_{e}}=\deg(v)$ $\displaystyle\mathbf{Var}(X_{i})$ $\displaystyle=\sum_{e\ni v}p_{e}\left(\frac{1}{p_{e}^{2}}-\frac{1}{p_{e}}\right)$ $\displaystyle\leq\sum_{e\ni v}\frac{1}{p_{e}}$ $\displaystyle\leq\sum_{(u,v)\ni v}\frac{n\min(\deg(u),\deg(v))}{\beta}\quad\textrm{by assumption}$ $\displaystyle\leq\sum_{(u,v)\ni v}\frac{n\deg(v)}{\beta}$ $\displaystyle=\frac{n\deg(v)^{2}}{\beta}$ Since the $X_{i}$ are independent, the variance of $D_{v}$ is just $\mathbf{Var}(D_{v})=\frac{1}{q^{2}}\sum_{i=1}^{q}\mathbf{Var}(X_{i})\leq\frac{n\deg(v)^{2}}{\beta q}.$ We now apply Bennett’s inequality for sums of i.i.d. variables (see, e.g., [20]), which says $\mathbb{P}[|D_{v}-\mathbb{E}D_{v}|>\mathbb{E}D_{v}]\leq\exp\left(\frac{-(\mathbb{E}D_{v})^{2}}{\mathbf{Var}(D_{v})(1+\frac{\mathbb{E}D_{v}}{q})}\right)$ We know that $\mathbb{E}D_{v}=\mathbb{E}X_{i}=\deg(v)$. Substituting our estimate for $\mathbf{Var}(D_{v})$ and setting $q\geq 4n\log n/\beta$ gives: $\displaystyle\mathbb{P}[D_{v}>2\deg(v)]$ $\displaystyle\leq\exp\left(\frac{-\deg(v)^{2}}{\frac{n\deg(v)^{2}}{\beta q}(1+\frac{\deg(v)}{q})}\right)$ $\displaystyle\leq\exp\left(\frac{-\beta q}{2n}\right)\quad\textrm{since $1+\frac{\deg(v)}{q}\leq 2$}$ $\displaystyle\leq\exp\left(-2\log n\right)=1/n^{2}.$ Taking a union bound over all $v$ gives the desired result.∎ Sampling with probabilities $p^{\prime}_{e}=p^{\prime}_{(u,v)}=\frac{1}{2}\left(\frac{\|Zb_{e}^{T}\|^{2}w_{e}}{\sum_{e}\|Zb_{e}^{T}\|^{2}w_{e}}+\frac{1}{n\min(\deg(u),\deg(v))}\right)$ satisfies the requirements of both Corollary 6 (with $\alpha=2$) and Lemma 11 (with $\beta=1/2$) and yields a sparsifier with the desired property. ###### Theorem 12. There is an $\widetilde{O}(m/\epsilon^{2})$ time algorithm which on input $G=(V,E,w),\epsilon>0$ produces a weighted subgraph $H=(V,\tilde{E},\tilde{w})$ of $G$ with $O(n\log n/\epsilon^{2})$ edges which, with probability at least $1/2$, satisfies both (2) and (7). ## References * [1] D. Achlioptas. Database-friendly random projections. In PODS ’01, pages 274–281, 2001. * [2] D. Achlioptas and F. McSherry. Fast computation of low rank matrix approximations. In STOC ’01, pages 611–618, 2001. * [3] S. Arora, E. Hazan, and S. Kale. A fast random sampling algorithm for sparsifying matrices. In APPROX-RANDOM ’06, volume 4110 of Lecture Notes in Computer Science, pages 272–279. Springer, 2006. * [4] Joshua D. Batson, Daniel A. Spielman, and Nikhil Srivastava. Twice-Ramanujan sparsifiers. In STOC ’09: Proceedings of the 41st annual ACM symposium on Theory of computing, pages 255–262, New York, NY, USA, 2009. ACM. * [5] A. A. Benczúr and D. R. Karger. Approximating s-t minimum cuts in $\tilde{O}(n^{2})$ time. In STOC ’96, pages 47–55, 1996. * [6] B. Bollobas. Modern Graph Theory. Springer, July 1998. * [7] A. K. Chandra, P. Raghavan, W. L. Ruzzo, and R. Smolensky. The electrical resistance of a graph captures its commute and cover times. In STOC ’89, pages 574–586, 1989. * [8] F. R. K. Chung. Spectral Graph Theory. CBMS Regional Conference Series in Mathematics. American Mathematical Society, 1997. * [9] P. Doyle and J. Snell. Random walks and electric networks. Math. Assoc. America., Washington, 1984. * [10] P. Drineas and R. Kannan. Fast monte-carlo algorithms for approximate matrix multiplication. In FOCS ’01, pages 452–459, 2001. * [11] P. Drineas and R. Kannan. Pass efficient algorithms for approximating large matrices. In SODA ’03, pages 223–232, 2003. * [12] A. Firat, S. Chatterjee, and M. Yilmaz. Genetic clustering of social networks using random walks. Computational Statistics & Data Analysis, 51(12):6285–6294, August 2007. * [13] F. Fouss, A. Pirotte, J.-M. Renders, and M. Saerens. Random-walk computation of similarities between nodes of a graph with application to collaborative recommendation. Knowledge and Data Engineering, IEEE Transactions on, 19(3):355–369, 2007. * [14] A. Frieze, R. Kannan, and S. Vempala. Fast monte-carlo algorithms for finding low-rank approximations. J. ACM, 51(6):1025–1041, 2004. * [15] Chris Godsil and Gordon Royle. Algebraic Graph Theory. Graduate Texts in Mathematics. Springer, 2001. * [16] A. V. Goldberg and R. E. Tarjan. A new approach to the maximum flow problem. In STOC ’86, pages 136–146, 1986. * [17] S. Guattery and G. L. Miller. Graph embeddings and Laplacian eigenvalues. SIAM J. Matrix Anal. Appl., 21(3):703–723, 2000. * [18] W. Johnson and J. Lindenstrauss. Extensions of Lipschitz mappings into a Hilbert space. Contemp. Math., 26:189–206, 1984. * [19] R. Khandekar, S. Rao, and U. Vazirani. Graph partitioning using single commodity flows. In STOC ’06, pages 385–390, 2006. * [20] G. Lugosi. Concentration-of-measure inequalities, 2003. Available at http://www.econ.upf.edu/$\sim$lugosi/anu.ps. * [21] M. Rudelson. Random vectors in the isotropic position. J. of Functional Analysis, 163(1):60–72, 1999. * [22] M. Rudelson and R. Vershynin. Sampling from large matrices: An approach through geometric functional analysis. J. ACM, 54(4):21, 2007. * [23] D. A. Spielman and S.-H. Teng. Nearly-linear time algorithms for graph partitioning, graph sparsification, and solving linear systems. In STOC ’04, pages 81–90, 2004. Full version available at http://arxiv.org/abs/cs.DS/0310051. * [24] D. A. Spielman and S.-H. Teng. Nearly-linear time algorithms for preconditioning and solving symmetric, diagonally dominant linear systems. Available at http://www.arxiv.org/abs/cs.NA/0607105, 2006. * [25] D. A. Spielman and S.-H. Teng. Spectral Sparsification of Graphs. Available at http://arxiv.org/abs/0808.4134, 2008.

arxiv-papers

2024-09-04T02:48:54.189837

0803.0988

# Faster Lossy Generalized Flow via Interior Point Algorithms††thanks: This material is based upon work supported by the National Science Foundation under Grant Nos. CCF-0707522 and CCF-0634957. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. Samuel I. Daitch Department of Computer Science Yale University Daniel A. Spielman Program in Applied Mathematics and Department of Computer Science Yale University ###### Abstract We present asymptotically faster approximation algorithms for the generalized flow problems in which multipliers on edges are at most $1$. For this lossy version of the maximum generalized flow problem, we obtain an additive $\epsilon$ approximation of the maximum flow in time $\widetilde{\mathcal{O}}\left(m^{3/2}\log^{2}(U/\epsilon)\right)$, where $m$ is the number of edges in the graph, all capacities are integers in the range $\left\\{1,\dotsc,U\right\\}$, and all loss multipliers are ratios of integers in this range. For minimum cost lossy generalized flow with costs in the range $\left\\{1,\dotsc,U\right\\}$, we obtain a flow that has value within an additive $\epsilon$ of the maximum value and cost at most the optimal cost. In many parameter ranges, these algorithms improve over the previously fastest algorithms for the generalized maximum flow problem by a factor of $m^{1/2}$ and for the minimum cost generalized flow problem by a factor of approximately $m^{1/2}/\epsilon^{2}$. The algorithms work by accelerating traditional interior point algorithms by quickly solving the linear equations that arise in each step. The contributions of this paper are twofold. First, we analyze the performance of interior point algorithms with approximate linear system solvers. This analysis alone provides an algorithm for the standard minimum cost flow problem that runs in time $\widetilde{\mathcal{O}}\left(m^{3/2}\log^{2}U\right)$—an improvement of approximately $\widetilde{\mathcal{O}}\left(n/m^{1/2}\right)$ over previous algorithms. Second, we examine the linear equations that arise when using an interior point algorithm to solve generalized flow problems. We observe that these belong to the family of symmetric M-matrices, and we then develop $\widetilde{\mathcal{O}}\left(m\right)$-time algorithms for solving linear systems in these matrices. These algorithms reduce the problem of solving a linear system in a symmetric M-matrix to that of solving $\mathcal{O}\left(\log n\right)$ linear systems in symmetric diagonally- dominant matrices, which we can do in time $\widetilde{\mathcal{O}}\left(m\right)$ using the algorithm of Spielman and Teng. All of our algorithms operate on numbers of bit length at most $\mathcal{O}\left(\log nU/\epsilon\right)$. ## 1 Introduction Interior-point algorithms are one of the most popular ways of solving linear programs. These algorithms are iterative, and their complexity is dominated by the cost of solving a system of linear equations at each iteration. Typical complexity analyses of interior point algorithms apply worst-case bounds on the running time of linear equations solvers. However, in most applications the linear equations that arise are quite special and may be solved by faster algorithms. Each family of optimization problem leads to a family of linear equations. For example, the maximum flow and minimum cost flow problems require the solution of linear systems whose matrices are symmetric and diagonally-dominant. The generalized versions of these flow problems result in symmetric M-matrices. The generalized maximum flow problem is specified by a directed graph $(V,E)$, an inward capacity $c(e)>0$ and a multiplier $\gamma(e)>0$ for each edge $e$, and source and sink vertices $s$ and $t$. For every unit flowing into edge $e$, $\gamma(e)$ flows out. In lossy generalized flow problems, each multiplier $\gamma(e)$ is restricted to be at most $1$. In the generalized maximum flow problem, one is asked to find the flow $f:E\rightarrow{\rm I\kern-2.0ptR}^{+}$ that maximizes the flow into $t$ given an unlimited supply at $s$, subject to the capacity constraints on the amount of flow entering each edge. In the generalized minimum cost flow problem, one also has a cost function $q(e)\geq 0$, and is asked to find the maximum flow of minimum cost (see [AMO93]). In the following chart, we compare the complexity of our algorithms with the fastest algorithms of which we are aware. The running times are given for networks in which all capacities and costs are positive integers less than $U$ and every loss factor is a ratio of two integers less than $U$. For the standard flow problems, our algorithms are exact, but for the generalized flow problems our algorithms find additive $\epsilon$ approximations, while the other approximation algorithms have multiplicative error $(1+\epsilon)$. However, we note that our algorithms only require arithmetic with numbers of bit-length $\mathcal{O}\left(\log(nU/\epsilon)\right)$, whereas we suspect that the algorithms obtaining multiplicative approximations might require much longer numbers. In the chart, $C$ refers to the value of the flow. Exact algorithms | Approximation algorithms | Our algorithm ---|---|--- Generalized Maximum Flow | | $\mathcal{O}\left(m^{2}(m+n\log n)\log U\right)$ [GJO97] | $\widetilde{\mathcal{O}}\left(m^{2}/\epsilon^{2}\right)$ [FW02] | $\widetilde{\mathcal{O}}\left(m^{1.5}\log^{2}(U/\epsilon)\right)$ $\mathcal{O}\left(m^{1.5}n^{2}\log(nU)\right)$ [Vai89] | $\widetilde{\mathcal{O}}\left(m(m+n\log\log B)\log\epsilon^{-1}\right)$ | | [GFNR98][TW98][FW02] | Generalized Minimum Cost Flow | | $\mathcal{O}\left(m^{1.5}n^{2}\log(nU)\right)$ [Vai89] | $\widetilde{\mathcal{O}}\left(m^{2}\log\log B/\epsilon^{2}\right)$ [FW02] | $\widetilde{\mathcal{O}}\left(m^{1.5}\log^{2}(U/\epsilon)\right)$ Maximum Flow | | $\mathcal{O}\left(\min(n^{3/2},m^{1/2})m\log(n^{2}/m)\log U\right)$ | | $\widetilde{\mathcal{O}}\left(m^{1.5}\log^{2}U\right)$ [GR98] | | Minimum Cost Flow | | $\mathcal{O}\left(nm\log(n^{2}/m)\log(nC)\right)$ [GT87] | | $\widetilde{\mathcal{O}}\left(m^{1.5}\log^{2}U\right)$ $\mathcal{O}\left(nm(\log\log U)\log(nC)\right)$ [AGOT92] | | $\mathcal{O}\left((m\log n)(m+n\log n)\right)$ [Orl88] | | ### 1.1 The solution of systems in M-matrices A symmetric matrix $M$ is diagonally dominant if each diagonal is at least the sum of the absolute values of the other entries in its row. A symmetric matrix $M$ is an $M$-matrix if there is a positive diagonal matrix $D$ for which $DMD$ is diagonally dominant. Spielman and Teng [ST04, ST06] showed how to solve linear systems in diagonally dominant matrices to $\epsilon$ accuracy in time $\widetilde{\mathcal{O}}\left(m\log\epsilon^{-1}\right)$. We show how to solve linear systems in $M$-matrices by first computing a diagonal matrix $D$ for which $DMD$ is diagonally dominant, and then applying the solver of Spielman and Teng. Our algorithm for finding the matrix $D$ applies the solver of Spielman and Teng an expected $\mathcal{O}\left(\log n\right)$ times. While iterative algorithms are known that eventually produce such a diagonal matrix $D$, they have no satisfactory complexity analysis [Li02, LLH+98, BCPT05]. ### 1.2 Analysis of interior point methods In our analysis of interior-point methods, we examine the complexity of the short-step dual path following algorithm of Renegar [Ren88] as analyzed by Ye [Ye97]. The key observations required by our complexity analysis are that none of the slack variables become too small during the course of the algorithm and that the algorithm still works if one $\mathcal{O}\left(1/\sqrt{m}\right)$-approximately solves each linear system in the matrix norm (defined below). Conveniently, this is the same type of approximation produced by our algorithm and that of Spielman and Teng. This is a very crude level of approximation, and it means that these algorithms can be applied very quickly. While other analyses of the behavior of interior point methods with inexact solvers have appeared [Ren96], we are unaware of any analyses that are sufficiently fine for our purposes. This analysis is given in detail in Appendix C. ### 1.3 Outline of the paper In Section 2, we describe the results of our analysis of interior point methods using apprximate solvers. In Section 3, we describe the formulation of the generalized flow problems as linear programs, and discuss how to obtain the solutions from the output of an interior-point algorithm. In Section 4, we give our algorithm for solving linear systems in M-matrices. ## 2 Interior-Point Algorithm using an Approximate Solver Our algorithm uses numerical methods to solve a linear program formulation of the generalized flow problems. The fastest interior-point methods for linear programs, such as that of Renegar [Ren88] require only $\mathcal{O}\left(\sqrt{n}\right)$ iterations to approach the solution, where each iteration takes a step through the convex polytope by solving a system of linear equations. In this paper, we consider stepping through the linear program using an only an approximate solver, i.e. an algorithm $\boldsymbol{\mathit{x}}=\mathtt{Solve}(M,\boldsymbol{\mathit{b}},\epsilon)$ that returns a solution satisfying $\left\|\boldsymbol{\mathit{x}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\leq\epsilon\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ where the matrix norm $\left\|\cdot\right\|_{M}$ is given by $\left\|\boldsymbol{\mathit{v}}\right\|_{M}=\sqrt{\boldsymbol{\mathit{v}}^{T}M\boldsymbol{\mathit{v}}}$. As mentioned above, we have analyzed the Renegar [Ren88] version of the dual path-folllowing algorithm, along the lines of the analysis that found in [Ye97], but modified to account for the use of an approximate solver. In particular, using the approximate solver we implement an interior-point algorithm with the following properties: ###### Theorem 2.1. $\boldsymbol{\mathit{x}}=\mathtt{InteriorPoint}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\lambda_{min},T,\boldsymbol{\mathit{y}}^{0},\epsilon)$ takes input that satisfies * • $A$ is an $n\times m$ matrix; $\boldsymbol{\mathit{b}}$ is a length $n$ vector; $\boldsymbol{\mathit{c}}$ is a length $m$ vector * • $AA^{T}$ is positive definite, and $\lambda_{min}>0$ is a lower bound on the eigenvalues of $AA^{T}$ * • $T>0$ is an upper bound on the absolute values of the coordinates in the dual linear program, i.e. $\left\|\boldsymbol{\mathit{y}}\right\|_{\infty}\leq T$ and $\left\|\boldsymbol{\mathit{s}}\right\|_{\infty}\leq T$ for all $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})$ that satisfy $\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}\geq 0$ * • initial point $\boldsymbol{\mathit{y}}^{0}$ is a length $n$ vector where $A^{T}\boldsymbol{\mathit{y}}^{0}<\boldsymbol{\mathit{c}}$ * • error parameter $\epsilon$ satisfies $0<\epsilon<1$ and returns $\boldsymbol{\mathit{x}}>0$ such that $\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|\leq\epsilon$ and $\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}<z^{*}+\epsilon$. Let us define * • $U$ is the largest absolute value of any entry in $A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}}$ * • $s^{0}_{min}$ is the smallest entry of $\boldsymbol{\mathit{s}}^{0}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{0}$ Then the algorithm makes $\mathcal{O}\left(\sqrt{m}\log\frac{TUm}{\lambda_{min}s^{0}_{min}\epsilon}\right)$ calls to the approximate solver, of the form $\mathtt{Solve}\left(AS^{-2}A^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\cdot,\epsilon^{\prime}\right)$ where $S$ is a positive diagonal matrix with condition number $\mathcal{O}\left(\frac{T^{2}Um^{2}}{\epsilon}\right)$, and $\boldsymbol{\mathit{v}},\epsilon^{\prime}$ satisfy $\log\frac{\left\|\boldsymbol{\mathit{v}}\right\|}{\epsilon^{\prime}}=\mathcal{O}\left(\log\frac{TUm}{s^{0}_{min}\epsilon}\right)$ In Appendix C, we present a complete description of this algorithm, with analysis and proof of correctness. ## 3 Solving Generalized Flow We consider network flows on a directed graph $(V,E)$ with $V=[n]$, $E=\\{e_{1},\dotsb,e_{m}\\}$, source $s\in V$ and sink $t\in V$. Edge $e_{j}$ goes from vertex $v_{j}$ to vertex $w_{j}$. and has inward capacity $c(e_{j})$, flow multiplier $\gamma(e_{j})<1$, and cost $q(e_{j})$. We assume without loss of generality that $t$ has a single in-edge, which we denote as $e_{t}$, and no out-edges. The generalized max-flow approximation algorithm will produce a flow that sends no worse than $\epsilon$ less than the maximum possible flow to the sink. The generalized min-cost approximation algorithm will produce a flow that, in addition to being within $\epsilon$ of a maximum flow, also has cost no greater than the minimum cost of a maximum flow (see [FW02]). ### 3.1 Fixing Approximate Flows The interior-point algorithm described in the previous section produces an output that may not exactly satisfy the linear constraints $A\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}$. In particular, when we apply the algorithm to a network flow linear program, the output may only be an approximate flow: ###### Definition 3.1. An $\epsilon$-approximate flow approximately satisfies all capacity constraints and flow conservation constraints. In particular, every edge may have flow up to $\epsilon$ over capacity, and every vertex besides $s$ and $t$ may have up to $\epsilon$ excess or deficit flow. An exact flow satisfies all capacity constraints and has exact flow conservation at all vertices except $s$ and $t$. We are going to modify the graph slightly before running the interior-point algorithm, so that it will be easier to obtain an exact flow from the approximate flow given by the interior-point algorithm. Let us compute the least-lossy-paths tree $T$ rooted at $s$. This is the tree that contains, for each $v\in V-\\{s,t\\}$, the path $\pi_{s,v}$ from $s$ to $v$ that minimizes $L(v)=\nolinebreak\prod_{e\in\pi_{s,v}}\gamma(e)^{-1}$, the factor by which the flow along the path is diminished. We can find this tree in time $\widetilde{\mathcal{O}}\left(m\right)$, using Dijkstra’s algorithm to solve the single-source shortest-paths problem with edge weights $-\log\gamma(e)$. Next, we delete from the graph all vertices $v$ such that $L(v)>\frac{\epsilon}{2mnU}$. Note that in a maximum-flow, it is not possible to have more than $\frac{\epsilon}{2n}$ flowing into such a $v$, since at most $mU$ can flow out of $s$. Thus, deleting each such $v$ cannot decrease the value of the maximum flow by more than $\frac{\epsilon}{2n}$. In total, we may decrease the value of the maximum flow by at most $\frac{\epsilon}{2}$. We define $\epsilon_{FLOW}=\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}$. In the subsequent sections, we show how to use the interior-point method to obtain an $\epsilon_{FLOW}$-approximate flow that has a value within $\epsilon_{4}$ of the maximum flow. Assuming that the graph had been preprocessed as above, we may convert the approximate flow into an exact flow: ###### Lemma 3.2. Suppose all vertices $v\in V-\left\\{s,t\right\\}$ satisfy $L(v)\leq\frac{\epsilon}{2mnU}$. In $\widetilde{\mathcal{O}}\left(m\right)$ time, we are able to convert an $\epsilon_{FLOW}$-approximate flow that has a value within $\frac{\epsilon}{4}$ of the maximum flow into an exact flow that has a value within $\frac{\epsilon}{2}$ of the maximum flow. The cost of this exact flow is no greater than the cost of the approximate flow. ###### Proof. Let us first fix the flows so that no vertex has more flow out than in. We use the least-lossy-paths tree $T$, starting at the leaves of the tree and working towards $s$. To balance the flow at a vertex $v$ we increase the flow on the tree edge into $v$. After completing this process, for each $v$ we will have added a path of flow that delivers at most $\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}$ additional units of flow to $v$. Since $L(v)\leq\frac{\epsilon}{2mnU}$, no such path requires more than $\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}\cdot\frac{2mnU}{\epsilon}=\frac{\epsilon}{32mnU^{2}}$ flow on an edge, and so in total we have added no more than $\frac{\epsilon}{32mU^{2}}$ to each edge. Next, let us fix the flows so that no vertex has more flow in than out. We follow a similar procedure as above, except now we may use any spanning tree rooted at and directed towards $t$. Starting from the leaves, we balance the vertices by increasing flow out the tree edge. Since the network is lossy, the total amount added to each edge is at most $\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}\cdot n\leq\frac{\epsilon^{2}}{64m^{2}nU^{3}}$. Recall that we started with each edge having flow up to $\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}$ over capacity. After balancing the flows at the vertices, each edge may now be over capacity by as much as $\frac{\epsilon}{32mU^{2}}+\frac{\epsilon^{2}}{64m^{2}nU^{3}}+\frac{\epsilon^{2}}{64m^{2}n^{2}U^{3}}\leq\frac{\epsilon}{16mU^{2}}$ Since the edge capacities are at least 1, the flow on an edge may be as much as $(1+\frac{\epsilon}{16mU^{2}})$ times the capacity. Furthermore, while balancing the flows we may have added as much as $\frac{\epsilon}{16mU^{2}}\cdot mU=\frac{\epsilon^{2}}{16U}$ to the total cost of the flow. Assuming that the value of approximate flow was at least $\frac{\epsilon}{4}$, its cost must also have been at least $\frac{\epsilon}{4}$, and so we have increased the cost by a multiplicative factor of at most $(1+\frac{\epsilon}{4U})$. (If the approximate flow had value less than $\frac{\epsilon}{4}$, then the empty flow trivially solves this flow rounding problem.) By scaling the entire flow down by a multiplicative factor of $(1+\frac{\epsilon}{4U})^{-1}$, we solve the capacity violations, and also reduce the cost of the exact flow to be no greater than that of the approximate flow. Since the value of a flow can be at most $U$, the flow scaling decreases the value of the flow by no more than $\epsilon/4$, as required. ∎ The above procedure produces an exact flow that is within $\epsilon/2$ of the maximum flow in the preprocessed graph, and therefore is within $\epsilon$ of the maximum flow in the original graph. Furthermore, the cost of the flow is no greater than the minimum cost of a maximum flow in the original graph. Thus to solve a generalized flow problem, it remains for us to describe how to use the interior-point algorithm to generate a $\epsilon_{FLOW}$-approximate flow that has a value within $\epsilon/4$ of the maximum flow, and, for the min-cost problem, also has cost no greater than the the minimum cost of a maximum flow. ### 3.2 Generalized Max-Flow We formulate the maximum flow problem as a linear program as follows: Let $A$ be the $(n-2)\times m$ matrix whose nonzero entries are $A_{v_{j},j}=-1$ and $A_{w_{j},j}=\gamma(e_{j})$, but without rows corrsponding to $s$ and $t$. Let $\boldsymbol{\mathit{c}}$ be the length $m$ vector containing the edge capcities. Let $\boldsymbol{\mathit{u}}_{t}$ be the length $m$ unit vector with a 1 entry for edge $e_{t}$. Let the vectors $\boldsymbol{\mathit{x}}_{1}$ and $\boldsymbol{\mathit{x}}_{2}$ respectively denote the flow into each edge and the unused inward capacity of each edge. The max-flow linear program, in canonical form, is: $\displaystyle\min_{\boldsymbol{\mathit{x}}_{i}}-\boldsymbol{\mathit{u}}_{t}^{T}\boldsymbol{\mathit{x}}_{1}$ $\displaystyle\text{s.t.}\quad\left[\begin{matrix}A&\\\ I&I\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{x}}_{1}\\\ \boldsymbol{\mathit{x}}_{2}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}0\\\ \boldsymbol{\mathit{c}}\end{matrix}\right]$ $\displaystyle\text{and }\quad\boldsymbol{\mathit{x}}_{i}$ $\displaystyle\geq 0$ The constraint $A\boldsymbol{\mathit{x}}_{1}=0$ ensures that flow is conserved at every vertex except $s$ and $t$, while the constraint $\boldsymbol{\mathit{x}}_{1}+\boldsymbol{\mathit{x}}_{2}=\boldsymbol{\mathit{c}}$ ensures that the capacities are obeyed. Now, the dual of the above linear program is not bounded, which is a problem for our interior-point algorithm. To fix this, we modify the linear program slightly: $\displaystyle\min_{\boldsymbol{\mathit{x}}_{i}}\left(-\boldsymbol{\mathit{u}}_{t}^{T}\boldsymbol{\mathit{x}}_{1}+\frac{4U}{\epsilon_{FLOW}}({\mbox{\boldmath$1$}}_{m}^{T}\boldsymbol{\mathit{x}}_{3}+{\mbox{\boldmath$1$}}_{n-2}^{T}\boldsymbol{\mathit{x}}_{4}+{\mbox{\boldmath$1$}}_{n-2}^{T}\boldsymbol{\mathit{x}}_{5})\right)\qquad\text{s.t.}\quad\boldsymbol{\mathit{x}}_{i}$ $\displaystyle\geq 0$ $\displaystyle\text{and}\quad\left[\begin{matrix}A&&&I&-I\\\ I&I&-I&&\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{x}}_{1}\\\ \boldsymbol{\mathit{x}}_{2}\\\ \boldsymbol{\mathit{x}}_{3}\\\ \boldsymbol{\mathit{x}}_{4}\\\ \boldsymbol{\mathit{x}}_{5}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}0\\\ \boldsymbol{\mathit{c}}\end{matrix}\right]$ (We use ${\mbox{\boldmath$1$}}_{k}$ to denote the all-ones vector of length $k$.) ###### Lemma 3.3. This modified linear program has the same optimum value as the original linear program. ###### Proof. Let us examine the new variables in the modified program and note that $\boldsymbol{\mathit{x}}_{3}$ has the effect of modifying the capacities, while $\boldsymbol{\mathit{x}}_{4}$ and $\boldsymbol{\mathit{x}}_{5}$ create excess or deficit of flow at the vertices. Since we have a lossy network, a unit modification of any of these values cannot change the value of the flow by more than 1, and therefore must increase the value of the modified linear program. Thus, at the optimum we have $\boldsymbol{\mathit{x}}_{3}=\boldsymbol{\mathit{x}}_{4}=\boldsymbol{\mathit{x}}_{5}=0$ and so the solution is the same as that of the original linear program. ∎ The modified linear program has the following equivalent dual linear program: $\displaystyle\max_{\boldsymbol{\mathit{y}}_{i}}\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{y}}_{2}\qquad\text{s.t.}\quad\boldsymbol{\mathit{s}}_{i}\geq 0$ $\displaystyle\text{and}\quad\left[\begin{matrix}A^{T}&I\\\ &I\\\ &-I\\\ I&\\\ -I&\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{y}}_{1}\\\ \boldsymbol{\mathit{y}}_{2}\end{matrix}\right]+\left[\begin{matrix}\boldsymbol{\mathit{s}}_{1}\\\ \boldsymbol{\mathit{s}}_{2}\\\ \boldsymbol{\mathit{s}}_{3}\\\ \boldsymbol{\mathit{s}}_{4}\\\ \boldsymbol{\mathit{s}}_{5}\end{matrix}\right]=\left[\begin{matrix}-\boldsymbol{\mathit{u}}_{t}\\\ {\mbox{\boldmath$0$}}\\\ (4U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}\\\ (4U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-2}\\\ (4U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-2}\end{matrix}\right]$ ###### Lemma 3.4. The above dual linear program is bounded. In particular, the coordinates of all feasible dual points have absolute value at most $(nU+1)\cdot\frac{4U}{\epsilon_{FLOW}}+1$. ###### Proof. Of the five constraints in the dual linear program, the last four give $\frac{4U}{\epsilon_{FLOW}}$ as an explicit bound on the absolute value of $\boldsymbol{\mathit{y}}$ coordinates. It then follows that $\frac{8U}{\epsilon_{FLOW}}$ is a upper bound on the coordinates of $\boldsymbol{\mathit{s}}_{2},\boldsymbol{\mathit{s}}_{3},\boldsymbol{\mathit{s}}_{4},\boldsymbol{\mathit{s}}_{5}$, and the coordinates of $\boldsymbol{\mathit{s}}_{1}=-\boldsymbol{\mathit{u}}_{t}-A^{T}\boldsymbol{\mathit{y}}_{1}-\boldsymbol{\mathit{y}}_{2}$ can be at most $(nU+1)\cdot\frac{4U}{\epsilon_{FLOW}}+1$. ∎ We refer to the $\boldsymbol{\mathit{s}}_{i}$ variables as the slacks. Recall that we must provide the interior-point algorithm with an initial dual feasible point $\boldsymbol{\mathit{y}}^{0}$ such that the corresponding slacks $\boldsymbol{\mathit{s}}^{0}$ are bounded away from zero. We choose the following initial point, and note that the slacks are bounded from below by $\frac{U}{\epsilon_{FLOW}}$: $\displaystyle\left[\begin{matrix}\boldsymbol{\mathit{y}}^{0}_{1}\\\ \boldsymbol{\mathit{y}}^{0}_{2}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}{\mbox{\boldmath$0$}}\\\ -(2U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ $\displaystyle\left[\begin{matrix}\boldsymbol{\mathit{s}}^{0}_{1}\\\ \boldsymbol{\mathit{s}}^{0}_{2}\\\ \boldsymbol{\mathit{s}}^{0}_{3}\\\ \boldsymbol{\mathit{s}}^{0}_{4}\\\ \boldsymbol{\mathit{s}}^{0}_{5}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}(2U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}-\boldsymbol{\mathit{u}}_{t}\\\ (2U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}\\\ (2U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}\\\ (4U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-2}\\\ (4U/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-2}\end{matrix}\right]$ We must also provide the interior-point algorithm with a lower bound on the eigenvalues of the matrix $\left[\begin{matrix}A&&&I&-I\\\ I&I&-I&&\end{matrix}\right]\left[\begin{matrix}A^{T}&I\\\ &I\\\ &-I\\\ I&\\\ -I&\end{matrix}\right]=\left[\begin{matrix}AA^{T}+2I&A\\\ A^{T}&3I\end{matrix}\right]$ Note that we may subtract $2I$ from the above matrix and still have a positive definite matrix, so $\lambda_{min}=2$ is certainly a lower bound on the eigenvalues. Using the above values for $\boldsymbol{\mathit{y}}^{0}$ and $\lambda_{min}$, and the bound on the dual coordinates given in Lemma 3.4, we now call $\mathtt{InteriorPoint}$ on the modified max-flow linear program, using error parameter $\frac{\epsilon_{FLOW}}{2}$. In the solution returned by the interior-point algorithm, the vector $\boldsymbol{\mathit{x}}_{1}$ assigns a flow value to each edge such that the flow constraints are nearly satisfied: ###### Lemma 3.5. $\boldsymbol{\mathit{x}}_{1}$ is an $\epsilon_{FLOW}$-approximate flow with value within $\epsilon_{FLOW}/2$ of the maximum flow. ###### Proof. Observe that the amount flowing into $t$ is at least $-1$ times the value of the modified linear program. Since the interior-point algorithm generates a solution to the modified linear program within $\epsilon_{FLOW}/2$ of the optimum value, which is $-1$ times the maximum flow, the amount flowing into $t$ surely must be within $\epsilon_{FLOW}/2$ of the maximum flow. Now, let us note more precisely that the modified linear program aims to minimize the objective function computed by subtracting the amount flowing into $t$ from $4U/\epsilon_{FLOW}$ times the sum of the entries of $\boldsymbol{\mathit{x}}_{3}$, $\boldsymbol{\mathit{x}}_{4}$, and $\boldsymbol{\mathit{x}}_{5}$. Since the minimum value of this objective function must be negative, and the solution returned by the interior-point algorithm has a value within $\epsilon_{FLOW}/2$ of the minimum, the value of this solution must be less than $\epsilon_{FLOW}/2<U$. The amount flowing into $t$ is also at most $U$, so no entry of $\boldsymbol{\mathit{x}}_{3},\boldsymbol{\mathit{x}}_{4},\boldsymbol{\mathit{x}}_{5}$ can be greater than $2U/(4U/\epsilon_{FLOW})=\epsilon_{FLOW}/2$. The interior-point algorithm guarantees that $\displaystyle\left\|A\boldsymbol{\mathit{x}}_{1}+\boldsymbol{\mathit{x}}_{4}-\boldsymbol{\mathit{x}}_{5}\right\|$ $\displaystyle<\frac{\epsilon_{FLOW}}{2}$ and $\displaystyle\left\|\boldsymbol{\mathit{x}}_{1}+\boldsymbol{\mathit{x}}_{2}-\boldsymbol{\mathit{x}}_{3}-\boldsymbol{\mathit{c}}\right\|$ $\displaystyle<\frac{\epsilon_{FLOW}}{2}$ and so we may conclude that $\displaystyle\left\|A\boldsymbol{\mathit{x}}_{1}\right\|$ $\displaystyle<\epsilon_{FLOW}$ and $\displaystyle\boldsymbol{\mathit{x}}_{1}\leq\boldsymbol{\mathit{c}}+\epsilon_{FLOW}$ Indeed, this is precisely what is means for $\boldsymbol{\mathit{x}}_{1}$ to describe an $\epsilon_{FLOW}$-approximate flow. ∎ ### 3.3 Generalized Min-Cost Flow As a first step in solving the generlized min-cost flow problem, we solve the generalized max-flow linear program as described above, to find a value $F$ that is within $\frac{\epsilon}{8}$ of the maximum flow. We now formulate a linear program for finding the minimum cost flow that delivers $F$ units of flow to $t$: $\displaystyle\min_{\boldsymbol{\mathit{x}}_{i}}\boldsymbol{\mathit{q}}^{T}\boldsymbol{\mathit{x}}_{1}\qquad\qquad\text{s.t.}\quad\boldsymbol{\mathit{x}}_{i}$ $\displaystyle\geq 0$ $\displaystyle\text{and}\quad\left[\begin{matrix}A&\\\ I&I\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{x}}_{1}\\\ \boldsymbol{\mathit{x}}_{2}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}F\cdot\boldsymbol{\mathit{e}}_{t}\\\ \boldsymbol{\mathit{c}}\end{matrix}\right]$ where $\boldsymbol{\mathit{q}}$ is the length $n$ vector containing the edge costs, and $\boldsymbol{\mathit{e}}_{t}$ is the length $n-1$ vector that assigns 1 to vertex $t$ and 0 to all the other vertices except $s$. $A$ is the same matrix as in the max-flow linear program, except that we include the row corresponding to $t$, which translates to a new constraint that $F$ units must flow into $t$. We must again modify the linear program so that the dual will be bounded: $\displaystyle\min_{\boldsymbol{\mathit{x}}_{i}}\left(\boldsymbol{\mathit{q}}^{T}\boldsymbol{\mathit{x}}_{1}+\left(\frac{4mU^{2}}{\epsilon_{FLOW}}\right)\left({\mbox{\boldmath$1$}}_{m}^{T}\boldsymbol{\mathit{x}}_{3}+{\mbox{\boldmath$1$}}_{n-1}^{T}\boldsymbol{\mathit{x}}_{4}+{\mbox{\boldmath$1$}}_{n-1}^{T}\boldsymbol{\mathit{x}}_{5}\right)\right)\qquad\text{s.t.}\quad\boldsymbol{\mathit{x}}_{i}$ $\displaystyle\geq 0$ $\displaystyle\text{and}\quad\left[\begin{matrix}A&&&I&-I\\\ I&I&-I&&\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{x}}_{1}\\\ \boldsymbol{\mathit{x}}_{2}\\\ \boldsymbol{\mathit{x}}_{3}\\\ \boldsymbol{\mathit{x}}_{4}\\\ \boldsymbol{\mathit{x}}_{5}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}F\cdot\boldsymbol{\mathit{e}}_{t}\\\ \boldsymbol{\mathit{c}}\end{matrix}\right]$ ###### Lemma 3.6. This modified linear program has the same optimum value as the original linear program. ###### Proof. We examine the new variables and note that $\boldsymbol{\mathit{x}}_{3}$ modifies the capacities, while $\boldsymbol{\mathit{x}}_{4}$ and $\boldsymbol{\mathit{x}}_{5}$ create excess supply (or demand) at the vertices. A unit modification to any of these values can at best create a new path for one unit of flow to arrive at the sink. This new path has cost at least 1, and it can replace an path in the optimum flow of cost at most $nU$, for a net improvement in the cost of the flow of at most $nU-1$, which is less than $\frac{4mU^{2}}{\epsilon_{FLOW}}$. Thus the value of the modified linear program can only increase when these new variables are set to non-zero values. ∎ Now, the dual linear program is: $\displaystyle\max_{\boldsymbol{\mathit{y}}_{i}}\left(F\cdot\boldsymbol{\mathit{e}}_{t}^{T}\boldsymbol{\mathit{y}}_{1}+\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{y}}_{2}\right)\qquad\text{s.t.}\quad\boldsymbol{\mathit{s}}_{i}\geq 0$ $\displaystyle\text{and}\quad\left[\begin{matrix}A^{T}&I\\\ &I\\\ &-I\\\ I&\\\ -I&\end{matrix}\right]\left[\begin{matrix}\boldsymbol{\mathit{y}}_{1}\\\ \boldsymbol{\mathit{y}}_{2}\end{matrix}\right]+\left[\begin{matrix}\boldsymbol{\mathit{s}}_{1}\\\ \boldsymbol{\mathit{s}}_{2}\\\ \boldsymbol{\mathit{s}}_{3}\\\ \boldsymbol{\mathit{s}}_{4}\\\ \boldsymbol{\mathit{s}}_{5}\end{matrix}\right]=\left[\begin{matrix}\boldsymbol{\mathit{q}}\\\ {\mbox{\boldmath$0$}}\\\ (4mU^{2}/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{m}\\\ (4mU^{2}/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-1}\\\ (4mU^{2}/\epsilon_{FLOW})\cdot{\mbox{\boldmath$1$}}_{n-1}\end{matrix}\right]$ ###### Lemma 3.7. The above dual linear program is bounded. In particular, the coordinates of all feasible dual points have absolute value at most $(nU+1)\cdot\frac{4mU^{2}}{\epsilon_{FLOW}}$. ###### Proof. Of the five constraints in the dual linear program, the last four give $\frac{4mU^{2}}{\epsilon_{FLOW}}$ as an explicit bound on the absolute value of $\boldsymbol{\mathit{y}}$ coordinates. It then follows that $\frac{8mU^{2}}{\epsilon_{FLOW}}$ is a upper bound on the coordinates of $\boldsymbol{\mathit{s}}_{2},\boldsymbol{\mathit{s}}_{3},\boldsymbol{\mathit{s}}_{4},\boldsymbol{\mathit{s}}_{5}$, and the coordinates of $\boldsymbol{\mathit{s}}_{1}=\boldsymbol{\mathit{q}}-A^{T}\boldsymbol{\mathit{y}}_{1}-\boldsymbol{\mathit{y}}_{2}$ can be at most $(nU+1)\cdot\frac{4mU^{2}}{\epsilon_{FLOW}}$. ∎ Let us also note that $\boldsymbol{\mathit{y}}^{0}=\left[\begin{matrix}{\mbox{\boldmath$0$}}\\\ -(mU^{2}/\epsilon_{FLOW}){\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ is an initial interior dual point with all slacks at least $\frac{mU^{2}}{\epsilon_{FLOW}}$. Using the above initial point, the bound on the dual coordinates from Lemma 3.7, and $\lambda_{min}=2$ as in the previous section, we run $\mathtt{InteriorPoint}$ on the modified min-cost linear program, with error parameter $\frac{\epsilon_{FLOW}}{2}$. In the solution returned by the interior-point algorithm, the vector $\boldsymbol{\mathit{x}}_{1}$ assigns a flow value to each edge such that the flow constraints are nearly satisfied: ###### Lemma 3.8. $\boldsymbol{\mathit{x}}_{1}$ is an $\epsilon_{FLOW}$-approximate flow with value within $\frac{5\epsilon}{32}$ of the maximum flow. ###### Proof. Note that any flow in total cannot cost more that $mU^{2}$, even if all edges are filled to maximum capacity. Therefore the value of the solution output by the interior-point algorithm can be at most $mU^{2}+\frac{\epsilon_{FLOW}}{2}<2mU^{2}$, and so in particular no entry of $\boldsymbol{\mathit{x}}_{3},\boldsymbol{\mathit{x}}_{4},\boldsymbol{\mathit{x}}_{5}$ can be greater than $\frac{\epsilon_{FLOW}}{2}$. Now, the interior-point algorithm guarantees that $\displaystyle\left\|A\boldsymbol{\mathit{x}}_{1}+\boldsymbol{\mathit{x}}_{4}-\boldsymbol{\mathit{x}}_{5}-F\cdot\boldsymbol{\mathit{e}}_{t}\right\|$ $\displaystyle<\frac{\epsilon_{FLOW}}{2}$ and $\displaystyle\left\|\boldsymbol{\mathit{x}}_{1}+\boldsymbol{\mathit{x}}_{2}-\boldsymbol{\mathit{x}}_{3}-\boldsymbol{\mathit{c}}\right\|$ $\displaystyle<\frac{\epsilon_{FLOW}}{2}$ and so we may conclude that $\displaystyle\left\|A\boldsymbol{\mathit{x}}_{1}-F\cdot\boldsymbol{\mathit{e}}_{t}\right\|$ $\displaystyle<\epsilon_{FLOW}$ and $\displaystyle\boldsymbol{\mathit{x}}_{1}$ $\displaystyle\leq\boldsymbol{\mathit{c}}+\epsilon_{FLOW}$ These inequalities imply that this is a $\epsilon_{FLOW}$-approximate flow, and additionally that at least $F-\epsilon_{FLOW}$ is flowing into $t$. Since $F$ is within $\frac{\epsilon}{8}$ of the maximum flow, the amount flowing into $t$ must be within $\frac{\epsilon}{8}+\epsilon_{FLOW}<\frac{5\epsilon}{32}$ of the maximum flow. ∎ By scaling down the $\boldsymbol{\mathit{x}}_{1}$ flow slightly, we obtain a flow that does not exceed the minimum cost of a maximum flow: ###### Lemma 3.9. $\boldsymbol{\mathit{x}}_{1}^{\prime}=(1-\frac{\epsilon}{12U})\boldsymbol{\mathit{x}}_{1}$ is an $\epsilon_{FLOW}$-approximate flow with value within $\frac{\epsilon}{4}$ of the maximum flow, and with cost at most the minimum cost of a maximum flow. ###### Proof. We may assume that the value of flow $\boldsymbol{\mathit{x}}_{1}$ is at least $\frac{3\epsilon}{32}$, because otherwise the maximum flow would have to be at most $\frac{3\epsilon}{32}+\frac{5\epsilon}{32}=\frac{\epsilon}{4}$, and so the empty flow would trivially be within $\frac{\epsilon}{4}$ of the maximum. Therefore, the minimum cost of a maximum flow must also at be least $\frac{3\epsilon}{32}$. The interior-point algorithm guarantees that the cost of $\boldsymbol{\mathit{x}}_{1}$ does not exceed this optimum cost by more than $\frac{\epsilon_{FLOW}}{2}$, and so must also not exceed the optimum cost by a multiplicative factor of more than $(1+\frac{16\epsilon_{FLOW}}{3\epsilon})<(1+\frac{\epsilon}{12U})$. Thus. $\boldsymbol{\mathit{x}}_{1}^{\prime}=(1-\frac{\epsilon}{12U})\boldsymbol{\mathit{x}}_{1}$ must have cost below the optimum. Furthermore, since the value of the flow $\boldsymbol{\mathit{x}}_{1}$ can be at most $U$, scaling down by $(1-\frac{\epsilon}{12U})$ cannot decrease the value of the flow by more than $\frac{\epsilon}{12}$. Therefore, the value of the value $\boldsymbol{\mathit{x}}_{1}^{\prime}$ is within $\frac{\epsilon}{12}+\frac{5\epsilon}{32}<\frac{\epsilon}{4}$ of the maximum. ∎ ### 3.4 Running Time The linear systems in the above linear programs take the form $\bar{A}=\left[\begin{matrix}A&&&I&-I\\\ I&I&-I&&\end{matrix}\right]$ so the running time of the interior-point method depends on our ability to approximately solve systems of the form $\bar{A}S^{-2}\bar{A}^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}$, where diagonal matrix $S$ and vector $\boldsymbol{\mathit{v}}$ are as described in Theorem 2.1. As it turns out, this is not much more difficult than solving a linear system in $AS_{1}^{-2}A^{T}$, where $S_{1}$ is the upper left submatrix of $S$. The matrix $AS_{1}^{-2}A^{T}$ is a symmetric $M$-matrix. In the next section, we describe how to approximately solve systems in such matrices in expected time $\widetilde{\mathcal{O}}\left(m\log\frac{\kappa}{\epsilon}\right)$, where $\kappa$ is the condition number of the matrix. We then extend this result to solve the systems $\bar{A}S^{-2}\bar{A}^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}$ in time $\widetilde{\mathcal{O}}\left(m\log\frac{\kappa\left\|\boldsymbol{\mathit{v}}\right\|}{\epsilon}\right)$, where $\kappa$ is the condition number of $\bar{A}S^{-2}\bar{A}^{T}$. ###### Theorem 3.10. Using out interior-point algorithm, we can solve the generalized max flow and generalized min-cost flow problems in time $\widetilde{\mathcal{O}}\left(m^{3/2}\log^{2}(U/\epsilon)\right)$ ###### Proof. According to Theorem 2.1, the interior-point algorithm requires $\mathcal{O}\left(\sqrt{m}\log\frac{TUm}{\lambda_{min}s^{0}_{min}\epsilon}\right)$ calls to the solver. Recall that $T$ is an bound on the coordinates of the dual linear program, and $s^{0}_{min}$ is the smallest slack at the initial point. Above, we gave both of these values to be polynomial in $\frac{mU}{\epsilon}$, for both the max- flow and min-cost linear programs. We also gave $\lambda_{min}=2$ as a lower bound on the eigenvalues of $\bar{A}\bar{A}^{T}$. Thus, the total number of solves is $\widetilde{\mathcal{O}}\left(\sqrt{m}\log\frac{U}{\epsilon}\right)$. Again referring to Theorem 2.1, we find that the condition number of $\bar{A}S^{-2}\bar{A}^{T}$ is be polynomial in $\frac{mU}{\epsilon}$, as is the expression $\frac{\left\|\boldsymbol{\mathit{v}}\right\|}{\epsilon}$. We conclude that each solve takes time $\widetilde{\mathcal{O}}\left(m\log\frac{U}{\epsilon}\right)$. The preprocessing only took time $\widetilde{\mathcal{O}}\left(m\right)$ so we obtain a total running time of $\widetilde{\mathcal{O}}\left(m^{3/2}\log^{2}(U/\epsilon)\right)$. ∎ ### 3.5 Standard Min-Cost Flow In this section we describe how to use interior-point algorithms to give an exact solution to the standard (i.e. no multipliers on edges) min-cost flow problem. We use the following property of the standard flow problem: ###### Theorem 3.11 (see [Sch03, Theorem 13.20]). Given a flow network with integer capacities, and a positive integer $F$, let $\Omega_{FLOW}$ be the set of flow vectors $\boldsymbol{\mathit{x}}$ that flow $F$ units into $t$ and satisfy all capacity and flow conservation constraints. Then $\Omega_{FLOW}$ is a convex polytope in which all vertices have integer coordinates. Our goal is to find the flow in $\Omega_{FLOW}$ of minimum cost. Since the cost function is linear, if there is a unique minimum-cost flow of value $F$, it must occur at a vertex of $\Omega_{FLOW}$. By Theorem 3.11 this must be an integer flow, and we could find this flow exactly by running the interior- point algorithm until it is clear to which integer flow we are converging. Unfortunately, the minimum-cost flow may not be unique. However, by applying the Isolation Lemma of Mulmuley, Vazirani, and Vazarani [MVV87], we can modify the cost function slightly so that the minimum-cost flow is unique, and is also a minimum-cost flow under the original cost function. Let us first state a modified version of the Isolation Lemma: ###### Lemma 3.12 (see [KS01, Lemma 4]). Given any collection of linear functions on $m$ variables with integer cooefficients in the range $\\{0,\dots,U\\}$. If each variable is independently set uniformly at random to a value from the set $\\{0,\dots,2mU\\}$, then with probability at least $1/2$ there is a unique function in the collection that takes minumum value. We now describe how to force the minimum-cost flow to be unique: ###### Lemma 3.13. Given a flow network with capacities and costs in the set $\left\\{1,2,\dots,U\right\\}$, and a positive integer $F$, modify the cost of each edge independently by adding a number uniformly at random from the set $\left\\{\frac{1}{4m^{2}U^{2}},\frac{2}{4m^{2}U^{2}},\dots,\frac{2mU}{4m^{2}U^{2}}\right\\}$. Then with probability at least $1/2$, the modified network has a unique minimum-cost flow of value $F$, and this flow is also a minimum-cost flow of value $F$ in the original network. ###### Proof. The modified cost of a flow at a vertex of $\Omega_{FLOW}$ is a linear function of $m$ independent variables chosen uniformly at random from the set $\left\\{\frac{1}{4m^{2}U^{2}},\frac{2}{4m^{2}U^{2}},\dots,\frac{2mU}{4m^{2}U^{2}}\right\\}$. where the coefficients are the coordinates of the flow vector, which by Lemma 3.11 are integers in the range $\\{0,\dots,U\\}$. So the Isolation Lemma tells us that with probablity at least $1/2$, there is a unique vertex of $\Omega_{FLOW}$ with minumum modified cost. Now, any vertex that was not originally of minimum cost must have been more expensive than the minimum cost by an integer. Since the sum of the flows on all edges can be at most $mU$, and no edge had its cost increased by more than $\frac{1}{2mU}$, the total cost of any flow cannot have increased by more than $1/2$. Thus, a vertex that was not originally of minimum cost cannot have minimum modified cost. ∎ We may now give an exact algorithm for standard minimum-cost flow. Note that this algorithm works for any integer flow value, but in particular we may easily find the exact max-flow value by running the interior-point max-flow algorithm with an error of $1/2$, since we know the max-flow value is an integer. ###### Lemma 3.14. To solve the standard minimum-cost flow problem in expected time $\widetilde{\mathcal{O}}\left(m^{3/2}\log^{2}U\right)$, perturb the edge costs as in Lemma 3.13, then run the min-cost flow interior point algorithm with an error of $\frac{1}{12m^{2}U^{3}}$, and round the flow on each edge to the nearest integer. ###### Proof. Let us prove correctness assuming that the modified costs do isolate a unique minimum-cost flow. The running time then follows directly from Theorem 3.10, and the fact from Lemma 3.13 that after a constant number of tries we can expect the modified costs to yield a unique minimum-cost flow. We first note that the modified edge costs are integer multiples of $\delta=\frac{1}{4m^{2}U^{2}}$. Therefore, by Theorem 3.11 the cost of the minumum-cost flow is at least $\delta$ less than the cost at any other vertex of $\Omega_{FLOW}$. Now, the flow returned by the interior-point algorithm can be expressed as a weighted average of the vertices of $\Omega_{FLOW}$. Since the cost of this flow is within $\frac{1}{12m^{2}U^{3}}=\frac{\delta}{3U}$ of the minimum cost, this weighted average must assign a combined weight of at most $\frac{1}{3U}$ to the non-minimum-cost vertices. Therefore, the flow along any edge differs by at most $1/3$ from the minimum-cost flow. So by rounding to the nearest integer flow, we obtain the minimum-cost flow. ∎ ## 4 Solving linear systems in symmetric M-Matrices A symmetric $M$-matrix is a positive definite symmetric matrix with non- positive off-diagonals (see, e.g. [HJ91, Axe96, BP94]). Every $M$-matrix has a factorization of the form $M=AA^{T}$ where each column of $A$ has at most $2$ nonzero entries [BCPT05]. Given such a factorization of an $M$-matrix, we we will show how to solve linear systems in the $M$-matrix in nearly-linear time. Throughout this section, $M$ will be an $n\times n$ symmetric $M$-matrix and $A$ will be a $n\times m$ matrix with 2 nonzero entries per column such that $M=AA^{T}$. Note that $M$ has $\mathcal{O}\left(m\right)$ non-zero entries. Our algorithm will make use of the Spielman-Teng $\widetilde{\mathcal{O}}\left(m\right)$ expected time approximate solver for linear systems in symmetric diagonally-dominant matrices, where we recall that a symmetric matrix is diagonally-dominant if each diagonal is at least the sum of the absolute values of the other entries in its row. It is strictly diagonally-dominant if each diagonal execceds each corresponding sum. We will use the following standard facts about symmetric $M$-matrices, which can be found, for example, in [HJ91]: ###### Fact 4.1. If $M=\left[\begin{matrix}M_{11}&M_{12}\\\ M_{12}^{T}&M_{22}\end{matrix}\right]$ is a symmetric $M$-matrix with $M_{11}$ a principal minor, then: 1. 1. $M$ is invertible and $M^{-1}$ is a nonnegative matrix. 2. 2. $M_{12}$ is a nonpositive matrix. 3. 3. $M_{11}$ is an $M$-matrix. 4. 4. The Schur complement $S=M_{22}-M_{12}^{T}M_{11}^{-1}M_{12}$ is an $M$-matrix. 5. 5. If all eigenvalues of $M$ fall in the range $[\lambda_{min},\lambda_{max}]$, then so do all diagonal entries of $S$. 6. 6. For any positive diagonal matrix $D$, $DMD$ is an $M$-matrix. 7. 7. There exists a positive diagonal matrix $D$ such that $DMD$ is strictly diagonally-dominant. Our algorithm will work by finding a diagonal matrix $D$ for which $DMD$ is diagonally-dominant, providing us with a system to which we may apply the solver of Spielman and Teng. Our algorithm builds $D$ by an iterative process. In each iteration, it decreases the number of rows that are not dominated by their diagonals by an expected constant factor. The main step of each iteration involves the solution of $\mathcal{O}\left(\log n\right)$ diagonally-dominant linear systems. For simplicity, we first explain how our algorithm would work if we made use of an algorithm $\boldsymbol{\mathit{x}}=\mathtt{ExactSolve}(M,\boldsymbol{\mathit{b}})$ that exactly solves the system $M\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}$, for diagonally-dominant $M$. We then explain how we may substitute an approximate solver. The key to our analysis is the following lemma, which says that if we multiply an $M$-matrix by a random diagonal matrix, then a constant fraction of the diagonals probably dominate their rows. ###### Lemma 4.2 (Random Scaling Lemma). Given an $n\times n$ $M$-matrix $M$, and positive real values $\zeta\leq 1$ and $r\leq\frac{1}{4}$, let $D$ be a random diagonal $n\times n$ matrix where each diagonal entry $d_{i}$ is chosen independently and uniformly from the interval $(0,1)$. Let $T\subset[n]$ be the set of rows of $MD$ with sums at least $r$ times the pre-scaled diagonal, i.e. $T=\\{i\in[n]:(MD{\mbox{\boldmath$1$}})_{i}\geq rm_{ii}\\}$ With probability at least $\frac{1-4r}{4r+7}$, we have $\left|T\right|\geq\left(\frac{1}{8}-\frac{r}{2}\right)\left(1-\beta-\frac{2}{3\zeta}\right)n$ where $\beta$ is the fraction of the diagonal entries of $M$ that are less than $\zeta$ times the average diagonal entry. Note in particular that for $r=0$, $T$ is the set of rows dominated by their diagonals. We will use the Random Scaling Lemma to decrease the number of rows that are not dominated by their diagonals. We will do this by preserving the rows that are dominated by their diagonals, and applying this lemma to the rest. Without loss of generality we write $M=\left[\begin{matrix}M_{11}&M_{12}\\\ M_{12}^{T}&M_{22}\end{matrix}\right]=\left[\begin{matrix}A_{1}A_{1}^{T}&A_{1}A_{2}^{T}\\\ A_{2}A_{1}^{T}&A_{2}A_{2}^{T}\end{matrix}\right]$, where the rows in the top section of $M$ are the ones that are already diagonally-dominant, so in particular $M_{11}$ is diagonally-dominant. Let $S=M_{22}-M_{12}^{T}M_{11}^{-1}M_{12}$ be the Schur complement and let $S_{D}$ be the matrix containing only the diagonal entries of $S$. We construct a random diagonal matrix $D_{R}$ of the same size as $M_{22}$ by choosing each diagonal element independently and uniformly from $(0,1)$. We then create diagonal matrix $D=\left[\begin{matrix}D_{1}&\\\ &D_{2}\end{matrix}\right]$ where $D_{2}=S_{D}^{-1/2}D_{R}$ and the diagonal entries of $D_{1}$ are given by $-M_{11}^{-1}M_{12}D_{2}{\mbox{\boldmath$1$}}$. We know that the diagonal entries of $D_{1}$ are positive because Fact 4.1 tells us that $M_{11}^{-1}$ is nonnegative and $M_{12}$ is nonpositive. We now show that the first set of rows of $DMD$ are diagonally-dominant, and a constant fraction of the rest probably become so as well. Since $M$ is an $M$-matrix and $D$ is positive diagonal, $DMD$ has no positive off-diagonals. Therefore, the diagonally-dominant rows of $DMD$ are the rows with nonnegative row sums. The row sums of $DMD$ are: $\displaystyle DMD{\mbox{\boldmath$1$}}$ $\displaystyle=\left[\begin{matrix}D_{1}M_{11}D_{1}{\mbox{\boldmath$1$}}+D_{1}M_{12}D_{2}{\mbox{\boldmath$1$}}\\\ D_{2}M_{12}^{T}D_{1}{\mbox{\boldmath$1$}}+D_{2}M_{22}D_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}0\\\ D_{2}SD_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]=\left[\begin{matrix}0\\\ D_{R}S_{D}^{-1/2}SS_{D}^{-1/2}D_{R}{\mbox{\boldmath$1$}}\end{matrix}\right]$ Note that the diagonal entries of $S_{D}^{-1/2}SS_{D}^{-1/2}$ are all 1. Thus by invoking Lemma 4.2 with $r=0$ and $\zeta=1$, we find that there is a $1/7$ probability that at least $1/24$ of the row sums in the bottom section of $DMD$ become nonnegative. Furthermore, we see that row sums in the top section remain nonnegative. The only problem with this idea is that in each iteration it could take $\widetilde{\mathcal{O}}\left(mn\right)$ time to compute the entire matrix $S$. Fortunately, we actually only need to compute the diagonals of $S$, (i.e. the matrix $S_{D}$). In fact, we only actually need a diagonal matrix $\Sigma$ that approximates $S_{D}$. As long the diagonals of $\Sigma^{-1/2}S\Sigma^{-1/2}$ fall in a relatively narrow range, we can still use the Random Scaling Lemma to get a constant fraction of improvement at each iteration. To compute these approximate diagonal values quickly, we use the random projection technique of Johnson and Lindenstrauss [JL84]. In Appendix A, we prove the following variant of their result, that deals with random projections into a space of constant dimension: ###### Theorem 4.3. For all constants $\alpha,\beta,\gamma,p\in(0,1)$, there is a positive integer $k=k_{JL}(\alpha,\beta,\gamma,p)$ such that the following holds: For any vectors $\boldsymbol{\mathit{v}}_{1},\dotsc,\boldsymbol{\mathit{v}}_{n}\in{\rm I\kern-2.0ptR}^{m}$ let $R$ be a $k\times m$ matrix with entries chosen independently at random from the standard normal distribution, and let $\boldsymbol{\mathit{w}}_{i}=\sqrt{\frac{1}{k}}R\boldsymbol{\mathit{v}}_{i}$. With probability at least $p$ both of the following hold: 1. 1. $\sum_{i=1}^{n}\frac{\left\|\boldsymbol{\mathit{v}}_{i}\right\|^{2}}{\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}\leq(1+\gamma)n$ 2. 2. $\left|\left\\{i:\frac{\left\|\boldsymbol{\mathit{v}}_{i}\right\|^{2}}{\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}<1-\alpha\right\\}\right|\leq\beta n$ Let us note that $S=A_{2}(I-A_{1}^{T}M_{11}^{-1}A_{1})A_{2}^{T}=A_{2}(I-A_{1}^{T}M_{11}^{-1}A_{1})^{2}A_{2}^{T}$ because $(I-A_{1}^{T}M_{11}^{-1}A_{1})$ is a projection matrix. So if we let $\boldsymbol{\mathit{a}}_{i}$ denote the $i$th row of $A_{2}$, we can write the $i$th diagonal of $S$ as $s_{ii}=\|(I-A_{1}^{T}M_{11}^{-1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|^{2}$. Then if we use Theorem 4.3 to create a random projection matrix $R$, $\|R(I-A_{1}^{T}M_{11}^{-1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|^{2}$ gives a good approximation to $s_{ii}$. Moreover, we can use one call to $\mathtt{ExactSolve}$ to compute each of the constant number of rows of the matrix $P=R(I-A_{1}^{T}M_{11}^{-1}A_{1})$. Since $A_{2}$ has $\mathcal{O}\left(m\right)$ entries, we can compute $PA_{2}^{T}$ in $\mathcal{O}\left(m\right)$ time, and obtain all the approximations $\|P\boldsymbol{\mathit{a}}_{i}^{T}\|^{2}$ in $\mathcal{O}\left(m\right)$ time, yielding the desired approximations of all $s_{ii}$ values. $\boldsymbol{\mathit{x}}=\mathtt{ExactMMatrixSolve}(A,\boldsymbol{\mathit{b}})$ Given: $n\times m$ matrix $A$, where $M=AA^{T}$ is an M-matrix and $A$ has at most 2 non-zeros per column. Returns: $\boldsymbol{\mathit{x}}$ satisfying $M\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}$ 1. Set $D:=I$. 2. Until $DMD$ is diagonally dominant do: a. Permute so that $DMD=\left[\begin{matrix}D_{1}M_{11}D_{1}&D_{1}M_{12}D_{2}\\\ D_{2}M_{12}^{T}D_{1}&D_{2}M_{22}D_{2}\end{matrix}\right]=\left[\begin{matrix}D_{1}A_{1}A_{1}^{T}D_{1}&D_{1}A_{1}A_{2}^{T}D_{2}\\\ D_{2}A_{2}A_{1}^{T}D_{1}&D_{2}A_{2}A_{2}^{T}D_{2}\end{matrix}\right]$ has the diagonally dominant rows in the top section. Let $\boldsymbol{\mathit{a}}_{1},\dotsc\boldsymbol{\mathit{a}}_{\nu}$ be the rows of $A_{2}$. b. Set $k=k_{JL}(\frac{1}{100},\frac{1}{5},\frac{1}{100},\frac{1}{3})$, and let $R$ be a random $k\times m$ matrix with independent standard normal entries. Let $\boldsymbol{\mathit{r}}_{i}$ be the $i$th row of $R$. c. For $i=1,\dotsc,k$, compute $\boldsymbol{\mathit{q}}_{i}^{T}=\mathtt{ExactSolve}(D_{1}M_{11}D_{1},D_{1}A_{1}\boldsymbol{\mathit{r}}_{i}^{T})$. d. Set $Q=\left[\begin{matrix}\boldsymbol{\mathit{q}}_{1}^{T}&\dotsb&\boldsymbol{\mathit{q}}_{k}^{T}\end{matrix}\right]^{T}$. e. Let $\Sigma$ be the $\nu\times\nu$ diagonal matrix with entries $\sigma_{i}=\|(R-QD_{1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|^{2}$. f. Let $D_{R}$ be a uniform random $\nu\times\nu$ diagonal matrix with diagonal entries in $(0,1)$. g. Set $D^{\prime}_{2}=\Sigma^{-1/2}D_{R}$ h. Set $D^{\prime}_{1}$ to be the matrix with diagonal $D_{1}\cdot\mathtt{ExactSolve}(D_{1}M_{11}D_{1},-D_{1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}})$ i. Set $D:=\left[\begin{matrix}D^{\prime}_{1}&\\\ &D^{\prime}_{2}\end{matrix}\right]$ 3. Return $\boldsymbol{\mathit{x}}=D\cdot\mathtt{ExactSolve}(DMD,D\boldsymbol{\mathit{b}})$ Figure 1: Algorithm for solving a linear system in a symmetric M-matrix. To speed up the algorithm we will replace the exact solver with the Spielman Teng approximate solver. Our suggested algorithm, still using an exact solver, is given in Figure 1. To make this algorithm fast, we replace the calls to the exact solver with calls to the approximate solver $\mathtt{STSolve}$ of Spielman and Teng: ###### Theorem 4.4 (Spielman-Teng [ST04, ST06]). The algorithm $\boldsymbol{\mathit{x}}=\mathtt{STSolve}(M,\boldsymbol{\mathit{b}},\epsilon)$ takes as input a symmetric diagonally-dominant $n\times n$ matrix $M$ with $m$ non-zeros, a column vector $\boldsymbol{\mathit{b}}$, and an error parameter $\epsilon>0$, and returns in expected time $\widetilde{\mathcal{O}}\left(m\log(1/\epsilon)\right)$ a column vector $\boldsymbol{\mathit{x}}$ satisfying $\|\boldsymbol{\mathit{x}}-M^{-1}\boldsymbol{\mathit{b}}\|_{M}\leq\epsilon\|M^{-1}\boldsymbol{\mathit{b}}\|_{M}$ We define the algorithm $\mathtt{MMatrixSolve}(A,\boldsymbol{\mathit{b}},\epsilon,\lambda_{min},\lambda_{max})$ as a modification of the algorithm $\mathtt{ExactMatrixSolve}$ in Figure 1. For this algorithm we need to provide upper and lower bounds $\lambda_{max},\lambda_{min}$ on the eigenvalues of the matrix $A$, and the running time will depend on $\kappa=\lambda_{max}/\lambda_{min}$. The modifications are that we need to set parameters: $\displaystyle\delta=(1/24)\lambda_{min}^{1/2}\kappa^{-1/2}n^{-1}\qquad\epsilon_{1}=.005(1.01\kappa mn)^{-1/2}\qquad\epsilon_{2}=(1/72)\kappa^{-5/2}n^{-2}$ and substitute the calls to ExactSolve in lines $2c$, $2h$ and $3$ respectively with * • $\mathtt{STSolve}(D_{1}M_{11}D_{1},D_{1}A_{1}\boldsymbol{\mathit{r}}_{i}^{T},\epsilon_{1})$ * • $\mathtt{STSolve}(D_{1}M_{11}D_{1},D_{1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}},\epsilon_{2})$ * • $\mathtt{STSolve}(DMD,D\boldsymbol{\mathit{b}},\epsilon)$. We may note that the final call to $\mathtt{STSolve}$ guarantees that $\|D^{-1}\boldsymbol{\mathit{x}}-D^{-1}M^{-1}\boldsymbol{\mathit{b}}\|_{DMD}\leq\epsilon\|D^{-1}M^{-1}\boldsymbol{\mathit{b}}\|_{DMD}$ or equivalently $\|\boldsymbol{\mathit{x}}-M^{-1}\boldsymbol{\mathit{b}}\|_{M}\leq\epsilon\|M^{-1}\boldsymbol{\mathit{b}}\|_{M}$ so the output fulfills the specification of an approximate solver, provided that the algorithm terminates. We can in fact bound the running time of this algorithm as follows: ###### Theorem 4.5. The expected running time of the algorithm $\mathtt{MMatrixSolve}$ is $\widetilde{\mathcal{O}}\left(m\log\frac{\kappa}{\epsilon}\right)$. ###### Proof. The running time is dominated by the calls to the Spielman-Teng solver. There are $\mathcal{O}\left(1\right)$ such solves per iterations, each of which take time $\widetilde{\mathcal{O}}\left(m\log\kappa\right)$, and at the conclusion of the algorithm, there is one final call of time $\widetilde{\mathcal{O}}\left(m\log\epsilon^{-1}\right)$. So, to prove the running time, it suffices for us to give a $\mathcal{O}\left(\log m\right)$ bound on the expected number of iterations. In particular, it suffices to show that in each iteration, the number of non- diagonally-domainant rows in $DMD$ decreases by a constant fraction with constant probability. In analyzing a single iteration, we let $D=\left[\begin{matrix}D_{1}&\\\ &D_{2}\end{matrix}\right]$ denote the diagonal scaling at the start of the iteration, and we let $D^{\prime}=\left[\begin{matrix}D^{\prime}_{1}&\\\ &D^{\prime}_{2}\end{matrix}\right]$ denote the new diagonal scaling. In Appendix A, we prove: ###### Lemma 4.6. $D^{\prime}$ is a positive diagonal matrix. This implies that $D^{\prime}MD^{\prime}$ has no positive off-diagonals, thereby enabling us to check which rows of $D^{\prime}MD^{\prime}$ are diagonally-dominant by looking for rows with nonnegative row sums. We again let $S=M_{22}-M_{12}^{T}M_{11}^{-1}M_{12}$ denote the Schur complement, and let $S_{D}$ denote the matrix containing the diagonal entries of $S$. Let us also define $\tilde{S}=\Sigma^{-1/2}S\Sigma^{-1/2}$. We know from Facts 4.1.4 and 4.1.6 that $\tilde{S}$ is an $M$-matrix. Let $\tilde{S}_{D}$ be the matrix containing the diagonal entries of $\tilde{S}$. In Appendix A, we show that the row sums of $MD^{\prime}$ are related to $\tilde{S}$ as follows: ###### Lemma 4.7. $MD^{\prime}{\mbox{\boldmath$1$}}\geq\left[\begin{matrix}0\\\ \Sigma^{1/2}(\tilde{S}D_{R}-\frac{1}{6}\tilde{S}_{D}){\mbox{\boldmath$1$}}\end{matrix}\right]$ The upper part of the above inequality tells us that all the row sums that were nonnegative in $DMD$ remain nonnegative in $D^{\prime}MD^{\prime}$. From the lower part of the inequality and by invoking the Random Scaling Lemma on the matrix $\tilde{S}$ with $r=\frac{1}{6}$, we find that with probabilty at least $\frac{1}{23}$, the fraction of remaining rows of $D^{\prime}MD^{\prime}$ that now have positive row sums is at least $\frac{1}{24}\left(1-\beta-\frac{2}{3\zeta}\right)$, where for some $\zeta<1$, $\beta$ is the fraction of the diagonal entries of $\tilde{S}$ that are less than $\zeta$ times the average diagonal entry. Indeed we prove in Appendix A: ###### Lemma 4.8. With probability at least $\frac{1}{9}$, at most $\frac{1}{5}$ of the diagonal entries of $\tilde{S}$ are smaller than $\left(\frac{99}{101}\right)^{3}$ times the average diagonal entry. So with probability at least $\frac{1}{9}\cdot\frac{1}{23}$, the fraction of rows with negative row sums in $DMD$ that now have positive row sums in $D^{\prime}MD^{\prime}$ is at least $\frac{1}{24}\left(1-\frac{1}{5}-\frac{2}{3}\left(\frac{101}{99}\right)^{3}\right)>0$. Thus, we may conclude that $\mathtt{MMatrixSolve}$ is expected to terminate after $\mathcal{O}\left(\log n\right)$ iterations, as claimed. ∎ ## 5 Final Remarks The reason that our interior-point algorithm currently cannot produce an exact solution to generalized flow problems is the dependence of our M-matrix solver on the condition number of the matrix, even when approximating in the matrix norm. It would be of interest to eliminate this dependence. It would also be nice to extend the result to networks with gains. The main obstacle is that the resulting linear programs may be ill-conditioned. ## References * [AGOT92] R.K. Ahuja, A.V. Goldberg, J.B. Orlin, and R.E. Tarjan. Finding minimum-cost flows by double scaling. Mathematical Programming, 53:243–266, 1992. * [AMO93] Ravindra K. Ahuja, Thomas L. Magnanti, and James B. Orlin. Network Flows. Prentice Hall, Inc., Englewood Cliffs, New Jersey, 1993. * [Axe96] Owe Axelsson. Iterative Solution Methods. Cambridge University Press, 1996. * [BCPT05] Erik G. Boman, Doron Chen, Ojas Parekh, and Sivan Toledo. On factor width and symmetric h-matrices. Linear Algebra and its Applications, 405:239–248, August 2005. * [BP94] Abraham Berman and Robert J. Plemmons. Nonnegative Matrices in the Mathematical Sciences. Society for Industrial and Applied Mathematics, Philadelphia, PA, USA, 1994. * [FW02] L.K. Fleischer and K.D. Wayne. Fast and simple approximation schemes for generalized flow. Mathematical Programming, 91(2):215–238, 2002. * [GFNR98] T.L. Griffith, M.A. Fuller, G.B. Northcraft, and T. Radzik. Faster algorithms for the generalized network flow problem. Mathematics of Operations Research, 23(1):69–199, 1998. * [GJO97] D. Goldfarb, Z. Jin, and J. Orlin. Polynomial-time highest-gain augmenting path algorithms for the generalized circulation problem. Mathematics of Operations Research, 22:793–802, 1997. * [GR98] A.V. Goldberg and S. Rao. Beyond the flow decomposition barrier. Journal of the ACM, 1998. * [GT87] A.V. Goldberg and R.E. Tarjan. Solving minimum cost flow problem by successive approximation. In Proceedings of the 19th ACM Symposium on the Theory of Computing, pages 7–18, 1987. * [HJ91] R. A. Horn and C. R. Johnson. Topics in Matrix Analysis. Cambridge University Press, 1991. * [JL84] W.B. Johnson and J. Lindenstrauss. Extensions of lipschitz mappings into a hilbert space. Contemp. Math., 26:189–206, 1984. * [KS01] Adam R. Klivans and Daniel A. Spielman. Randomness efficient identity testing of multivariate polynomials. In Proceedings of the 33th ACM Symposium on the Theory of Computing, pages 216–223, 2001. * [Li02] Lei Li. On the iterative criterion for generalized diagonally dominant matrices. SIAM Journal on Matrix Analysis, 24(1):17–24, 2002. * [LLH+98] Bishan Li, Lei Li, Masunori Harada, Hiroshi Niki, and Michael J. Tsatsomeros. An iterative criterion for h-matrices. Linear Algebra and its Applications, 271:179–190, 1998. * [MVV87] K. Mulmuley, U.V. Vazirani, and V.V. Vazirani. Matching is as easy as matrix inversion. Combinatorica, 7(1):105–113, 1987. * [Orl88] J.B. Orlin. A faster strongly polynomial minimum cost flow algorithm. In Proceedings of the 20th ACM Symposium on the Theory of Computing, pages 377–387, 1988. * [Ren88] J. Renegar. A polynomial-time algorithm based on Newton’s method for linear programming. Mathematical Programming, 40:59–93, 1988. * [Ren96] J. Renegar. Condition numbers, the barrier method, and the conjugate-gradient method. SIAM Journal on Optimization, 6:879–912, 1996. * [Sch03] Alexander Schrijver. Combinatorial Optimization: Polyhedra and Efficiency. Number 24 in Algorithms and Combinatorics. Springer-Verlag, Berlin, 2003\. * [ST04] Daniel A. Spielman and Shang-Hua Teng. Nearly-linear time algorithms for graph partitioning, graph sparsification, and solving linear systems. In Proceedings of the thirty-sixth annual ACM Symposium on Theory of Computing (STOC-04), pages 81–90, New York, June 13–15 2004. ACM Press. Full version available at http://arxiv.org/abs/cs.DS/0310051. * [ST06] Daniel A. Spielman and Shang-Hua Teng. Nearly-linear time algorithms for preconditioning and solving symmetric, diagonally dominant linear systems. Available at http://www.arxiv.org/abs/cs.NA/0607105, 2006. * [TW98] E. Tardos and K.D. Wayne. Simple generalized maximum flow algorithms. In 6th International Integer Programming and Combinatorial Optimization Conference, pages 310–324, 1998. * [Vai89] P.M. Vaidya. Speeding up linear programming using fast matrix multiplication. In 30th Annual IEEE Symposium on Foundations of Computer Science, pages 332–337, 1989. * [Ye97] Y. Ye. Interior Point Algorithms: Theory and Analysis. John Wiley, New York, 1997. ## Appendix A Proofs for Section 4 ###### Lemma 4.3. Given vectors $\boldsymbol{\mathit{v}}_{1},\dotsc,\boldsymbol{\mathit{v}}_{n}\in{\rm I\kern-2.0ptR}^{m}$ and constants $\alpha,\beta,\gamma,p\in(0,1)$, for positive constant integer $k=k_{JL}(\alpha,\beta,\gamma,p)$, let $R$ be a $k\times m$ matrix with entries chosen independently at random from the standard normal distribution, and let $\boldsymbol{\mathit{w}}_{i}=\sqrt{\frac{1}{k}}R\boldsymbol{\mathit{v}}_{i}$. With probability at least $p$ both of the following hold: 1. (i) $\sum_{i=1}^{n}\frac{\left\|\boldsymbol{\mathit{v}}_{i}\right\|^{2}}{\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}\leq(1+\gamma)n$ 2. (ii) $\left|\left\\{i:\frac{\left\|\boldsymbol{\mathit{v}}_{i}\right\|^{2}}{\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}<1-\alpha\right\\}\right|\leq\beta n$ ###### Proof. Let $Z_{i}=\frac{\left\|\boldsymbol{\mathit{v}}_{i}\right\|^{2}}{\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}$ and $Z=\sum_{i=1}^{n}Z_{i}$. Let $\boldsymbol{\mathit{r}}_{1},\dots,\boldsymbol{\mathit{r}}_{k}$ be the rows of $R$, and let $w_{ij}=k^{-1/2}\left\langle\boldsymbol{\mathit{r}}_{j},\boldsymbol{\mathit{v}}_{i}\right\rangle$ be the $j$th entry of $\boldsymbol{\mathit{w}}_{i}$. Without loss of generality, we assume that all the $\boldsymbol{\mathit{v}}_{i}$ are unit vectors. Thus for any given $i$, the expressions $k^{1/2}w_{i1},\dots,k^{1/2}w_{ik}$ are independent standard normal random variables. So the expression $\frac{Z_{i}}{k}=\frac{1}{k\left\|\boldsymbol{\mathit{w}}_{i}\right\|^{2}}=\frac{1}{\sum_{j=1}^{k}(\sqrt{k}w_{ij})^{2}}$ has inverse-chi-square distribution, with mean $\frac{1}{k-2}$ and variance $\frac{2}{(k-2)^{2}(k-4)}$. Therefore, $Z$ has mean $\frac{kn}{k-2}$ and variance at most $\frac{2k^{2}n^{2}}{(k-2)^{2}(k-4)}$, because $\mbox{\bf Var}\left[Z\right]=\mbox{\bf Var}\left[\sum_{i=1}^{n}Z_{i}\right]=\sum_{i,j=1}^{n}\mbox{\bf Cov}\left[Z_{i}Z_{j}\right]\leq\sum_{i,j=1}^{n}\sqrt{\mbox{\bf Var}\left[Z_{i}\right]\mbox{\bf Var}\left[Z_{j}\right]}=n^{2}\mbox{\bf Var}\left[Z_{i}\right]=k^{2}n^{2}\mbox{\bf Var}\left[\frac{Z_{i}}{k}\right]$ So using Cantelli’s inequality, we may conclude that $\mbox{\bf Pr}\left[Z>(1+\gamma)n\right]<\frac{\mbox{\bf Var}\left[Z\right]}{\mbox{\bf Var}\left[Z\right]+(1+\gamma-\frac{k}{k-2})^{2}n^{2}}\leq\frac{2}{2+(k-4)(1-\frac{2}{k})^{2}(\gamma-\frac{2}{k-2})^{2}}$ (1) By the same reasoning, $\frac{k}{Z_{i}}$ has chi-square distribution, with mean $k$ and variance $2k$. So using Cantelli’s inequality, we find that $\mbox{\bf Pr}\left[Z_{i}<1-\alpha\right]=\mbox{\bf Pr}\left[\frac{k}{Z_{i}}>\frac{k}{1-\alpha}\right]<\frac{\mbox{\bf Var}\left[k/Z_{i}\right]}{\mbox{\bf Var}\left[k/Z_{i}\right]+(\frac{k}{1-\alpha}-k)^{2}}=\frac{2}{2+(\frac{\alpha}{1-\alpha})^{2}k}$ Thus, the set $\left\\{i:Z_{i}<1-\alpha\right\\}$ has expected cardinality less than $\frac{2}{2+(\frac{\alpha}{1-\alpha})^{2}k}n$. So using Markov’s inequality, we conclude that $\mbox{\bf Pr}\left[\left|\left\\{i:Z_{i}<1-\alpha\right\\}\right|>\beta n\right]<\frac{2}{\beta(2+(\frac{\alpha}{1-\alpha})^{2}k)}$ (2) Combining inequalities 1 and 2 via the union bound, we find the probability that (i) and (ii) both occur is at least $1-\frac{2}{2+(k-4)(1-\frac{2}{k})^{2}(\gamma-\frac{2}{k-2})^{2}}-\frac{2}{\beta(2+(\frac{\alpha}{1-\alpha})^{2}k)}$ which is greater than $p$ for sufficiently large $k$. ∎ ###### Lemma 4.2 (Random Scaling Lemma). Given an $n\times n$ M-matrix $M$, and positive real values $\zeta\leq 1$ and $r\leq\frac{1}{4}$, let $D$ be a random diagonal $n\times n$ matrix where each diagonal entry $d_{i}$ is chosen independently and uniformly from the interval $(0,1)$. Let $T\subset[n]$ be the set of rows of $MD$ with sums at least $r$ times the pre-scaled diagonal, i.e. $T=\\{i\in[n]:(MD{\mbox{\boldmath$1$}})_{i}\geq rm_{ii}\\}$ With probability at least $\frac{1-4r}{4r+7}$, we have $\left|T\right|\geq\left(\frac{1}{8}-\frac{r}{2}\right)\left(1-\beta-\frac{2}{3\zeta}\right)n$ where $\beta$ is the fraction of the diagonal entries of $M$ that are less than $\zeta$ times the average diagonal entry. ###### Proof. Let $M_{O}$ denote the matrix containing only the off-diagonal elements of $M$. Thus, $M_{O}$ has no positive entries. Let $B$ be the set of rows of $M$ in which the diagonal entry is less than $\zeta$ times the average diagonal entry. Thus $\left|B\right|=\beta n$. We define a subset $J$ of rows of $M$ whose sums are not too far from being positive. In particular, we let $J$ be the set of rows in which the sum of the off-diagonal entries is no less than $-\frac{3}{2}$ times the diagonal entry: $J=\left\\{i\in[n]:(M_{O}{\mbox{\boldmath$1$}}_{n})_{i}\geq-\frac{3}{2}m_{ii}\right\\}$ Let us prove that $J$ cannot be too small. Let $S$ be the sum of the diagonal entries of $M$. We have: $\displaystyle S$ $\displaystyle={\mbox{\boldmath$1$}}_{n}^{T}M{\mbox{\boldmath$1$}}_{n}-\sum_{i\in[n]}(M_{O}{\mbox{\boldmath$1$}}_{n})_{i}$ $\displaystyle\geq-\sum_{i\in[n]}(M_{O}{\mbox{\boldmath$1$}}_{n})_{i}\qquad\text{(because $M$ is positive definite)}$ $\displaystyle\geq-\sum_{i\in[n]-(J\cup B)}(M_{O}{\mbox{\boldmath$1$}}_{n})_{i}\qquad\text{(because $M_{O}$ is non- positive)}$ $\displaystyle\geq\frac{3}{2}\sum_{i\in[n]-(J\cup B)}m_{ii}\qquad\text{(by definition of $J$)}$ $\displaystyle\geq\frac{3}{2}\frac{\zeta S}{n}\left|[n]-(J\cup B)\right|\qquad\text{(by definition of $B$)}$ $\displaystyle\geq\frac{3}{2}\frac{\zeta S}{n}(n-\left|J\right|-\beta n)$ So we see that $\left|J\right|\geq(1-\beta-\frac{2}{3\zeta})n$ Next, let us show that the rows in $J$ have a high probability of being in $T$. Consider the $i$th row sum of $M_{O}D$: $(M_{O}D{\mbox{\boldmath$1$}})_{i}=\sum_{j\neq i}d_{j}m_{ij}=\frac{1}{2}\sum_{j\neq i}m_{ij}+\sum_{j\neq i}(d_{j}-\frac{1}{2})m_{ij}=\frac{1}{2}(M_{O}{\mbox{\boldmath$1$}})_{i}+\sum_{j\neq i}(d_{j}-\frac{1}{2})m_{ij}$ Since each $(d_{j}-\frac{1}{2})$ is symmetrically distributed around zero, we may conclude that $\frac{1}{2}(M_{O}{\mbox{\boldmath$1$}})_{i}$ is the median value of $(M_{O}D{\mbox{\boldmath$1$}})_{i}$. We may also note that $(MD{\mbox{\boldmath$1$}})_{i}=(M_{O}D{\mbox{\boldmath$1$}})_{i}+d_{i}m_{ii}$, and that the values of $(M_{O}D{\mbox{\boldmath$1$}})_{i}$ and $d_{i}m_{ii}$ are independent. We thus have, for $i\in J$: $\displaystyle\mbox{\bf Pr}\left[(MD{\mbox{\boldmath$1$}})_{i}\geq rm_{ii}\right]$ $\displaystyle\geq\mbox{\bf Pr}\left[(M_{O}D{\mbox{\boldmath$1$}})_{i}\geq\frac{1}{2}(M_{O}{\mbox{\boldmath$1$}})_{i}\right]\cdot\mbox{\bf Pr}\left[d_{i}m_{ii}\geq rm_{ii}-\frac{1}{2}(M_{O}{\mbox{\boldmath$1$}})_{i}\right]$ $\displaystyle=\frac{1}{2}\cdot\mbox{\bf Pr}\left[d_{i}m_{ii}\geq rm_{ii}-\frac{1}{2}(M_{O}{\mbox{\boldmath$1$}})_{i}\right]$ $\displaystyle\geq\frac{1}{2}\cdot\mbox{\bf Pr}\left[d_{i}m_{ii}\geq rm_{ii}+\frac{1}{2}\cdot\frac{3}{2}m_{ii}\right]\quad\text{(by definition of $J$)}$ $\displaystyle=\frac{1}{2}\cdot\mbox{\bf Pr}\left[d_{i}\geq r+\frac{3}{4}\right]$ $\displaystyle=\frac{1}{4}-r$ Thus the expected size of $J-T$ is at most $\left(r+\frac{3}{4}\right)\left|J\right|$. So we find $\displaystyle\mbox{\bf Pr}\left[\left|T\right|>\left(\frac{1}{8}-\frac{r}{2}\right)\left|J\right|\right]$ $\displaystyle\geq\mbox{\bf Pr}\left[\left|J\cap T\right|>\left(\frac{1}{8}-\frac{r}{2}\right)\left|J\right|\right]$ $\displaystyle=\mbox{\bf Pr}\left[\left|J-T\right|<\left(\frac{r}{2}+\frac{7}{8}\right)\left|J\right|\right]$ $\displaystyle\geq 1-\frac{r+\frac{3}{4}}{\frac{r}{2}+\frac{7}{8}}\qquad\text{(by Markov's inequality)}$ $\displaystyle=\frac{1-4r}{4r+7}$ The lemma then follows from the lower bound on $\left|J\right|$ proven above. ∎ ###### Lemma 4.6. $D^{\prime}$ is a positive diagonal matrix. ###### Proof. $D^{\prime}_{2}=\Sigma^{-1/2}D_{R}$ is trivially positive diagonal by construction. To check that $D^{\prime}_{1}$ is positive, we use Lemma A.1, which implies that $D^{\prime}_{1}{\mbox{\boldmath$1$}}>-M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}_{n-\nu}+\delta\left(M_{11}^{-1}{\mbox{\boldmath$1$}}_{n-\nu}-\frac{3}{4}\lambda_{max}^{-1}{\mbox{\boldmath$1$}}_{n-\nu}\right)$ To see why the above expression is positive, recall from Fact 4.1 that $M_{11}^{-1}$ and $-M_{12}$ are positive matrices. Furthermore, note that the diagonals of $M_{11}^{-1}$ are at least $\lambda_{max}^{-1}$. ∎ ###### Lemma A.1. $\left\|D^{\prime}_{1}{\mbox{\boldmath$1$}}-M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|<\frac{3}{4}\delta\lambda_{max}^{-1}$ ###### Proof. Recall from the algorithm that $D_{1}^{-1}D^{\prime}_{1}{\mbox{\boldmath$1$}}=\mathtt{STSolve}(D_{1}M_{11}D_{1},D_{1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}},\epsilon_{2})$ Therefore, $\mathtt{STSolve}$ guarantees that $\left\|D_{1}^{-1}D^{\prime}_{1}{\mbox{\boldmath$1$}}-D_{1}^{-1}M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|_{D_{1}M_{11}D_{1}}\leq\epsilon_{2}\left\|D_{1}^{-1}M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|_{D_{1}M_{11}D_{1}}$ or equivalently $\left\|D^{\prime}_{1}{\mbox{\boldmath$1$}}-M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|_{M_{11}}\leq\epsilon_{2}\left\|M_{11}^{-1}(-M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}+\delta I){\mbox{\boldmath$1$}}\right\|_{M_{11}}$ which in turn implies that $\left\|D^{\prime}_{1}{\mbox{\boldmath$1$}}-M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|\leq\epsilon_{2}\kappa^{1/2}\left\|M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|$ We can then see $\displaystyle\left\|D^{\prime}_{1}{\mbox{\boldmath$1$}}-M_{11}^{-1}(-M_{12}D^{\prime}_{2}+\delta I){\mbox{\boldmath$1$}}\right\|$ $\displaystyle\leq\epsilon_{2}\kappa^{1/2}\left\|-M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}+\delta M_{11}^{-1}{\mbox{\boldmath$1$}}\right\|$ $\displaystyle\leq\epsilon_{2}\kappa^{1/2}\left\|M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}\right\|+\delta\epsilon_{2}\kappa^{1/2}\left\|M_{11}^{-1}{\mbox{\boldmath$1$}}\right\|$ $\displaystyle\leq\epsilon_{2}\kappa^{1/2}\left\|M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}\right\|+\delta\epsilon_{2}\kappa^{1/2}\lambda_{min}^{-1}n^{1/2}$ $\displaystyle\leq\epsilon_{2}\kappa n^{1/2}\left\|D^{\prime}_{2}{\mbox{\boldmath$1$}}\right\|+\delta\epsilon_{2}\kappa^{1/2}\lambda_{min}^{-1}n^{1/2}\qquad\text{(by Lemma \ref{lem:aux2})}$ $\displaystyle\leq\epsilon_{2}\kappa n^{1/2}\left\|\Sigma^{-1/2}{\mbox{\boldmath$1$}}\right\|+\delta\epsilon_{2}\kappa^{1/2}\lambda_{min}^{-1}n^{1/2}\qquad\text{($D^{\prime}_{2}<\Sigma^{-1/2}$ by construction)}$ $\displaystyle\leq 2\epsilon_{2}\kappa n^{1/2}\left\|S_{D}^{-1/2}{\mbox{\boldmath$1$}}\right\|+\delta\epsilon_{2}\kappa^{1/2}\lambda_{min}^{-1}n^{1/2}\quad\text{($\Sigma^{-1/2}\leq 2S_{D}^{-1/2}$ by Lemma \ref{lem:scaledschur})}$ $\displaystyle\leq 2\epsilon_{2}\kappa\lambda_{min}^{-1/2}n+\delta\epsilon_{2}\kappa^{1/2}\lambda_{min}^{-1}n^{1/2}\qquad\text{($S_{D}^{-1/2}{\mbox{\boldmath$1$}}<\lambda_{min}^{-1/2}{\mbox{\boldmath$1$}}$ by Fact \ref{fact:mmatrix}.\ref{fact:schurdiag})}$ $\displaystyle=\left(2\delta^{-1}\epsilon_{2}\kappa^{2}\lambda_{min}^{1/2}n+\epsilon_{2}\kappa^{1/2}n^{1/2}\right)\delta\lambda_{max}^{-1}$ $\displaystyle=\left(\frac{2}{3}+\frac{1}{72}\kappa^{-2}n^{-3/2}\right)\delta\lambda_{max}^{-1}$ $\displaystyle<\frac{3}{4}\delta\lambda_{max}^{-1}\ $ ∎ ###### Lemma 4.8. With probability at least $\frac{1}{9}$, at most $\frac{1}{5}$ of the diagonal entries of $\tilde{S}$ are smaller than $\left(\frac{99}{101}\right)^{3}$ times the average diagonal entry. ###### Proof. Recall that the diagonal entries of $\tilde{S}$ are $\tilde{s}_{ii}=\frac{s_{ii}}{\sigma_{i}}$, where $s_{ii}=\left\|(I-A_{1}^{T}M_{11}^{-1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\right\|^{2}$ and $\sigma_{i}=\left\|(R-QD_{1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\right\|^{2}$. Let us define $w_{i}=\frac{1}{k}\left\|R(I-A_{1}^{T}M_{11}^{-1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\right\|^{2}$, where $k=k_{JL}\left(\frac{1}{100},\frac{1}{5},\frac{1}{100},\frac{1}{3}\right)$. By Lemma 4.3, there is at least $\frac{1}{3}$ probability that $\displaystyle\frac{1}{\nu}\sum_{i=1}^{\nu}\frac{s_{ii}}{w_{i}}$ $\displaystyle\leq 1.01$ and $\displaystyle\left|\left\\{i:\frac{s_{ii}}{w_{i}}\leq.99\right\\}\right|\leq\frac{1}{5}\nu$ So, by Lemma A.2 below, there is at least a $\frac{1}{3}-\frac{2}{9km}>\frac{1}{9}$ probability that the average diagonal entry of $\tilde{S}$ is at most $\frac{1}{\nu}\sum_{i=1}^{\nu}\frac{s_{ii}}{\sigma_{i}}=\frac{1}{\nu}\sum_{i=1}^{\nu}\frac{s_{ii}}{w_{i}}\cdot\frac{w_{i}}{\sigma_{i}}\leq\frac{1.01}{k(.99)^{2}}$ and similarly we have the following bound on the number of small diagonal entries: $\displaystyle\left|\left\\{i:\frac{s_{ii}}{\sigma_{i}}\leq\frac{.99}{k(1.01)^{2}}\right\\}\right|\leq\frac{1}{5}\nu$ ∎ ###### Lemma A.2. With probability at least $1-\frac{2}{9km}$ it holds for all $i$ that $\frac{1}{k(1.01)^{2}}\leq\frac{w_{i}}{\sigma_{i}}\leq\frac{1}{k(.99)^{2}}$ ###### Proof. We have: $\displaystyle\left|\sigma_{i}^{1/2}-k^{1/2}w_{i}^{1/2}\right|$ $\displaystyle=\left|\|(R-QD_{1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|-\|(R-RA_{1}^{T}M_{11}^{-1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|\right|$ $\displaystyle\leq\|((R-QD_{1}A_{1})-(R-RA_{1}^{T}M_{11}^{-1}A_{1}))\boldsymbol{\mathit{a}}_{i}^{T}\|$ $\displaystyle=\|(RA_{1}^{T}M_{11}^{-1}A_{1}-QD_{1}A_{1})\boldsymbol{\mathit{a}}_{i}^{T}\|$ $\displaystyle\leq\|\boldsymbol{\mathit{a}}_{i}\|\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}A_{1}-\boldsymbol{\mathit{q}}_{j}D_{1}A_{1}\|^{2}}$ $\displaystyle=\|\boldsymbol{\mathit{a}}_{i}\|\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}D_{1}^{-1}-\boldsymbol{\mathit{q}}_{j}\|_{D_{1}M_{11}D_{1}}^{2}}$ $\displaystyle\leq\lambda_{max}^{1/2}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}D_{1}^{-1}-\boldsymbol{\mathit{q}}_{j}\|_{D_{1}M_{11}D_{1}}^{2}}$ ($\left\|\boldsymbol{\mathit{a}}_{i}\right\|^{2}$ is $i$th diagonal of $M_{22}$, so cannot exceed $M_{22}$’s largest eigenvalue) $\displaystyle\leq\lambda_{max}^{1/2}(\frac{s_{ii}}{\lambda_{min}})^{1/2}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}D_{1}^{-1}-\boldsymbol{\mathit{q}}_{j}\|_{D_{1}M_{11}D_{1}}^{2}}\quad\text{(using Fact \ref{fact:mmatrix}.\ref{fact:schurdiag})}$ $\displaystyle=(\kappa s_{ii})^{1/2}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}D_{1}^{-1}-\boldsymbol{\mathit{q}}_{j}\|_{D_{1}M_{11}D_{1}}^{2}}$ $\displaystyle\leq(\kappa s_{ii})^{1/2}\epsilon_{1}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}D_{1}^{-1}\|_{D_{1}M_{11}D_{1}}^{2}}\quad\text{(by guarantee of $\mathtt{STSolve}$)}$ $\displaystyle=.005\cdot s_{ii}^{1/2}(1.01mn)^{-1/2}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}A_{1}^{T}M_{11}^{-1}A_{1}\|^{2}}$ $\displaystyle=.005\cdot s_{ii}^{1/2}(1.01mn)^{-1/2}\sqrt{\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}\|^{2}}\quad\text{(because $A_{1}^{T}M_{11}A_{1}$ is a projection matrix)}$ $\displaystyle\leq.01\cdot s_{ii}^{1/2}k^{1/2}(1.01n)^{-1/2}$ The above inequality does not hold with probability at most $\frac{2}{9km}$, based on the fact that expression $\sum_{j=1}^{k}\|\boldsymbol{\mathit{r}}_{j}\|^{2}$ has chi-square distribution with $mk$ degrees of freedom. $\displaystyle\leq.01\cdot k^{1/2}w_{i}^{1/2}\qquad\text{(Lemma \ref{lem:randproj} implies that $s_{ii}\leq 1.01\cdot nw_{i}$)}$ So we conclude that $\left|\sqrt{\frac{\sigma}{w_{i}}}-k^{1/2}\right|\leq.01\cdot k^{1/2}$ ∎ ###### Lemma 4.7. $MD^{\prime}{\mbox{\boldmath$1$}}\geq\left[\begin{matrix}0\\\ \Sigma^{1/2}(\tilde{S}D_{R}-\frac{1}{6}\tilde{S}_{D}){\mbox{\boldmath$1$}}\end{matrix}\right]$ ###### Proof. $\displaystyle MD^{\prime}{\mbox{\boldmath$1$}}$ $\displaystyle=\left[\begin{matrix}M_{11}D^{\prime}_{1}{\mbox{\boldmath$1$}}+M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}\\\ M_{12}^{T}D^{\prime}_{1}{\mbox{\boldmath$1$}}+M_{22}D^{\prime}_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}0\\\ SD^{\prime}_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]+M\left[\begin{matrix}D^{\prime}_{1}{\mbox{\boldmath$1$}}+M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}-\delta M_{11}^{-1}{\mbox{\boldmath$1$}}\\\ 0\end{matrix}\right]+\left[\begin{matrix}\delta{\mbox{\boldmath$1$}}\\\ \delta M_{12}^{T}M_{11}^{-1}{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle\geq\left[\begin{matrix}0\\\ SD^{\prime}_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]-\lambda_{max}\|D^{\prime}_{1}{\mbox{\boldmath$1$}}+M_{11}^{-1}M_{12}D^{\prime}_{2}{\mbox{\boldmath$1$}}-\delta M_{11}^{-1}{\mbox{\boldmath$1$}}\|{\mbox{\boldmath$1$}}+\left[\begin{matrix}\delta{\mbox{\boldmath$1$}}\\\ -\delta\|M_{12}^{T}M_{11}^{-1}{\mbox{\boldmath$1$}}\|{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle\geq\left[\begin{matrix}0\\\ SD^{\prime}_{2}{\mbox{\boldmath$1$}}\end{matrix}\right]-\frac{3}{4}\delta{\mbox{\boldmath$1$}}+\left[\begin{matrix}\delta{\mbox{\boldmath$1$}}\\\ -\delta\kappa^{1/2}n{\mbox{\boldmath$1$}}\end{matrix}\right]\quad\text{(using Lemmas \ref{lem:aux1} and \ref{lem:aux2})}$ $\displaystyle\geq\left[\begin{matrix}0\\\ SD^{\prime}_{2}{\mbox{\boldmath$1$}}-2\delta\kappa^{1/2}n{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}0\\\ S\Sigma^{-1/2}D_{R}{\mbox{\boldmath$1$}}-\frac{1}{12}\lambda_{min}^{1/2}{\mbox{\boldmath$1$}}\end{matrix}\right]$ $\displaystyle\geq\left[\begin{matrix}0\\\ S\Sigma^{-1/2}D_{R}{\mbox{\boldmath$1$}}-\frac{1}{12}S_{D}^{1/2}{\mbox{\boldmath$1$}}\end{matrix}\right]\quad\text{(using Fact \ref{fact:mmatrix}.\ref{fact:schurdiag})}$ $\displaystyle\geq\left[\begin{matrix}0\\\ S\Sigma^{-1/2}D_{R}{\mbox{\boldmath$1$}}-\frac{1}{6}S_{D}^{1/2}\tilde{S}_{D}^{1/2}{\mbox{\boldmath$1$}}\end{matrix}\right]\quad\text{(using Lemma \ref{lem:scaledschur})}$ ∎ ###### Lemma A.3. For all positive vectors $\boldsymbol{\mathit{v}}$, $\|M_{12}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}\|\leq\kappa^{1/2}n^{1/2}\left\|\boldsymbol{\mathit{v}}\right\|$ ###### Proof. Define $c=\lambda_{min}^{-1}\kappa^{-1/2}n^{-1/2}\left\|\boldsymbol{\mathit{v}}\right\|=\lambda_{max}^{-1}\kappa^{1/2}n^{-1/2}\left\|\boldsymbol{\mathit{v}}\right\|$. $\displaystyle\|M_{12}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}\|\leq\|M_{12}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}\|_{1}$ $\displaystyle=-{\mbox{\boldmath$1$}}^{T}M_{12}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}\qquad\text{(by Fact \ref{fact:mmatrix}, $M_{11}^{-1}$ and $-M_{12}$ are nonnegtive)}$ $\displaystyle=\frac{1}{2c}\left(\boldsymbol{\mathit{v}}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}+c^{2}{\mbox{\boldmath$1$}}^{T}M_{22}{\mbox{\boldmath$1$}}-\left[\begin{matrix}\boldsymbol{\mathit{v}}^{T}M_{11}^{-1}&c{\mbox{\boldmath$1$}}^{T}\end{matrix}\right]M\left[\begin{matrix}M_{11}^{-1}\boldsymbol{\mathit{v}}\\\ c{\mbox{\boldmath$1$}}\end{matrix}\right]\right)$ $\displaystyle\leq\frac{1}{2c}\left(\boldsymbol{\mathit{v}}^{T}M_{11}^{-1}\boldsymbol{\mathit{v}}+c^{2}{\mbox{\boldmath$1$}}^{T}M_{22}{\mbox{\boldmath$1$}}\right)$ $\displaystyle\leq\frac{1}{2}\left(\frac{\left\|\boldsymbol{\mathit{v}}\right\|^{2}}{c\lambda_{min}}+c\lambda_{max}n\right)=\kappa^{1/2}n^{1/2}\left\|\boldsymbol{\mathit{v}}\right\|$ ∎ ## Appendix B Solving Matrices from the Interior-Point Method In the interior-point algorithm, we need to solve matrices of the form $M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}=\left[\begin{matrix}AD_{1}^{2}A^{T}+D_{2}^{2}&AD_{1}^{2}\\\ D_{1}^{2}A^{T}&D_{1}^{2}+D_{3}^{2}\end{matrix}\right]+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}$ where $A$ is an $n\times m$ matrix with entries bounded by $U$ in absolute value, $AA^{T}$ is an M-matrix, and $D_{1},D_{2},D_{3}$ are positive diagonal matrices. We show how to do this using our $\mathtt{MMatrixSolve}$ algorithm. Consider the Schur complement of $M$: $M_{S}=(AD_{1}^{2}A^{T}+D_{2}^{2})-AD_{1}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}D_{1}^{2}A^{T}=AD_{1}^{2}D_{3}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}A^{T}+D_{2}^{2}=A_{S}A_{S}^{T}$ where $A_{S}=\left[\begin{matrix}AD_{1}D_{3}(D_{1}^{2}+D_{3}^{2})^{-1/2}&D_{2}\end{matrix}\right]$. Note that $M_{S}$ is also an M-matrix, and that the eigenvalues of $M_{S}$ fall in the range $[d_{min}^{2},d_{max}^{2}(U\sqrt{nm}+1)]$ where $d_{min}$ and $d_{max}$ are respectively the smallest and largest diagonal entry in $D_{1},D_{2},D_{3}$. We can build an solver for systems in $M$ from a solver for systems in $M_{S}$, by using the following easily verifiable property of the Schur complement: ###### Lemma B.1. For $M=\left[\begin{matrix}M_{11}&M_{12}\\\ M_{12}^{T}&M_{22}\end{matrix}\right]$ and Schur complement $M_{S}=M_{11}-M_{12}M_{22}^{-1}M_{12}^{T}$, we have $\left\|\left[\begin{matrix}\boldsymbol{\mathit{x}}_{1}\\\ \boldsymbol{\mathit{x}}_{2}\end{matrix}\right]-M^{-1}\left[\begin{matrix}\boldsymbol{\mathit{b}}_{1}\\\ \boldsymbol{\mathit{b}}_{2}\end{matrix}\right]\right\|_{M}=\left\|\boldsymbol{\mathit{x}}_{1}-M_{S}^{-1}(\boldsymbol{\mathit{b}}_{1}-M_{12}M_{22}^{-1}\boldsymbol{\mathit{b}}_{2})\right\|_{M_{S}}+\left\|\boldsymbol{\mathit{x}}_{2}-M_{22}^{-1}(\boldsymbol{\mathit{b}}_{2}-M_{12}^{T}\boldsymbol{\mathit{x}}_{1})\right\|_{M_{22}}$ Then, to solve systems in $M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}$, we can use the Sherman-Morrison formula: $(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}=M^{-1}-\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}$ In particular, we give the following algorithm, which runs in time $\widetilde{\mathcal{O}}\left(m\log\frac{\kappa\left\|\boldsymbol{\mathit{v}}\right\|}{\epsilon}\right)$: $\boldsymbol{\mathit{x}}=\mathtt{Solve}(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\boldsymbol{\mathit{b}},\epsilon)$ where $M=\left[\begin{matrix}AD_{1}^{2}A^{T}+D_{2}^{2}&AD_{1}^{2}\\\ D_{1}^{2}A^{T}&D_{1}^{2}+D_{3}^{2}\end{matrix}\right]$ and $\boldsymbol{\mathit{b}}=\left[\begin{matrix}\boldsymbol{\mathit{b}}_{1}\\\ \boldsymbol{\mathit{b}}_{2}\end{matrix}\right]$ and $\boldsymbol{\mathit{v}}=\left[\begin{matrix}\boldsymbol{\mathit{v}}_{1}\\\ \boldsymbol{\mathit{v}}_{2}\end{matrix}\right]$ • Define $\epsilon_{1}=\frac{\epsilon}{2}(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1}$ and $\epsilon_{2}=\min\left\\{\frac{1}{2},\frac{\epsilon}{14}(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1}\right\\}$ • $\boldsymbol{\mathit{y}}^{\prime}=\mathtt{MMatrixSolve}\left(A_{S},\boldsymbol{\mathit{b}}_{1}-AD_{1}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}\boldsymbol{\mathit{b}}_{2},\epsilon_{1},d_{min}^{2},d_{max}^{2}(U\sqrt{nm}+1)\right)$ • $\boldsymbol{\mathit{y}}=\left[\begin{matrix}\boldsymbol{\mathit{y}}^{\prime}\\\ (D_{1}^{2}+D_{3}^{2})^{-1}(\boldsymbol{\mathit{b}}_{2}-D_{1}^{2}A^{T}\boldsymbol{\mathit{y}}^{\prime})\end{matrix}\right]$ • $\boldsymbol{\mathit{z}}^{\prime}=\mathtt{MMatrixSolve}\left(A_{S},\boldsymbol{\mathit{v}}_{1}-AD_{1}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}\boldsymbol{\mathit{v}}_{2},\epsilon_{2},d_{min}^{2},d_{max}^{2}(U\sqrt{nm}+1)\right)$ • $\boldsymbol{\mathit{z}}=\left[\begin{matrix}\boldsymbol{\mathit{z}}^{\prime}\\\ (D_{1}^{2}+D_{3}^{2})^{-1}(\boldsymbol{\mathit{v}}_{2}-D_{1}^{2}A^{T}\boldsymbol{\mathit{z}}^{\prime})\end{matrix}\right]$ • Return $\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{y}}-\frac{\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}$ ###### Lemma B.2. $\boldsymbol{\mathit{x}}=\mathtt{Solve}(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\boldsymbol{\mathit{b}},\epsilon)$ satisfies $\left\|\boldsymbol{\mathit{x}}-(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}<\epsilon\left\|(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}$ ###### Proof. We first show that $\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\leq\epsilon_{1}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$: $\displaystyle\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ $\displaystyle=\left\|\boldsymbol{\mathit{y}}^{\prime}-M_{S}^{-1}(\boldsymbol{\mathit{b}}_{1}-AD_{1}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}\boldsymbol{\mathit{b}}_{2})\right\|_{M_{S}}\quad\text{(by Lemma \ref{lem:schurnorm})}$ $\displaystyle\leq\epsilon_{1}\left\|M_{S}^{-1}(\boldsymbol{\mathit{b}}_{1}-AD_{1}^{2}(D_{1}^{2}+D_{3}^{2})^{-1}\boldsymbol{\mathit{b}}_{2})\right\|_{M_{S}}\quad\text{(guaranteed by $\mathtt{MMatrixSolve}$)}$ $\displaystyle=\epsilon_{1}\left(\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}-\left\|M_{22}^{-1}\boldsymbol{\mathit{b}}_{2}\right\|_{M_{22}}\right)\quad\text{(by Lemma \ref{lem:schurnorm})}$ $\displaystyle\leq\epsilon_{1}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ By the same reasoning, $\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\leq\epsilon_{2}\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}$. Next, let us define the inner product $\left\langle\boldsymbol{\mathit{v}}_{1},\boldsymbol{\mathit{v}}_{2}\right\rangle_{M}=\boldsymbol{\mathit{v}}_{1}^{T}M\boldsymbol{\mathit{v}}_{2}$. We will use repeatedly the inequality $\left|\left\langle\boldsymbol{\mathit{v}}_{1},\boldsymbol{\mathit{v}}_{2}\right\rangle_{M}\right|\leq\left\|\boldsymbol{\mathit{v}}_{1}\right\|_{M}\left\|\boldsymbol{\mathit{v}}_{2}\right\|_{M}$ Recall that we return the value $\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{y}}-\frac{\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}$. So we begin by analyzing the expressions $\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}$ and $\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}$: $\displaystyle\left\|\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}-M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ $\displaystyle\leq\left\|\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}-\boldsymbol{\mathit{z}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\left\|\boldsymbol{\mathit{z}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}-M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ $\displaystyle=\left|\left\langle\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}},M^{-1}\boldsymbol{\mathit{b}}\right\rangle_{M}\right|\left\|\boldsymbol{\mathit{z}}\right\|_{M}+\left|\left\langle M^{-1}\boldsymbol{\mathit{v}},M^{-1}\boldsymbol{\mathit{b}}\right\rangle_{M}\right|\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}$ $\displaystyle\leq\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\left\|\boldsymbol{\mathit{z}}\right\|_{M}+\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}$ $\displaystyle=\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\left(\left\|\boldsymbol{\mathit{z}}\right\|_{M}+\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\right)$ $\displaystyle\leq\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\left(\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}+2\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\right)$ $\displaystyle\leq\epsilon_{2}(\epsilon_{2}+2)\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ (3) $\displaystyle|\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}-\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}|$ $\displaystyle=\left|\left\langle M^{-1}\boldsymbol{\mathit{v}},\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\rangle_{M}\right|$ $\displaystyle\leq\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}\left\|\boldsymbol{\mathit{z}}-M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}$ $\displaystyle\leq\epsilon_{2}\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}$ $\displaystyle=\epsilon_{2}(\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})$ (4) We thus have: $\displaystyle\left\|\boldsymbol{\mathit{x}}-(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ (5) $\displaystyle=\left\|\left(\boldsymbol{\mathit{y}}-\frac{\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}\right)-\left(M^{-1}\boldsymbol{\mathit{b}}-\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\right)\right\|_{M}$ $\displaystyle\leq\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\left\|\frac{\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}-\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}\right\|_{M}+\left\|\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}-\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\right\|_{M}$ $\displaystyle=\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{1}{1+\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}\left(\left\|\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}-M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{|\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}-\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}|}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\left\|M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\right)$ $\displaystyle\leq\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{1}{\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}}\left(\left\|\boldsymbol{\mathit{z}}\boldsymbol{\mathit{z}}^{T}\boldsymbol{\mathit{b}}-M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{|\boldsymbol{\mathit{v}}^{T}\boldsymbol{\mathit{z}}-\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}|}{\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\left\|M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\right)$ $\displaystyle\leq\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{1}{(1-\epsilon_{2})\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\left(\epsilon_{2}(\epsilon_{2}+2)\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\epsilon_{2}\left\|M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\right)$ (by equations 3 and 4) $\displaystyle=\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{\epsilon_{2}(\epsilon_{2}+2)\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\epsilon_{2}\left|\left\langle M^{-1}\boldsymbol{\mathit{v}},M^{-1}\boldsymbol{\mathit{b}}\right\rangle_{M}\right|\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}}{(1-\epsilon_{2})\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}}$ $\displaystyle\leq\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{\epsilon_{2}(\epsilon_{2}+2)\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\epsilon_{2}\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}}{(1-\epsilon_{2})\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2}}$ $\displaystyle=\left\|\boldsymbol{\mathit{y}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}+\frac{\epsilon_{2}(\epsilon_{2}+3)}{1-\epsilon_{2}}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ $\displaystyle\leq\left(\epsilon_{1}+\frac{\epsilon_{2}(\epsilon_{2}+3)}{1-\epsilon_{2}}\right)\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ $\displaystyle\leq\epsilon(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}$ (6) So we conclude $\displaystyle\left\|\boldsymbol{\mathit{x}}-(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}$ $\displaystyle\leq(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{1/2}\left\|\boldsymbol{\mathit{x}}-(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\qquad\text{by Lemma \ref{lem:rankoneupdate}(i)}$ $\displaystyle\leq\epsilon(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1/2}\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\qquad\text{by equation (\ref{eq:almostdone})}$ $\displaystyle=\epsilon(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1/2}\left\|\boldsymbol{\mathit{b}}\right\|_{M^{-1}}$ $\displaystyle\leq\epsilon\left\|\boldsymbol{\mathit{b}}\right\|_{(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}}\qquad\text{by Lemma \ref{lem:rankoneupdate}(ii)}$ $\displaystyle=\epsilon\left\|(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{b}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}$ ∎ ###### Lemma B.3. For all vectors $\boldsymbol{\mathit{v}}$, $\boldsymbol{\mathit{w}}$, and symmetric positive definite $M$: (i) $\displaystyle\left\|\boldsymbol{\mathit{w}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}$ $\displaystyle\leq\left\|\boldsymbol{\mathit{w}}\right\|_{M}(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{1/2}$ (ii) $\displaystyle\left\|\boldsymbol{\mathit{w}}\right\|_{(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}}$ $\displaystyle\geq\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}(1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}})^{-1/2}$ ###### Proof of (i). $\displaystyle\left\|\boldsymbol{\mathit{w}}\right\|_{M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}}$ $\displaystyle=(\boldsymbol{\mathit{w}}^{T}(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})\boldsymbol{\mathit{w}})^{1/2}$ $\displaystyle=(\boldsymbol{\mathit{w}}^{T}M\boldsymbol{\mathit{w}}+(\boldsymbol{\mathit{w}}^{T}\boldsymbol{\mathit{v}})^{2})^{1/2}$ $\displaystyle=(\left\|\boldsymbol{\mathit{w}}\right\|_{M}^{2}+\left\langle\boldsymbol{\mathit{w}},M^{-1}\boldsymbol{\mathit{v}}\right\rangle_{M}^{2})^{1/2}$ $\displaystyle\leq(\left\|\boldsymbol{\mathit{w}}\right\|_{M}^{2}+\left\|\boldsymbol{\mathit{w}}\right\|_{M}^{2}\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2})^{1/2}$ $\displaystyle=\left\|\boldsymbol{\mathit{w}}\right\|_{M}(1+\left\|M^{-1}\boldsymbol{\mathit{v}}\right\|_{M}^{2})^{1/2}$ ∎ ###### Proof of (ii). $\displaystyle\left\|\boldsymbol{\mathit{w}}\right\|_{(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}}$ $\displaystyle=(\boldsymbol{\mathit{w}}^{T}(M+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T})^{-1}\boldsymbol{\mathit{w}})^{1/2}$ $\displaystyle=\left(\boldsymbol{\mathit{w}}^{T}\left(M^{-1}-\frac{M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\right)\boldsymbol{\mathit{w}}\right)^{1/2}$ $\displaystyle=\left(\boldsymbol{\mathit{w}}^{T}M^{-1}\boldsymbol{\mathit{w}}-\frac{\boldsymbol{\mathit{w}}^{T}M^{-1}\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{w}}}{1+\boldsymbol{\mathit{v}}^{T}M^{-1}\boldsymbol{\mathit{v}}}\right)^{1/2}$ $\displaystyle=\left(\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}^{2}-\frac{\left\langle\boldsymbol{\mathit{w}},\boldsymbol{\mathit{v}}\right\rangle_{M^{-1}}^{2}}{1+\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}}\right)^{1/2}$ $\displaystyle\geq\left(\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}^{2}-\frac{\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}^{2}\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}}{1+\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}}\right)^{1/2}$ $\displaystyle=\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}\left(1-\frac{\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}}{1+\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}}\right)^{1/2}$ $\displaystyle=\left\|\boldsymbol{\mathit{w}}\right\|_{M^{-1}}\left(1+\left\|\boldsymbol{\mathit{v}}\right\|_{M^{-1}}^{2}\right)^{-1/2}$ ∎ ## Appendix C Interior-Point Method using an Approximate Solver Throughout this section, we take $\mathtt{Solve}$ to be an algorithm such that $\boldsymbol{\mathit{x}}=\mathtt{Solve}(M,\boldsymbol{\mathit{b}},\epsilon)$ satisfies $\left\|\boldsymbol{\mathit{x}}-M^{-1}\boldsymbol{\mathit{b}}\right\|_{M}\leq\epsilon\left\|M^{-1}\boldsymbol{\mathit{b}}\right\|$ We use the notational convention that $S$ denotes the diagonal matrix whose diagonal is $\boldsymbol{\mathit{s}}$. The same applies for $X$ and $\boldsymbol{\mathit{x}}$, etc. ${\mbox{\boldmath$1$}}_{k}$ denotes the all-ones vector of length $k$. $\mathring{\Omega}$ denotes the interior of polytope $\Omega$. We are given a canonical primal linear program $\boldsymbol{\mathit{z}}^{*}=\min_{\boldsymbol{\mathit{x}}}\left\\{\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}:A\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}};\boldsymbol{\mathit{x}}\geq 0\right\\}$ which has the same solution as the dual linear program $\boldsymbol{\mathit{z}}^{*}=\max_{(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})}\left\\{\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}:A^{T}\boldsymbol{\mathit{y}}+\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}};\boldsymbol{\mathit{s}}\geq 0\right\\}$ where $A$ is an $n\times m$ matrix, $\boldsymbol{\mathit{x}},\boldsymbol{\mathit{s}},\boldsymbol{\mathit{c}}$ are length $m$, and $\boldsymbol{\mathit{y}},\boldsymbol{\mathit{b}}$ are length $n$, and $m\geq n$. (Unfortunately, this use of $n$ and $m$ is reversed from the standard linear programming convention. We do this to be consistent with the standard graph-theory convention that we use throughout the paper.) We let $\Omega^{D}$ denote the dual polytope $\Omega^{D}=\left\\{(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}}):A^{T}\boldsymbol{\mathit{y}}+\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}};\boldsymbol{\mathit{s}}\geq 0\right\\}$ so we can write the solution to the linear program as $\boldsymbol{\mathit{z}}^{*}=\max_{(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\Omega^{D}}\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}$. In this appendix, we present an $\mathtt{InteriorPoint}$ algorithm based on that of Renegar [Ren88], modified to use an approximate solver. Our analysis follows that found in [Ye97]. ###### Theorem 2.1. $\boldsymbol{\mathit{x}}=\mathtt{InteriorPoint}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\lambda_{min},T,\boldsymbol{\mathit{y}}^{0},\epsilon)$ takes input that satisfy * • $A$ is an $n\times m$ matrix; $\boldsymbol{\mathit{b}}$ is a length $n$ vector; $\boldsymbol{\mathit{c}}$ is a length $m$ vector * • $AA^{T}$ is positive definite, and $\lambda_{min}>0$ is a lower bound on the eigenvalues of $AA^{T}$ * • $T>0$ is an upper bound on the absolute values of the dual coordinates, i.e. $\left\|\boldsymbol{\mathit{y}}\right\|_{\infty}<T$ and $\left\|\boldsymbol{\mathit{s}}\right\|_{\infty}<T$ for all $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})$ that satisfy $\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}\geq 0$ * • initial point $\boldsymbol{\mathit{y}}^{0}$ is a length $n$ vector where $A^{T}\boldsymbol{\mathit{y}}^{0}<\boldsymbol{\mathit{c}}$ * • error parameter $\epsilon$ satisfies $0<\epsilon<1$ and returns $\boldsymbol{\mathit{x}}>0$ satisfying $\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|\leq\epsilon$ and $z^{*}<\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}<z^{*}+\epsilon$. Let us define * • $U$ is the largest absolute value of any entry in $A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}}$ * • $s^{0}_{min}$ is the smallest entry of $\boldsymbol{\mathit{s}}^{0}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{0}$ Then the algorithm makes $\mathcal{O}\left(\sqrt{m}\log\frac{TUm}{\lambda_{min}s^{0}_{min}\epsilon}\right)$ calls to the approximate solver, of the form $\mathtt{Solve}\left(AS^{-2}A^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\cdot,\epsilon^{\prime}\right)$ where $S$ is a positive diagonal matrix with condition number $\mathcal{O}\left(\frac{T^{2}Um^{2}}{\epsilon}\right)$, and $\boldsymbol{\mathit{v}},\epsilon^{\prime}$ satisfy $\log\frac{\left\|\boldsymbol{\mathit{v}}\right\|}{\epsilon^{\prime}}=\mathcal{O}\left(\log\frac{TUm}{s^{0}_{min}\epsilon}\right)$ ### C.1 The Analytic Center Standard interior-point methods focus on a particular point in the interior of the dual polytope. This point, called the analytic center, is the point that maximizes the product of the slacks, i.e. the product of the elements of $\boldsymbol{\mathit{s}}$. For the purpose of our analysis, we use the following equivalent definition of the analytic center: ###### Fact C.1 (see [Ye97, §3.1]). The analytic center of $\Omega^{D}=\left\\{(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}}):A^{T}\boldsymbol{\mathit{y}}+\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}};\boldsymbol{\mathit{s}}\geq 0\right\\}$ is the unique point $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}})\in\mathring{\Omega}^{D}$ that satisfies $\eta_{A}(\accentset{*}{\boldsymbol{\mathit{s}}})=0$, where we define $\displaystyle\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})$ $\displaystyle=S^{-1}(I-S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}){\mbox{\boldmath$1$}}_{m}$ $\displaystyle\eta_{A}(\boldsymbol{\mathit{s}})$ $\displaystyle=\left\|S\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})-{\mbox{\boldmath$1$}}_{m}\right\|=\left\|S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}_{m}\right\|=\left\|AS^{-1}{\mbox{\boldmath$1$}}_{m}\right\|_{(AS^{-2}A^{T})^{-1}}$ These definitions of $\boldsymbol{\mathit{x}}_{A}$ and $\eta_{A}$ satisfying the following properties: ###### Lemma C.2. Let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}})$ be the analytic center of $\Omega^{D}$. For any point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\mathring{\Omega}^{D}$ we have 1. (i) $A\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})=0$ 2. (ii) $\eta_{A}(\boldsymbol{\mathit{s}})<1$ implies $\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})>0$ 3. (iii) $\boldsymbol{\mathit{x}}_{A}(\accentset{*}{\boldsymbol{\mathit{s}}})=\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}$ 4. (iv) For all $\boldsymbol{\mathit{x}}$ satisfying $A\boldsymbol{\mathit{x}}=0$, it holds that $\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|\geq\eta_{A}(\boldsymbol{\mathit{s}})$ The first three properties are straightforward from the definition. We present a proof of the last: ###### Proof of C.2(iv). Note that $S\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})-{\mbox{\boldmath$1$}}_{m}$ is orthogonal to $S(\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}))$, because $\displaystyle\left\langle S\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})-{\mbox{\boldmath$1$}}_{m},S(\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}))\right\rangle$ $\displaystyle=\left\langle S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}_{m},S(\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}))\right\rangle$ $\displaystyle=\left\langle(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}_{m},A(\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}))\right\rangle$ $\displaystyle=\left\langle(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}_{m},0\right\rangle$ $\displaystyle=0$ We thus have $\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|=\left\|S\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})-{\mbox{\boldmath$1$}}_{m}+S(\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}))\right\|\geq\left\|S\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})-{\mbox{\boldmath$1$}}_{m}\right\|=\eta_{A}(\boldsymbol{\mathit{s}})$ ∎ It will be useful to note that the slacks of the analytic center cannot be too small. We can bound the slacks of the analytic center away from zero as follows: ###### Lemma C.3 (compare [Ye97, Thm 2.6]). Let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}})$ be the analytic center of $\Omega^{D}$. For every $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\Omega^{D}$, we have $\accentset{*}{\boldsymbol{\mathit{s}}}>\frac{1}{m}\boldsymbol{\mathit{s}}$ ###### Proof. $\displaystyle\left\|\accentset{*}{S}^{-1}\boldsymbol{\mathit{s}}\right\|_{\infty}\leq{\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}\boldsymbol{\mathit{s}}$ $\displaystyle={\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}\accentset{*}{\boldsymbol{\mathit{s}}}+{\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}(\boldsymbol{\mathit{s}}-\accentset{*}{\boldsymbol{\mathit{s}}})$ $\displaystyle=m+{\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}(\boldsymbol{\mathit{s}}-\accentset{*}{\boldsymbol{\mathit{s}}})$ $\displaystyle=m+{\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}\left((\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}})-(\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}})\right)$ $\displaystyle=m+{\mbox{\boldmath$1$}}_{m}^{T}\accentset{*}{S}^{-1}A^{T}(\accentset{*}{\boldsymbol{\mathit{y}}}-\boldsymbol{\mathit{y}})$ $\displaystyle=m$ where we know from Lemmas C.2(i) and C.2(iii) that $A\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}=0$ ∎ Let us note that a point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\mathring{\Omega}^{D}$ that satisfies $\eta_{A}(\boldsymbol{\mathit{s}})<1$ is close to the analytic center, in the sense that the slacks $\boldsymbol{\mathit{s}}$ are bounded by a constant ratio from the slacks of the analytic center: ###### Lemma C.4 ([Ye97, Thm 3.2(iv)]). Suppose $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\mathring{\Omega}^{D}$ satisfies $\eta_{A}(\boldsymbol{\mathit{s}})=\eta<1$ and let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}})$ be the analytic center of $\Omega^{D}$. Then $\left\|S^{-1}\accentset{*}{\boldsymbol{\mathit{s}}}-{\mbox{\boldmath$1$}}_{m}\right\|\leq\frac{\eta}{1-\eta}$. If $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\mathring{\Omega}^{D}$ is sufficiently close to the analytic center (as measured by $\eta_{A}$), then with a single call to the approximate solver, we can take a Newton-type step to find a point even closer to the analytic center. This $\mathtt{NewtonStep}$ procedure is presented in Figure 2. $\boldsymbol{\mathit{y}}^{+}=\mathtt{NewtonStep}(A,\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}})$ • Let $\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}$ • Let $\boldsymbol{\mathit{d}}_{y}=\mathtt{Solve}(AS^{-2}A^{T},-AS^{-1}{\mbox{\boldmath$1$}}_{m},\epsilon_{3})$ where $\epsilon_{3}=\frac{1}{20(\sqrt{m}+1)}$ • Return $\boldsymbol{\mathit{y}}^{+}=\boldsymbol{\mathit{y}}+(1-\epsilon_{3})\boldsymbol{\mathit{d}}_{y}$ Figure 2: Procedure for stepping closer to the analytic center In the first part of the following lemma, we prove that the point returned by $\mathtt{NewtonStep}$ is indeed still inside the dual polytope. In the second part, we show how close the new point is to the analytic center: ###### Lemma C.5 (compare [Ye97, Thm 3.3]). Suppose $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})\in\mathring{\Omega}^{D}$ satisfies $\eta_{A}(\boldsymbol{\mathit{s}})=\eta<1$. Let $\boldsymbol{\mathit{y}}^{+}=\mathtt{NewtonStep}(A,\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}})$ and $\boldsymbol{\mathit{s}}^{+}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{+}$ Then (i) $\boldsymbol{\mathit{s}}^{+}>0$ and (ii) $\eta_{A}(\boldsymbol{\mathit{s}}^{+})\leq\eta^{2}+\frac{1}{20}\eta$ ###### Proof. (i) The solver guarantees that $\left\|\boldsymbol{\mathit{d}}_{y}+(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}\right\|_{AS^{-2}A^{T}}\leq\epsilon_{3}\left\|(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}\right\|_{AS^{-2}A^{T}}=\epsilon_{3}\cdot\eta$ or equivalently $\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}+S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}\right\|\leq\epsilon_{3}\left\|S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}\right\|=\epsilon_{3}\cdot\eta$ (7) and so $\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}\right\|\leq(1+\epsilon_{3})\left\|S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}\right\|=(1+\epsilon_{3})\eta<1+\epsilon_{3}$ We thus have $\displaystyle\left\|S^{-1}\boldsymbol{\mathit{s}}^{+}-{\mbox{\boldmath$1$}}\right\|$ $\displaystyle=\left\|S^{-1}\left(\boldsymbol{\mathit{s}}-(1-\epsilon_{3})A^{T}\boldsymbol{\mathit{d}}_{y}\right)-{\mbox{\boldmath$1$}}\right\|$ $\displaystyle=(1-\epsilon_{3})\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}\right\|$ $\displaystyle\leq(1-\epsilon_{3})(1+\epsilon_{3})<1$ Thus $S^{-1}\boldsymbol{\mathit{s}}^{+}$ is positive and so is $\boldsymbol{\mathit{s}}^{+}$. (ii) Let $\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}})$ and $\boldsymbol{\mathit{x}}^{+}=\boldsymbol{\mathit{x}}_{A}(\boldsymbol{\mathit{s}}^{+})$. We have $\displaystyle\eta_{A}(\boldsymbol{\mathit{s}}^{+})$ $\displaystyle\leq\left\|X\boldsymbol{\mathit{s}}^{+}-{\mbox{\boldmath$1$}}_{m}\right\|\qquad\text{(by Lemma \ref{lem:etamin})}$ $\displaystyle=\left\|X(\boldsymbol{\mathit{s}}-(1-\epsilon_{3})A^{T}\boldsymbol{\mathit{d}}_{y})-{\mbox{\boldmath$1$}}_{m}\right\|$ $\displaystyle=\left\|(1-\epsilon_{3})XS(X\boldsymbol{\mathit{s}}-{\mbox{\boldmath$1$}}_{m}-S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y})-(1-\epsilon_{3})(XS-I)(X\boldsymbol{\mathit{s}}-{\mbox{\boldmath$1$}}_{m})+\epsilon_{3}(X\boldsymbol{\mathit{s}}-{\mbox{\boldmath$1$}}_{m})\right\|$ $\displaystyle\leq(1-\epsilon_{3})\left\|XS(S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}-S\boldsymbol{\mathit{x}}+{\mbox{\boldmath$1$}}_{m})\right\|+(1-\epsilon_{3})\left\|(XS-I)(X\boldsymbol{\mathit{s}}-{\mbox{\boldmath$1$}}_{m})\right\|+\epsilon_{3}\left\|(X\boldsymbol{\mathit{s}}-{\mbox{\boldmath$1$}}_{m})\right\|$ $\displaystyle\leq(1-\epsilon_{3})\left\|S\boldsymbol{\mathit{x}}\right\|\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}-S\boldsymbol{\mathit{x}}+{\mbox{\boldmath$1$}}_{m}\right\|+(1-\epsilon_{3})\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|^{2}+\epsilon_{3}\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|$ (using the relation $\left\|V\boldsymbol{\mathit{w}}\right\|\leq\left\|\boldsymbol{\mathit{v}}\right\|_{\infty}\left\|\boldsymbol{\mathit{w}}\right\|\leq\left\|\boldsymbol{\mathit{v}}\right\|\left\|\boldsymbol{\mathit{w}}\right\|$) $\displaystyle\leq(1-\epsilon_{3})(\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|+\left\|{\mbox{\boldmath$1$}}_{m}\right\|)\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}-S\boldsymbol{\mathit{x}}+{\mbox{\boldmath$1$}}_{m}\right\|+(1-\epsilon_{3})\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|^{2}+\epsilon_{3}\left\|S\boldsymbol{\mathit{x}}-{\mbox{\boldmath$1$}}_{m}\right\|$ $\displaystyle=(1-\epsilon_{3})(\eta+\sqrt{m})\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}-S\boldsymbol{\mathit{x}}+{\mbox{\boldmath$1$}}_{m}\right\|+(1-\epsilon_{3})\eta^{2}+\epsilon_{3}\eta$ $\displaystyle=(1-\epsilon_{3})(\eta+\sqrt{m})\left\|S^{-1}A^{T}\boldsymbol{\mathit{d}}_{y}+S^{-1}A^{T}(AS^{-2}A^{T})^{-1}AS^{-1}{\mbox{\boldmath$1$}}_{m}\right\|+(1-\epsilon_{3})\eta^{2}+\epsilon_{3}\eta$ $\displaystyle\leq(1-\epsilon_{3})(\eta+\sqrt{m})\epsilon_{3}\eta+(1-\epsilon_{3})\eta^{2}+\epsilon_{3}\eta\qquad\text{(by equation \ref{eq:newtonsolve})}$ $\displaystyle\leq\epsilon_{3}(\eta+\sqrt{m})\eta+(1-\epsilon_{3})\eta^{2}+\epsilon_{3}\eta$ $\displaystyle=\eta^{2}+\epsilon_{3}(\sqrt{m}+1)\eta$ $\displaystyle=\eta^{2}+\frac{1}{20}\eta$ ∎ ### C.2 The Path-Following Algorithm In a path-following algorithm, we modify the dual polytope $\Omega^{D}=\left\\{(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}}):A^{T}\boldsymbol{\mathit{y}}+\boldsymbol{\mathit{s}}=\boldsymbol{\mathit{c}};\boldsymbol{\mathit{s}}\geq 0\right\\}$ by adding an additional contraint $\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}\geq z$, where $z\leq z^{*}$. As we let $z$ approach $z^{*}$, the center of the polytope approaches the solution to the dual linear program. Letting $s_{gap}=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z$ denote the new slack variable, we define the modified polytope: $\Omega^{D}_{\boldsymbol{\mathit{b}},z}=\left\\{\left(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap}\right):\left[\begin{matrix}A^{T}\boldsymbol{\mathit{y}}+\boldsymbol{\mathit{s}}\\\ -\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}+s_{gap}\end{matrix}\right]=\left[\begin{matrix}\boldsymbol{\mathit{c}}\\\ -z\end{matrix}\right];\boldsymbol{\mathit{s}},s_{gap}\geq 0\right\\}$ Using a trick of Renegar, when we define the analytic center of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$, we consider there to be $m$ copies of the slack $s_{gap}$, as follows: ###### Definition C.6. The analytic center of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$ is the point $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$, that satisfies $\tilde{\eta}(\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})=0$, where we define $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})=\eta_{\tilde{A}}(\tilde{\boldsymbol{\mathit{s}}})\qquad\text{where $\tilde{A}=\left[\begin{matrix}A&-\boldsymbol{\mathit{b}}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]$ and $\tilde{\boldsymbol{\mathit{s}}}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ s_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$}$ The central path is the set of analytic centers of the polytopes $\left\\{\Omega^{D}_{\boldsymbol{\mathit{b}},z}\right\\}_{z\leq z^{*}}$ A path-following algorithm steps through a sequence of points near the central path, as $z$ increases towards $z^{*}$. It is useful to note that given any point on the central path, we may easily construct a feasible primal solution $\boldsymbol{\mathit{x}}$, as follows: ###### Lemma C.7. Let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})$ be the analytic center of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$. Then the vector $\boldsymbol{\mathit{x}}=\frac{\accentset{*}{s}_{gap}}{m}\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}$ satisfies $A\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}$. More generally, for any $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$, the vector $\boldsymbol{\mathit{x}}=\frac{s_{gap}}{m}S^{-1}{\mbox{\boldmath$1$}}_{m}$ satisfies $\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}=\frac{s_{gap}}{m}\cdot\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})$ ###### Proof. We prove the second assertion: $\displaystyle\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}$ $\displaystyle=\frac{s_{gap}}{m}\left\|AS^{-1}{\mbox{\boldmath$1$}}_{m}-ms_{gap}^{-1}\boldsymbol{\mathit{b}}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}$ $\displaystyle=\frac{s_{gap}}{m}\left\|\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}$ $\displaystyle=\frac{s_{gap}}{m}\cdot\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})$ The first assertion now follows from the definition of analytic center. ∎ Let us now describe how to take steps along the central path using our approximate solver. In Figure 3, we present the procedure $\mathtt{Shift}$, which takes as input a value $z<z^{*}$ and a point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ satisfying $\eta(\boldsymbol{\mathit{s}},s_{gap})\leq\frac{1}{10}$. The output is a new value $z^{+}$ that is closer to $z^{*}$, and a new point $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z^{+}}$ satisfying $\eta(\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\leq\frac{1}{10}$. The procedure requires a single call to the solver. $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},s_{gap}^{+},z^{+})=\mathtt{Shift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z)$ • Let $z^{+}=z+\frac{s_{gap}}{10\sqrt{m}}$ • Let $\boldsymbol{\mathit{y}}^{+}=\mathtt{NewtonStep}(\tilde{A},\tilde{\boldsymbol{\mathit{c}}},\boldsymbol{\mathit{y}})$ where $\tilde{A}=\left[\begin{matrix}A&-\boldsymbol{\mathit{b}}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]$ and $\tilde{\boldsymbol{\mathit{c}}}=\left[\begin{matrix}\boldsymbol{\mathit{c}}\\\ -z^{+}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ • Let $\left[\begin{matrix}\boldsymbol{\mathit{s}}^{+}\\\ s_{gap}^{+}\end{matrix}\right]=\left[\begin{matrix}\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{+}\\\ \boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{+}-z^{+}\end{matrix}\right]$ Figure 3: Procedure for taking a step along the central path Let us examine this procedure more closely. After defining the incremented value $z^{+}$, if we let $s_{gap}^{\prime}=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z^{+}=s_{gap}-(z^{+}-z)$, then $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap}^{\prime})$ is a point in the shifted polytope $\Omega^{D}_{\boldsymbol{\mathit{b}},z^{+}}$. However this point may be slightly farther away from the central path. One call to the $\mathtt{NewtonStep}$ procedure suffices to obtain a new point $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z^{+}}$ that is sufficiently close to the central path, satisfying $\tilde{\eta}(\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\leq\frac{1}{10}$. We prove this formally: ###### Lemma C.8 (compare [Ye97, Lem 4.5]). Given $z<z^{*}$ and $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ where $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq\frac{1}{10}$, let $s_{gap}^{\prime}=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z^{+}$ and $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},s_{gap}^{+},z^{+})=\mathtt{Shift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z)$. Then 1. (i) $z^{+}<z^{*}$ 2. (ii) $s_{gap}^{\prime}>0$ and $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap}^{\prime})<\frac{21}{100}$ 3. (iii) $\boldsymbol{\mathit{s}}^{+},s_{gap}^{+}>0$ and $\tilde{\eta}(\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})<\frac{1}{10}$ ###### Proof. (i) $z^{+}=z+\frac{\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z}{10\sqrt{m}}<z+(\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z)=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}<z^{*}$ (ii) We note that $s^{\prime}_{gap}=s_{gap}-(z^{+}-z)=\left(1-\frac{1}{10\sqrt{m}}\right)s_{gap}>0$. Let us write $\tilde{\boldsymbol{\mathit{s}}}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ s_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ and $\tilde{\boldsymbol{\mathit{s}}}^{\prime}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ s_{gap}^{\prime}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ and note that $\tilde{\boldsymbol{\mathit{s}}}-\tilde{\boldsymbol{\mathit{s}}}^{\prime}=\left[\begin{matrix}0\\\ (s_{gap}-s^{\prime}_{gap}){\mbox{\boldmath$1$}}_{m}\end{matrix}\right]=\left[\begin{matrix}0\\\ (z^{+}-z){\mbox{\boldmath$1$}}_{m}\end{matrix}\right]=\left[\begin{matrix}0\\\ \frac{s_{gap}}{10\sqrt{m}}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ (8) Let us define $\tilde{\boldsymbol{\mathit{x}}}=\left[\begin{matrix}\boldsymbol{\mathit{x}}\\\ \boldsymbol{\mathit{x}}_{gap}\end{matrix}\right]=\boldsymbol{\mathit{x}}_{\tilde{A}}(\tilde{\boldsymbol{\mathit{s}}})$. So we have $\displaystyle\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap}^{\prime})=\eta_{\tilde{A}}(\tilde{\boldsymbol{\mathit{s}}}^{\prime})$ $\displaystyle\leq\left\|\tilde{S}^{\prime}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|\qquad\text{(by Lemma \ref{lem:etamin})}$ $\displaystyle\leq\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\left\|(\tilde{S}^{\prime}-\tilde{S})\tilde{\boldsymbol{\mathit{x}}}\right\|$ $\displaystyle=\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10\sqrt{m}}\left\|s_{gap}\boldsymbol{\mathit{x}}_{gap}\right\|\qquad\text{(by Equation \ref{eq:sdiff})}$ $\displaystyle\leq\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10\sqrt{m}}\left(\left\|s_{gap}\boldsymbol{\mathit{x}}_{gap}-{\mbox{\boldmath$1$}}_{m}\right\|+\left\|{\mbox{\boldmath$1$}}_{m}\right\|\right)$ $\displaystyle=\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10\sqrt{m}}\left\|s_{gap}\boldsymbol{\mathit{x}}_{gap}-{\mbox{\boldmath$1$}}_{m}\right\|+\frac{1}{10}$ $\displaystyle\leq\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10}\left\|s_{gap}\boldsymbol{\mathit{x}}_{gap}-{\mbox{\boldmath$1$}}_{m}\right\|+\frac{1}{10}$ $\displaystyle\leq\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10}\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}-{\mbox{\boldmath$1$}}_{2m}\right\|+\frac{1}{10}$ $\displaystyle=\frac{11}{10}\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})+\frac{1}{10}$ (9) $\displaystyle\leq\frac{11}{10}\cdot\frac{1}{10}+\frac{1}{10}=\frac{21}{100}$ (iii) By Lemma C.5, we have $\boldsymbol{\mathit{s}}^{+},s_{gap}^{+}>0$ and $\tilde{\eta}(\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\leq\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap}^{\prime})^{2}+\frac{1}{20}\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap}^{\prime})\leq\left(\frac{21}{100}\right)^{2}+\frac{1}{20}\cdot\frac{21}{100}<\frac{1}{10}$ ∎ $\boldsymbol{\mathit{x}}=\mathtt{InteriorPoint}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0},\epsilon)$ • Compute $(\boldsymbol{\mathit{y}}^{C},z^{C})=\mathtt{FindCentralPath}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0})$ and $\left[\begin{matrix}\boldsymbol{\mathit{s}}^{C}\\\ s_{gap}^{C}\end{matrix}\right]=\left[\begin{matrix}\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{C}\\\ \boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{C}-z^{C}\end{matrix}\right]$ • Set $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z):=(\boldsymbol{\mathit{y}}^{C},\boldsymbol{\mathit{s}}^{C},s_{gap}^{C},z^{C})$ • While $s_{gap}>\frac{\epsilon}{3}$: – Set $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z):=\mathtt{Shift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z)$ • Compute $\boldsymbol{\mathit{v}}=\mathtt{Solve}(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T},\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m},\epsilon_{4})$ $\displaystyle\text{where }\tilde{A}$ $\displaystyle=\left[\begin{matrix}A&-\boldsymbol{\mathit{b}}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]$ $\displaystyle\text{and }\epsilon_{4}=\min\left(1,\frac{s_{min}}{TU}\cdot\frac{m^{1/2}}{n}\right)$ $\displaystyle\text{and }\tilde{\boldsymbol{\mathit{s}}}$ $\displaystyle=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ s_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ and $s_{min}$ is the smallest entry of $\tilde{\boldsymbol{\mathit{s}}}$ • Return $\boldsymbol{\mathit{x}}=\frac{\boldsymbol{\mathit{x}}^{\prime}}{mx_{gap}^{\prime}}$ where $\left[\begin{matrix}\boldsymbol{\mathit{x}}^{\prime}\\\ x^{\prime}_{gap}\end{matrix}\right]=\left[\begin{matrix}S^{-1}{\mbox{\boldmath$1$}}_{m}-S^{-2}A^{T}\boldsymbol{\mathit{v}}\\\ s_{gap}^{-1}+s_{gap}^{-2}\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{v}}\end{matrix}\right]$ Figure 4: Dual path-following interior-point algorithm using an approximate solver We now present the complete path-following $\mathtt{InteriorPoint}$ algorithm, implemented using an approximate solver, in Figure 4. For now we postpone describing the $\mathtt{FindCentralPath}$ subroutine, which gives an initial point near the central path. In particular, it produces a $z^{C}<z^{*}$ and $(\boldsymbol{\mathit{y}}^{C},\boldsymbol{\mathit{s}}^{C},s_{gap}^{C})\in\Omega^{D}_{\boldsymbol{\mathit{b}},z^{C}}$ satisfying $\tilde{\eta}(\boldsymbol{\mathit{s}}^{C},s_{gap}^{C})\leq\frac{1}{10}$. Once we have this initial central path point, Lemma C.8 tells us that after each call to $\mathtt{Shift}$ we have a new value $z<z^{*}$ and new central path point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ that satisfies $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq\frac{1}{10}$. Later we will analyze the number of calls to $\mathtt{Shift}$ before the algortihm terminates. First, let us confirm that the algorithm returns the correct output: ###### Lemma C.9. The output of $\boldsymbol{\mathit{x}}=\mathtt{InteriorPoint}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0},\epsilon)$ satisfies * (i) $\boldsymbol{\mathit{x}}>0$ * (ii) $\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|\leq\frac{\epsilon}{12\sqrt{2}\cdot Tn^{1/2}}$ * (iii) $\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}<z^{*}+\epsilon$ ###### Proof. (i) To assist in our proof, let us define $\tilde{\boldsymbol{\mathit{x}}}^{\prime}=\left[\begin{matrix}\boldsymbol{\mathit{x}}^{\prime}\\\ x^{\prime}_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ and note that $\tilde{\boldsymbol{\mathit{x}}}^{\prime}=\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}-\tilde{S}^{-2}\tilde{A}^{T}\boldsymbol{\mathit{v}}$. We have $\displaystyle\left\|\tilde{S}\tilde{\boldsymbol{\mathit{x}}}^{\prime}-{\mbox{\boldmath$1$}}_{2m}\right\|$ $\displaystyle=\left\|\tilde{S}^{-1}\tilde{A}^{T}\boldsymbol{\mathit{v}}\right\|$ $\displaystyle=\left\|\boldsymbol{\mathit{v}}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}$ $\displaystyle\leq\left\|(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}+\left\|\boldsymbol{\mathit{v}}-(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}$ $\displaystyle\leq(1+\epsilon_{4})\left\|(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}\qquad\text{(by guarantee of {Solve})}$ $\displaystyle=(1+\epsilon_{4})\cdot\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq 2\cdot\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq 2\cdot\frac{1}{10}\leq\frac{1}{5}$ (10) Since $\tilde{\boldsymbol{\mathit{s}}}$ is positive, we conclude that $\tilde{\boldsymbol{\mathit{x}}}^{\prime}$ must also be positive, and so must be $\boldsymbol{\mathit{x}}$. (ii) We have $\displaystyle\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|$ $\displaystyle=\frac{1}{mx_{gap}^{\prime}}\left\|A\boldsymbol{\mathit{x}}^{\prime}-mx_{gap}^{\prime}\boldsymbol{\mathit{b}}\right\|$ $\displaystyle=\frac{1}{mx_{gap}^{\prime}}\left\|\tilde{A}\tilde{\boldsymbol{\mathit{x}}}^{\prime}\right\|$ $\displaystyle=\frac{1}{mx_{gap}^{\prime}}\left\|\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}-\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}\boldsymbol{\mathit{v}}\right\|$ Observe that the largest eigenvalue of the matrix $\tilde{A}\tilde{A}^{T}=AA^{T}+m\boldsymbol{\mathit{b}}\boldsymbol{\mathit{b}}^{T}$ is less than the trace, which is at most $2nmU^{2}$. Thus, the largest eigenvalue of $\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}$ is at most $2nmU^{2}s_{min}^{-2}$. So we proceed $\displaystyle\leq\frac{(2nmU^{2}s_{min}^{-2})^{1/2}}{mx_{gap}^{\prime}}\left\|\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}-\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}\boldsymbol{\mathit{v}}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}$ $\displaystyle=\frac{(2n)^{1/2}U}{m^{1/2}x_{gap}^{\prime}s_{min}}\left\|(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}-\boldsymbol{\mathit{v}}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}$ $\displaystyle\leq\frac{(2n)^{1/2}U}{m^{1/2}x_{gap}^{\prime}s_{min}}\cdot\epsilon_{4}\left\|(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}}\qquad\text{(by guarantee of {Solve})}$ $\displaystyle=\frac{(2n)^{1/2}U}{m^{1/2}x_{gap}^{\prime}s_{min}}\cdot\epsilon_{4}\cdot\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})$ $\displaystyle\leq\frac{(2n)^{1/2}U}{m^{1/2}x_{gap}^{\prime}s_{min}}\cdot\epsilon_{4}\cdot\frac{1}{10}$ $\displaystyle\leq\frac{(2n)^{1/2}U}{m^{1/2}x_{gap}^{\prime}s_{min}}\cdot\frac{m^{1/2}s_{min}}{nTU}\cdot\frac{1}{10}$ $\displaystyle=\frac{1}{5\sqrt{2}\cdot Tn^{1/2}}\cdot\frac{1}{x_{gap}^{\prime}}$ $\displaystyle\leq\frac{1}{5\sqrt{2}\cdot Tn^{1/2}}\cdot\frac{5}{4}\cdot s_{gap}\qquad\text{(from equation \ref{eq:sxprime}, we know $s_{gap}x_{gap}^{\prime}\geq\frac{4}{5}$)}$ $\displaystyle\leq\frac{1}{5\sqrt{2}\cdot Tn^{1/2}}\cdot\frac{5}{4}\cdot\frac{\epsilon}{3}$ $\displaystyle\leq\frac{\epsilon}{12\sqrt{2}\cdot Tn^{1/2}}$ (11) (iii) We have $\displaystyle\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}$ $\displaystyle=\frac{\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}^{\prime}}{mx^{\prime}_{gap}}$ $\displaystyle\geq\frac{5\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}^{\prime}}{6m}\cdot s_{gap}\qquad\text{(from equation \ref{eq:sxprime}, we know $s_{gap}x_{gap}^{\prime}\leq\frac{6}{5}$)}$ $\displaystyle=\frac{5\left\|S\boldsymbol{\mathit{x}}^{\prime}\right\|_{1}}{6m}\cdot s_{gap}$ $\displaystyle\geq\frac{5}{6m}\left(\left\|{\mbox{\boldmath$1$}}_{m}\right\|_{1}-\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|_{1}\right)\cdot s_{gap}$ $\displaystyle=\frac{5}{6m}\left(m-\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|_{1}\right)\cdot s_{gap}$ $\displaystyle\geq\frac{5}{6m}\left(m-\sqrt{m}\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|\right)\cdot s_{gap}$ $\displaystyle\geq\frac{5}{6m}\left(m-\frac{1}{5}\sqrt{m}\right)\cdot s_{gap}\qquad\text{(by equation \ref{eq:sxprime})}$ $\displaystyle\geq\frac{5}{6m}\cdot\frac{4}{5}\cdot m\cdot s_{gap}$ $\displaystyle=\frac{2}{3}\cdot s_{gap}$ (12) $\displaystyle\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}$ $\displaystyle=\frac{\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}^{\prime}}{mx^{\prime}_{gap}}$ $\displaystyle\leq\frac{5\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}^{\prime}}{4m}\cdot s_{gap}\qquad\text{(from equation \ref{eq:sxprime}, we know $s_{gap}x_{gap}^{\prime}\geq\frac{4}{5}$)}$ $\displaystyle=\frac{5\left\|S\boldsymbol{\mathit{x}}^{\prime}\right\|_{1}}{4m}\cdot s_{gap}$ $\displaystyle\leq\frac{5}{4m}\left(\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|_{1}+\left\|{\mbox{\boldmath$1$}}_{m}\right\|_{1}\right)\cdot s_{gap}$ $\displaystyle=\frac{5}{4m}\left(\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|_{1}+m\right)\cdot s_{gap}$ $\displaystyle\leq\frac{5}{4m}\left(\sqrt{m}\left\|S\boldsymbol{\mathit{x}}^{\prime}-{\mbox{\boldmath$1$}}_{m}\right\|+m\right)\cdot s_{gap}$ $\displaystyle\leq\frac{5}{4m}\left(\frac{1}{5}\sqrt{m}+m\right)\cdot s_{gap}\qquad\text{(by equation \ref{eq:sxprime})}$ $\displaystyle\leq\frac{5}{4m}\cdot\frac{6}{5}\cdot m\cdot s_{gap}$ $\displaystyle=\frac{3}{2}\cdot s_{gap}$ (13) We then have $\displaystyle\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}-z^{*}$ $\displaystyle>\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}$ $\displaystyle=(\boldsymbol{\mathit{c}}^{T}-\boldsymbol{\mathit{y}}^{T}A)\boldsymbol{\mathit{x}}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})$ $\displaystyle=\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})$ $\displaystyle\geq\frac{2}{3}\cdot s_{gap}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})\qquad\text{(by Equation \ref{eq:sxupperbound})}$ $\displaystyle\leq\dots\frac{2}{3}\cdot\epsilon+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})$ $\displaystyle\leq\frac{2}{3}\cdot\epsilon+\left\|\boldsymbol{\mathit{y}}\right\|\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|$ $\displaystyle\leq\frac{2}{3}\cdot\epsilon+Tn^{1/2}\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|$ $\displaystyle\leq\frac{2}{3}\cdot\epsilon+\frac{1}{12\sqrt{2}}\cdot\epsilon\qquad\text{(by Lemma \ref{lem:intptoutput}(ii))}$ $\displaystyle<\epsilon$ $\displaystyle\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}-z^{*}$ $\displaystyle<\boldsymbol{\mathit{c}}^{T}\boldsymbol{\mathit{x}}-z$ $\displaystyle=(\boldsymbol{\mathit{c}}^{T}-\boldsymbol{\mathit{y}}^{T}A)\boldsymbol{\mathit{x}}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})+\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z$ $\displaystyle=\boldsymbol{\mathit{s}}^{T}\boldsymbol{\mathit{x}}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})+s_{gap}$ $\displaystyle\leq\frac{5}{2}\cdot s_{gap}+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})\qquad\text{(by Equation \ref{eq:sxupperbound})}$ $\displaystyle\leq\frac{5}{6}\cdot\epsilon+\boldsymbol{\mathit{y}}^{T}(A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}})$ $\displaystyle\leq\frac{5}{6}\cdot\epsilon+\left\|\boldsymbol{\mathit{y}}\right\|\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|$ $\displaystyle\leq\frac{5}{6}\cdot\epsilon+Tn^{1/2}\left\|A\boldsymbol{\mathit{x}}-\boldsymbol{\mathit{b}}\right\|$ $\displaystyle\leq\frac{5}{6}\cdot\epsilon+\frac{1}{12\sqrt{2}}\cdot\epsilon\qquad\text{(by Lemma \ref{lem:intptoutput}(ii))}$ $\displaystyle<\epsilon$ ∎ Next, we analyze the number of $\mathtt{Shift}$ iterations until the algorithm terminates. We can measure the progress of the algorithm with the potential function $B(z)$: $B(z)=\sum_{j=1}^{m}\log\accentset{*}{\boldsymbol{\mathit{s}}}_{j}+m\log\accentset{*}{s}_{gap}\qquad\text{where $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})$ is the analytic center of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$}$ Soon, we will show how a decrease in $B(z)$ implies that $s_{gap}$ is decreasing and thus the algorithm is making progress. Let us first show that the value of $B(z)$ decreases by $\Omega(\sqrt{m})$ after each iteration. ###### Lemma C.10 (compare [Ye97, Lem 4.6]). Given $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ satisfiying $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})<\frac{1}{10}$, let $(\boldsymbol{\mathit{y}}^{+},z^{+})=\mathtt{Shift}(\boldsymbol{\mathit{y}},z)$. Then $B(z^{+})\leq B(z)-\Theta(\sqrt{m})$. ###### Proof. Let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})$ and $(\accentset{*}{\boldsymbol{\mathit{y}}}^{+},\accentset{*}{\boldsymbol{\mathit{s}}}^{+},\accentset{*}{s}_{gap}^{+})$ respectively be the analytic centers of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$ and $\Omega^{D}_{\boldsymbol{\mathit{b}},z^{+}}$. Following Lemma C.7, we define $\boldsymbol{\mathit{x}}=\frac{\accentset{*}{s}_{gap}}{m}\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}$ that satisfies $A\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}$. We have $\displaystyle e^{\frac{B(z^{+})-B(z)}{2m}}$ $\displaystyle=\sqrt{\left(\prod_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}\right)^{\frac{1}{m}}\cdot\frac{\accentset{*}{s}_{gap}^{+}}{\accentset{*}{s}_{gap}}}$ $\displaystyle\leq\frac{1}{2m}\sum_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}+\frac{1}{2}\frac{\accentset{*}{s}_{gap}^{+}}{\accentset{*}{s}_{gap}}$ $\displaystyle\leq 1+\frac{1}{2m}\sum_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}-\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}+\frac{1}{2}\frac{\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}}$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\cdot\frac{\accentset{*}{s}_{gap}}{m}(\accentset{*}{\boldsymbol{\mathit{s}}}^{+}-\accentset{*}{\boldsymbol{\mathit{s}}})\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}+\frac{1}{2\accentset{*}{s}_{gap}}(\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap})$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left((\accentset{*}{\boldsymbol{\mathit{s}}}^{+}-\accentset{*}{\boldsymbol{\mathit{s}}})^{T}\boldsymbol{\mathit{x}}+(\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap})\right)$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left(\left((\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}}^{+})-(\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}})\right)^{T}\boldsymbol{\mathit{x}}+(\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap})\right)$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left((\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{\boldsymbol{\mathit{y}}}^{+})^{T}A\boldsymbol{\mathit{x}}+(\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap})\right)$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left((\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{\boldsymbol{\mathit{y}}}^{+})^{T}\boldsymbol{\mathit{b}}+(\accentset{*}{s}_{gap}^{+}-\accentset{*}{s}_{gap})\right)\qquad\text{(by Lemma \ref{lem:constructx})}$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left((\boldsymbol{\mathit{b}}^{T}\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{s}_{gap})-(\boldsymbol{\mathit{b}}^{T}\accentset{*}{\boldsymbol{\mathit{y}}}^{+}-\accentset{*}{s}_{gap}^{+})\right)$ $\displaystyle=1+\frac{1}{2\accentset{*}{s}_{gap}}\left(z-z^{+}\right)$ (14) $\displaystyle=1-\frac{s_{gap}}{20\sqrt{m}\cdot\accentset{*}{s}_{gap}}$ $\displaystyle\leq 1-\frac{9}{200\sqrt{m}}\qquad\text{($\frac{\accentset{*}{s}_{gap}}{s_{gap}}\leq\frac{10}{9}$ by Lemma \ref{lem:nearcenter})}$ $\displaystyle\leq e^{-\frac{9}{200\sqrt{m}}}$ We conclude that $B(z^{+})-B(z)\leq-\frac{9}{100}\sqrt{m}$. ∎ Let us now show that a decrease in the potential function $B(z)$ implies a decrease in the value of $s_{gap}$: ###### Lemma C.11 (compare [Ye97, Prop 4.2]). Given $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ and $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z^{+}}$ where $z^{+}>z$ and $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq\eta$ and $\tilde{\eta}(\boldsymbol{\mathit{s}}^{+},s_{gap}^{+})\leq\eta$ for $\eta<1$. Then $\frac{s_{gap}}{s_{gap}^{+}}\leq(1-2\eta)\cdot\left(e^{\frac{B(z_{1})-B(z_{2})}{m}}-1\right)$ ###### Proof. Let $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{s}_{gap})$ and $(\accentset{*}{\boldsymbol{\mathit{y}}}^{+},\accentset{*}{\boldsymbol{\mathit{s}}}^{+},\accentset{*}{s}_{gap}^{+})$ respectively be the analytic centers of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$ and $\Omega^{D}_{\boldsymbol{\mathit{b}},z^{+}}$. We define $\boldsymbol{\mathit{x}}=\frac{\accentset{*}{s}_{gap}}{m}\accentset{*}{S}^{-1}{\mbox{\boldmath$1$}}_{m}$ and $\boldsymbol{\mathit{x}}^{+}=\frac{\accentset{*}{s}_{gap}^{+}}{m}(\accentset{*}{S}^{+})^{-1}{\mbox{\boldmath$1$}}_{m}$, which by Lemma C.7 satisfy $A\boldsymbol{\mathit{x}}=\boldsymbol{\mathit{b}}=A\boldsymbol{\mathit{x}}^{+}$. We have $\displaystyle e^{\frac{B(z)-B(z^{+})}{m}}$ $\displaystyle=\left(\prod_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}\right)^{\frac{1}{m}}\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle\leq\frac{1}{m}\left(\sum_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{m}\sum_{j=1}^{m}\frac{\accentset{*}{\boldsymbol{\mathit{s}}}_{j}-\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}{\accentset{*}{\boldsymbol{\mathit{s}}}^{+}_{j}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}(\accentset{*}{\boldsymbol{\mathit{s}}}-\accentset{*}{\boldsymbol{\mathit{s}}}^{+})^{T}\boldsymbol{\mathit{x}}^{+}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}\left((\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}})-(\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}}^{+})\right)^{T}\boldsymbol{\mathit{x}}^{+}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}(\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{\boldsymbol{\mathit{y}}}^{+})^{T}A\boldsymbol{\mathit{x}}^{+}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}(\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{\boldsymbol{\mathit{y}}}^{+})^{T}\boldsymbol{\mathit{b}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}(\accentset{*}{\boldsymbol{\mathit{y}}}-\accentset{*}{\boldsymbol{\mathit{y}}}^{+})^{T}A\boldsymbol{\mathit{x}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}\left((\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}})-(\boldsymbol{\mathit{c}}-A^{T}\accentset{*}{\boldsymbol{\mathit{y}}}^{+})\right)^{T}\boldsymbol{\mathit{x}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}(\accentset{*}{\boldsymbol{\mathit{s}}}-\accentset{*}{\boldsymbol{\mathit{s}}}^{+})^{T}\boldsymbol{\mathit{x}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle\leq\left(1+\frac{1}{\accentset{*}{s}_{gap}^{+}}\accentset{*}{\boldsymbol{\mathit{s}}}^{T}\boldsymbol{\mathit{x}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle=\left(1+\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}\right)\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle\leq\left(1+\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}\right)^{2}$ So $\displaystyle\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}$ $\displaystyle\geq e^{\frac{B(z)-B(z^{+})}{2m}}-1$ Using Lemma C.4, we may conclude $\frac{s_{gap}}{s_{gap}^{+}}=\frac{s_{gap}}{\accentset{*}{s}_{gap}}\cdot\frac{\accentset{*}{s}_{gap}}{\accentset{*}{s}_{gap}^{+}}\cdot\frac{\accentset{*}{s}_{gap}^{+}}{s_{gap}^{+}}\geq\frac{1-\frac{\eta}{1-\eta}}{1+\frac{\eta}{1-\eta}}\cdot\left(e^{\frac{B(z)-B(z^{+})}{2m}}-1\right)$ ∎ ###### Corollary C.12. The $\mathtt{InteriorPoint}$ algorithm makes $\mathcal{O}\left(\sqrt{m}\log\frac{s_{gap}^{C}}{\epsilon}\right)$ calls to $\mathtt{Shift}$. ###### Proof. Recall that the algorithm will terminate only when the value of $s_{gap}$ has decreased from its initial value of $s_{gap}^{C}$ to below $\frac{\epsilon}{3}$. Thus, Lemma C.11 ensures us that $s_{gap}$ will be smaller than $\frac{\epsilon}{3}$ once $B(z)$ has decreased by $\Omega\left(m\log\frac{s_{gap}^{C}}{\epsilon}\right)$. According to Lemma C.10, this occurs after $\mathcal{O}\left(\sqrt{m}\log\frac{s_{gap}^{C}}{\epsilon}\right)$ $\mathtt{Shift}$ iterations. ∎ ### C.3 Finding the Central Path It remains for us to describe how to initialize the path-following algorithm by finding a point near the central path. Essentially, this is accomplished by running the path-following algorithm in reverse. Instead of stepping towards the optimum given by $\boldsymbol{\mathit{b}}$, we step away from the optimum given by the vector $\underaccent{\bar}{\bb}=A(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}$ that depends on our initial feasible point $(\boldsymbol{\mathit{y}}^{0},\boldsymbol{\mathit{s}}^{0})\in\mathring{\Omega}^{D}$. Our analysis parallels that in the previous section. The following function $\underaccent{\bar}{\eta}$ measures the proximity of a point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})\in\mathring{\Omega}^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}}$ to the central path given by $\underaccent{\bar}{\bb}$: $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})=\eta_{\tilde{\underaccent{\bar}{A}}}(\tilde{\underaccent{\bar}{\ss}})\qquad\text{where $\tilde{\underaccent{\bar}{A}}=\left[\begin{matrix}A&-\underaccent{\bar}{\bb}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]$ and $\tilde{\underaccent{\bar}{\ss}}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ \underaccent{\bar}{s}_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$}$ To initialize the algorithm, we observe that $(\boldsymbol{\mathit{y}}^{0},\boldsymbol{\mathit{s}}^{0},m)\in\Omega^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}^{0}}$ is on the $\underaccent{\bar}{\bb}$ central path, where we define $\underaccent{\bar}{z}^{0}=\underaccent{\bar}{\bb}^{T}\boldsymbol{\mathit{y}}^{0}-m$: ###### Lemma C.13. $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}}^{0},m)=0$ ###### Proof. Defining $\tilde{\underaccent{\bar}{\ss}}^{0}=\left[\begin{matrix}\boldsymbol{\mathit{s}}^{0}\\\ m{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$, we have $\tilde{\underaccent{\bar}{A}}(\tilde{\underaccent{\bar}{S}}^{0})^{-1}{\mbox{\boldmath$1$}}_{2m}=\left[\begin{matrix}A&-\underaccent{\bar}{\bb}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]\left[\begin{matrix}(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}\\\ m^{-1}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]=A(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}-\underaccent{\bar}{\bb}\cdot\frac{{\mbox{\boldmath$1$}}_{m}^{T}{\mbox{\boldmath$1$}}_{m}}{m}=\underaccent{\bar}{\bb}-\underaccent{\bar}{\bb}=0$ Thus, $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}}^{0},m)=\left\|\tilde{\underaccent{\bar}{A}}(\tilde{\underaccent{\bar}{S}}^{0})^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(\tilde{\underaccent{\bar}{A}}(\tilde{\underaccent{\bar}{S}}^{0})^{-2}\tilde{\underaccent{\bar}{A}}^{T})^{-1}}=0$ ∎ $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z)=\mathtt{FindCentralPath}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0})$ • Define $\displaystyle\left[\begin{matrix}\boldsymbol{\mathit{s}}^{0}\\\ \underaccent{\bar}{s}_{gap}^{0}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{0}\\\ m\end{matrix}\right]$ $\displaystyle\underaccent{\bar}{\bb}$ $\displaystyle=A(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}$ $\displaystyle\underaccent{\bar}{z}^{0}$ $\displaystyle=\underaccent{\bar}{\bb}^{T}\boldsymbol{\mathit{y}}^{0}-\underaccent{\bar}{s}_{gap}^{0}$ • Set $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z}):=(\boldsymbol{\mathit{y}}^{0},\boldsymbol{\mathit{s}}^{0},\underaccent{\bar}{s}_{gap}^{0},\underaccent{\bar}{z}^{0})$ • While $\underaccent{\bar}{s}_{gap}<40\lambda_{min}^{-1/2}Tm\left\|\underaccent{\bar}{\bb}\right\|$: – Set $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z}):=\mathtt{Unshift}\left(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z}\right)$ • Return $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}})$ and $z=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-40\lambda_{min}^{-1/2}Tm\left\|\boldsymbol{\mathit{b}}\right\|$ and $s_{gap}=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z$ $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},\underaccent{\bar}{s}_{gap}^{+},\underaccent{\bar}{z}^{+})=\mathtt{Unshift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z})$ • Let $\underaccent{\bar}{z}^{+}=\underaccent{\bar}{z}-\frac{\underaccent{\bar}{s}_{gap}}{10\sqrt{m}}$ • Let $\boldsymbol{\mathit{y}}^{+}=\mathtt{NewtonStep}(\tilde{\underaccent{\bar}{A}},\tilde{\underaccent{\bar}{\cc}},\boldsymbol{\mathit{y}})$ where $\tilde{\underaccent{\bar}{A}}=\left[\begin{matrix}A&-\underaccent{\bar}{\bb}{\mbox{\boldmath$1$}}_{m}^{T}\end{matrix}\right]$ and $\tilde{\underaccent{\bar}{\cc}}=\left[\begin{matrix}\boldsymbol{\mathit{c}}\\\ -\underaccent{\bar}{z}^{+}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ • Let $\left[\begin{matrix}\boldsymbol{\mathit{s}}^{+}\\\ \underaccent{\bar}{s}_{gap}^{+}\end{matrix}\right]=\left[\begin{matrix}\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{+}\\\ \underaccent{\bar}{\bb}^{T}\boldsymbol{\mathit{y}}^{+}-\underaccent{\bar}{z}^{+}\end{matrix}\right]$ Figure 5: Algorithm for finding point near central path given feasible interior point We present the $\mathtt{FindCentralPath}$ algorithm in Figure 5. Starting with $\underaccent{\bar}{z}=\underaccent{\bar}{z}^{0}$, we take steps along the $\underaccent{\bar}{\bb}$ central path, decreasing $\underaccent{\bar}{z}$ until it is sufficiently small that the analytic center of $\Omega^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}}$ is close to the analytic center of $\Omega^{D}$, and therfore also close to the analytic center of $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$ for some sufficiently small $z$. Let us show that the $\mathtt{Unshift}$ procedure indeed takes steps near the $\underaccent{\bar}{\bb}$ central path: ###### Lemma C.14 (compare Lemmas C.8). Given $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})\in\mathring{\Omega}^{D}_{\underaccent{\bar}{\bb},z}$ satisfying $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})\leq\frac{1}{40}$. Let $\underaccent{\bar}{s}_{gap}^{\prime}=\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-\underaccent{\bar}{z}^{+}$ and $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},\underaccent{\bar}{s}_{gap}^{+},\underaccent{\bar}{z}^{+})=\mathtt{Unshift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z})$ . Then 1. (i) $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}^{\prime}_{gap})\leq\frac{51}{400}$ 2. (ii) $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}}^{+},\underaccent{\bar}{s}^{+}_{gap})\leq\frac{1}{40}$. ###### Proof of C.14(i). Following the proof of Lemma C.8(i) through equation 9, we have $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap}^{\prime})\leq\frac{11}{10}\cdot\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})+\frac{1}{10}\leq\frac{11}{10}\cdot\frac{1}{40}+\frac{1}{10}=\frac{51}{400}$ ∎ ###### Proof of C.14(ii). By Lemma C.5, we have $\tilde{\eta}(\boldsymbol{\mathit{s}}^{+},\underaccent{\bar}{s}_{gap}^{+})\leq\tilde{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap}^{\prime})^{2}+\frac{1}{20}\tilde{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap}^{\prime})\leq\left(\frac{51}{400}\right)^{2}+\frac{1}{20}\cdot\frac{51}{400}<\frac{1}{40}$ ∎ Next, let us prove that the point returned by $\mathtt{FindCentralPath}$ is indeed near the original central path (i.e. the path given by $\boldsymbol{\mathit{b}}$): ###### Lemma C.15. For $\boldsymbol{\mathit{y}}^{0}$ satisfying $A^{T}\boldsymbol{\mathit{y}}^{0}<\boldsymbol{\mathit{c}}$, let $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap},z)=\mathtt{FindCentralPath}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0})$. Then $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\mathring{\Omega}^{D}_{\boldsymbol{\mathit{b}},z}$ and $\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})\leq\frac{1}{10}$. ###### Proof. Using the values at the end of the algorithm, we write $\tilde{\boldsymbol{\mathit{s}}}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ s_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$ and $\tilde{\underaccent{\bar}{\ss}}=\left[\begin{matrix}\boldsymbol{\mathit{s}}\\\ \underaccent{\bar}{s}_{gap}{\mbox{\boldmath$1$}}_{m}\end{matrix}\right]$. To begin, we note $\displaystyle\underaccent{\bar}{s}_{gap}$ $\displaystyle\geq 40\lambda_{min}^{-1/2}Tm\left\|\underaccent{\bar}{\bb}\right\|$ $\displaystyle=40m(T^{-2}\lambda_{min})^{-1/2}\left\|\underaccent{\bar}{\bb}\right\|$ $\displaystyle\geq 40m\left\|\underaccent{\bar}{\bb}\right\|_{(AS^{-2}A^{T})^{-1}}$ (15) where the last inequality follows because the smallest eigenvalue of $AS^{-2}A^{T}$ is at least $T^{-2}\lambda_{min}$. Similarly, $s_{gap}=40\lambda_{min}^{-1/2}Tm\left\|\boldsymbol{\mathit{b}}\right\|\geq 40m\left\|\boldsymbol{\mathit{b}}\right\|_{(AS^{-2}A^{T})^{-1}}$ (16) We have $\displaystyle\tilde{\eta}(\boldsymbol{\mathit{s}},s_{gap})$ $\displaystyle=\left\|\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(\tilde{A}\tilde{S}^{-2}\tilde{A}^{T})^{-1}}$ $\displaystyle\leq\left\|\tilde{A}\tilde{S}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(AS^{-2}A^{T})^{-1}}$ (because $\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}-AS^{-2}A^{T}=ms_{gap}^{-2}\boldsymbol{\mathit{b}}\boldsymbol{\mathit{b}}^{T}$ is positive semidefinite) $\displaystyle=\left\|\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-1}{\mbox{\boldmath$1$}}_{2m}-ms_{gap}^{-1}\boldsymbol{\mathit{b}}+m\underaccent{\bar}{s}_{gap}^{-1}\underaccent{\bar}{\bb}\right\|_{(AS^{-2}A^{T})^{-1}}$ $\displaystyle\leq\left\|\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(AS^{-2}A^{T})^{-1}}+m\underaccent{\bar}{s}_{gap}^{-1}\left\|\underaccent{\bar}{\bb}\right\|_{(AS^{-2}A^{T})^{-1}}+ms_{gap}^{-1}\left\|\boldsymbol{\mathit{b}}\right\|_{(AS^{-2}A^{T})^{-1}}$ $\displaystyle\leq\left\|\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(AS^{-2}A^{T})^{-1}}+\frac{1}{40}+\frac{1}{40}\qquad\text{(by equations \ref{eq:unshift-hatsgap} and \ref{eq:unshift-sgap})}$ $\displaystyle\leq\left(1+m\underaccent{\bar}{s}_{gap}^{-1}\left\|\underaccent{\bar}{\bb}\right\|_{(AS^{-2}A^{T})^{-1}}\right)^{1/2}\left\|\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-2}\tilde{\underaccent{\bar}{A}}^{T})^{-1}}+\frac{1}{40}+\frac{1}{40}$ (by Lemma B.3, using the fact that $\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-2}\tilde{\underaccent{\bar}{A}}^{T}-AS^{-2}A^{T}=m\underaccent{\bar}{s}_{gap}^{-2}\underaccent{\bar}{\bb}\underaccent{\bar}{\bb}^{T}$) $\displaystyle\leq\left(1+\frac{1}{40}\right)^{1/2}\left\|\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-1}{\mbox{\boldmath$1$}}_{2m}\right\|_{(\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-2}\tilde{\underaccent{\bar}{A}}^{T})^{-1}}+\frac{1}{40}+\frac{1}{40}\qquad\text{(by equation \ref{eq:unshift-hatsgap})}$ $\displaystyle=\left(1+\frac{1}{40}\right)^{1/2}\cdot\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})+\frac{1}{40}+\frac{1}{40}$ $\displaystyle\leq 2\cdot\underaccent{\bar}{\eta}(\tilde{\underaccent{\bar}{\ss}},\underaccent{\bar}{s}_{gap})+\frac{1}{20}$ $\displaystyle\leq 2\cdot\frac{1}{40}+\frac{1}{20}\qquad\text{(by Lemma \ref{lem:unshift}(ii))}$ $\displaystyle=\frac{1}{10}$ ∎ To measure the progress of the $\mathtt{FindCentralPath}$ algorithm, we define $\underaccent{\bar}{B}(\underaccent{\bar}{z})$: $\underaccent{\bar}{B}(\underaccent{\bar}{z})=\sum_{j=1}^{m}\log\accentset{*}{\boldsymbol{\mathit{s}}}_{j}+m\log\accentset{*}{\underaccent{\bar}{s}}_{gap}\qquad\text{where $(\accentset{*}{\boldsymbol{\mathit{y}}},\accentset{*}{\boldsymbol{\mathit{s}}},\accentset{*}{\underaccent{\bar}{s}}_{gap})$ is the analytic center of $\Omega^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}}$}$ ###### Lemma C.16 (compare Lemma C.10). Given $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})\in\mathring{\Omega}^{D}_{\underaccent{\bar}{\bb},z}$ satisfying $\underaccent{\bar}{\eta}(\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap})\leq\frac{1}{40}$. Let $(\boldsymbol{\mathit{y}}^{+},\boldsymbol{\mathit{s}}^{+},\underaccent{\bar}{s}_{gap}^{+},\underaccent{\bar}{z}^{+})=\mathtt{Unshift}(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},\underaccent{\bar}{s}_{gap},\underaccent{\bar}{z})$. Then $\underaccent{\bar}{B}(\underaccent{\bar}{z}^{+})\geq\underaccent{\bar}{B}(\underaccent{\bar}{z})+\Theta(\sqrt{m})$. ###### Proof. We will follow the proof of Lemma C.10, with some minor changes. Before we proceed, let us recall the definition $\underaccent{\bar}{z}^{+}=\underaccent{\bar}{z}-\frac{\underaccent{\bar}{s}_{gap}}{10\sqrt{m}}$ to note that $\underaccent{\bar}{s}_{gap}^{\prime}=\underaccent{\bar}{s}_{gap}+\underaccent{\bar}{z}-\underaccent{\bar}{z}^{+}=10\sqrt{m}(\underaccent{\bar}{z}-\underaccent{\bar}{z}^{+})+\underaccent{\bar}{z}-\underaccent{\bar}{z}^{+}\leq 11\sqrt{m}(\underaccent{\bar}{z}-\underaccent{\bar}{z}^{+})$ (17) Now, we switch the places of $z$ and $z^{+}$, and follow the proof of Lemma C.10 up to Equation 14: $\displaystyle e^{\frac{\underaccent{\bar}{B}(\underaccent{\bar}{z})-\underaccent{\bar}{B}(\underaccent{\bar}{z}^{+})}{2m}}$ $\displaystyle\leq 1+\frac{1}{2\accentset{*}{\underaccent{\bar}{s}}_{gap}^{+}}\cdot(\underaccent{\bar}{z}^{+}-\underaccent{\bar}{z})$ We continue: $\displaystyle\leq 1-\frac{\underaccent{\bar}{s}_{gap}^{\prime}}{22\sqrt{m}\cdot\accentset{*}{\underaccent{\bar}{s}}_{gap}^{+}}\qquad\text{(by Equation \ref{eq:stosprime})}$ $\displaystyle\leq 1-\frac{1}{22\sqrt{m}}\cdot\frac{349}{400}\qquad\text{(by Lemmas \ref{lem:unshift} and \ref{lem:nearcenter})}$ $\displaystyle\leq e^{-\frac{169}{4400\sqrt{m}}}$ ∎ ###### Corollary C.17. The $\mathtt{FindCentralPath}$ algorithm makes $O\left(\sqrt{m}\log\frac{TUm}{\lambda_{min}s^{0}_{min}}\right)$ calls to $\mathtt{Unshift}$, where $s^{0}_{min}$ is the smallest entry of $\boldsymbol{\mathit{s}}^{0}=\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{0}$. ###### Proof. Recall that the algorithm will terminate only when the value of $\underaccent{\bar}{s}_{gap}$ has increased from its initial value of $m$ to at least $40\lambda_{min}^{-1/2}Tm\left\|\underaccent{\bar}{\bb}\right\|$. So, by Lemma C.11, this will have happened once $\underaccent{\bar}{B}(z)$ has increased by $\Omega\left(m\log\left(\lambda_{min}^{-1/2}T\left\|\underaccent{\bar}{\bb}\right\|\right)\right)$ . According to Lemma C.16, this occurs after $O\left(\sqrt{m}\log\left(\lambda_{min}^{-1/2}T\left\|\underaccent{\bar}{\bb}\right\|\right)\right)$ iterations. To complete the proof, we note that $\left\|\underaccent{\bar}{\bb}\right\|=\left\|A(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}\right\|\leq\frac{n^{1/2}mU}{s^{0}_{min}}$ ∎ ### C.4 Calls to the Solver In each call to $\mathtt{Unshift}$, we solve one system in a matrix of the form $\tilde{\underaccent{\bar}{A}}\tilde{\underaccent{\bar}{S}}^{-2}\tilde{\underaccent{\bar}{A}}^{T}=AS^{-2}A^{T}+m\underaccent{\bar}{s}_{gap}^{-2}\underaccent{\bar}{\bb}\underaccent{\bar}{\bb}^{T}$ and in each call to $\mathtt{Shift}$, we solve one system in a matrix of the form $\tilde{A}\tilde{S}^{-2}\tilde{A}^{T}=AS^{-2}A^{T}+ms_{gap}^{-2}\boldsymbol{\mathit{b}}\boldsymbol{\mathit{b}}^{T}$ At the end of the interior-point algorithm we have one final call of the latter form. In order to say something about the condition number of the above matrices, we must bound the slack vector $\boldsymbol{\mathit{s}}$. We are given an upper bound of $T$ on the elements of $\boldsymbol{\mathit{s}}$, so it remains to prove a lower bound: ###### Lemma C.18. Throughout the $\mathtt{InteriorPoint}$ algorithm, $\boldsymbol{\mathit{s}}\geq\frac{\epsilon}{48nmTU}\boldsymbol{\mathit{s}}^{0}$ ###### Proof. At all times duing the algorithm, we know from Lemma C.4 that the elements of $\boldsymbol{\mathit{s}}$ are bounded by a constant factor from the slacks at the current central path point $\accentset{*}{\boldsymbol{\mathit{s}}}$. In particular, taking into account Lemmas C.8 and C.14, we surely have $\boldsymbol{\mathit{s}}\geq\frac{1}{2}\accentset{*}{\boldsymbol{\mathit{s}}}$. So let us bound from below the elements of $\accentset{*}{\boldsymbol{\mathit{s}}}$. During the $\mathtt{FindCentralPath}$ subroutine, as we decrease $\underaccent{\bar}{z}$ and expand the polytope $\Omega^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}}$, clearly the initial point $\boldsymbol{\mathit{s}}^{0}$ remains in the interior of $\Omega^{D}_{\underaccent{\bar}{\bb},\underaccent{\bar}{z}}$ throughout. Thus, by Lemma C.3, we have $\accentset{*}{\boldsymbol{\mathit{s}}}\geq\frac{1}{2m}\boldsymbol{\mathit{s}}^{0}$, and so $\boldsymbol{\mathit{s}}\geq\frac{1}{2}\accentset{*}{\boldsymbol{\mathit{s}}}\geq\frac{1}{4m}\boldsymbol{\mathit{s}}^{0}$. Unfortunately, during the main part of the algorithm, as we increase $z$ and shrink the polytope $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$, the initial point may not remain inside the polytope. In particular, once we have $z\geq\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{0}$, the initial point is no longer in $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$, but we may define a related point $(\boldsymbol{\mathit{y}}^{z},\boldsymbol{\mathit{s}}^{z},s_{gap}^{z})$ that is in $\Omega^{D}_{\boldsymbol{\mathit{b}},z}$. Given our current point $(\boldsymbol{\mathit{y}},\boldsymbol{\mathit{s}},s_{gap})\in\Omega^{D}_{\boldsymbol{\mathit{b}},z}$ for $z\geq\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{0}$, let us define $r=\frac{\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z}{2(\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{0})}$ and note that $0<r<\frac{1}{2}$. We then define $\displaystyle\boldsymbol{\mathit{y}}^{z}$ $\displaystyle=r\boldsymbol{\mathit{y}}^{0}+(1-r)\boldsymbol{\mathit{y}}$ $\displaystyle\left[\begin{matrix}\boldsymbol{\mathit{s}}^{z}\\\ s_{gap}^{z}\end{matrix}\right]$ $\displaystyle=\left[\begin{matrix}\boldsymbol{\mathit{c}}-A^{T}\boldsymbol{\mathit{y}}^{z}\\\ \boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{z}-z\end{matrix}\right]=\left[\begin{matrix}r\boldsymbol{\mathit{s}}^{0}+(1-r)\boldsymbol{\mathit{s}}\\\ \frac{1}{2}(\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-z)\end{matrix}\right]>0$ Therefore Lemma C.3 gives $\accentset{*}{\boldsymbol{\mathit{s}}}\geq\frac{1}{2m}\boldsymbol{\mathit{s}}^{z}=\frac{r\boldsymbol{\mathit{s}}^{0}+(1-r)\boldsymbol{\mathit{s}}}{2m}\geq\frac{r}{2m}\boldsymbol{\mathit{s}}^{0}$ We then find $r=\frac{s_{gap}}{2(\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}-\boldsymbol{\mathit{b}}^{T}\boldsymbol{\mathit{y}}^{0})}\geq\frac{s_{gap}}{4nTU}\geq\frac{\epsilon}{24nTU}$ The last inequality follows because, when $s_{gap}$ decreased below $\frac{\epsilon}{3}$ on the final step, using Lemma C.4 we find that it certainly could not have decreased by more than a factor of $\frac{1}{2}$. We conclude $\boldsymbol{\mathit{s}}\geq\frac{1}{2}\accentset{*}{\boldsymbol{\mathit{s}}}\geq\frac{r}{4m}\boldsymbol{\mathit{s}}^{0}\geq\frac{\epsilon}{48nmTU}\boldsymbol{\mathit{s}}^{0}$ ∎ We may now summarize the calls to the solver as follows: ###### Theorem C.19. The $\mathtt{InteriorPoint}(A,\boldsymbol{\mathit{b}},\boldsymbol{\mathit{c}},\boldsymbol{\mathit{y}}^{0},\epsilon)$ algorithm makes $\mathcal{O}\left(\sqrt{m}\log\frac{TUm}{\lambda_{min}s^{0}_{min}\epsilon}\right)$ calls to the approximate solver, of the form $\mathtt{Solve}\left(AS^{-2}A^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\cdot,\Theta(m^{-1/2})\right)$ and one call of the form $\mathtt{Solve}\left(AS^{-2}A^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T},\cdot,\Omega\left(\frac{s^{0}_{min}\epsilon}{m^{1/2}n^{2}T^{2}U^{2}}\right)\right)$ where $S$ is a positive diagonal matrix with condition number $\mathcal{O}\left(\frac{T^{2}Um^{2}}{\epsilon}\right)$, and $\boldsymbol{\mathit{v}}$ satisfies $\left\|\boldsymbol{\mathit{v}}\right\|=\mathcal{O}\left(\frac{U(mn)^{1/2}}{s^{0}_{min}\epsilon}\right)$ ###### Proof. From Lemmas C.17 and C.12, the total number of solves is $\mathcal{O}\left(\sqrt{m}\left(\log\frac{TUm}{\lambda_{min}s^{0}_{min}}+\log\frac{s_{gap}^{C}}{\epsilon}\right)\right)$, where we know from the $\mathtt{FindCentralPath}$ algorithm that $s_{gap}^{C}=40\frac{Tm\left\|\boldsymbol{\mathit{b}}\right\|}{\lambda_{min}^{1/2}}=\mathcal{O}\left(\frac{TUmn^{1/2}}{\lambda_{min}^{1/2}}\right)$ As we noted above, all solves are in matrices that take the form $AS^{-2}A^{T}+\boldsymbol{\mathit{v}}\boldsymbol{\mathit{v}}^{T}$, where $\displaystyle\boldsymbol{\mathit{v}}$ $\displaystyle=m^{1/2}s_{gap}^{-1}\boldsymbol{\mathit{b}}$ or $\displaystyle\boldsymbol{\mathit{v}}$ $\displaystyle=m^{1/2}\underaccent{\bar}{s}_{gap}^{-1}A(S^{0})^{-1}{\mbox{\boldmath$1$}}_{m}$ We know that $s_{gap}=\Omega(\epsilon)$ and $\underaccent{\bar}{s}_{gap}=\Omega(m)$, so we obtain the respective bounds $\displaystyle\left\|\boldsymbol{\mathit{v}}\right\|$ $\displaystyle=\mathcal{O}\left(\frac{U(mn)^{1/2}}{\epsilon}\right)$ $\displaystyle\left\|\boldsymbol{\mathit{v}}\right\|$ $\displaystyle=\mathcal{O}\left(\frac{U(mn)^{1/2}}{s^{0}_{min}}\right)$ The condition number of $S$ comes from Lemma C.18 and the upper bound of $T$ on the slacks. The error parameter for the solver is $\Theta\left(m^{-1/2}\right)$ from the all $\mathtt{NewtonStep}$ calls. In the final solve, the error parameter is $\frac{s_{min}m^{1/2}}{TUn}\geq\frac{m^{1/2}}{TUn}\cdot\frac{s^{0}_{min}\epsilon}{48nmTU}$, again invoking Lemma C.18. ∎

arxiv-papers

2024-09-04T02:48:54.199620

0803.0999

# Bounding the Size and Probability of Epidemics on Networks Joel C. Miller 655 W 12th Ave, Vancouver, BC V5Z 4R4, Canada ###### Abstract We consider an infectious disease spreading along the edges of a network which may have significant clustering. The individuals in the population have heterogeneous infectiousness and/or susceptibility. We define the _out- transmissibility_ of a node to be the marginal probability that it would infect a randomly chosen neighbor given its infectiousness and the distribution of susceptibility. For a given distribution of out- transmissibility, we find the conditions which give the upper [or lower] bounds on size and probability of an epidemic, under weak assumptions on the transmission properties, but very general assumptions on the network. We find similar bounds for a given distribution of in-transmissibility (the marginal probability of being infected by a neighbor). We also find conditions giving global upper bounds on size and probability. The distributions leading to these bounds are network-independent. In the special case of networks with high girth (locally tree-like), we are able to prove stronger results. In general the probability and size of epidemics are maximal when the population is homogeneous and minimal when the variance of in- or out-transmissibility is maximal. ###### keywords: Epidemiology, Networks, Attack Rate, Probability, Transmissibility Joel C. Miller 92D3060K35 ## 1 Introduction The spread of infectious disease is governed by many different factors which vary on the individual level. Heterogeneity in the population comes from a number of sources including, but not limited to, genetic diversity, previous infections, vaccination history, or existence of co-infections. In this paper we investigate the effects of heterogeneity on disease spread, focusing on the effect of simultaneous heterogeneities in infectiousness and susceptibility. Figure 1: The spread of disease in a network. The outbreak begins with a single infected individual [large empty circles] and then spreads along edges to others. The infected nodes recover with immunity [large filled circles]. Eventually the outbreak dies out. We consider the spread of infectious diseases in networks as shown in figure 1. Individuals in the population are modeled as nodes and potentially infectious contacts are modeled as edges between the corresponding nodes. We consider the spread of an SIR disease, that is the nodes are divided into three compartments: _Susceptible_ , _Infected_ , and _Recovered_. A susceptible node may be infected by an infected neighbor. Following infection, the newly infected node may infect some, all, or none of its neighbors and then recover. After recovery, a node cannot be reinfected. Typically in a large network outbreaks are either small or large (in a sense made more formal in section 2). We are primarily interested in what controls the probability of large outbreaks and the fraction of nodes infected in a large outbreak. Before discussing earlier results, we introduce some terminology. The _transmissibility_ $T_{uv}$ is the probability that an infection of node $u$ would result in direct infection of the neighbor $v$. The _in- transmissibility_ $T_{in}(v)$ is the marginal probability that a neighbor of $v$ would infect $v$ given the characteristics of $v$, and the _out- transmissibility_ $T_{out}(u)$ is the marginal probability that $u$ would infect a neighbor given the characteristics of $u$. Both the in- and out- transmissibility necessarily have the same average, $\left\langle T\right\rangle$. These definitions will be made more precise in section 2. Most network-based epidemic models assume homogeneous transmissibility $T_{uv}=\left\langle T\right\rangle$ between all pairs of neighboring nodes. Models that do allow heterogeneities generally show that they reduce the probability or size of epidemics [4, 16, 29, 21, 13]. For an arbitrary network with homogeneous susceptibility [$T_{in}(v)=\left\langle T\right\rangle$ for all $v$], but heterogeneous infectiousness, [16] showed that epidemics are most likely and largest if infectiousness is homogeneous [$T_{out}(u)=\left\langle T\right\rangle$ for all $u$]. It was noted by [29] that the same argument shows that with homogeneous susceptibility epidemics are least likely and smallest if infectiousness is maximally heterogeneous ($T_{out}=0$ for a fraction $1-\left\langle T\right\rangle$ of the population and $T_{out}=1$ for the remainder). The recent work of [21, 13] considered the effect of heterogeneity on a specific subclass of unclustered networks (variously called Molloy–Reed networks [23] or Configuration Model networks [26]), finding similar results. One of these, [21], studied simultaneous heterogeneities in susceptibility and infectiousness, showing that for given $\left\langle T\right\rangle$, the same cases give the upper and lower bounds on probability while epidemics are largest if susceptibility is homogeneous and smallest if susceptibility is maximally heterogeneous. We are unaware of any work which has considered simultaneous heterogeneities in infectiousness and susceptibility in networks with clustering or even in unclustered networks more general than Molloy–Reed networks. In this paper we investigate the spread of epidemics in which $T_{in}$ and $T_{out}$ can simultaneously be heterogeneous, using techniques from [16] and [21]. We will consider both clustered and general unclustered networks. Clustered networks are more difficult because of the existence of short cycles, and so a stronger assumption will be made for them. Often only the distribution of $T_{out}$ (or, more rarely, of $T_{in}$) would be available early in an outbreak. If we know the distribution of $T_{out}$, it does not in general uniquely determine the distribution of $T_{in}$ and so we cannot fully predict the final details of an outbreak. Our focus is on identifying the best and worst case scenarios given the distribution of $T_{out}$ (or $T_{in}$), thus helping to provide policy makers with knowledge of what to expect and how best to mitigate it. Mathematical theories modeling the spread of infectious diseases have been developed in a number of fields [14, 1, 2, 8]. The techniques used include differential equations, stochastic models, agent-based simulations, and network-based approaches. The differential equations approaches may be thought of as a mean-field approximation to a subclass of network models, while the stochastic and agent-based approaches can be made formally equivalent to network-based methods [13, 24]. Consequently results for networks will apply to other models as well. Network epidemic models have primarily been studied by the statistics community [30, 16, 17, 4, 5, 22, 3] and the statistical physics/applied mathematics communities [25, 20, 19, 28, 9, 27, 18, 10, 11]. In general the statistics community has produced more rigorous results, but has considered more restricted classes of networks. The physics and applied mathematics communities have considered a wider range of networks, but the results are less rigorous. The interaction between these fields has been relatively sparse, leading to repeated discoveries of some results and a lack of cohesion in the topics studied. We attempt to bring some of these different approaches together in this paper. This paper is structured as follows: in section 2 we introduce the model and clarify definitions. In section 3 we consider epidemics spreading on general networks. In section 4 we find stronger results for networks with no short cycles. Finally in section 5 we discuss extensions and implications of our results. ## 2 Epidemics in networks We consider the spread of disease on a network $G$. An _outbreak_ begins when a single node (the _index case_) chosen uniformly from the population is infected. The disease spreads from an infected node $u$ to a neighboring susceptible node $v$ with a probability equal to the _transmissibility_ $T_{uv}$. Each infected node attempts to infect each of its neighbors and then recovers (and is no longer susceptible or infected). The outbreak ends when no infected nodes remain. This section, like Gaul, is divided into three parts. First we describe the neighbor-to-neighbor transmissibility $T_{uv}$. This will depend on the characteristics of both $u$ and $v$. We then introduce the concept of an Epidemic Percolation Network, which is a tool to study the routes of transmission in a given network. We finally discuss tools which will be used to make the concept of a “large network” rigorous. ### 2.1 Transmissibility Following [21], we assume that the factors influencing infectiousness of node $u$ and susceptibility of node $v$ may be summarized in $\mathcal{I}_{u}$ and $\mathcal{S}_{v}$. In general, these may be vector-valued functions (though with few exceptions they are taken to be scalars in the literature). For example, $\mathcal{I}_{u}$ may represent $u$’s viral load, duration of infection, and willingness or ability to leave work if sick, while $\mathcal{S}_{v}$ may represent $v$’s previous vaccination history, genetic predisposition to infection, and previous exposure to related infections. If $u$ and $v$ are neighbors, the transmissibility $T_{uv}$ is then $T_{uv}=T(\mathcal{I}_{u},\mathcal{S}_{v})$ for some function $T$. The function $T(\mathcal{I},\mathcal{S})$ is the probability of transmission from a node with infectiousness $\mathcal{I}$ to a node with susceptibility $\mathcal{S}$ assuming that the nodes are joined by an edge. We may think of $T_{uv}$ as defined only for neighboring nodes, or we may take $T_{uv}=\chi_{\\{u,v\\}}T(\mathcal{I}_{u},\mathcal{S}_{v})$ where $\chi_{\\{u,v\\}}=0$ if $\\{u,v\\}$ is not an edge and $1$ if it is. We assume that $\mathcal{I}$ and $\mathcal{S}$ are assigned independently, using the probability density functions $P(\mathcal{I})$ and $P(\mathcal{S})$ (although we use the same symbol $P$ for both, we assume that the two functions are different). Particularly if $\mathcal{I}$ and $\mathcal{S}$ are vectors, we may not be able to clearly define which of two nodes is “more infectious” (_i.e._ , there may not be a well-defined ordering). For example, with a sexually transmitted disease, we might have $u_{1}$ and $u_{2}$ infected, with $u_{1}$ having a high viral load and regular condom use (with occasional lapses), while $u_{2}$ has a low viral load but no condom use. Let us assume they have contacts with susceptibles $v_{1}$ and $v_{2}$ where $v_{1}$ has a high level of resistance, and thus will only be infected by a large dose, while $v_{2}$ has no immune protection and thus will be infected by even a small dose. Under these assumptions, $u_{1}$ is more likely to infect $v_{1}$, while $u_{2}$ is more likely to infect $v_{2}$. Which is “more infectious” depends on the test susceptible considered. The probability that $u$ infects a neighbor (prior to knowing $\mathcal{S}$ for that neighbor) is given by the _out-transmissibility_ of $u$ $T_{out}(u)=\int T(\mathcal{I}_{u},\mathcal{S})P(\mathcal{S})d\mathcal{S}$ and the probability that $v$ would be infected by a neighbor is given by the _in-transmissibility_ of $v$ $T_{in}(v)=\int T(\mathcal{I},\mathcal{S}_{v})P(\mathcal{I})d\mathcal{I}$ At times it will be convenient to use $T_{out}(\mathcal{I})$ and $T_{in}(\mathcal{S})$ [rather than $T_{out}(u)$ and $T_{in}(v)$] to denote the out- and in-transmissibility of arbitrary nodes with $\mathcal{I}$ and $\mathcal{S}$ respectively. When the concepts of being “more infectious” and “more susceptible” are clearly defined, $T_{in}(\mathcal{S})$ and $T_{out}(\mathcal{I})$ are invertible functions. However, because the ordering is not well-defined in general, they may not be invertible. If they are invertible, it is often convenient to change variables and set $\mathcal{I}_{u}=T_{out}(u)$ or $\mathcal{S}_{v}=T_{in}(v)$. We will do this frequently in section 3, where we restrict our attention to cases where the ordering described above is well-defined. From $P(\mathcal{S})$ and $P(\mathcal{I})$, we may find the distributions of $T_{in}$ and $T_{out}$. We use $Q_{in}(T_{in})$ to denote the probability density function for the in-transmissibility $T_{in}$ and $Q_{out}(T_{out})$ to denote the probability density function for the out-transmissibility $T_{out}$. The averages $\int_{0}^{1}T_{in}Q_{in}(T_{in})dT_{in}$ and $\int_{0}^{1}T_{out}Q_{out}(T_{out})dT_{out}$ are both equal to $\left\langle T\right\rangle$. Given distributions of $\mathcal{I}$ and $\mathcal{S}$ and the function $T$, there is always a $Q_{in}$ and $Q_{out}$ pair that result. Also, given a $Q_{in}$ or a $Q_{out}$ it is always possible to find $P(\mathcal{I})$, $P(\mathcal{S})$, and $T$ that are consistent. For example, given any $Q_{in}$, for each node $v$ we assign a $T_{in}(v)$ from $Q_{in}$ and set $\mathcal{S}_{v}=T_{in}(v)$. Then $T(\mathcal{I},\mathcal{S})=\mathcal{S}$ is consistent with $Q_{in}$ and yields $Q_{out}(T_{out})=\delta(T_{out}-\left\langle T\right\rangle)$. This means that for any distribution of in-transmissibility, it is possible that the infectiousness of nodes is homogeneous. Although it is possible to find a $Q_{in}$ for any $Q_{out}$ (and _vice versa_) not all pairs $Q_{out}$ and $Q_{in}$ are compatible. For example, if $Q_{in}=(1-\left\langle T\right\rangle)\delta(T_{in})+\left\langle T\right\rangle\delta(T_{in}-1)$ (_i.e._ , susceptibility is maximally heterogeneous) then the out- transmissibility must be homogeneously distributed; no other distribution is possible. This particular example will be important in Section 4. Although in principle $\mathcal{I}$ and $\mathcal{S}$ may be vector-valued, they frequently are assumed to be scalars with the transmissibility between two neighbors given by (for example [7, 21]) $T_{uv}=T(\mathcal{I}_{u},\mathcal{S}_{v})=1-\exp(-\alpha\mathcal{I}_{u}\mathcal{S}_{v})$ (1) A number of disease models yield this form. For example: let $\alpha$ be the rate at which virus from an infected person reaches a susceptible person. Let $\mathcal{I}_{u}$ be the infectious period of $u$. Let $\mathcal{S}_{v}$ be the probability that a virus reaching $v$ causes infection. Then the probability $p$ that $v$ has not become infected satisfies $\dot{p}=-\alpha\mathcal{S}_{v}p$. Integrating this over the infectious period $\mathcal{I}_{u}$ of $u$ yields equation (1). We need one final concept related to the transmissibility. Let a node $u$ be given, and let $V=\\{v_{1},\ldots,v_{m}\\}$ be a subset of the neighbors of $u$. Assume we know $\vec{\mathcal{S}}=(\mathcal{S}_{v_{1}},\ldots,\mathcal{S}_{v_{m}})$, but not $\mathcal{I}_{u}$. Define $\phi_{in}(V,\vec{\mathcal{S}})=\int\prod_{v\in V}[1-T(\mathcal{I},\mathcal{S}_{v})]P(\mathcal{I})d\mathcal{I}\,.$ (2) This is the probability that $u$ will not infect any node in $V$ given knowledge of $\mathcal{S}$ for each $v\in V$, but marginalized over the possible values of $\mathcal{I}$ for $u$. We may similarly define $\displaystyle\psi_{in}(V)$ $\displaystyle=\int\phi_{in}(V,\vec{\mathcal{S}})P(\vec{\mathcal{S}})d\vec{\mathcal{S}}\,,$ (3) $\displaystyle=\int(1-T_{out})^{|V|}Q_{out}(T_{out})dT_{out}\,.$ (4) This is the probability that $u$ will not infect any $v\in V$ marginalized over $\mathcal{S}$ of $v\in V$ and the values of $\mathcal{I}$ for $u$. If $|V|=1$, then $\psi_{in}(V)=1-\left\langle T\right\rangle$, which will be important later when we consider unclustered networks. ### 2.2 Epidemic Percolation Networks Given a network $G$, the distributions $P(\mathcal{I})$ and $P(\mathcal{S})$, and the function $T(\mathcal{I},\mathcal{S})$, we assign $\mathcal{I}$ and $\mathcal{S}$ to each node of $G$. We then create a new directed network $\mathcal{E}$ which is an _Epidemic Percolation Network_ (EPN) [12] as follows: the nodes of $\mathcal{E}$ are the nodes of $G$. For each edge $\\{u,v\\}$ of $G$, we place directed edges $(u,v)$ and $(v,u)$ into $\mathcal{E}$ with probability $T_{uv}$ and $T_{vu}$ respectively. The original network $G$ gives the paths a disease _could_ follow, while a realization of $\mathcal{E}$ gives the paths the disease _will_ follow (if given the chance) for a simulation. The out-component of a given node $u$ found by assigning $\mathcal{I}$ and $\mathcal{S}$ and generating an EPN comes from the same distribution as the nodes infected by the dynamic epidemic process described earlier with $u$ as the index case. The processes are formally equivalent. To motivate some definitions, we assume sufficiently high transmissibility that there are nodes in $\mathcal{E}$ with giant in- or out-components [6]. We define $H_{out}$ to be those nodes with a giant in-component, and $H_{in}$ to be those nodes with a giant out-component in $\mathcal{E}$. We define $H_{scc}=H_{in}\cap H_{out}$. $H_{scc}$ will almost surely be a strongly connected component. $H_{in}$ is the in-component of $H_{scc}$ and $H_{out}$ is its out-component. In general, infection of any $u\in H_{in}$ results in infection of all nodes in $H_{out}$ and occasionally a few other nodes (if $u\not\in H_{scc}$). We define such an outbreak to be an _epidemic_. If $u\not\in H_{in}$, then a small _self-limiting_ outbreak occurs. For large values of $N=|G|$, the probability of an epidemic is given by $\mathcal{Y}=\mathbb{E}[|H_{in}|]/N$ and the expected fraction infected in an epidemic is given by $\mathcal{A}=\mathbb{E}[|H_{out}|]/N+\mathcal{O}\left(\log N/N\right)$. As $N$ grows, $|H_{out}|/N$ approaches $\mathbb{E}[|H_{out}|]/N$, and so the size of a single epidemic in a large population closely approximates expected size of epidemics (note that if we include non-epidemic outbreaks in the average, this does not hold). If the directions of arrows in the EPNs are reversed, then $H_{in}$ and $H_{out}$ interchange roles. Consequently, replacing $T_{uv}=T(\mathcal{I}_{u},\mathcal{S}_{v})$ with $\hat{T}_{uv}=T(\mathcal{I}_{v},\mathcal{S}_{u})$ interchanges the size and probability. As such, results derived for the probability of an epidemic also apply to the size. ### 2.3 Large Networks The results we derive will be appropriate in the limit of “large networks”. However, in practice we are usually interested in a single given network. Unfortunately $|G|\to\infty$ is a vague concept when we are given a single, finite network. There are many ways to increase its size, with different impacts on epidemics. In this section, we define what is meant by $|G|\to\infty$ in a way that allows us to produce rigorous results. Consider a sequence of networks $\\{G_{n}\\}$ which satisfy $|G_{n}|\to\infty$ as $n\to\infty$. We define an open ball $B_{d}(u)$ to be a network centered at a node $u$ such that all nodes $v\in B_{d}(u)$ are at most a distance $d$ from $u$. Given a network $G$, we define $P_{G}(B_{d}(u))$ to be the probability that if we choose a node $\hat{u}$ randomly from $G$, then the set of nodes of distance at most $d$ from $\hat{u}$ is isomorphic to $B_{d}(u)$ (with the isomorphism mapping $\hat{u}$ to $u$). We define _sequential convergence of local statistics_ to mean that given any $d$ and $B_{d}(u)$, $P_{G_{n}}(B_{d}(u))=P_{G_{d}}(B_{d}(u))$ for all $n\geq d$. For the results developed later, all that is strictly needed is that $P_{G_{n}}(B_{d}(u))$ converges as $n\to\infty$, but the stronger statement that for $n\geq d$ they do not change makes the proofs simpler. This means that for large enough $n$, networks have the same “small-scale” structure, and the size of what is considered “small-scale” increases with $|G|$. We restrict our attention to sequences which have sequential convergence of local statistics. For a given EPN, we define $H_{in}(d)$ and $H_{out}(d)$ to be the set of nodes from which a path of length (at least) $d$ begins or ends respectively. At large $d$, these will correspond to the $H_{in}$ and $H_{out}$ described earlier. We define $\mathcal{Y}_{d}(G)$ and $\mathcal{A}_{d}(G)$ to be the probability that a randomly chosen node from $G$ is in $H_{in}(d)$ and $H_{out}(d)$ respectively. Sequential convergence means that $\mathcal{Y}_{d}(G_{n})=\mathcal{Y}_{d}(G_{d})$ and $\mathcal{A}_{d}(G_{n})=\mathcal{A}_{d}(G_{d})$ for $n\geq d$. We finally define $\displaystyle\mathcal{Y}$ $\displaystyle=\lim_{d\to\infty}\mathcal{Y}_{d}(G_{d})\,,$ $\displaystyle\mathcal{A}$ $\displaystyle=\lim_{d\to\infty}\mathcal{A}_{d}(G_{d})\,.$ $\mathcal{Y}$ measures the probability of an epidemic and $\mathcal{A}$ measures the fraction infected. We will prove our results in the limit $n\to\infty$ by showing that $H_{in}(d)$ and $H_{out}(d)$ for a given $G_{n}$ are maximal or minimal under different conditions. This means that our results are generally true for arbitrary finite networks. The reason we use the large $n$ limit is because for networks which are small it is unclear what constitutes a giant component in an EPN, or similarly, for a network with some unusual structure on a size comparable to the network size (for example a network made up of a few disconnected components), a giant component may not be uniquely defined. Using the large $n$ limit avoids these problems. We could avoid the need for a limit by instead assuming the existence of a giant strongly connected component in the EPN and showing that the same conditions maximize or minimize the probability a node is in the in- or out-component of this giant strongly connected component. ## 3 Bounds in general networks We begin by considering the spread of infectious diseases on arbitrary networks. We begin with a simple lemma which we will need in this section and the next. ###### Lemma 3.1 (Edge Reversal) Given $T_{uv}=T(\mathcal{I}_{u},\mathcal{S}_{v})$, if we interchange the roles of infectiousness and susceptibility so that $T_{uv}=T(\mathcal{I}_{v},\mathcal{S}_{u})$ for all edges, then $\mathcal{Y}$ and $\mathcal{A}$ interchange roles. ###### Proof 3.2 If we replace $T_{uv}=T(\mathcal{I}_{u},\mathcal{S}_{v})$ with $\hat{T}_{uv}=T(\mathcal{I}_{v},\mathcal{S}_{u})$, then the new EPNs correspond to reversing the direction of edges in the original EPNs. Since reversing the direction of edges in an EPN interchanges $H_{in}(d)$ and $H_{out}(d)$, this interchanges $\mathcal{Y}$ and $\mathcal{A}$, and finishes the proof. ∎ We now make a simplifying assumption which we will need for networks with short cycles. [Ordering Assumption] If $T(\mathcal{I}_{1},\mathcal{S}_{1})>T(\mathcal{I}_{2},\mathcal{S}_{1})$ for any $\mathcal{S}_{1}$, then $T(\mathcal{I}_{1},\mathcal{S})\geq T(\mathcal{I}_{2},\mathcal{S})$ for all $\mathcal{S}$. Further, strict inequality occurs for a set of positive measure. Similarly if $T(\mathcal{I}_{1},\mathcal{S}_{1})>T(\mathcal{I}_{1},\mathcal{S}_{2})$ for any $\mathcal{I}_{1}$, then $T(\mathcal{I},\mathcal{S}_{1})\geq T(\mathcal{I},\mathcal{S}_{2})$ for all $\mathcal{I}$ with strict inequality for a set of positive measure. The ordering assumption is a statement about the functional form of $T(\mathcal{I},\mathcal{S})$. It places no restrictions on the network. The assumption holds for equation (1), but as noted earlier there are many scenarios where it fails. The ordering assumption implies that $T_{out}(\mathcal{I})$ and $T_{in}(\mathcal{S})$ are invertible mappings. It also allows us to assume that $\mathcal{I}$ is a scalar quantity ordered such that $\mathcal{I}_{u}\geq\mathcal{I}_{u^{\prime}}\quad\Leftrightarrow\quad T(\mathcal{I}_{u},\mathcal{S})\geq T(\mathcal{I}_{u^{\prime}},\mathcal{S})\>\>\>\forall\mathcal{S}\quad\Leftrightarrow\quad T_{out}(u)\geq T_{out}(u^{\prime})$ and further $\mathcal{I}_{u}>\mathcal{I}_{u^{\prime}}\Leftrightarrow T_{out}(u)>T_{out}(u^{\prime})$. We may make similar conclusions about $\mathcal{S}$. There will be more than one way to represent $\mathcal{I}$ or $\mathcal{S}$ as scalars. It will frequently (but not always) be convenient to identify $\mathcal{I}$ with $T_{out}(\mathcal{I})$ and $\mathcal{S}$ with $T_{in}(\mathcal{S})$. Previous work by [16] considered the spread of infectious diseases on networks for which the only heterogeneity came from variation in duration of infection. Hence all nodes have the same $T_{in}$, and variation occurs only in $T_{out}$. This model satisfies the ordering assumption. In this section we generalize the results of [16] by allowing $T_{in}$ and $T_{out}$ to be heterogeneous simultaneously. We will drop the ordering assumption in section 4 where we consider networks with no short cycles. Even in this section, many of the results hold without the ordering assumption, but the proofs are less clean. The assumption is only strictly needed for Theorems 3.5, 3.9, and 3.11. We are now ready to show that increased heterogeneity generally decreases the size and probability of epidemics. We show that for a given $Q_{in}$ [resp. $Q_{out}$], both $\mathcal{Y}$ and $\mathcal{A}$ are maximal when $T_{out}$ [resp. $T_{in}$] is homogeneous. They are minimal when the variance of $T_{out}$ [resp. $T_{in}$] is maximal subject to the constraint of $Q_{in}$ [resp. $Q_{out}$]. We can also derive conditions for a global upper bound on $\mathcal{Y}$ and $\mathcal{A}$. The upper bounds occur when $T_{uv}=\left\langle T\right\rangle$ for all neighbors $u$ and $v$. We hypothesize a lower bound, but cannot prove it in networks with short cycles. To make the notation cleaner in the following lemma, we identify $\mathcal{S}$ with $T_{in}$ and so we may use $T(\mathcal{I},T_{in})$ in place of $T(\mathcal{I},\mathcal{S})$. ###### Lemma 3.3 Assume a sequence of networks $\\{G_{n}\\}$ with sequential convergence of local statistics and a susceptibility distribution $Q_{in}(T_{in})$. Assume the ordering assumption holds and consider a distribution of infectiousness $P_{1}(\mathcal{I})$ with transmissibility given by $T_{1}(\mathcal{I},T_{in})$, that is consistent with $Q_{in}$. Let $\phi_{in,1}(V,\vec{\mathcal{S}})$ be as in equation (2). Let $\mathcal{A}_{1}$ and $\mathcal{Y}_{1}$ be the corresponding attack rate and epidemic probability. Similarly choose another $P_{2}(\mathcal{I})$, $T_{2}(\mathcal{I},T_{in})$ with corresponding $\mathcal{A}_{2}$, $\mathcal{Y}_{2}$, and $\phi_{2}(V,\vec{\mathcal{S}})$. Assume that $\phi_{in,1}(V,\vec{\mathcal{S}})\leq\phi_{in,2}(V,\vec{\mathcal{S}})$ for all $V$ and $\vec{\mathcal{S}}$. Then $\mathcal{A}_{1}\geq\mathcal{A}_{2}$ and $\mathcal{Y}_{1}\geq\mathcal{Y}_{2}$. ###### Proof 3.4 Let $d\geq 0$ be given. Take $G_{n}$, $n\geq d$. We will show that a node in an EPN created from $G_{n}$ using the first distribution is more likely to be in $H_{out}(d)$ than a node in an EPN created using the second distribution. Choose any node $u$ from $G_{n}$. Partition the nodes of $G_{n}$ into disjoint sets $\\{u\\}$, $U_{1}$, and $U_{2}$. To the nodes in $U_{1}$ we assign $\mathcal{I}$ from $P_{1}(\mathcal{I})$ and to the nodes in $U_{2}$ we assign $\mathcal{I}$ from $P_{2}(\mathcal{I})$. We assign $T_{in}$ to all nodes from $Q_{in}(T_{in})$. We will consider the effects of adding $u$ to $U_{1}$ versus adding it to $U_{2}$. Consider a partial EPN $\mathcal{E}$ created by assigning edges $(w,v)$ from all $w\neq u$, using $T_{1}(\mathcal{I}_{w},T_{i}(v))$ if $w\in U_{1}$ and $T_{2}(\mathcal{I}_{w},T_{i}(v))$ if $w\in U_{2}$. Now consider an arbitrary node $u^{\prime}$ (which may be $u$) which is not already in $H_{in}(d)$, but which would join $H_{in}(d)$ if the appropriate edges were added from $u$. Let $V$ be the set of neighbors $v$ of $u$ for which adding the edge $(u,v)$ would allow a path from $u^{\prime}$ to $u$ to be extended to a path of length $d$. We consider extensions of $\mathcal{E}$ formed by placing $u$ into $U_{1}$ or $U_{2}$. The probability that $u^{\prime}$ would be in $H_{in}(d)$ in the extended EPN is equal to the probability that $u$ has at least one edge to some node in $V$. This probability is at least as high if $u\in U_{1}$ as if $u\in U_{2}$ by our assumption $\phi_{in,1}(V,\vec{S})\leq\phi_{in,2}(V,\vec{S})$. Consequently the probability of $u^{\prime}$ to be in $H_{in}(d)$ is maximal if $u\in U_{1}$. Induction on $|U_{1}|$ shows that $\mathcal{Y}_{d}(G_{n})$ is largest if all nodes are in $U_{1}$. We now show that $u\in U_{1}$ increases $\mathcal{A}_{d}$ compared with $u\in U_{2}$. We can prove that placing $u$ in $U_{1}$ versus $U_{2}$ can only increase the probability of a node to be at the end of a length $d$ path. The proof proceeds largely as above. Consider the same partial EPN $\mathcal{E}$ defined above. Let $u^{\prime}$ be a node which is not in $H_{out}(d)$ but would be if an edge from $u$ to any $v\in V$ (note that $u\neq u^{\prime}$). The probability that $u^{\prime}$ will be in $H_{out}(d)$ is $\phi_{in,1}(V,\vec{\mathcal{S}})$ or $\phi_{in,2}(V,\vec{\mathcal{S}})$ depending on whether $u$ is assigned $\mathcal{I}$ from $P_{1}$ or $P_{2}$. Because $\phi_{in,1}(V,\vec{\mathcal{S}})\leq\phi_{in,2}(V,\vec{\mathcal{S}})$ it follows that $\mathcal{A}_{d}$ is largest if $u\in U_{1}$. Induction on $|U_{1}|$ shows $\mathcal{A}_{d}(G_{n})$ is maximal if all nodes are in $U_{1}$. Taking $d\to\infty$, it follows then that $\mathcal{Y}$ and $\mathcal{A}$ are maximal if all nodes are in $U_{1}$, and so the proof is finished. ∎ We begin by showing that for fixed distribution of in-transmissibility, the size and probability are largest when the out-transmissibility is homogeneous. ###### Theorem 3.5 Let $Q_{in}(T_{in})$ be given. Assume that the ordering assumption holds and that $\\{G_{n}\\}$ satisfies sequential convergence of statistics. Set $\mathcal{S}_{v}=T_{in}(v)$. Then $\mathcal{Y}$ and $\mathcal{A}$ are maximized when $T(\mathcal{I},T_{in})=T_{in}$. ###### Proof 3.6 By the ordering assumption, we may take $\mathcal{I}$ to be scalar with $\mathcal{I}_{1}>\mathcal{I}_{2}$ iff $T_{out}(\mathcal{I}_{1})>T_{out}(\mathcal{I}_{2})$. This allows us to use Chebyshev’s “other” inequality [15]: if $h_{1}$ and $h_{2}$ are decreasing functions of $x$ and $p$ is a probability density function, $\int h_{1}(x)h_{2}(x)p(x)\,dx\geq\left[\int h_{1}(x)p(x)\,dx\right]\left[\int h_{2}(x)p(x)\,dx\right]$ By induction $\int[\prod h_{j}(x)]p(x)\,dx\geq\prod\int h_{j}(x)p(x)\,dx$ for any number of decreasing functions $h_{j}$. Applying this to the decreasing function $h_{j}(\mathcal{I})=1-T(\mathcal{I},T_{in}(v_{j}))$ we have $\displaystyle\phi_{in}(V,\vec{\mathcal{S}})$ $\displaystyle=\int\left[\prod_{v\in V}h_{j}(\mathcal{I})\right]P(\mathcal{I})\,d\mathcal{I}\,,$ $\displaystyle\geq\prod_{v\in V}1-T_{in}(v)\,,$ with equality if $T(\mathcal{I},T_{in})=T_{in}$. Thus by Lemma 3.3, $\mathcal{A}$ and $\mathcal{Y}$ are maximal, completing the proof. ∎ We have proven the upper bounds given $Q_{in}(T_{in})$ occur when $T_{out}$ is homogeneous. We now show the lower bounds occur when $T_{out}$ is maximally heterogeneous. Because of the ordering assumption, we may take $\mathcal{S}_{v}=T_{in}(v)$. ###### Theorem 3.7 Let $Q_{in}(T_{in})$ be given, assume the ordering assumption holds, and assume that $\\{G_{n}\\}$ satisfies sequential convergence of statistics. Take $\mathcal{I}$ to be chosen uniformly from $[0,1]$. Setting $T(\mathcal{I},T_{in})=\begin{cases}0&T_{in}<\mathcal{I}\\\ 1&T_{in}>\mathcal{I}\end{cases}$ (5) minimizes $\mathcal{Y}$ and $\mathcal{A}$. ###### Proof 3.8 Given equation (5), we have $\phi_{in}(V,\vec{\mathcal{S}})=\min_{v\in V}\\{1-T_{in}(v)\\}$. We need to prove that for any arbitrary transmission function $\hat{T}(\mathcal{I},T_{in})$ satisfying the ordering assumption and consistent with $T_{in}$, $\phi_{in}(V,\vec{\mathcal{S}})\leq\min_{v\in V}\\{1-T_{in}(v)\\}$. To do this, let $\hat{T}$ be given, $T_{in}$ assigned to $v_{1}$, …, $v_{n}$ and assume $v_{1}$, …, $v_{n}$ are ordered such that $T_{in}(v_{1})\geq T_{in}(v_{2})\geq\cdots\geq T_{in}(v_{n})$. Then $\displaystyle\phi_{in}(V,\vec{\mathcal{S}})$ $\displaystyle=\int\prod_{j=1}^{n}[1-\hat{T}(\mathcal{I},T_{in}(v_{j}))]P(\mathcal{I})d\mathcal{I}\,,$ $\displaystyle\leq\int[1-\hat{T}(\mathcal{I},T_{in}(v_{1}))]P(\mathcal{I})d\mathcal{I}\,,$ $\displaystyle\leq 1-T_{in}(v_{1})\,.$ This shows that for any $\hat{T}$, $\phi_{in}(V,\vec{\mathcal{S}})$ is at most the value it takes for (5). Thus Lemma 3.3 shows that $\mathcal{Y}$ and $\mathcal{A}$ are minimal, completing the proof. ∎ We derived the results above with fixed $Q_{in}$. Lemma 3.1 shows that the equivalent results must hold for $Q_{out}$. ###### Theorem 3.9 Let $Q_{out}(T_{out})$ be given. Assume that the ordering assumption holds and that $\\{G_{n}\\}$ satisfies sequential convergence of statistics. Set $\mathcal{I}_{u}=T_{out}(u)$. * • If $T(T_{out},\mathcal{S})=T_{out}$ Then $\mathcal{Y}$ and $\mathcal{A}$ are maximized. * • If $\mathcal{S}$ is chosen uniformly in $[0,1]$ and $T(T_{out},\mathcal{S})=\begin{cases}0&T_{out}<\mathcal{S}\\\ 1&T_{out}>\mathcal{S}\,,\end{cases}$ then $\mathcal{Y}$ and $\mathcal{A}$ are minimized. ###### Proof 3.10 This follows immediately from Lemma 3.1 with Theorems 3.5 and 3.7. ∎ We now give a global upper bound for both $\mathcal{Y}$ and $\mathcal{A}$. ###### Theorem 3.11 Let $\left\langle T\right\rangle$ be given. Under the ordering assumption with sequential convergence of statistics for $\\{G_{n}\\}$, the maximum of $\mathcal{Y}$ and $\mathcal{A}$ occur when $T_{uv}=\left\langle T\right\rangle$ for all neighboring nodes. ###### Proof 3.12 Consider a $P(\mathcal{I})$, $P(\mathcal{S})$, and $T(\mathcal{I},\mathcal{S})$ which yields a global maximum for either $\mathcal{Y}$ or $\mathcal{A}$. If $T_{in}$ is not homogeneous, then we can find a new infection process which preserves the same $Q_{out}(T_{out})$ with homogeneous $T_{in}$ which can only increase $\mathcal{Y}$ or $\mathcal{A}$. A repeated application preserving the new homogeneous in-transmissibility, but now making $T_{out}$ also homogeneous again can only increase $\mathcal{Y}$ or $\mathcal{A}$. $T_{out}$ and $T_{in}$ are then homogeneous. This completes the proof.∎ We finish with a conjecture about global lower bounds. ###### Conjecture 3.13 Under the ordering assumption with sequential convergence of statistics for $\\{G_{n}\\}$, the minimum of $\mathcal{Y}$ occurs when $Q_{out}(T_{out})=\left\langle T\right\rangle\delta(T_{out}-1)+(1-\left\langle T\right\rangle)\delta(T_{out})$. The minimum of $\mathcal{A}$ occurs when $Q_{in}(T_{in})=\left\langle T\right\rangle\delta(T_{in}-1)+(1-\left\langle T\right\rangle)\delta(T_{in})$. Note that if $Q_{out}(T_{out})=\left\langle T\right\rangle\delta(T_{out}-1)+(1-\left\langle T\right\rangle)\delta(T_{out})$, then $Q_{in}(T_{in})=\delta(T_{in}-\left\langle T\right\rangle)$ is homogeneous. ### 3.1 Discussion The results of this section have focused on extending earlier results of Kuulasmaa [16] who considered a population with homogeneous susceptibility and heterogeneities in infectiousness due entirely to variation in duration of infection. We have extended these results to cover a wide range of heterogeneities in infectiousness and susceptibility (simultaneously), under the assumption that infectiousness and susceptibility are assigned independently. In order to extend the proof used by Kuulasmaa, we have been forced to make the ordering assumption, which effectively means that if we order people by how infectious they would be to one test susceptible individual, the order is the same as we would find for another test susceptible individual. We do not have any counter-examples to these theorems in the case where the ordering assumption fails, and so it is not clear that it is needed. In section 4 we will see that similar results hold in unclustered networks without needing the ordering assumption. Our results show that in general increasing the heterogeneity of the population is useful for either decreasing the size or decreasing the probability that an epidemic occurs. Given $Q_{in}$ [resp. $Q_{out}$], both $\mathcal{Y}$ and $\mathcal{A}$ are maximized if $T_{out}$ [resp. $T_{in}$] is homogeneous and minimized if it is maximally heterogeneous. Similarly, given just $\left\langle T\right\rangle$, we find that the global maxima of $\mathcal{Y}$ and $\mathcal{A}$ occur when $T=\left\langle T\right\rangle$. Perhaps surprisingly, the conditions leading to upper and lower bounds are independent of the network, though the size of the variation between these bounds is network-dependent. Although we can prove lower bounds given $Q_{in}$ [or $Q_{out}$], we cannot prove global lower bounds given $\left\langle T\right\rangle$. We hypothesize that the global lower bound for $\mathcal{Y}$ occurs when $Q_{out}$ is maximally heterogeneous and the global lower bound for $\mathcal{A}$ occurs when $Q_{in}$ is maximally heterogeneous. In the next section we will see that these are the lower bounds for an unclustered population. However, we have not found a rigorous proof for general networks. In the proof of the upper bound, we took a given $Q_{out}$ and found $Q_{in}$ that maximizes $\mathcal{Y}$ and $\mathcal{A}$. We then held that $Q_{in}$ fixed and found $Q_{out}$ to maximize, arriving at the upper bound. However, applying a similar technique to the lower bound fails because given any $Q_{out}$, if we find a minimizing $Q_{in}$, attempting to then minimize with $Q_{in}$ fixed simply returns the original $Q_{out}$. The difficulty results from the fact that increasing heterogeneity in $T_{out}$ restricts the amount of heterogeneity in $T_{in}$ and _vice versa_. ## 4 Bounds in unclustered networks Most studies of infectious diseases spreading on networks have been made for networks for which the effect of short cycles may be neglected [25]. These investigations have generally used Molloy–Reed networks [23] (also known as the configuration model [26]). The theory we develop here applies to these networks, but also to more general networks which may have degree-degree correlations, or even longer range correlations. When we study networks with no short cycles, we are able to prove stronger results and abandon the ordering assumption. We find that $\mathcal{Y}$ depends on the network and $Q_{out}(T_{out})$ only, while $\mathcal{A}$ depends on the network and $Q_{in}(T_{in})$ only. We can prove global upper and (unlike in the general case) lower bounds on $\mathcal{Y}$ and $\mathcal{A}$. [Unclustered Assumption] Given a sequence of networks $\\{G_{n}\\}$, we assume that $G_{n}$ has girth greater than $2n$. This assumption means that $B_{d}(u)$ chosen from any $G_{n}$ with $n\geq d$ must be cycle free. In particular, there is no alternate path between a node and a neighbor. It was this complication that forced the use of the ordering assumption earlier, and since the complication no longer exists, we drop the ordering assumption. The unclustered assumption will also allow us to use $\psi_{in}(V)$ rather than $\phi_{in}(V,\vec{S})$. Thus we only require the marginal probability of the set of nodes $V$ not to be infected to satisfy an inequality, rather than the inequality be satisfied for every possible set of susceptibilities. We must bear in mind that knowing $T_{in}$ or $T_{out}$ no longer uniquely determines $\mathcal{I}$ or $\mathcal{S}$. ###### Lemma 4.1 Let the sequence $\\{G_{n}\\}$ satisfy the unclustered assumption with sequential convergence of statistics. Take $P_{1}(\mathcal{I})$, $P_{1}(\mathcal{S})$ and $T_{1}(\mathcal{I},\mathcal{S})$. Let $\psi_{in,1}(V)$ be as in equation (4). Similarly take $P_{2}(\mathcal{I})$, $P_{2}(\mathcal{S})$, and $T_{2}(\mathcal{I},\mathcal{S})$ with corresponding $\psi_{in,2}(V)$. If $\psi_{in,1}(V)\leq\psi_{in,2}(V)$ then $\mathcal{Y}_{1}\geq\mathcal{Y}_{2}$. ###### Proof 4.2 This proof is similar to that of Lemma 3.3. Let $d\geq 0$ be given. Take $G_{n}$, $n\geq d$. Choose a node $u$ from $G_{n}$ and partition the nodes of $G_{n}$ into $\\{u\\}$, $U_{1}$, and $U_{2}$. To the nodes in $U_{1}$ we assign $\mathcal{I}$ from $P_{1}(\mathcal{I})$ and to the nodes of $U_{2}$ we assign $\mathcal{I}$ from $P_{2}(\mathcal{I})$. To each node $w$ (including $u$), we assign two susceptibilities, $\mathcal{S}_{w,1}$ and $\mathcal{S}_{w,2}$ such that $\mathcal{S}_{w,1}$ comes from $P_{1}(\mathcal{S})$ and $\mathcal{S}_{w,2}$ comes from $P_{2}(\mathcal{S})$. We create a partial EPN $\mathcal{E}$ as follows. For each $v\in U_{1}$, we assign edges $(v,w)$ using $T_{1}(\mathcal{I}_{v},\mathcal{S}_{w,1})$, and for $v\in U_{2}$ we assign them using $T_{2}(\mathcal{I}_{v},\mathcal{S}_{w,2})$. We do not yet assign edges from $u$ (but edges may point to $u$). Consider any $u^{\prime}$ not in $H_{in}(d)$ which would join $H_{in}(d)$ if an edge was added from $u$ to any $v\in V$. By assumption, $\psi_{in,1}(V)\leq\psi_{in,2}(V)$ and so the probability is greatest if $\mathcal{I}_{u}$ is chosen from $P_{1}(\mathcal{I})$. It follows that $\mathcal{Y}_{1}\geq\mathcal{Y}_{2}$. This completes the proof. ∎ This proof can be modified to work on clustered networks without the ordering assumption, so Lemma 3.3 does not require the ordering assumption. However, the proof is more technical and provides little additional insight, particularly because the main results following from Lemma 3.3 do require the ordering assumption. ###### Theorem 4.3 Let the sequence $\\{G_{n}\\}$ satisfy the unclustered assumption with sequential convergence of local statistics. Let $Q_{in}(T_{in})$ be fixed. Then $\mathcal{A}$ is fixed. ###### Proof 4.4 We follow the technique used to prove $\mathcal{A}$ is larger for one distribution than the other in Lemma 3.3. However, in following that proof, the lack of clustering means $|V|=1$. Since for any distribution $\psi(V)=1-\left\langle T\right\rangle$ when $|V|=1$, all distributions must give the same $\mathcal{A}$, and the proof is finished. ∎ ###### Theorem 4.5 If the assumptions of Theorem 4.3 hold except that $Q_{out}$ is fixed rather than $Q_{in}$, then $\mathcal{Y}$ is fixed. ###### Proof 4.6 This follows immediately from Lemma 3.1 and Theorem 4.3. ∎ ###### Theorem 4.7 Let $Q_{in}$ be given. Assume that $\\{G_{n}\\}$ satisfies the unclustered assumption with sequential convergence of statistics. $\mathcal{Y}$ is maximized when $T(\mathcal{I},\mathcal{S})=T_{in}(\mathcal{S})$. Although this result is analogous to Theorem 3.5, the proof is fundamentally altered because we no longer have the ordering assumption. ###### Proof 4.8 We first note that if $T(\mathcal{I},\mathcal{S})=T_{in}(\mathcal{S})$, then $T_{out}=\left\langle T\right\rangle$ for all nodes. Now consider an arbitrary function $T(\mathcal{I},\mathcal{S})$ with $P(\mathcal{I})$ and $P(\mathcal{S})$ to satisfy $Q_{in}(T_{in})$. The function $(1-T_{out})^{|V|}$ in equation (4) is convex, so by Jensen’s inequality $\psi_{in}$ is minimized by $T_{out}=\left\langle T\right\rangle$. Lemma 4.1 completes the proof. ∎ ###### Theorem 4.9 Let $Q_{in}$ be given. Assume $\\{G_{n}\\}$ satisfies the unclustered assumption with sequential convergence of statistics. $\mathcal{Y}$ is minimized when $\mathcal{I}$ is chosen uniformly from $[0,1]$ and $T(\mathcal{I},\mathcal{S})=\begin{cases}0&\mathcal{I}>T_{in}(\mathcal{S})\\\ 1&\mathcal{I}<T_{in}(\mathcal{S})\,.\end{cases}$ ###### Proof 4.10 Following the proof of Theorem 3.7, we may show that $\phi_{in}$ is maximized (subject to $Q_{in}$) exactly when these assumptions hold. Thus from equation (3) $\psi_{in}$ is also maximized when these assumptions hold. Lemma 4.1 completes the proof. ∎ ###### Theorem 4.11 Let $Q_{out}$ be given. Assume $\\{G_{n}\\}$ satisfies the unclustered assumption with sequential convergence of statistics. * • $\mathcal{A}$ is maximized when $T(\mathcal{I},\mathcal{S})=T_{out}(\mathcal{I})$. * • $\mathcal{A}$ is minimized when $\mathcal{S}$ is chosen uniformly from $[0,1]$ and $T(\mathcal{I},\mathcal{S})=\begin{cases}0&\mathcal{S}>T_{out}(\mathcal{I})\\\ 1&\mathcal{S}<T_{out}(\mathcal{I})\,.\end{cases}$ ###### Proof 4.12 This follows from Theorems 4.7 and 4.9 with Lemma 3.1. ∎ Before proving our final result, we introduce a lemma. ###### Lemma 4.13 Let $f(x)$ be a convex function on $[0,1]$ and $\rho(x)$ be a probability density function on $[0,1]$, with expected value $\rho_{0}$. Then $\int f(x)\rho(x)dx\leq(1-\rho_{0})f(0)+\rho_{0}f(1)\,.$ ###### Proof 4.14 The definition of convexity gives $f(x)\leq(1-x)f(0)+xf(1)\,.$ Thus $\int f(x)\rho(x)dx\leq\int[(1-x)f(0)\rho(x)+xf(1)\rho(x)]dx\leq(1-\rho_{0})f(0)+\rho_{0}f(1)\,.$ ∎ ###### Theorem 4.15 Let $\\{G_{n}\\}$ be a sequence of networks satisfying the unclustered assumption with sequential convergence of statistics. Assume that $\left\langle T\right\rangle$ is given: * • The global upper bound for both $\mathcal{Y}$ and $\mathcal{A}$ occurs when $T_{uv}=\left\langle T\right\rangle$ for all pairs of neighbors. * • The global lower bound for $\mathcal{Y}$ occurs when $Q_{out}(T_{out})=\left\langle T\right\rangle\delta(T_{out}-1)+(1-\left\langle T\right\rangle)\delta(T_{out})$. * • The global lower bound for $\mathcal{A}$ occurs when $Q_{in}(T_{in})=\left\langle T\right\rangle\delta(T_{in}-1)+(1-\left\langle T\right\rangle)\delta(T_{in})$. ###### Proof 4.16 The proof of the upper bound is identical to that of Theorem 3.11. We prove the lower bound for $\mathcal{Y}$. The lower bound for $\mathcal{A}$ follows from Lemma 3.1. We have $\psi_{in}(V)=\int(1-T_{out})^{|V|}Q_{out}(T_{out})dT_{out}$. We now seek to find $Q_{out}$ which maximizes $\psi_{in}$ in order to apply Lemma 4.1. Since $(1-T_{out})^{|V|}$ is a convex function, we may apply Lemma 4.13 with $Q_{out}$ playing the role of $\rho$. The maximum occurs when $Q_{out}(T_{out})=\left\langle T\right\rangle\delta(T_{out}-1)+(1-\left\langle T\right\rangle)\delta(T_{out})$ and so Lemma 4.1 finishes the proof. ∎ Although the upper bound for both $\mathcal{Y}$ and $\mathcal{A}$ occurs when $T_{uv}=\left\langle T\right\rangle$ for all pairs, our earlier results show that for unclustered networks $\mathcal{Y}$ depends only on $Q_{out}(T_{out})$ and the network, and so as long as $T_{out}(u)=\left\langle T\right\rangle$ for all nodes $u$, we achieve the upper bound on $\mathcal{Y}$ (but not on $\mathcal{A}$). Symmetrically, if $T_{in}(v)=\left\langle T\right\rangle$ for all nodes $v$, we achieve the upper bound on $\mathcal{A}$. Note that the lower bound for $\mathcal{A}$ requires that $T_{out}(u)=\left\langle T\right\rangle$ for all $u$, and so the population is homogeneously infectious. It follows from Theorem 4.9 that $\mathcal{Y}$ is then maximal. Similarly the lower bound for $\mathcal{Y}$ requires that $\mathcal{A}$ be maximal. ### 4.1 Discussion The results of this section generalize those of [21] which considered the special case of Molloy–Reed networks. These results prove that the same scenarios give upper and lower bounds in unclustered networks with a wide range of correlations including assortative or disassortative mixing (high degree nodes preferentially joining with high or low degree nodes respectively), or longer range correlations. Although we proved these under the assumption that no short cycles exist, the results remain useful in networks with either few short cycles, or for situations in which the transmissibility is low enough that the short cycles are only rarely followed. Because of the lack of short cycles, the ordering assumption is not needed. This means that our results apply to a much wider class of disease transmission mechanisms, but at the cost of restricting the network. Again we find that which conditions give the upper or lower bound is network- independent. The amount of variation there is between these bounds, however, is network-dependent. The main distinction from clustered networks is that $\mathcal{Y}$ depends only on the network structure and $Q_{out}(T_{out})$. That is, $\mathcal{Y}$ is independent of $Q_{in}(T_{in})$. Similarly $\mathcal{A}$ depends only on the structure and $Q_{in}(T_{in})$. We note that unless the effect of clustering is very large, the dependence of $\mathcal{Y}$ on in- transmissibility and $\mathcal{A}$ on out-transmissibility will be weak. Curiously the global lower bound for $\mathcal{A}$ found in Theorem 4.15 requires that $Q_{out}(T_{out})=\delta(T_{out}-\left\langle T\right\rangle)$, and so the population is homogeneously susceptible. It follows from Theorem 4.9 that $\mathcal{Y}$ is then maximal. Similarly the lower bound for $\mathcal{Y}$ requires that $\mathcal{A}$ be maximal. This has important implications for policy design because strategies to reduce $T$ tend to have a heterogeneous impact on either $\mathcal{S}$ or $\mathcal{I}$. ## 5 Conclusions We have extended earlier work on the effect of heterogeneity in infectiousness on the spread of epidemics through networks. Our extensions allow for heterogeneity in susceptibility as well. Many of the results are similar. In general we find that the size and probability of epidemics are reduced if the population is more heterogeneous. Unfortunately, increasing heterogeneity in susceptibility restricts the level of heterogeneity possible in infectiousness. In the extreme case where susceptibility is maximally heterogeneous, infectiousness must be homogeneous. Perhaps surprisingly, we have found that the distributions leading to upper and lower bounds on $\mathcal{Y}$ and $\mathcal{A}$ are network-independent. Early in an outbreak, it is likely that we may gain some information about $Q_{out}(T_{out})$. For example, in the early stages of the SARS epidemic, it was known that a number of people were highly infectious, while the rest were only mildly infectious and so $Q_{out}(T_{out})$ was highly heterogeneous. However, there was little information on $Q_{in}$. Once given the distribution of $Q_{out}$, the results here show which distributions of $Q_{in}$ give the largest or smallest $\mathcal{Y}$ and $\mathcal{A}$. Our results further suggest that the distribution of infectiousness found for SARS is consistent with a low epidemic probability. It is difficult to extrapolate from observations what the sizes would have been without the interventions put into place, but the fact that a number of isolated cases occurred throughout the world without sparking local epidemics suggest that the probability of an epidemic from each introduction was low, consistent with our predictions. Our results further suggest that in order to prevent an epidemic, it is best to take measures that will have a heterogeneous impact on infectiousness, but in order to affect the size of an epidemic, it is best to take measures that will have a heterogeneous impact on susceptibility. In terms of actual interventions, we compare two strategies aimed at controlling a disease which is initially spreading with homogeneous $T$: in the first we devote resources to vaccinating half of the population, while in the second we devote them to identifying and removing half of the infected population. Both strategies reduce $\left\langle T\right\rangle$ by a factor of $2$. In the first, the susceptibility is highly heterogeneous, but the probability an infected node infects a randomly chosen neighbor has simply gone down by a factor of $2$, and so it remains homogeneous. Assuming that the unclustered approximation is valid, this maximizes the impact on size, but the impact on probability is minimized. In contrast, the second strategy maximizes the impact on probability, but minimizes the impact on size. ## Acknowledgments This work was supported in part by the Division of Mathematical Modeling at the UBC CDC and by DOE at LANL under Contract DE-AC52-06NA25396 and the DOE Office of ASCR program in Applied Mathematical Sciences. ## References * [1] Abbey, H. (1952). An examination of the Reed-Frost theory of epidemics. Human Biology 24, 201–233. * [2] Anderson, R. M. and May, R. M. (1991). Infectious Diseases of Humans. Oxford University Press, Oxford. * [3] Andersson, H. (1999). Epidemic models and social networks. Math. Scientist 24, 128–147. * [4] Ball, F. (1985). Deterministic and stochastic epidemics with several kinds of susceptibles. Advances in Applied Probability 17, 1–22. * [5] Ball, F. and O’Neill, P. (1999). The distribution of general final state random variables for stochastic epidemic models. Journal of Applied Probability 36, 473–491. * [6] Broder, A., Kumar, R., Maghoul, F., Raghavan, P., Rajagopalan, S., Stata, R., Tomkins, A. and Wiener, J. (2000). Graph structure in the web. Computer Networks 33, 309–320. * [7] Del Valle, S. Y., Hyman, J. M., Hethcote, H. W. and Eubank, S. G. (2007). Mixing patterns between age groups in social networks. Social Networks 29, 539–554. * [8] Eubank, S., Guclu, H., Kumar, V. S. A., Marathe, M. V., Srinivasan, A., Toroczkai, Z. and Wang, N. (2004). Modelling disease outbreaks in realistic urban social networks. Nature 429, 180–184. * [9] Hastings, M. B. (2006). Systematic series expansions for processes on networks. Physical Review Letters 96, 148701\. * [10] Keeling, M. J. (2005). The implications of network structure for epidemic dynamics. Theoretical Population Biolology 67, 1–8. * [11] Keeling, M. J. and Eames, K. T. D. (2005). Networks and epidemic models. Journal of the Royal Society Interface 2, 295–307. * [12] Kenah, E. and Robins, J. M. (2007). Network-based analysis of stochastic SIR epidemic models with random and proportionate mixing. Journal of Theoretical Biology. * [13] Kenah, E. and Robins, J. M. (2007). Second look at the spread of epidemics on networks. Physical Review E 76, 36113\. * [14] Kermack, W. O. and McKendrick, A. G. (1927). A contribution to the mathematical theory of epidemics. Royal Society of London Proceedings Series A 115, 700–721. * [15] Kingman, J. F. C. (1978). Uses of exchangeability. The Annals of Probability 6, 183–197. * [16] Kuulasmaa, K. (1982). The spatial general epidemic and locally dependent random graphs. Journal of Applied Probability 19, 745–758. * [17] Kuulasmaa, K. and Zachary, S. (1984). On spatial general epidemics and bond percolation processes. Journal of Applied Probability 21, 911–914. * [18] Madar, N., Kalisky, T., Cohen, R., ben Avraham, D. and Havlin, S. (2004). Immunization and epidemic dynamics in complex networks. The European Physical Journal B 38, 269–276. * [19] Meyers, L. A. (2007). Contact network epidemiology: Bond percolation applied to infectious disease prediction and control. Bulletin of the American Mathematical Society 44, 63–86. * [20] Meyers, L. A., Newman, M. and Pourbohloul, B. (2006). Predicting epidemics on directed contact networks. Journal of Theoretical Biology 240, 400–418. * [21] Miller, J. C. (2007). Epidemic size and probability in populations with heterogeneous infectivity and susceptibility. Physical Review E 76, 010101\. * [22] Mollison, D. (1977). Spatial contact models for ecological and epidemic spread. Journal of the Royal Statistical Society. Series B (Methodological) 39, 283–326. * [23] Molloy, M. and Reed, B. (1995). A critical point for random graphs with a given degree sequence. Random structures & algorithms 6, 161–179. * [24] Neal, P. (2007). Copuling of two SIR epidemic models with variable susceptibilities and infectivities. Journal of Applied Probability 44, 41–57. * [25] Newman, M. E. J. (2002). Spread of epidemic disease on networks. Physical Review E 66, 16128\. * [26] Newman, M. E. J. (2003). The structure and function of complex networks. SIAM Review 45, 167–256. * [27] Pastor-Satorras, R. and Vespignani, A. (2001). Epidemic spreading in scale-free networks. Physical Review Letters 86, 3200–3203. * [28] Serrano, M. and Boguñá, M. (2006). Percolation and epidemic thresholds in clustered networks. Physical Review Letters 97, 088701\. * [29] Trapman, P. (2007). On analytical approaches to epidemics on networks. Theoretical Population Biology 71, 160–173. * [30] van den Berg, J., Grimmett, G. R. and Schinazi, R. B. (1998). Dependent random graphs and spatial epidemics. The Annals of Applied Probability 8, 317–336.

arxiv-papers

2024-09-04T02:48:54.211165

0803.1117

# Production of heavy and superheavy nuclei in massive fusion reactions Zhao-Qing Feng1111Corresponding author. _E-mail address:_ fengzhq@impcas.ac.cn, Gen-Ming Jin1, Jun-Qing Li1, Werner Scheid2 1 _Institute of Modern Physics, Chinese Academy of Sciences, Lanzhou 730000, China_ 2 _Institut für Theoretische Physik der Universität, 35392 Giessen, Germany_ Abstract Within the framework of a dinuclear system (DNS) model, the evaporation- residue excitation functions and the quasi-fission mass yields in the 48Ca induced fusion reactions are investigated systematically and compared with available experimental data. Maximal production cross sections of superheavy nuclei based on stable actinide targets are obtained. Isotopic trends in the production of the superheavy elements Z=110, 112-118 based on the actinide isotopic targets are analyzed systematically. Optimal evaporation channels and combinations as well as the corresponding excitation energies are proposed. The possible factors that influencing the isotopic dependence of the production cross sections are analyzed. The formation of the superheavy nuclei based on the isotopes U with different projectiles are also investigated and calculated. _PACS:_ 25.70.Jj, 24.10.-i, 25.60.Pj _Keywords:_ DNS model; evaporation-residue excitation functions; 48Ca induced fusion reactions; isotopic trends ## 1 Introduction The synthesis of heavy or superheavy nuclei is a very important subject in nuclear physics motivated with respect to the island of stability which is predicted theoretically, and has obtained much experimental research with the fusion-evaporation reactions [1, 2]. The existence of the superheavy nucleus (SHN) ($Z\geq 106$) is due to strong binding shell effects against the large Coulomb repulsion. However, the shell effects get reduced with increasing the excitation energy of the formed compound nucleus. Combinations with a doubly magic nucleus or nearly magic nucleus are usually chosen owing to the larger reaction $Q$ values. Reactions with 208Pb or 209Bi targets were first proposed by Oganessian et al. to synthesize SHN [3]. Six new elements with Z=107-112 were synthesized in cold fusion reactions for the first time and investigated at GSI (Darmstadt, Germany) with the heavy-ion accelerator UNILAC and the SHIP separator [1, 4]. Recently, experiments on the synthesis of element 113 in the 70Zn+209Bi reaction have been performed successfully at RIKEN (Tokyo, Japan) [5]. However, it is difficulty to produce heavier SHN in the cold fusion reactions because of the smaller production cross sections that are lower than 1 pb for $Z>113$. Other possible ways to produce SHN are very needed to be investigated in experimentally and theoretically. Recently, the superheavy elements Z=113-116, 118 were synthesized at FLNR in Dubna (Russia) with the double magic nucleus 48Ca bombarding actinide nuclei [6, 7, 8]. New heavy isotopes 259Db and 265Bh have also been synthesized at HIRFL in Lanzhou (China) [9]. Further experimental works are necessary in order to testify the new synthesized SHN. A reasonable understanding of the formation of SHN in the massive fusion reactions is still a challenge for theory. In accordance with the evolution of two heavy colliding nuclei, the dynamical process of the compound nucleus formation and decay is usually divided into three reaction stages, namely the capture process of the colliding system to overcome the Coulomb barrier, the formation of the compound nucleus to pass over the inner fusion barrier, and the de-excitation of the excited compound nucleus by neutron emission against fission. The transmission in the capture process depends on the incident energy and relative angular momentum of the colliding nuclei, which is the same as that in the fusion of light and medium mass systems. The complete fusion of the heavy system after capture in competition with quasi-fission is very important in the estimation of the SHN production. The concept of the ”extra-push” energy explains for the fusion of two heavy colliding nuclei in the macroscopic dynamical model [10, 11]. At present it is still difficult to make an accurate description of the fusion dynamics. After the capture and the subsequent evolution to form the compound nucleus, the thermal compound nucleus will decay by the emission of light particles and $\gamma$ rays against fission. The three stages will affect the formation of evaporation residues observed in laboratories. The evolution of the whole process of massive heavy-ion collisions is very complicated at near- barrier energies. Most of the theoretical methods on the formation of SHN have a similar viewpoint in the description of the capture and the de-excitation stages, but there are different description of the compound nucleus formation process. There are mainly two sorts of models, depending on whether the compound nucleus is formed along the radial variable (internuclear distance) or by nucleon transfer in a touching configuration which is usually the minimum position of the interaction potential after capture of the colliding system. Several transport models have been established to understand the fusion mechanism of two heavy colliding nuclei leading to SHN formation, such as the macroscopic dynamical model [10, 11], the fluctuation-dissipation model [12], the concept of nucleon collectivization [13] and the dinuclear system model [14, 15]. Recently, the improved isospin-dependent quantum molecular dynamics (ImIQMD) model was also proposed to investigate the fusion dynamics of SHN [16, 17]. With these models experimental data can be reproduced to a certain extent, and some new results have been predicted. However, these models differ from each other, and sometimes different physical ideas are used. Further improvements of these models have to be made. Here we use a dinuclear system (DNS) model [15, 18], in which the nucleon transfer is coupled with the relative motion by solving a set of microscopically derived master equations, and a barrier distribution of the colliding system is introduced in the model. We present a new and extended investigation of the production of superheavy nuclei in the 48Ca induced fusion reactions and in other combinations. In Section 2 we give a simple description on the DNS model. Calculated results of fusion dynamics and SHN production are given in Section 3. In Section 4 conclusions are discussed. ## 2 Dinuclear system model The dinuclear system [19] is a molecular configuration of two touching nuclei which keep their own individuality [14]. Such a system has an evolution along two main degrees of freedom: (i) the relative motion of the nuclei in the interaction potential to form the DNS and the decay of the DNS (quasi-fission process) along the R degree of freedom (internuclear motion), (ii) the transfer of nucleons in the mass asymmetry coordinate $\eta=(A_{1}-A_{2})/(A_{1}+A_{2})$ between two nuclei, which is a diffusion process of the excited systems leading to the compound nucleus formation. Off- diagonal diffusion in the surface $(A_{1},R)$ is not considered since we assume the DNS is formed at the minimum position of the interaction potential of two colliding nuclei. In this concept, the evaporation residue cross section is expressed as a sum over partial waves with angular momentum $J$ at the centre-of-mass energy $E_{c.m.}$, $\sigma_{ER}(E_{c.m.})=\frac{\pi\hbar^{2}}{2\mu E_{c.m.}}\sum_{J=0}^{J_{max}}(2J+1)T(E_{c.m.},J)P_{CN}(E_{c.m.},J)W_{sur}(E_{c.m.},J).$ (1) Here, $T(E_{c.m.},J)$ is the transmission probability of the two colliding nuclei overcoming the Coulomb potential barrier in the entrance channel to form the DNS. In the same manner as in the nucleon collectivization model [13], the transmission probability $T$ is calculated by using the empirical coupled channel model, which can reproduce very well available experimental capture cross sections [13, 15]. The $P_{CN}$ is the probability that the system will evolve from a touching configuration into the compound nucleus in competition with quasi-fission of the DNS and fission of the heavy fragment. The last term is the survival probability of the formed compound nucleus, which can be estimated with the statistical evaporation model by considering the competition between neutron evaporation and fission [15]. We take the maximal angular momentum as $J_{max}=30$ since the fission barrier of the heavy nucleus disappears at high spin [20]. In order to describe the fusion dynamics as a diffusion process in mass asymmetry, the analytical solution of the Fokker-Planck equation [14] and the numerical solution of the master equations [21, 22] have been used, which were also used to treat deep inelastic heavy-ion collisions [23]. Here, the fusion probability is obtained by solving a set of master equations numerically in the potential energy surface of the DNS. The time evolution of the distribution function $P(A_{1},E_{1},t)$ for fragment 1 with mass number $A_{1}$ and excitation energy $E_{1}$ is described by the following master equations [18, 21], $\displaystyle\frac{dP(A_{1},E_{1},t)}{dt}=\sum_{A_{1}^{\prime}}W_{A_{1},A_{1}^{\prime}}(t)\left[d_{A_{1}}P(A_{1}^{\prime},E_{1}^{\prime},t)-d_{A_{1}^{\prime}}P(A_{1},E_{1},t)\right]-$ $\displaystyle\left[\Lambda^{qf}(\Theta(t))+\Lambda^{fis}(\Theta(t))\right]P(A_{1},E_{1},t).$ (2) Here $W_{A_{1},A_{1}^{\prime}}$ is the mean transition probability from the channel $(A_{1},E_{1})$ to $(A_{1}^{\prime},E_{1}^{\prime})$, and $d_{A_{1}}$ denotes the microscopic dimension corresponding to the macroscopic state $(A_{1},E_{1})$. The sum is taken over all possible mass numbers that fragment $A_{1}^{\prime}$ may take (from $0$ to $A=A_{1}+A_{2}$), but only one nucleon transfer is considered in the model with $A_{1}^{\prime}=A_{1}\pm 1$. The excitation energy $E_{1}$ is the local excitation energy $\varepsilon^{\ast}_{1}$ with respect to fragment $A_{1}$, which is determined by the dissipation energy from the relative motion and the potential energy of the corresponding DNS and will be shown later in Eqs.(8) and (9). The dissipation energy is described by the parametrization method of the classical deflection function [24, 25]. The motion of nucleons in the interacting potential is governed by the single-particle Hamiltonian [15, 21]: $H(t)=H_{0}(t)+V(t)$ (3) with $\displaystyle H_{0}(t)$ $\displaystyle=$ $\displaystyle\sum_{K}\sum_{\nu_{K}}\varepsilon_{\nu_{K}}(t)a_{\nu_{K}}^{{\dagger}}(t)a_{\nu_{K}}(t),$ $\displaystyle V(t)$ $\displaystyle=$ $\displaystyle\sum_{K,K^{\prime}}\sum_{\alpha_{K},\beta_{K^{\prime}}}u_{\alpha_{K},\beta_{K^{\prime}}}(t)a_{\alpha_{K}}^{{\dagger}}(t)a_{\beta_{K^{\prime}}}(t)=\sum_{K,K^{\prime}}V_{K,K^{\prime}}(t).$ (4) Here the indices $K,K^{\prime}$ $(K,K^{\prime}=1,2)$ denote the fragments $1$ and $2$. The quantities $\varepsilon_{\nu_{K}}$ and $u_{\alpha_{K},\beta_{K^{\prime}}}$ represent the single particle energies and the interaction matrix elements, respectively. The single particle states are defined with respect to the centers of the interacting nuclei and are assumed to be orthogonalized in the overlap region. So the annihilation and creation operators are dependent on time. The single particle matrix elements are parameterized by $u_{\alpha_{K},\beta_{K^{\prime}}}(t)=U_{K,K^{\prime}}(t)\left\\{\exp\left[-\frac{1}{2}\left(\frac{\varepsilon_{\alpha_{K}}(t)-\varepsilon_{\beta_{K^{\prime}}}(t)}{\Delta_{K,K^{\prime}}(t)}\right)^{2}\right]-\delta_{\alpha_{K},\beta_{K^{\prime}}}\right\\},$ (5) which contain some parameters $U_{K,K^{\prime}}(t)$ and $\Delta_{K,K^{\prime}}(t)$. The detailed calculation of these parameters and the mean transition probabilities were described in Refs. [15, 21]. The evolution of the DNS along the variable R leads to the quasi-fission of the DNS. The quasi-fission rate $\Lambda^{qf}$ can be estimated with the one- dimensional Kramers formula [26, 27]: $\Lambda^{qf}(\Theta(t))=\frac{\omega}{2\pi\omega^{B_{qf}}}\left(\sqrt{\left(\frac{\Gamma}{2\hbar}\right)^{2}+(\omega^{B_{qf}})^{2}}-\frac{\Gamma}{2\hbar}\right)\exp\left(-\frac{B_{qf}(A_{1},A_{2})}{\Theta(t)}\right).$ (6) Here the quasi-fission barrier is counted from the depth of the pocket of the interaction potential. The local temperature is given by the Fermi-gas expression $\Theta=\sqrt{\varepsilon^{\star}/a}$ corresponding to the local excitation energy $\varepsilon^{\star}$ and level density parameter $a=A/12$ $MeV^{-1}$. In Eq.(6) the frequency $\omega^{B_{qf}}$ is the frequency of the inverted harmonic oscillator approximating the interaction potential of two nuclei in R around the top of the quasi-fission barrier, and $\omega$ is the frequency of the harmonic oscillator approximating the potential in R around the bottom of the pocket. The quantity $\Gamma$, which denotes the double average width of the contributing single-particle states, determines the friction coefficients: $\gamma_{ii^{\prime}}=\frac{\Gamma}{\hbar}\mu_{ii^{\prime}}$, with $\mu_{ii^{\prime}}$ being the inertia tensor. Here we use constant values $\Gamma=2.8$ MeV, $\hbar\omega^{B_{qf}}=2.0$ MeV and $\hbar\omega=3.0$ MeV for the following reactions. The Kramers formula is derived with the quasi- stationary condition of the temperature $\Theta(t)<B_{qf}(A_{1},A_{2})$. However, the numerical calculation in Ref. [27] indicated that Eq.(6) is also useful for the condition of $\Theta(t)>B_{qf}(A_{1},A_{2})$. In the reactions of synthesizing SHN, there is the possibility of the fission of the heavy fragment in the DNS. Because the fissility increases with the charge number of the nucleus, the fission of the heavy fragment can affect the quasi-fission and fusion when the DNS evolves towards larger mass asymmetry. The fission rate $\Lambda^{fis}$ can also be treated with the one-dimensional Kramers formula [26] $\Lambda^{fis}(\Theta(t))=\frac{\omega_{g.s.}}{2\pi\omega_{f}}\left(\sqrt{\left(\frac{\Gamma_{0}}{2\hbar}\right)^{2}+\omega_{f}^{2}}-\frac{\Gamma_{0}}{2\hbar}\right)\exp\left(-\frac{B_{f}(A_{1},A_{2})}{\Theta(t)}\right),$ (7) where the $\omega_{g.s.}$ and $\omega_{f}$ are the frequencies of the oscillators approximating the fission-path potential at the ground state and on the top of the fission barrier for nucleus $A_{1}$ or $A_{2}$ (larger fragment), respectively. Here, we take $\hbar\omega_{g.s.}=\hbar\omega_{f}=1.0$ MeV, $\Gamma_{0}=2$ MeV. The fission barrier is calculated as the sum of a macroscopic part and the shell correction energy used in Refs. [15, 28]. The fission of the heavy fragment does not favor the diffusion of the system to a light fragment distribution. Therefore, it leads to a slight decrease of the fusion probability. In the relaxation process of the relative motion, the DNS will be excited by the dissipation of the relative kinetic energy. The excited system opens a valence space $\Delta\varepsilon_{K}$ in fragment $K(K=1,2)$, which has a symmetrical distribution around the Fermi surface. Only the particles in the states within this valence space are actively involved in excitation and transfer. The averages on these quantities are performed in the valence space: $\Delta\varepsilon_{K}=\sqrt{\frac{4\varepsilon^{\ast}_{K}}{g_{K}}},\varepsilon^{\ast}_{K}=\varepsilon^{\ast}\frac{A_{K}}{A},g_{K}=\frac{A_{K}}{12},$ (8) where the $\varepsilon^{\ast}$ is the local excitation energy of the DNS, which provides the excitation energy for the mean transition probability. There are $N_{K}=g_{K}\Delta\varepsilon_{K}$ valence states and $m_{K}=N_{K}/2$ valence nucleons in the valence space $\Delta\varepsilon_{K}$, which gives the dimension $d(m_{1},m_{2})=\left(\begin{array}[]{c}N_{1}\\\ m_{1}\end{array}\right)\left(\begin{array}[]{c}N_{2}\\\ m_{2}\end{array}\right)$. The local excitation energy is defined as $\varepsilon^{\ast}=E_{x}-\left(U(A_{1},A_{2})-U(A_{P},A_{T})\right).$ (9) Here the $U(A_{1},A_{2})$ and $U(A_{P},A_{T})$ are the driving potentials of fragments $A_{1}$, $A_{2}$ and fragments $A_{P}$, $A_{T}$ (at the entrance point of the DNS), respectively. The detailed calculation of the driving potentials can be seen in Ref. [18]. The excitation energy $E_{x}$ of the composite system is converted from the relative kinetic energy loss, which is related to the Coulomb barrier $B$ [29] and determined for each initial relative angular momentum $J$ by the parametrization method of the classical deflection function [24, 25]. So $E_{x}$ is coupled with the relative angular momentum. After reaching the reaction time in the evolution of $P(A_{1},E_{1},t)$, all those components on the left side of the B.G. (Businaro-Gallone) point contribute to the formation of the compound nucleus. The hindrance in the diffusion process by nucleon transfer to form the compound nucleus is the inner fusion barrier $B_{fus}$, which is defined as the difference of the driving potential at the B.G. point and at the entrance position. Nucleon transfers to more symmetric fragments undergo quasi-fission. The formation probability of the compound nucleus at the Coulomb barrier $B$ (here a barrier distribution $f(B)$ is considered) and angular momentum $J$ is given by $P_{CN}(E_{c.m.},J,B)=\sum_{A_{1}=1}^{A_{BG}}P(A_{1},E_{1},\tau_{int}(E_{c.m.},J,B)).$ (10) Here the interaction time $\tau_{int}(E_{c.m.},J,B)$ is obtained using the deflection function method [30], which means the time duration for nucleon transfer from the capture stage to the formation of the complete fused system with the order of 10-20 s. We obtain the fusion probability as $P_{CN}(E_{c.m.},J)=\int f(B)P_{CN}(E_{c.m.},J,B)dB,$ (11) where the barrier distribution function is taken in asymmetric Gaussian form [13, 15]. So the fusion cross section is written as $\sigma_{fus}(E_{c.m.})=\frac{\pi\hbar^{2}}{2\mu E_{c.m.}}\sum_{J=0}^{\infty}(2J+1)T(E_{c.m.},J)P_{CN}(E_{c.m.},J).$ (12) The survival probability of the excited compound nucleus cooled by the neutron evaporation in competition with fission is expressed as follows: $W_{sur}(E_{CN}^{\ast},x,J)=P(E_{CN}^{\ast},x,J)\prod\limits_{i=1}^{x}\left(\frac{\Gamma_{n}(E_{i}^{\ast},J)}{\Gamma_{n}(E_{i}^{\ast},J)+\Gamma_{f}(E_{i}^{\ast},J)}\right)_{i},$ (13) where the $E_{CN}^{\ast},J$ are the excitation energy and the spin of the compound nucleus, respectively. The $E_{i}^{\ast}$ is the excitation energy before evaporating the $i$th neutron, which has the relation $E_{i+1}^{\ast}=E_{i}^{\ast}-B_{i}^{n}-2T_{i},$ (14) with the initial condition $E_{1}^{\ast}=E_{CN}^{\ast}$. The energy $B_{i}^{n}$ is the separation energy of the $i$th neutron. The nuclear temperature $T_{i}$ is given by $E_{i}^{\ast}=aT_{i}^{2}-T_{i}$ with the level density parameter $a$. $P(E_{CN}^{\ast},x,J)$ is the realization probability of emitting $x$ neutrons. The widths of neutron evaporation and fission are calculated using the statistical model. The details can be found in Ref. [15]. The level density is expressed by the back-shifted Bethe formula [31] with the spin cut-off model as $\rho(E^{\ast},J)=K_{rot}K_{vib}\frac{2J+1}{24\sqrt{2}\sigma^{3}}a^{-1/4}(E^{\ast}-\Delta)^{-5/4}\exp[2\sqrt{a(E^{\ast}-\Delta)}]\exp[-\frac{(J+1/2)^{2}}{2\sigma^{2}}],$ (15) where the $K_{rot}$ and $K_{vib}$ are the coefficients of the rotational and vibrational enhancements. The pairing energy is given by $\Delta=\chi\frac{12}{\sqrt{A}}$ (16) in MeV($\chi$=-1, 0 and 1 for odd-odd, odd-even and even-even nuclei, respectively). The spin cut-off parameter is calculated by the formula: $\sigma^{2}=T\zeta_{r.b}/\hbar^{2},$ (17) where the rigid-body moment of inertia has the relation $\zeta_{r.b}=0.4MR^{2}$ with the mass $M$ and the radius $R$ of the nucleus. The level density parameter is related to the shell correction energy $E_{sh}(Z,N)$ and the excitation energy $E^{\ast}$ of the nucleus as $a(E^{\ast},Z,N)=\tilde{a}(A)[1+E_{sh}(Z,N)f(E^{\ast}-\Delta)/(E^{\ast}-\Delta)].$ (18) Here, $\tilde{a}(A)=\alpha A+\beta A^{2/3}b_{s}$ is the asymptotic Fermi-gas value of the level density parameter at high excitation energy. The shell damping factor is given by $f(E^{\ast})=1-\exp(-\gamma E^{\ast})$ (19) with $\gamma=\tilde{a}/(\epsilon A^{4/3})$. All the used parameters are listed in Table 1. In Fig.1 we give the level density parameters of different nuclides at the ground state calculated by using Eq.(18) and compared them with two empirical formulas $a(A)=A/8$, and $A/12$. It can be seen that the strong shell effects appear in the level density. With this procedure introduced above, we calculated the angular momentum dependence of the capture, fusion and survival probabilities as shown in Fig.2 for the reaction 48Ca+208Pb at incident energies 172.36 MeV and 192.36 MeV, respectively. The values of the three stages decrease obviously with increasing the relative angular momentum. So in the following estimation of the production cross sections, we cut off the maximal angular momentum at $J_{max}=30$, which is taken as the same value that used in the cold fusion reactions [18]. ## 3 Results and discussions ### 3.1 Fusion-fission reactions and quasi-fission mass yields As a test of the parameters for the estimation of the transmission of two colliding nuclei and the de-excitation of the thermal compound nucleus, we analyzed the fusion-fission reactions for the selected systems shown in Fig.3 assuming $P_{CN}=1$. The capture and evaporation residue cross sections are compared with the available experimental data [32, 33, 34, 35]. For these systems the quasi-fission does not dominate in the sub-barrier region, which also means that $P_{CN}\sim 1$. The evaporation residues are mainly determined through the capture of the light projectile by the target nucleus and the survival probabilities of the formed compound nucleus. The experimental data can be reproduced rather well within the error bars. Some discrepancies may come from the quasi-fission in the above barrier region and from the input quantities, such as the neutron separation energy, shell correction and mass. The rotational and the vibrational enhancement in the level density can also affect the survival probabilities of the excited compound nucleus [36]. Here we take unity for both coefficients as shown in Table 1 because the height of the fission barrier is also sensitive to the survival of the compound nucleus by fitting the experimental evaporation residue excitation functions in the fusion-fission reactions. Since the electrostatic energy of the composite systems formed by two heavy colliding nuclei is very large, so although the two nuclei may be captured by the nuclear potential, they almost always separate after mass transfer from the heavier nucleus to the lighter one rather fusing. This process is called quasi-fission [37, 38], which is the main feature in the massive fusion reactions and can inhibit fusion by several degree of freedom. Recently, experiment has performed nice works by measuring the quasi-fission and fusion- fission mass yields [39]. In the DNS model, the quasi-fission mass yields are expressed as [26] $Y_{q-f}(A_{1})=\sum_{J=0}^{J_{max}}\int_{0}^{\tau_{int}}P(A_{1},E_{1},t)\Lambda^{qf}(\Theta(t))dt.$ (20) In Fig.4 we show a comparison of the calculated quasi-fission mass yields and the experimental data for the two 48Ca induced reaction systems. The trends of the distribution can be reproduced by the DNS model. At the domain of the medium-mass fragments A1=ACN/2-30$\sim$ACN/2+30, The experimental data are higher than the calculated values, which may be come from the contribution of the fusion-fission fragments. ### 3.2 Evaporation residue cross sections The evaporation residues observed in laboratories by the consecutive $\alpha$ decay are mainly produced by the complete fusion reactions, in which the fusion dynamics and the structure properties of the compound nucleus affect their production. Within the framework of the DNS model, we calculated the evaporation residue cross sections producing SHN Z=110, 112, 113, 115 with 232Th, 238U, 237Np and 243Am targets in the 48Ca induced reactions as shown in Fig.5, and compared them with the Dubna data [7, 40, 41] as well as with the recent GSI data [42] for 238U targets in the 3n channel. Compared with the Dubna data for the system 48Ca+238U, the GSI results show that the formation cross sections in the 3n channel have a slight decrease at the same excitation energy, which is in a good agreement with our calculated results. The calculations were carried out before getting the experimental data [41] for the reaction 48Ca+237Np, and a good agreement with the data is also found [43]. The excitation energy of the compound nucleus is obtained by $E^{\ast}_{CN}=E_{c.m.}+Q$, where the $E_{c.m.}$ is the incident energy in the center-of-mass system. The $Q$ value is given by $Q=\Delta M_{P}+\Delta M_{T}-\Delta M_{C}$, and the corresponding mass excesses $\Delta M_{i}$ ($i=P,T,C$) are taken the data from Ref. [44] for the projectile, target and compound nucleus denoted with the symbols $P$, $T$ and $C$, respectively. Usually, the neutron-rich projectile-target combinations are in favor of synthesizing SHN experimentally, which can enhance the survival probability $W_{sur}$ in Eq.(1) of the formed compound nucleus because of the smaller neutron separation energy. Differently to the cold fusion reactions [18], the maximal production cross sections from Ds to 115 especially in the 2n-5n channels are not changed much although the heavier SHNs are synthesized. Within the error bars the experimental data can be reproduced rather well. With the same procedure, we analyzed the evaporation residue excitation functions with targets 242,244Pu and 245,248Cm that are used to synthesize the superheavy elements Z=114 and 116 in Dubna [40, 45] (Fig.6). Our calculations show that the target 244Pu has a larger production cross section than 242Pu because of the larger survival probability. In Fig.7 we also calculated the evaporation residue excitation functions to synthesize superheavy elements Z=117-120 using the actinide isotopes with longer half-lives 247Bk, 249Cf, 254Es and 257Fm. The 3n evaporation channel with an excitation energy of the formed compound nucleus around 30 MeV is favorable to produce SHN with Z$\geq$117 by using the actinide targets. Within the error bars, the positions of the maximal production cross sections are in good agreement with the available experimental results. Similar calculation of the evaporation residue excitation functions was also reported in Ref. [46]. The spectrum form of evaporating neutrons is mainly determined by the survival probability, in which the neutron separation energy and the shell correction play a very important role in the determination of the value. We considered the angular momentum influence in the calculation of the level density, but did not include it in the estimation of the fission barrier of the thermal compound nucleus. As pointed out in section 1, the fission barrier of SHN decreases rapidly with increasing excitation energy of the compound nucleus, where the rotation of the system affects the height of the barrier and also influences other crucial quantities such as the level density etc. In Fig.8 we show a comparison of the calculated maximal production cross sections of superheavy elements Z=102-120 in the cold fusion reactions by evaporating one neutron, in the 48Ca induced reactions with actinide targets by evaporating three neutrons, and the experimental data [1, 2, 4, 47]. The production cross sections decrease rapidly with increasing the charge number of the synthesized compound nucleus in the cold fusion reactions, such as from 0.2 $\mu b$ for the reaction 48Ca+208Pb to 1 pb for 70Zn+208Pb, and even below 0.1 pb for synthesizing Z$\geq$113 [18]. It seems to be difficult to synthesize superheavy elements Z$\geq$113 in the cold fusion reactions at the present facilities. The calculated results show that the 48Ca induced reactions have smaller production cross sections with 232Th target, but are in favor of synthesizing heavier SHN (Z$\geq$113) because of the larger cross sections. The experimental data also give such trends. In the DNS concept, the inner fusion barrier increases with reducing mass asymmetry in the cold fusion reactions, which leads to a decrease of the formation probability of the compound nucleus. However, the 48Ca induced reactions have not such increase of the inner fusion barrier for synthesizing heavier SHN. Because of the larger transmission and the higher fusion probability, we obtain larger production cross sections for synthesizing SHN (Z$\geq$113) in the 48Ca induced reactions although these reactions have the smaller survival probability than those in the cold fusion reactions. It is still a good way to synthesize heavier SHN by using the 48Ca induced reactions. Of course, further experimental data are anticipated to be obtained in the future. However, the actinide targets are difficulty to be handled in experiments synthesizing heavier SHN. ### 3.3 Isotopic dependence of the production cross sections Recent experimental data show that the production cross sections of the SHN depend on the isotopic combination of the target and projectile in the 48Ca induced fusion reactions. For example, the maximal cross section in the 3n channel is $3.7\pm^{3.6}_{1.8}$ pb for the reaction 48Ca+245Cm at the excitation energy 37.9 MeV; however, it is $1.2$ pb for the reaction 48Ca+248Cm although the later is a neutron-rich target [8, 40]. The isotopic trends of the production cross sections were also observed and investigated in cold fusion reactions [48, 18]. Further investigations on the isotopic trends in the 48Ca induced reactions are very necessary for predicting the optimal combinations, excitation energies (incident energies) and evaporation channels in the synthesis of SHN. In Fig.9 we show the calculated isotopic trends in producing superheavy elements Z=110, 112 with the isotopic actinides Th and U in the 3n channels, and compare them with the available experimental data performed in Dubna [40] (squares with error bars) and at GSI [42] (circles with error bars). The results show that the targets 230Th in the 4n channel and 235,238U in the 3n channel have the largest cross sections. The isotopic trends in synthesizing Z=113-116 with the actinide targets Np, Pu, Am and Cm are also calculated systematically, and compared with the existing data measured in Dubna [7, 40, 45] and the results of Adamian et al. [49] for the Pu isotopes as shown in Fig.10 and Fig.11. The isotopes 237Np, 241Pu, 242,243Am and 245,247Cm in the 3n channels, and 244Pu in the 4n channel as well as the isotope 250Cm are suitable for synthesizing SHN. Except for the 244Pu, our calculated cross sections are smaller than the ones of the Adamian et al. In the DNS model, the isotopic dependence of the production cross sections is mainly determined by both the fusion and survival probabilities. Of course, the transmission probability of two colliding nuclei can also be affected since the isotopes have initial quadrupole deformations. With the same procedure, we analyzed the dependence of the production cross sections on the isotopes Bk and Cf in the 3n channels for synthesizing the superheavy elements Z=117, 118 and compared them with the available experimental data [8] shown in Fig.12. The results show that the targets 248,249Bk and 251,252Cf are favorable for synthesizing the superheavy elements Z=117 and 118. The corresponding excitation energies are also given in the figures. In Fig.13 we show the dependence of the inner fusion barrier, the fission barrier of the compound nucleus, and the neutron separation energies of evaporating 3n and 4n on the mass numbers of the isotopic targets Cm in the 48Ca induced reactions. It is obvious that the combinations with the isotopes 245,247Cm have smaller inner fusion barriers, higher fission barriers and smaller 3n separation energies, which result in larger production cross sections producing the superheavy element Z=116. Although the lower fission barrier for the isotope 250Cm, it gives the smaller inner fusion barrier and neutron separation energies, which also leads to the larger cross sections in the 3n and 4n channels as shown in Fig.11. The shell correction and the neutron separation energies are taken from Ref. [44]. When the neutron number of the target increases, the DNS gets more asymmetrical and the fusion probability increases if the DNS does not consist of more stable nuclei (such as magic nuclei) because of a smaller inner fusion barrier. A smaller neutron separation energy and a larger shell correction lead to a larger survival probability. The compound nucleus with closed neutron shells has a larger shell correction energy and a larger neutron separation energy. The neutron- rich actinide target has larger fusion and survival probabilities due to the larger asymmetric initial combinations and smaller neutron separation energies. But such actinide isotopes are usually unstable with smaller half- lives. With the establishment of the high intensity radioactive-beam facilities, the neutron-rich SHN may be synthesized experimentally, which approaches the island of stability. ### 3.4 238U based reactions The uranium is the heaviest element existing in the nature. It has a larger mass asymmetry constructed as a target in the fusion reactions with the various neutron-rich light projectiles. The isotope 238U is the neutron- richest nucleus in the U isotopes and often chosen as the target for synthesizing SHN. In Fig.14 we give evaporation residue excitation functions of the reactions 40Ar, 50Ti, 54Cr, 64Ni+238U in the 2n-5n channels. The results show that the 4n channel in the reaction 40Ar+238U has the larger cross sections with 2.1 pb at an excitation energy 42 MeV. This reaction is being used to synthesize the superheavy nucleus Ds with HIRFL accelerator at Institute of Modern Physics in Lanzhou. The reactions 50Ti, 54Cr, 64Ni+238U lead to the cross section smaller than 0.1 pb. The isotopic trends based on the U isotopes are also investigated using the DNS model as shown in Fig.15. Calculations show that the isotopes 235U and 238U are favorable in producing SHN. The cross sections are reduced with increasing the mass numbers of the projectiles. Other reaction mechanisms to synthesize SHN have to be investigated with theoretical models, such as the massive transfer reactions, and the complete fusion reactions induced by weakly bound nuclei. Work in these directions is in progress within the framework of the DNS model. ## 4 Conclusions Using the DNS model, we systematically investigated the production of superheavy residues in fusion-evaporation reactions, in which the nucleon transfer leading to the formation of the superheavy compound nucleus is described with a set of microscopically derived master equations that are solved numerically and include the quasi-fission of the DNS and the fission of the heavy fragments. The fusion dynamics and the evaporation residue excitation functions in the 48Ca fusion reactions are systematically investigated. The calculated results are in good agreement with the available experimental data within the error bars. Isotopic trends in the production of superheavy elements are analyzed. It is shown that the isotopes 235,238U, 237Np, 241,244Pu, 242Am and 245,247,250Cm, 248,249Bk and 251,252Cf in the 3n channels, and 230Th, 244Pu, 248,250Cm in the 4n channels are favorable for producing the superheavy elements Z=110, 112 and 113-118, respectively. The evaporation residue excitation functions of the reactions 40Ar, 50Ti, 54Cr, 64Ni+238U in the 2n-5n channels and the isotopic trends with 40Ar, 48Ca, 50Ti, 54Cr, 58Fe and 64Ni bombarding U isotopes are also studied. ## 5 Acknowledgement One of us (Z.-Q. Feng) is grateful to Prof. H. Feldmeier, Dr. G.G. Adamian and Dr. N.V. Antonenko for fruitful discussions and help, and also thanks the hospitality during his stay in GSI. This work was supported by the National Natural Science Foundation of China under Grant No. 10805061, the special foundation of the president fellowship, the west doctoral project of Chinese Academy of Sciences, and major state basic research development program under Grant No. 2007CB815000. ## References * [1] S. Hofmann and G. Münzenberg, Rev. Mod. Phys. 72 (2000) 733; S. Hofmann, Rep. Prog. Phys. 61 (1998) 639. * [2] Yu.Ts. Oganessian, J. Phys. G 34 (2007) R165; Nucl. Phys. A 787 (2007) 343c. * [3] Yu.Ts. Oganessian, A.S. Iljnov, A.G. Demin, et al., Nucl. Phys. A 239 (1975) 353; Nucl. Phys. A 239 (1975) 157. * [4] G. Münzenberg, J. Phys. G 25 (1999) 717. * [5] K. Morita, K. Morimoto, D. Kaji, et al., J. Phys. Soc. Jpn. 73 (2004) 2593. * [6] Yu.Ts. Oganessian, A.G. Demin, A.S. Iljnov, et al., Nature 400 (1999) 242; Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 62 (2000) 041604(R). * [7] Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 69 (2004) 021601(R). * [8] Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 74 (2006) 044602. * [9] Z.G. Gan, Z. Qin, H.M. Fan, et al., Eur. Phys. J. A 10 (2001) 21; Z.G. Gan, J.S. Guo, X.L. Wu, et al., Eur. Phys. J. A 20 (2004) 385. * [10] W.J. Swiatecki, Prog. Part. Nucl. Phys. 4 (1980) 383\. * [11] S. Bjornholm and W.J. Swiatecki, Nucl. Phys. A 391 (1982) 471. * [12] Y. Aritomo, T. Wada, M. Ohta, and Y. Abe, Phys. Rev. C 59 (1999) 796. * [13] V.I. Zagrebaev, Phys. Rev. C 64 (2001) 034606; V.I. Zagrebaev, Y. Aritomo, M.G. Itkis, Yu.Ts. Oganessian, and M. Ohta, Phys. Rev. C 65 (2001) 014607. * [14] G.G. Adamian, N.V. Antonenko, W. Scheid et al., Nucl. Phys. A 627 (1997) 361; Nucl. Phys. A 633 (1998) 409. * [15] Z.Q. Feng, G.M. Jin, F. Fu, and J.Q. Li, Nucl. Phys. A 771 (2006) 50. * [16] N. Wang, Z.X. Li, X.Z. Wu, et al., Phys. Rev. C 69 (2004) 034608; N. Wang, Z.X. Li, X.Z. Wu, and E. G. Zhao, Mod. Phys. Lett. A 20 (2005) 2619. * [17] Z.Q. Feng, G.M. Jin and F.S. Zhang, Nucl. Phys. A 802 (2008) 91; Z.Q. Feng, F.S. Zhang, G.M. Jin, and X. Huang, Nucl. Phys. A 750 (2005) 232; Z.Q. Feng, G.M. Jin, F.S. Zhang, et al., Chin. Phys. Lett. 22 (2005) 3040. * [18] Z.Q. Feng, G.M. Jin, J.Q. Li, and W. Scheid, Phys. Rev. C 76 (2007) 044606. * [19] V. V. Volkov, Phys. Rep. 44 (1978) 93. * [20] P. Reiter, T.L. Khoo, T. Lauritsen, et al., Phys. Rev. Lett. 84 (2000) 3542. * [21] W. Li, N. Wang, J. Li, et al., Eur. Phys. Lett. 64 (2003) 750; J. Phys. G 32 (2006) 1143. * [22] A. Diaz-Torres, G.G. Adamian, N.V. Antonenko, and W. Scheid, Phys. Rev. C 64 (2001) 024604; A. Diaz-Torres, Phys. Rev. C 74 (2006) 064601. * [23] W. Nörenberg, Z. Phys. A 274 (1975) 241; S. Ayik, B. Schürmann and Nörenberg, Z. Phys. A 277 (1976) 299\. * [24] G. Wolschin and W. Nörenberg, Z. Phys. A 284 (1978) 209. * [25] J.Q. Li, X.T. Tang, G. Wolschin, Phys. Lett. B 105 (1981) 107. * [26] G.G. Adamian, N.V. Antonenko and W. Scheid, Phys. Rev. C 68 (2003) 034601. * [27] P. Grangé, Li Jun-Qing and H. A. Weidenmüller, Phys. Rev. C 27 (1983) 2063. * [28] G.G. Adamian, N.V. Antonenko, S.P. Ivanova, and W. Scheid, Phys. Rev. C 62 (2000) 064303. * [29] Z.Q. Feng, G.M. Jin, F. Fu, and J. Q. Li, High Ener. Phys. Nucl. Phys., 31 (2007) 366. * [30] J.Q. Li, G. Wolschin, Phys. Rev. C 27 (1983) 590. * [31] H. Bethe, Phys. Rev. 50 (1936) 332; Rev. Mod. Phys. 9 (1937) 69. * [32] E.V. Prokhorova, A.A. Bogachev, M.G. Itkis, et al., Nucl. Phys. A 802 (2008) 45. * [33] M. Dasgupta and D.J. Hinde, Nucl. Phys. A 734 (2004) 148. * [34] K. Nishio, H. Ikezoe, Y. Nagame, et al., Phys. Rev. Lett. 93 (2004) 162701. * [35] J.M. Gates, M.A. Garcia, K.E. Gregorich, et al., Phys. Rev. C 77 (2008) 034603. * [36] A.R. Junghans, M.de Jong, H.-G. Clerc, et al., Nucl. Phys. A 629 (1998) 635. * [37] B.B. Back, Phys. Rev. C 31 (1985) 2104. * [38] J. Toke, R. Bock, G.X. Dai, et al., Nucl. Phys. A 440 (1985) 327. * [39] M.G. Itkis, J. Äystö, S. Beghini, et al., Nucl. Phys. A 734 (2004) 136. * [40] Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 70 (2004) 064609. * [41] Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 76 (2007) 011601(R). * [42] S. Hofmann, D. Ackermann, S. Antalic, et al., Eur. Phys. J. A 32 (2007) 251. * [43] Z.Q. Feng, PhD thesis, Institute of Modern Physics, Chinese Academy of Sciences, 2007. * [44] P. Möller et al., At. Data Nucl. Data Tables 59 (1995) 185. * [45] Yu.Ts. Oganessian, V.K. Utyonkov, Yu.V. Lobanov, et al., Phys. Rev. C 69 (2004) 054607. * [46] V.I. Zagrebaev, Nucl. Phys. A 734 (2004) 164. * [47] K.E. Gregorich, T.N. Ginter, W. Loveland, et al., Eur. Phys. J. A 18 (2003) 633. * [48] S. Hofmann, V. Ninov, F.P. Heßberger, et al., Z. Phys. A 350 (1995) 277. * [49] G. G. Adamian, N. V. Antonenko, and W. Scheid, Phys. Rev. C 69 (2004) 014607. Table 1: Parameters used in the calculation of the level density. $K_{rot}$ $K_{vib}$ $b_{s}$ $\alpha$ $\beta$ $\epsilon$ 1 1 1 0.114 0.098 0.4 Figure 1: Calculated values of the level density parameters as a function of the atomic mass. Figure 2: Calculated capture, fusion and survival probabilities as functions of the relative angular momenta in the reaction 48Ca+208Pb at excitation energies of the compound nucleus of 20 MeV and 40 MeV, respectively. Figure 3: Comparison of the calculated fusion-fission excitation functions and the available experimental data for the reactions 16O+208Pb, 16O+238U, 36Ar+148Sm and 26Mg+238U. Figure 4: Calculated quasi- fission mass yields for the reactions 48Ca+244Pu and 48Ca+248Cm at excitation energies of the compound nuclei 42 MeV and 33 MeV, respectively, and compared them with the available experimental data [39]. Figure 5: The calculated evaporation residue excitation functions with 232Th, 238U, 237Np and 243Am targets in 48Ca induced reactions, and compared with the available experimental data [7, 40, 41]. Figure 6: The same as in Fig.5, but for the targets 242,244Pu and 245,248Cm to produce superheavy elements Z=114 and 116. Figure 7: The same as in Fig.5, but for the targets 247Bk, 249Cf, 254Es and 257Fm to synthesize superheavy elements Z=117-120. Figure 8: Maximal production cross sections of superheavy elements Z=102-120 in cold fusion reactions based on 208Pb and 209Bi targets with projectile nuclei 48Ca, 50Ti, 54Cr, 58Fe, 64Ni, 70Zn, 76Ge, 82Se, 86Kr and 88Sr, in 48Ca induced reactions with actinide targets by evaporating 3 neutrons, in comparison with available experimental data [1, 2, 4, 47]. Figure 9: Isotopic dependence of the calculated maximal production cross sections in the 3n evaporation channel and the corresponding excitation energies in the synthesis of superheavy elements Z=110 and 112 for the reactions 48Ca+ATh and 48Ca+AU, and compared with the experimental data [40, 42]. Figure 10: The same as in Fig.9, but for isotopic targets Np and Pu to produce superheavy elements Z=113 and 114. Figure 11: The same as in Fig.9, but for isotopic targets Am and Cm to synthesize superheavy elements Z=115 and 116 in 3n and 4n channels. Figure 12: The same as in Fig.9, but for isotopes Bk and Cf in 48Ca induced reactions. Figure 13: (a) the inner fusion barrier, (b) the fission barrier of the compound nucleus and (c) the neutron separation energy as a function of the mass numbers of the isotopic targets Cm in the reactions 48Ca+ACm. Figure 14: The evaporation residue excitation functions in the reactions 40Ar, 50Ti, 54Cr, 64Ni+238U. Figure 15: The production cross sections in the 3n channels as a function of the mass number of the isotopic targets U with projectiles 40Ar, 48Ca, 50Ti, 54Cr, 58Fe and 64Ni.

arxiv-papers

2024-09-04T02:48:54.219458

0803.1207

arxiv-papers

2024-09-04T02:48:54.224810

0803.1347

# On Hawking/Unruh Process: Where does the Radiation Come from? Tadas K. Nakamura ###### Abstract The energy source of the radiation in Unruh/Hawking process is investigated with emphasis on the particle number definition based on conservation laws. It has been shown that the particle radiation is not the result of pair creation by the gravitational force, but the result of difference in the conservation laws to define the particle number. The origin of the radiated energy in the distant future corresponds to the zero point oscillations with infinitely large wave numbers. This result implies the need of reconsideration on the scenario of black hole evaporation. CFAAS, Fukui Prefectural University, Fukui 910-1195, Japan (tadas@fpu.ac.jp) ## 1 Introduction The theory of quantum particle radiation by gravitational force [1, 2] is generally accepted as well established, however, several authors pointed out essential problems that might brow up the whole story [3, 4, 5] (see [6] for a review). What called trans-Plankian problem has been known from very early years [7]. In the last two decades, this problem has been investigated intensively [8, 9]. The derivation of Hawking/Unruh process is based on so called cis-Plankian physics, i.e., the theories of gravitation and quantum field we know at the present. It is believed these theories will break down beyond an extremely small scale, the Plank scale presumably, and we do not know what happens there. The unknown physics in that scale is called trans-Plankian physics. The calculation of Hawking/Unruh effect inevitably requires the wave modes with infinitesimally small wave length (see, e.g., [6]), therefore, we need to know the trans-Plankian physics to understand the radiation mechanism; this is the trans-Plankian problem. There has been attempts [9] to derive the radiation within the cis-Plankian scale, however, they are based on _ad hoc_ assumptions yet to be tested experimentally. Helfer [4] pointed out another issue we have to consider before the trans- Plankian problem. The radiation at later times must have plied up near the horizon at the time of black hole formation, and its backreaction is far from negligible to the black hole metric. This implies the theory of Hawking radiation is intrinsically inconsistent even within the framework of cis- Plankian physics, because such backreaction may completely destroy the black hole formation. This is a serious problem. Papers on the trans-Plankian problem so far assume that the Hawking radiation is well predicted by the cis-Plankian physics, and discuss what will happen if we have to consider trans-Plankian effects. However, if cis-Plankian physics itself fails to derive the Hawking radiation, then we do not have any reason to believe the existence of radiation. The purpose of the present paper is to formulate the Helfer’s conclusion [4] from a different point of view. We wish to show the theory of black hole evaporation is inconsistent even if the cis-Plankian physics is valid up to infinitesimally small scale. To this end, we need another unknown physics within the cis-Plankian regime: the effect of the zero-point oscillation to the gravity. We do not know the general theory for this, however, there can be two possibilities for the Hawking radiation. One is such that the radiation can carry away the black hole energy to cause its evaporation. In this case, the backreaction of the quantum filed is so large as to alter the black hole formation completely [4]. We will see in the present paper there can be another possibility that the field has no backreaction to the black hole geometry. In this case, however, there will be no black hole evaporation at all even with the existence of Hawking radiation. This may be plausible because there are considerable amount of observational evidence to believe the existence of black holes. We take an approach a little different from the conventional quantization with creation/annihilation operators in the present paper; the mathematical structure is equivalent, but its interpretation is not the same. The quantization with creation/annihilation operators takes two different steps, transition from the classic to quantum field and introduction of the particle picture, namely, at the same time. The essential step in the canonical quantization method is to replace a classical Poisson bracket with a quantum commutation relation regarding the field as a collection of quantum operators. If conserved quantities have discrete eigenvalues with equal intervals, then we can construct the particle picture. It is well known the latter is not always possible in a curved spacetime. It also should be noted these two steps do not have to be done at the same place even when we have the particle picture. The commutation relation must be given on a Chaucy surface on which the Poisson bracket is defined. The particle picture, in contrast, does not have to be on the same surface. It can be on some other spacelike surface, which does not have to be a Chaucy surface as long as there exist some conservation laws on it. In the present paper, the essential quantization, i.e., definition of the commutation relation, will be done on the surface of constant time in Minkowski/Kruscal coordinates. Then the particle picture is introduced based on that quantization, not only on the same surface but also on the surface of constant time in Rindler/Schwarzschild coordinates. The particle picture is based on conservation laws in general. What we directly measure is not the particle number itself, but some conserved quantity such as energy or electric charge. We imagine there are $n$ particles, each of which carries a certain amount of conserved quantity, if the total of the quantity has discrete values proportional to an integer $n$. This means the concept of “particle number” is defined by conserved quantities. If all the conserved quantities share the same $n$, then we can define one unique particle number, however, this is not the case. There can be several different definitions of particle numbers because there can be several different Killing vector fields that determine the conservation laws in a relativistic spacetime. Consequently one physical state can have different particle numbers, and this is what is happening in the Unruh/Hawking process. Particles are not created in the literal meaning of “creation”, which means the particle number increases as time goes on. Rather, what takes place is just a difference of particle numbers caused by the difference in their definitions. This is in agreement with the result of Belinski [3] calculated from another viewpoint. We will see in the present paper the radiation of particles comes from the vacuum state, i.e., zero particle state, of another kind of particle number. This is possible because a vacuum is not a completely empty space but has zero point oscillations. The continuous particle radiation can take place because the zero point oscillations exist up to infinitely large wave numbers, which means infinite amount of energy source. The present paper is organized in the following. In section 2 we first review some basic concepts to clarify the procedure of quantization used in the later sections. We examine in Section 3 the case of Unruh process in a flat spacetime, since it has the two different types of conservation laws clearly defined; we can understand the problem with this simple analogy. We apply the results obtained in Section 3 to the case of Schwarzschild black holes in Section 4, and a brief summary is given in Section 5. ## 2 Basics ### 2.1 Time and Energy The concept of energy is often used in a sloppy way, which sometimes leads to misconceptions. The integration of the energy-momentum tensor cannot be carried out in the curved spacetime in general, however, there can be well defined “energy” as a globally conserved quantity if there exists a Killing vector field. If a Killing vector $\xi_{\nu}$ is timelike, then the integration $\int_{\Sigma}\xi_{\nu}T^{\nu\mu}d\Sigma_{\mu}$ ($T^{\nu\mu}$: energy-momentum tensor) over an appropriate spacelike surface $\Sigma$ is conserved with respect to the time evolution in $\xi_{\nu}$. If there are several different timelike Killing vector fields, there can be the same number of corresponding energies; the energies defined by different Killing vector fields are different physical entities. Sometimes this difference in energies is not well understood and causes confusion; one good example is an intuitive explanation of the Hawking radiation found in popular science books. It goes like: (1) a vacuum is not an empty space but filled with instantaneous pair production of virtual particles; (2) a pair of virtual particles can exist within a short time period of $\Delta t\sim\hbar/\Delta E$ because of the uncertain principle; (3) when such virtual particles are created near the event horizon, one of the pair may fall into the black hole across the horizon during the time interval of $\Delta t$; (4) once a virtual particle crosses the horizon, its energy becomes negative; (5) then the other particle of the pair can have positive energy without violating the energy conservation law. This explanation does not specify the Killing vector field with which the time and energy are defined. If the Killing vector is something like the Schwarzschild time, then a particle takes infinitely long time to reach the horizon, and cannot cross the horizon during the period of $\Delta t$. If, on the other hand, the Killing vector is such that a particle can cross the horizon within a finite period, then the corresponding energy does not change the sign on the other side of the horizon. The pair production near the horizon is not likely to occur to cause the Hawking radiation in both cases. ### 2.2 Conservation Laws and Particle Numbers Usually the quantization process to investigate Hawking/Unruh process is based on the creation/annihilation operators defined by the negative/positive frequency modes. In the present paper we take one step backwards and perform the quantization by replacing the Poisson bracket with the commutation relation. Hereafter, let us use the word “quantization” with the meaning of the transition from the classical to quantum theory, and does not necessarily mean the particle picture. The particle picture is derived from conserved quantities after the quantization. If the quantum observable of a conserved quantity has the structure of a harmonic oscillator, its eigenvalues are proportional to $n+\frac{1}{2}$ with $n=0,1,2,\cdots$. Usually the constant $\frac{1}{2}$ is subtracted out by normal ordering, thus the quantity is proportional to $n$. When there are several conserved quantities that share the same $n$ for the same state, then we can interpret $n$ as the particle number. Suppose we establish quantization somehow, and find an observable $\hat{a}$ (hat mark indicates a quantum operator) and its Hermite conjugate $\hat{a}^{\dagger}$ have the following commutation relation $[\hat{a},\hat{a}^{\dagger}]=\hat{a}\,\hat{a}^{\dagger}-\hat{a}^{\dagger}\hat{a}=1\,.$ (1) We use the unit system with $G=\hbar=c=1$ throughout the present paper. The general theory of quantum harmonic oscillators tells us (see., e.g., [10]) an observable defined as $\hat{A}=\frac{A_{0}}{2}\left(\hat{a}\,\hat{a}^{\dagger}+\hat{a}^{\dagger}\hat{a}\right)$ (2) has the eigenstates $\left|n_{A}\right\rangle$ that satisfies $\hat{A}\left|n_{A}\right\rangle=A_{0}\left(n+\frac{1}{2}\right)\left|n_{A}\right\rangle\;,(n=0,1,2\cdots,)\,.$ (3) if $\hat{a}$ has the commutation relation of (1). For the above argument the observable $\hat{A}$ does not have to be related to the Hamiltonian explicitly (note: the Poisson bracket has something to do with the Hamiltonian implicitly), and there can be several choices for such observables. For example if we define a new observable $\hat{b}$ as $\hat{b}=\alpha\,\hat{a}+\beta\,\hat{a}^{\dagger}$ (4) with $\alpha^{2}-\beta^{2}=1$ then it also satisfies the commutation relation like (1) and thus $\hat{B}=\frac{1}{2}B_{0}(\hat{b}\,\hat{b}^{\dagger}+\hat{b}^{\dagger}\hat{b})$ has the eigenvalues $B_{0}(n+\frac{1}{2})$. It is easy to confirm its eigenstates $\left|n_{B}\right\rangle$ are not the eigenstates of $\hat{A}$, i.e., $\hat{A}\left|n_{B}\right\rangle\neq(n+\frac{1}{2})\left|n_{B}\right\rangle$, and vice versa. Both pairs $\hat{a}$, $\hat{a}^{\dagger}$ and $\hat{b}$, $\hat{b}^{\dagger}$ have the structure of annihilation/creation operators, however, it is not enough for the particle picture. To construct the particle picture with $\hat{a}$ and $\hat{a}^{\dagger}$, $\hat{A}$ must obey a conservation law in time, i.e., $\partial\hat{A}/\partial t=0$ (here we employ the Heisenberg picture) at least approximately. If $\hat{A}$ rapidly changes even without interactions, so does $n$, and it is not appropriate to regard $n$ as a particle number. When the Hamiltonian does not depend on time explicitly, the condition of $\partial\hat{A}/\partial t=0$ is equivalent to the following commutation relation: $[\hat{A},\hat{H}]=0\,.$ (5) Obviously the Hamiltonian itself satisfies this condition, therefore, the Hamiltonian is usually used to introduce the particle picture. Then the operators $\hat{a}$ and $\hat{a}^{\dagger}$ become the amplitudes of wave modes with positive and negative frequencies respectively, which are usually used in the procedure of quantization as the annihilation and creation operators. Therefore, the mathematical structure in the present paper is equivalent to the one in the conventional method. If there are other conserved observables with respect to the time $t$, then they share the same set of eigenstates with the Hamiltonian $\hat{H}$ because of the commutation relation (5). Therefore $n$ can be regarded as the particle number without specifying the conserved quantities, as long as the conservation laws are on the same time evolution of $t$. Especially, the ground sate of the observables is uniquely determined, and we call it “vacuum”. We can define the number operator as $\hat{N}=\hat{a}^{\dagger}\hat{a},$ (6) whose eigenvalue is the particle number $n$, and the vacuum means the eigenstate with $n=0$. However, in relativistic spacetimes there can be several different types of time evolution with different sets of conservation laws because several different Killing vector fields can exist; the Minkowski and Rindler times in a flat spacetime are a good example. If two different types of time evolution have their own conservation laws, then the conserved quantities that belong to different time evolution may have different sets of eigenstates. Consequently the ground state in one time evolution is not the ground state in another, in other words, they have different vacuum states. This is what causes Hawking/Unruh radiation as we will see in the next section. ## 3 Unruh process ### 3.1 Minkowski Coordinates Suppose the following real valued Klein-Goldon equation in a two dimensional Minkowski spacetime where $t$ and $x$ are the time and space coordinates: $\phi_{,tt}-\phi_{,xx}=0\,.$ (7) We write $\partial\phi/\partial t=\phi_{,t}$ etc. in shorthand. We take the Cauchy surface for the canonical dynamics as the one defined with $t=\textnormal{constant}$, then the Hamiltonian may be written as $H=\int_{-\infty}^{\infty}\frac{1}{2}\left[\phi_{,t}^{2}(t,x)+\phi_{,x}^{2}(t,x)\right]\,dx\,.$ (8) We expand the field as $\phi(x,t)=\int\left[a(k)\,u(k;x,t)+a^{*}(k)\,u^{*}(k;x,t)\right]dk\,,$ (9) with mode functions $u(k;x,t)=\frac{1}{\sqrt{4\pi\omega}}\,e^{-i\omega t+ikx}\,,$ (10) where $\omega=|k|$ and the asterisk indicates complex conjugate. Precisely speaking, $a_{k}$ diverges to infinity as we usually encounter in the Fourier transform; we assume some appropriate prescription, such as the distribution/hyperfunction formulation, has been applied to avoid this difficulty in this paper. The essential transition from the classical to the quantum field is done by replacing $a(k)$ and $a^{*}(k)$ with the operators satisfying the commutation relation of $\left[\hat{a}(k),\hat{a}(k^{\prime})^{\dagger}\right]=\hat{a}(k)\,\hat{a}^{\dagger}(k^{\prime})-\hat{a}^{\dagger}(k)\,\hat{a}(k^{\prime})=\delta(k-k^{\prime})\,.$ (11) The above quantization is based on the Cauchy surface of $t=\textnormal{constant}$. It should be noted that the quantization in this paper takes place only once at this point. Later we introduce the particle picture on the surface of constant Rindler time, but it is expressed by a linear superposition of $\hat{a}$ and $\hat{a}^{\dagger}$ and based on the commutation relation defined here. The Hamiltonian (8) can be expressed as a collection of quantum harmonic oscillators: $\hat{H}=\int\frac{\omega}{2}\left[\hat{a}(k)\,\hat{a}(k)^{\dagger}+\hat{a}^{\dagger}(k)\,\hat{a}(k)\right]dk\,.$ (12) Therefore, we can define the particle number as explained in the previous section, and ground sate of $\hat{H}$ is called “vacuum”. Now that we have the quantized operators $\hat{a}(k)$ and $\hat{a}^{\dagger}(k)$, we can calculate the field $\hat{\phi}(x,t)$ at any point of the spacetime as a quantum observable. Any classical quantity defined from the classical field $\phi$ can be quantized by replacing $\phi\rightarrow\hat{\phi}=\int\left[\hat{a}(k)\,u(k;x,t)+\hat{a}^{\dagger}(k)\,u^{*}(k;x,t)\right]dk\,.$ (13) ### 3.2 Rindler Coordinates The Hamiltonian $H$ in (8) is the energy with the conservation law based the Killing vector field of Minkowski time $\partial_{t}$. We examine in the following another conservation law resulting from another Killing vector field $\kappa x\partial_{t}-\kappa t\partial_{x}$, where $\kappa$ is a real constant that corresponds to the relativistic acceleration. The energy $M$ for this Killing vector field is written in the classical field theory as $M=\int_{\Sigma(\eta)}\frac{1}{\kappa(t^{2}-x^{2})}\left(t\phi_{x}^{2}+x\phi_{,t}^{2}\right)d\Sigma\,,$ (14) where $\Sigma(\eta)$ is a surface specified by $t/x=\tanh(\kappa\eta)$ and $x>0$. Clearly $M$ is not the same quantity as $H$, therefor, let us distinguish $M$ and $H$ by calling them “Rindler energy” and “Minkowski energy” respectively. The density of the Rindler energy is conserved locally, and there is no Rindler energy flow across the left and right Rindler wedges, thus we have $\partial M/\partial\eta=0$; once we calculate $M$ on a surface $\Sigma(\eta)$ with a given $\eta$, then the result holds for all $\eta$. When we choose $\eta=0$, we can express $M$ with the coefficients $a(k)$ of the Minkowski modes. Since $M$ is quadratic in $\phi$ we can write $M=\iint\left[A(k,k^{\prime})\,a(k)\,a(k^{\prime})+B(k,k^{\prime})\,a(k)\,a^{*}(k^{\prime})+C(k,k^{\prime})\,a^{*}(k)\,a^{*}(k^{\prime})\right]dk\,dk^{\prime}\,,$ (15) with coefficients $A$, $B$, and $C$ that do not depend on $\eta$ or $t$. The above quantity is quantized by replacing $a(k)\rightarrow\hat{a}(k)$ and $a^{*}(k)\rightarrow\hat{a}^{\dagger}(k)$ as $\hat{M}=\iint\left[A\,\hat{a}(k)\,\hat{a}(k^{\prime})+\frac{B}{2}\left(\hat{a}(k)\,\hat{a}^{\dagger}(k^{\prime})+\hat{a}^{\dagger}(k)\,\hat{a}(k^{\prime})\right)+C\,\hat{a}^{\dagger}(k)\,\hat{a}^{\dagger}(k^{\prime})\right]dk\,dk^{\prime}\,.$ (16) As noted before, this quantization is based on the Cauchy surface of $t=\textnormal{constant}$, not $\Sigma(\eta)$. Therefore, $\Sigma(\eta)$ does not have to be a Cauchy surface. What we wish to show in the following is that the particle numbers defined by $\hat{M}$ and $\hat{H}$ are not the same. Before that, we have to show that $\hat{M}$ surely can define the particle number. Suppose an operator $\hat{b}$ is defined as a linear superposition of $\hat{a}$ and $\hat{a}^{\dagger}$ as $\hat{b}(p)=\int\left[\alpha(p,k)\,\hat{a}(k)+\beta(p,k)\,\hat{a}^{\dagger}(k)\right]dk\,.$ (17) If $\hat{b}$ satisfies the commutation relation $[\hat{b}(p),\hat{b}^{\dagger}(p)]=\delta(p-p^{\prime})\,,$ (18) and $\hat{M}$ can be expressed with $\hat{b}$ as $\hat{M}=\frac{1}{2}\int\sigma\left[\hat{b}(p)\,\hat{b}^{\dagger}(p)+\hat{b}^{\dagger}(p)\,\hat{b}(p)\right]dp\,,$ (19) then we can define the particle number with $\hat{M}$ in the similar way as done in the previous subsection with $\hat{H}$. It is possible show the above two equations with direct calculation, however, it is easier to use the following Rindler coordinates $(\eta,\rho)$ for the right Rindler wedge i.e., region of $x>0,$ |x|>|t|: $t=\rho\sinh(\kappa\eta),\;x=\rho\cosh(\kappa\eta)\,.$ (20) Then the Rindler energy $\hat{M}$ may be written as $\hat{M}=\int_{0}^{\infty}\frac{1}{\kappa\rho}\left(\hat{\phi}_{,\eta}^{2}+\kappa^{2}\rho^{2}\hat{\phi}_{,\rho}^{2}\right)d\rho\,.$ (21) We introduce the eigenfunctions $v(p;\eta,\rho)=\frac{1}{\sqrt{4\pi\sigma}}\,\exp(-i\sigma\eta+i\kappa^{-1}p\ln(\kappa\rho))\,,$ (22) with $\sigma=|p|$. The wave function $\hat{\phi}$ can be expanded in the right Rindler wedge as $\hat{\phi}=\int\left[\hat{b}(p)\,v(p)+\hat{b}^{\dagger}(p)\,v^{*}(p)\right]dp\,,$ (23) then $\hat{M}$ in (21) can be cast into (19). In this context $\alpha$ and $\beta$ in (17) are equivalent to the Bogolubov coefficients that satisfy $\int\left[\alpha(p_{1},k)\alpha^{*}(p_{2},k)-\beta(p_{1},k)\beta^{*}(p_{2},k)\right]dk=\delta(p_{1}-p_{2})\,,$ (24) The above property combined with (11) yields the commutation relation of (18), therefore we can define particle numbers with $\hat{M}$. What we do next is to compare the eigenstates of $\hat{M}$ and $\hat{H}$. From (16) it is clear that $\hat{M}$ and $\hat{H}$ do not share the same set of eigenstates unless $A$ and $C$ vanishes. We can see from (17) and (15) that $A$ and $C$ vanished when $\beta=0$, however, $\beta$ can be evaluated by a straightforward integration (see, e.g., [11]), resulting $\left|\beta(p,k)\right|^{2}=\frac{1}{2\pi\omega}\,\frac{1}{e^{2\pi\sigma/\kappa}-1}\neq 0\,.$ (25) Therefore, the eigenstates of $\hat{M}$ are not the eigenstates of $\hat{H}$. This means that the ground state of $\hat{H}$ is not the ground state of $\hat{M}$, which can be stated in other words as “a vacuum defined by the Minkowski energy is not a vacuum defined by the Rindler energy”. The expected value of the particle number defined by Rindler energy (we call Rindler particle number hereafter) in the Minkowski vacuum can be calculated using (25), resulting the well known Plankian distribution [11]. What we have seen above is not surprising since different operators may have different sets of eigenstates. However, it contradicts with the picture of the pair production by the gravitational force often found in intuitive explanations like the one in Section 2. The particle number defined by Minkowski energy (Minkowski particle number hereafter) is zero for all $t$, and the Rindler particle number has the fixed Plankian distribution for all $\eta$. This is consistent with the time symmetry; the vacuum in the flat spacetime must be symmetric in time, both in $t$ and $\eta$, but the particle creation process is not symmetric. ### 3.3 Origin of the Radiation The Rindler particle number in the Unruh process is calculated by the expected value of the Rindler energy for the ground state of the Minkowski energy, i.e., $\left\langle 0_{H}\right|\hat{M}\left|0_{H}\right\rangle$. If the “vacuum” were a completely empty space, i.e., $\hat{a}\left|0_{H}\right\rangle=\hat{a}^{\dagger}\left|0_{H}\right\rangle=0$, then $\left\langle 0_{H}\right|\hat{M}\left|0_{H}\right\rangle$ would vanish. However, this is not true since the quantum ground state has zero point oscillation, and hence $\left\langle 0_{H}\right|\hat{M}\left|0_{H}\right\rangle\neq 0$. This means the “particles” found in the Unruh process comes from the zero point oscillation of the Minkowski modes. Then what we wish to know is the properties of zero point oscillations that contribute to the continuous radiation of Rindler energy. In the present paper we concentrate on right moving waves, i.e., $\omega k<0$ or $\sigma p<0$, since their analog in the Schwarzschild spacetime play the key role in the black hole evaporation. It should be noted, however, left moving waves are also problematic and should be examined in the next step. We first examine the properties of waves in the classical limit, and then apply the result to the quantum vacua. To begin with, we observe that the a eigenmode (22) has infinite Rindler energy in a finite region of $0\leq\rho<\rho_{c}$ with arbitrary position $\rho_{c}$ in the right Rindler wedge; this can be confirmed by the following direct integration: $\lim_{\varepsilon\rightarrow 0}\int_{\varepsilon}^{\rho_{c}}\frac{1}{\kappa\rho}\,\left(\hat{v}_{,\eta}^{2}+\kappa^{2}\rho^{2}\hat{v}_{,\rho}^{2}\right)\,d\rho\rightarrow\infty\,.$ (26) Also it is easy to confirm there is a constant rightward outflux of the Rindler energy at $\rho=\rho_{c}$ by direct calculation. This outflux comes from the region of $0\leq\rho<\rho_{c}$, but the total Rindler energy can be conserved because the amount of the Rindler energy in that region is infinitely large. Belinski [3] considered this fact as physically unacceptable and concluded the radiation results from $v(p)$ is just a mathematical illusion. The present paper takes a different interpretation. The infinite Rindler energy can be physically real as long as we believe zero point oscillations exist for any high frequency modes, because the collection of such oscillations has infinitely large Rindler energy even in a finite volume. To see this, we examine the behavior of the a wave packet in the following form (“$c.c.$” means complex conjugate): $\phi(\eta,\rho)=\exp\left(\frac{-(\rho-e^{\kappa\eta}\,\rho_{0})^{2}}{(e^{\kappa\eta}\,s_{0})^{2}}\right)\,\exp[-i\sigma\eta+i\kappa^{-1}p\ln(\kappa\rho)]+c.c.\,.$ (27) This wave packet was initially localized around $\rho=\rho_{0}$ with width $s_{0}$ at $\eta=0$, and propagates rightward. The width of the packet becomes larger and the wave number becomes smaller as a result of wave propagation. The wave packet can be expanded by the Minkowski modes $u(k;t,x)$ as $\phi(t,x)=\int\left[(\phi,u(k))\,u(k;t,x)+(\phi,u^{*}(k))\,u^{*}(k;t,x)\right]\,dk$ (28) with the Klein-Goldon inner products $(\phi_{1},\phi_{2})$, which can be calculated at $t=\eta=0$ as $(\phi_{1},\phi_{2})=\int\left[\phi_{1,t}\phi_{2}^{*}-\phi_{1}\phi_{2,t}^{*}\right]_{t=0}\,dx\,.$ (29) When $s\ll\rho_{0}\ln(\kappa\rho_{0})$ then we can approximate $\phi(\eta,\rho)\simeq\exp\left(\frac{-(\rho-\rho_{1})^{2}}{s_{1}^{2}}-ip\eta+\frac{ip}{\kappa}\left[\ln(\kappa\rho_{1})+\frac{1}{\rho_{1}}\,(\rho-\rho_{1})\right]\right)+c.c.\,,$ (30) where $s_{1}=s_{0}\,e^{\eta}$ and $\rho_{1}=\rho_{0}\,e^{\eta}$ are the width and center of the wave packet at a time $\eta$. Using the above approximation we obtain $\phi(t,x)=\frac{2}{s_{0}\sqrt{\pi}}\exp\left[i\kappa^{-1}p\ln(\kappa\rho_{0})\right]\int e^{-ik\rho_{0}}\exp\left[-s_{0}^{2}(k-p/\kappa\rho_{0})^{2}\right]e^{-i\omega t+ikx}dk+c.c.$ (31) from (28) with (29). The above expression means that the wave packet comes from the Minkowski modes with wave number around $k_{0}=p/\kappa\rho_{0}$ when $s_{0}\gg\kappa\rho_{0}/p$. Suppose we find a wave packet around $\rho_{1}$ at a given time $\eta=\eta_{1}\,(>0)$ in the Rindler space then its position at $\eta=0$ was $\rho_{0}=\rho_{1}\,e^{-\kappa\eta_{1}}$, therefore, the packet consists of the Minkowski modes with $k\sim k_{0}=p\,e^{\kappa\eta_{1}}/\kappa\rho_{0}$. When we regard the wave field at $\eta=\eta_{1}$ as a superposition of such wave packets, we see that the waves in the region of $0<\rho<\rho_{1}$ at $\eta=\eta_{1}$ consists of Minkowski modes with wave numbers larger than $k_{0}=p\,e^{\kappa\eta_{1}}/\kappa\rho_{0}$. Since $k_{0}\rightarrow\infty$ in the limit of $\eta_{1}\rightarrow\infty$, we understand the Rindler energy radiation at the distant future in $\eta$ comes from the Minkowski modes with infinitely large wave numbers. The Rindler coordinates represent an observer with constant acceleration, and the relative velocity of the accelerating observer to the rest frame becomes infinitely large in the limit of $\eta\rightarrow\infty$. Waves with finite wave numbers in this limit are infinitely red shifted, therefore its original wave number must have been infinitely large. Now let us apply the above observation to quantum vacua to see the origin of the Rindler particles. Suppose the quantum state is Minkowski vacuum, i.e., the ground state of the Minkowski energy. Then the state has zero point oscillation up to infinitely large wave numbers. Usually the energy of these zero point oscillation is subtracted out by normal ordering, and we regard there is no particle in the ground state. However, the ground state of the Minkowski energy is not the ground state of the Rindler energy, which means the existence of the Rindler particles. The continuous radiation of Rindler particles is possible for any large $\eta$ because the zero point Minkowski energy exist for modes with any large wave numbers. The radiated Rindler energy must have been piled up near $\rho=0$ at the initial time of $\eta=0$, since there is no particle creation as we have seen in the previous subsection. ## 4 Hawking Radiation Let us move on to the Hawking radiation from a Schwarzchild black hole in this section. We introduce the Schwarzchild coordinates $(t,r,\theta,\varphi)$ whose metric is $ds^{2}=\left(1-\frac{2M}{r}\right)dt^{2}-\left(1-\frac{2M}{r}\right)^{-1}dr^{2}-r^{2}d\theta^{2}-r^{2}\cos^{2}\theta d\varphi\,.$ (32) The Kruscal coordinates $(u,v,\theta,\varphi)$ are related to $(t,r,\theta,\varphi)$ as $\displaystyle u^{2}-v^{2}$ $\displaystyle=$ $\displaystyle 2M(2M-r)\exp\left(\frac{r}{2M}\right)\,,$ (33) $\displaystyle\left|\frac{u-v}{u+v}\right|$ $\displaystyle=$ $\displaystyle\exp\left(\frac{t}{2M}\right)\,.$ (34) The rest of the coordinates, $\theta$ and $\varphi$, are unchanged. It is generally accepted that the quantum properties of vacuum near the Schwarzchild event horizon is essentially the same as those in the Rindler spacetime [2, 12], therefore, the results we obtained in the previous section are basically valid by replacing Minkowski/Rindler coordinates with Kruscal/Schwarzchild coordinates (note: $t$ in the Schwarzchild spacetime corresponds to $\eta$, not $t$, in the flat spacetime). There are, however, two fundamental differences. One is the definition of the energy in Kruscal coordinates, and the other is the backreaction of the quantum fields to the black hole metric. The latter causes the essential problem in the scenario of the black hole evaporation. The first difference is about the energy that corresponds to the Minkowski energy. The Kruscal time $u$ is not a global Killing time, and thus there is no global energy conservation law like for the Minkowski energy. However, $u$ can be approximately regarded as a Killing time near the horizons. As we have seen in the previous section, the radiation in later time comes from the infinitely high frequency modes infinitesimally near the horizon, therefore the energy is conserved approximately for these waves. This is in parallel to the approximation of geometrical optics used by Hawking in his original paper [1]; geometrical optics assumes locally constant frequency, which means locally constant energy. Hereafter we assume the energy corresponds to the Kruscal time $u$ is approximately conserved, and treat it in the same way as for the Minkowski energy in the previous subsection. We call it Hartle-Hawking energy since its ground state is often called Hartle- Hawking vacuum. The energy defined with the Schwarzchild time is called Boulware energy hereafter for the same reason. We have another problem in the definition of the Hartle-Hawking energy in a Schwarzchild spacetime. Minkowski energy is defined as an integration over a surface of $t=\textnormal{constant}$ in a flat spacetime. If we introduce a similar definition for the Hartle-Hawking energy with Kruscal time $u=\textnormal{constant}$, the energy would include the part of the white hole in the extended Schwarzchild spacetime. This difficulty may be avoided by analyzing the black hole formation process by a star collapse, or the analytical continuation method proposed by Hartle and Hawking [13]. A detailed analysis on this point will be given in a forthcoming paper of the author. The second difference is far more serious; the energy of zero point oscillations may change the metric. Usually the zero point energy is subtracted out by normal ordering in the source term of the Einstein equation, and the vacuum does not have effect on the metric. This means only the excited state of the energy can cause gravitation. However, as we have seen in the previous section, the ground state of Hartle-Hawking energy is not the ground state of the Boulware energy, and vise versa. In a flat spacetime we consider the ground state of the Minkowski energy is the state of no gravitation, because the Minkowski coordinates are the “natural” coordinate system. We have seen there is infinite accumulation of the Rindler energy near $\rho=0$ for a Rindler mode $v(p;\eta,\rho)$. There must be an infinitely strong source of gravitational force at $\rho=0$ if we assume the ground state of the Rindler energy is the state of no gravitation, since the vacuum state defined by the Minkowski energy is the excited state of the Rindler energy. This is not plausible, and we can conclude the ground state of the Minkowski energy has no effect on the metric. In contrast, we do not know which coordinate system is “natural” to calculate the energy (stress-energy tensor) for a curved spacetime in general (see, e.g., [14]). The Schwarzchild coordinates are implicitly assumed to be “natural” in the scenario of the black hole evaporation, in other words, excited states of Boulware energy causes the gravity. The black hole evaporation is believed to be the result of the Hawking radiation that carries the energy away from the black hole, and the energy in this context is the Boulware energy; this means the Boulware energy can have effects on the black hole metric somehow. If this is true, however, the Rindler energy radiated at later times must have been exist just outside of the horizon from the beginning [3]. The energy does not come from inside the black hole, but comes from the Minkowski modes with extremely high wave numbers. This means the backreaction of the quantum field is far from negligible to the black hole metric [4]. On the contrary, we can imagine the gravity is caused by the exited state of Hartle-Hawking energy and its ground state has no effect on the metric. The black hole can exist in this case, however, it cannot evaporate. There exists a constant outflow of Boulware energy, but it is the ground state of the Hartle-Hawking energy and does not have a backreaction on the metric. Consequently the black hole metric is unchanged at all, just like the Unruh process does not change the flat metric. There can be other possibilities for the effect of the zero point energy to the metric, however, it is hard to imagine there is an extremely convenient case which is favorable for the scenario of black hole evaporation. ## 5 Summary What we have seen in the present paper are: 1. 1. The ground state of Minkowski/Hartle-Hawking energy is not the ground states of Rindler/Boulware energy, and this is what causes the Hawking/Unruh process. 2. 2. The quantum state is unchanged and particles pairs are not created in any coordinate system; the number of Minkowski/Hartle-Hawking particles is zero and number of Rindler/Boulware particles has time stationary Plankian distribution all through the time, where “time” means the Rindler/Schwarzchild time . 3. 3. The radiation of Rindler/Boulware energy in the distant future of Rindler/Schwarzchild time comes from the zero point oscillation of Minkowski/Hartle-Hawking energy with infinitely large wave frequencies. 4. 4. The effect of zero point energy to the metric is not known, however, we have the following two possibilities for a Schwarzchild spacetime. The scenario of black hole evaporation is inconsistent in both cases. 1. (a) If the Hawking radiation causes the black hole evaporation, it means the excitation in the Boulware energy can cause the metric change. The Boulware energy radiated later time was accumulated near the horizon at the initial time, whose existence essentially alter the Schwarzchild metric from the beginning. 2. (b) If, on the contrary, the Boulware energy of the Hawking radiation does not affect the metric, then the Schwarzchild metric can exist as we expect, but exists forever. There is no evaporation of the black hole. We started the present study by assuming that the cis-Plankian physics is valid for any small scale phenomena, and end up with the inconsistency of black hole evaporation. This fact means we have no reason to believe the black hole evaporation. A new theory of physics in trans-Plankian scale may save the evaporation, but may not; we can imagine anything, but cannot believe. What we can say for sure is that the physics we know at the present is not able to predict the black hole evaporation, if the calculations in the present paper are correct. We see the scenario of black hole evaporation is inconsistent, however, we do not know what is the consistent theory even within the cis-Plankian regime. The problem deeply depends on the renormalization procedure in curved spacetimes, to which we do not know the answer yet. It is often said Hawking process can be a touchstone for the theory of quantum gravity. The author of the present paper would like to say it also can be a touchstone for the renormalization theory, or theory on what is avoided by renormalization at the present, in curved spacetimes. ## References * [1] Hawking, S. W., Comm. Math. Phys. 43, 199 (1975). * [2] Unruh, W. G., Phys. Rev. D14, 870 (1976). * [3] Belinski, V. A., Phys. Lett. A209, 13 (1995); Phys Lett. A354, 249 (2007). * [4] Helfer, A. D., gr-qc/0008016. * [5] Helfer, A. D., Int. J. Mod. Phys. D13, 2299 (2004), gr-qc/0503052. * [6] Helfer, A. D., Rept.Prog.Phys. 66, 943 (2003), gr-qc/0304042. * [7] Gibbons, G. W., in _Proc. First Marcel Grossman Meeting on General Relativity_ , ed. R. Ruffini, 499 North-Holland (1977). * [8] Jacobson, T., Phys. Rev. D44, 1731 (1991); Phys. Rev. D48, 728 (1993); Prog The Phys Suppl 136, 1 (1999), hep-th/0001085. * [9] Unruh, W. G., Phys. Rev., D51, 2827 (1995); Brout, R., Massar, S., Parentani, R., and Spindel, Ph., Phys. Rev. D52, 4559 (1995), hep-thy/9606121; Corely, S. and Jacobson, T., Phys. Rev., D53, 6720, hep-th/9601973; Himemoto, Y. and Tanaka, T., Phys. Rev. D61, 064004, gr-qc/9904076; Saida, H. and Sakagami, M, Phys. Rev. D61, 084023, gr-qc/9905034. * [10] Messiah, A., Quantum Mechanics, North Holland, (1961). * [11] Wipf, A., in _Black Holes: Theory and Observation, Proc. 179th W. E. Heraeus Seminar_ , ed. F. W. Hehl, C. Kiefer, and R. J. K. Metzler, 385, Springer (1998), hep-th/9801025; DeWitt, B., The Global Approach to Quantum Field Theory, Oxford (2003). * [12] Fulling, S. A., J. Phys. A10, 917 (1977); Wald, R. M.,_Quantum Field in Curved Spacetime and Black Hole Thermodynamics,_ U. Chicago Press (1994). * [13] Hartle, J. B., and Hawking, S. W., Phys. Rev. D13, 2188 (1976). * [14] Birrell, N. D., and Davies, P. C. W., _Quantum Fields in Curved Space_ , Cambridge Univ. Press (1982); Fulling, S. A., _Aspects of Quantum Field Theory in Cureved Space-Time_ , Cambridge Univ. Press (1989).

arxiv-papers

2024-09-04T02:48:54.229400

0803.1476

11institutetext: Mark Marley 22institutetext: NASA Ames Research Center, Mail Stop 245-3, Moffett Field CA 94035, 22email: Mark.S.Marley@NASA.gov 33institutetext: S. Leggett 44institutetext: Gemini Observatory, 670 North A’ohoku Place Hilo, HI 96720 44email: skl@gemini.edu # The Future of Ultracool Dwarf Science with JWST Mark S. Marley and S.K. Leggett ###### Abstract Ultracool dwarfs exhibit a remarkably varied set of characteristics which hint at the complex physical processes acting in their atmospheres and interiors. Spectra of these objects not only depend upon their mass and effective temperature, but also their atmospheric chemistry, weather, and dynamics. As a consequence divining their mass, metallicity and age solely from their spectra has been a challenge. JWST, by illuminating spectral blind spots and observing objects with constrained masses and ages should finally unearth a sufficient number of ultracool dwarf Rosetta Stones to allow us to decipher the processes underlying the complex brown dwarf cooling sequence. In addition the spectra of objects invisible from the ground, including very low mass objects in clusters and nearby cold dwarfs from the disk population, will be seen for the first time. In combination with other ground- and space-based assets and programs, JWST will usher in a new golden era of brown dwarf science and discovery. ## 1 Introduction The explosive growth of brown dwarf and ultracool dwarf discoveries over the past dozen years has been so extraordinary that it is a rare paper in the field that does not open by remarking upon it. The first undisputed brown dwarf, Gl 229 B, was discovered as a companion to an M dwarf in 1995 (Nakajima et al. 1995). The Two-Micron All Sky Survey (2MASS, Skrutskie et al. 2006)) and the Sloan Digital Sky Survey (SDSS, York et al. 2000) subsequently revealed large numbers of ultracool low-mass field dwarfs. These surveys first led to the discovery of the isolated field late-L dwarfs (Kirkpatrick et al. 1999), then the mid-T dwarfs (Burgasser et al. 1999, Strauss et al. 1999) and finally the early-T dwarfs (Leggett et al. 2000). Today over 600 warm ($T_{\rm eff}\sim 2400$ to 1400 K) L and cool (600 to 1400 K) T dwarfs are known111See http://www.DwarfArchives.org and the quest for the elusive, even cooler “Y” dwarfs is ongoing. Note that collectively late M and later type dwarfs are often termed ‘Ultracool Dwarfs’ (or UCDs) to avoid having to distinguish whether particular warm objects in this group are above or below the hydrogen burning minimum mass, the requirement for bestowing the term ‘brown dwarf’. Brown dwarfs will continuously cool over time. The more massive UCDs will eventually arrive on the bottom of the hydrogen burning main sequence (Burrows et al. 1997). Figure 1: The most prominent signatures of the ultra cool dwarf spectral sequence are seen in these 0.65 to $14.5\,\rm\mu m$ spectra of a mid-M, L, and T dwarfs as well as Jupiter (adapted from Cushing et al. (2006)). The spectra have been normalized to unity at $1.3\,\rm\mu m$ and multiplied by constants. Major absorption bands are marked. The collision-induced opacity of $\rm H_{2}$ is indicated as a dashed line because it shows no distinct spectral features but rather a broad, smooth absorption. Jupiter’s flux shortward of $\sim 4\,\rm\mu m$ is predominantly scattered solar light; thermal emission dominates at longer wavelengths (near- and mid-infrared Jovian spectra from Rayner, Cushing & Vacca (in preparation) and Kunde et al. (2004), respectively). Ultracool dwarf science is exciting not only for the rapid pace of discovery, but for a host of other reasons as well. First, since brown dwarfs lack an internal energy source (beyond a brief period of deuterium burning), they cool off over time; they thus reach effective temperatures below those found in stars and enter the realm where chemical equilibrium favors such decidedly ‘unstellar’ atmospheric species as $\rm CH_{4}$ and $\rm NH_{3}$. Along with these more typical ‘planetary’ gasses, silicate and iron clouds are found in their atmospheres, leading to interesting, complex interactions between atmospheric chemical, radiative-transfer, dynamical, and meteorological processes. Ultracool dwarfs thus bridge the domain between the bottom of the stellar main sequence and giant planets. They are a laboratory for understanding processes that will also be important in the characterization of extrasolar giant planets. Second, they occupy the low mass end of the stellar initial mass function. Understanding the IMF requires that we understand the masses of individual field objects, which ultimately requires an understanding of their luminosity evolution as well as the dependence of their spectra on mass, gravity, effective temperature and metallicity. Finally, as terra incognita, brown dwarfs (some of our nearest stellar neighbors) have offered a series of surprises that test our ability to understand the universe around us. Figure 2: Important chemical equilibrium boundaries for substellar objects (modified from Lodders & Fegley 2006). Green, red, and blue lines denote various condensation boundaries for a solar abundance mixture of gasses in a substellar atmosphere. Light purple lines denote equilibrium boundaries between important gaseous species. Grey dashed lines show model atmospheric temperature-pressure profiles for M, L and T dwarfs (the latter specifically for Gl 229 B) as well as for Jupiter. As one moves upwards in the diagram along the model $(T,P)$ curves, the labeled species will condense at the intersection with the condensation curves and would be expected to be absent from the gas at lower temperatures further up along the model profiles. This figure can be compared with Figure 1 to understand why spectral features for various compounds are present or absent in each observed spectrum. The most distinctive features of the UCD spectral sequence are highlighted in Figure 1. At effective temperatures below those of late-M dwarfs, several chemical changes (illustrated in Figure 2) occur that strongly impact the spectral energy distribution. First, major diatomic metal species (particularly TiO and FeH) become incorporated into grains, leading to the gradual departure of hallmarks of the M spectral sequence (Kirkpatrick et al. 1999). Second, the formation of iron and silicate grains produces optically thick clouds that veil gaseous absorption bands and redden the near-IR $JHK$ colors of L dwarfs. The atmospheric temperature domain where these clouds are most important is $\sim 1500-2000\,\rm K$ (e.g. Ackerman & Marley 2001). At lower $T_{\rm eff}$, the clouds lie near or below the base of the wavelength- dependent photosphere, and only marginally affect the SEDs of T dwarfs. Finally, CH4 supplants CO as the dominant carbon-bearing molecule. This transition is first noted in the 3–4 $\mu$m spectra of mid-L dwarfs (Noll et al. 2000) and appears in both the $H$ and $K$ bands of T0 dwarfs (Geballe et al. 2002). Together, increasing CH4 absorption and sinking cloud decks cause progressively bluer near-IR colors of T dwarfs. For types T5 and later, significant collision-induced H2 opacity in the $K$ band enhances the trend toward bluer near-infrared colors. These changes in spectral features are used to assign spectral types to UCDs as briefly explained in §2. Given assigned spectral types, measurement of the bolometric luminosity of individual objects along with their parallaxes connect the spectral sequence to effective temperature. Figure 3 illustrates the effective temperature as a function of spectral type from late M through late T. While the general correlation of increasing spectral type with falling effective temperature is unmistakable, a remarkably rapid set of spectral changes (as expressed in the variation in spectral type) happens over a relatively small span of $T_{\rm eff}$ near 1400 K. As we will discuss, understanding this variation in expressed spectral signatures, the ‘L to T transition’, is a key subject of current brown dwarf research. Figure 3: Effective temperature as a function of infrared spectral type for ultracool dwarfs with known parallax (data from Golimowski et al. 2004, Vrba et al. 2004, and Luhman et al. 2007). Note the roughly constant effective temperature for dwarfs of spectral types from late L to early T. See Kirkpatrick (2007) for further discussion. Most of the scientific inquiry into these ultracool dwarfs has focused on their formation and youth, on the resultant initial mass function, and on the determination of their global properties, particularly mass, effective temperature, metallicity, and cloudiness. This review will focus primarily on the latter areas. Burrows et al. (2001) provide a much more in depth review and background to brown dwarf science and is an excellent starting point for those new to the subject. Kirkpatrick (2007) provides a more current look at outstanding issues in the field from an observational perspective. Here we first present a very brief review of the ultracool dwarf spectral types and the nature of the current datasets. We then move on to discuss the role that clouds and atmospheric mixing play in controlling the emitted spectra of these objects and the enigmatic L- to T-type transition. Because clouds control the spectral energy distribution of the L and early T dwarfs, and since clouds are inherently difficult to model, constraining gravity solely by comparison of observations to spectral data is particularly challenging. Finally we will close with a look forward to some of the ultracool dwarf science opportunities that will be enabled by JWST. Because of space limitations we neglect several other important avenues of UCD research, including studies of the IMF, very young objects, and of objects with unusual colors. ## 2 Spectral Type The currently known ultracool dwarfs span spectral types from late M, L0 through L9, and T0 through T9. The TiO and VO bands, which dominate the optical portions of late-M dwarf spectra, disappear in the L dwarfs (Kirkpatrick et al. 1999), where metallic oxides are replaced by metallic hydrides and where features due to neutral alkali metals are strong. To systematize such objects Kirkpatrick et al. established spectral indices and defined an optical classification scheme for L dwarfs, which is commonly used. The indices measure the strengths of TiO, VO, CrH, Rb and Cs features as well as a red color term, at wavelengths between 0.71 and 0.99 $\mu$m. With the discovery of the T dwarfs, which have very little flux in the optical, Geballe et al. (2002) defined a near-infrared classification scheme that encompassed both the L and T dwarfs. The indices measure the strength of the H2O and CH4 absorption features between 1.1 and $2\,\mu$m, and for the early L dwarfs a red color term is also used which is slightly modified from Kirkpatrick et al. (1999). Burgasser et al. (2002a) introduced a near-infrared classification scheme for T dwarfs that was very similar to that of Geballe et al. The two schemes were unified in Burgasser et al. (2006) and this near-infrared scheme is the commonly used scheme for typing T dwarfs. For L dwarfs, both the optical Kirkpatrick et al. and the near-infrared Geballe et al. schemes are used; these usually produce the same type (the Geballe et al. scheme was pinned to the Kirkpatrick et al. types), but for L dwarfs with unusual colors they can give significantly different types. Because of this, the classification scheme used for L dwarfs should always be specified. Leggett et al. (2007, and other work referenced therein) explore the possible signatures of the next spectral type, for which the letter Y has been suggested (Kirkpatrick 2005). It is likely that NH3 features will join the familiar water and methane absorption seen in T dwarfs as the effective temperatures approach 600 K. In actuality the situation is likely more complex: it is already known that the atmospheric NH3 abundance (as seen in Spitzer mid-infrared spectra of T dwarfs) is reduced by vertical mixing which drags N2 up from deeper layers in the atmosphere (e.g., Saumon et al. (2006)). If this mechanism continues to act at low effective temperatures (which is not a certainty (Hubeny & Burrows 2007)) the near-infrared ammonia features may be weaker than expected. Another outstanding problem with predicting the signature of the proposed Y dwarfs, is that the linelist for NH3 is very incomplete at 1.0-1.5 $\mu$m. Since the near-infrared flux of Y dwarfs is expectedly to rapidly ‘collapse’ with falling $T_{\rm eff}$ (Burrows et al. 2003), it may even be appropriate to ultimately type these objects with mid-, instead of near-infrared, spectra. In this case spectra obtained by Spitzer or JWST would be required for spectral typing. Regardless, dwarfs with $T_{\rm eff}\sim 650\,\rm K$ are now being found (e.g. Warren et al. 2007), and it is likely that soon temperatures where significant spectral changes occur will be reached. ## 3 Ultracool Dwarf Datasets The L and T dwarfs were discovered primarily as a result of the far-red and near-infrared Sloan Digital Sky Survey and 2 Micron All Sky Survey (e.g. Kirkpatrick et al. 1999, Strauss et al. 1999). This continues with current surveys - the Canada France Hawaii Brown Dwarf Survey and the UKIRT Infrared Deep Sky Survey (e.g., Lodieu et al. 2007) are identifying extreme-T dwarfs by their very red far-red and blue near-infrared colors. Spectral classification is carried out in the far-red or near-infrared. Hence the existing data for L and T dwarfs is primarily far-red and near-infrared imaging and spectroscopy. The spectroscopy has been medium- or low-resolution, both because that is all that is required for the spectral classification, but also because the dwarfs are faint. Some ground-based imaging and spectroscopy has been carried out at 3.0-5.0 $\mu$m (e.g. Noll et al. 2000, Golimowski et al. 2004). Such work is extremely challenging due to the very high and rapidly variable sky background at these wavelengths, and only the brightest dwarfs could be observed from the ground. This situation changed with the launch of the Spitzer Space Telescope. Roellig et al. (2004) and Patten et al. (2006) demonstrated the quality and quantity of mid-infrared imaging and spectroscopy of L and T dwarfs that Spitzer could produce, and such work continues through the current, final, cryogenic cycle. IRS spectral data, which span 6 to 15 $\mu$m, show a strong 11 $\mu$m NH3 absorption feature in T dwarfs, as well as H2O and CH4 absorption features in both L and T dwarfs (Figure 1). IRAC photometry covers the 3 to 8 $\mu$m wavelength range, and the 3 to 5 $\mu$m range may continue to be available in the warm-Spitzer era. These IRAC bandpasses include CH4, CO and H2O features, and signatures of vertical transport have been recognized in the photometry (Leggett et al. 2007). Warren et al. (2007) further suggest that the $H-$ [4.49] color may be a very good indicator of temperature for dwarfs cooler than 1000 K. For extreme-T dwarfs the near-infrared CH4 and H2O bands are so strong that it will be difficult to measure an increase in their strength, hence the mid-infrared may prove to be vital to interpreting the cold objects. ## 4 Ultracool Dwarf Atmospheres Ultracool dwarf emergent spectra are controlled by the variation in abundances of important atomic, molecular, and grain absorbers both with height in the atmosphere at a given age and over time as the objects cool. The major atomic and molecular absorption features are imprinted on spectra that have no true continuum222Sharp & Burrows (2007) and Freedman et al. (2008) discuss the atmospheric opacity sources in detail.. Flux emerges from a many-scale-height- thick range of depths in the atmosphere as a function of wavelength. For example (Fig. 4), in an early L dwarf, brightness temperatures333Brightness temperature (the temperature that a blackbody that emits radiation of the observed intensity at a given wavelength) is commonly used in planetary atmospheres studies to elucidate the temperature of the emitting level in an atmosphere. range from 1000 K in the depths of alkali absorption lines (Burrows et al. 2000) in the far red, to well over 2000 K in the molecular windows in between strong water absorption bands. By providing a continuum opacity source, clouds can limit the flux emerging in some molecular window regions, but not others. Furthermore the strength of some molecular absorption features, particularly $\rm CH_{4}$, CO, and $\rm NH_{3}$, can depend on the strength of mixing in the atmosphere. Thus a full description of an ultracool dwarf atmosphere hinges on the dwarf’s gravity, effective temperature, cloud properties, and mixing. In this section we summarize the important unsolved problems related to these atmospheres. Figure 4: Brightness temperature as a function of wavelength for atmosphere models which include (solid) or exclude (dotted) silicate and iron clouds (Ackerman & Marley 2001). Brightness temperature increases downward to suggest increasing depth in the atmosphere from which the wavelength-dependent flux emerges. The solid straight line indicates the base of the silicate cloud while the long dashed line denotes the ‘top’ of the cloud (the level in the atmosphere at which the cloud column extinction reaches 0.1). Shading suggests the decrease in cloud extinction with altitude. Since cloud particle radii exceed $10\,\rm\mu m$ in these models, the Mie extinction efficiency is not a strong function of wavelength over the range shown. Shown are models characteristic of (a) an early-type L dwarf with $T_{\rm eff}=1800\,\rm K$, (b) a late-type L dwarf with $T_{\rm eff}=1400\,\rm K$, and (c) a T dwarf with $T_{\rm eff}=900\,\rm K$ . All of these models are for solar composition and gravity appropriate for a 30 Jupiter-mass brown dwarf. Note that the spectral region just longward of $1\,\rm\mu m$ is particularly sensitive to the cloud opacity. ### 4.1 Clouds At high effective temperatures the column abundance of condensates (see Marley 2000 for a discussion of influences on cloud optical depth) is low and the difference between models computed with and without cloud opacity is slight (Figure 4). At lower temperatures, however, the cloud substantially alters the temperature profile of the atmosphere and provides a continuum opacity source that limits the depth to which the usual molecular windows probe into the atmosphere. By the effective temperature of the mid-T dwarfs, however, most of the flux emerges from above the cloud level and the clouds are again less important. For the effective temperature range of the mid to late L dwarfs and the early T dwarfs, however, clouds clearly play a very large role in controlling the vertical structure and emergent spectra of brown dwarfs. Thus any attempt to model brown dwarf atmospheres must include a treatment of clouds. However clouds are the leading source of uncertainty in terrestrial atmosphere models and are inherently difficult to model. Their influence depends on the variation of particle size, abundance, and composition with altitude, which in turn depend on the complex interaction of many microphysical processes (e.g., Ackerman & Marley 2001, Helling et al. 2006). Fits of model spectra to observational data are highly sensitive to the treatment of clouds in the underlying atmosphere model and the approaches taken by various modeling groups vary widely (see the comparison study in Helling et al. 2008a). For example Cushing et al. (2008) demonstrate reasonably accurate fits of model spectra to near- and mid-IR spectra of a sample of L and T dwarfs, but the precise values of effective temperature and gravity obtained from the fits depend entirely upon the cloud sedimentation efficiency (Ackerman & Marley 2001) assumed. Since the models are highly dependent on the cloud description, the derived effective temperature and gravities, while plausible, are nevertheless uncertain. A similar conclusion was reached by Helling et al. (2008b). Finding a selection of L dwarfs with known $g$ and $T_{\rm eff}$ that could serve as calibrators of the model spectra would be invaluable. L dwarf companions to main sequence stars with constrained ages, L dwarf binaries with resolved orbits, and L dwarfs in clusters of known ages are all promising targets for such work. ### 4.2 Characterizing Clouds The spectral range of the InfraRed Spectrometer (IRS) on Spitzer includes the $10\,\rm\mu m$ silicate feature which arises from the Si-O stretching vibration in silicate grains. The spectral shape and importance of the silicate feature depends on the particle size and composition of the silicate grains. According to phase-equilibrium arguments, in brown dwarf atmospheres the first expected silicate condensate is forsterite $\rm Mg_{2}SiO_{4}$ (Lodders 2002), at $T\sim 1700\,\rm K$ ($P=1\,\rm bar$). Since Mg and Si have approximately equal abundances in a solar composition atmosphere, the condensation of forsterite leaves substantial silicon, present as SiO, in the gas phase. In equilibrium, at temperatures about 50 to 100 K cooler than the forsterite condensation temperature, the gaseous SiO reacts with the forsterite to form enstatite, $\rm MgSiO_{3}$ (Lodders 2002). The precise vertical distribution of silicate species depends upon the interplay of the atmospheric dynamics and chemistry and such details have yet to be fully modeled, although efforts to improve the detailed cloud modeling continue (e.g., Cooper et al. 2003; Woitke & Helling 2004; Helling & Woitke 2006; Helling et al. 2008b). Figure 5: Top: Spitzer IRS spectrum of 2MASS 2224 (L4.5) and the best fitting model from Cushing et al. (2006). Middle: Optical absorption ($Q_{\rm abs}/a$) for amorphous enstatite ($\rm MgSiO_{3}$) and forsterite ($\rm Mg_{2}SiO_{3}$) for three different particle sizes, 0.1, 1, and $10\,\rm\mu m$. Bottom: Optical absorption for crystalline enstatite, also for three different particle sizes. The deviation of the model (shifted vertically) from the data suggests that additional small, and perhaps crystalline, silicate grains are required to adequately account for the observed spectrum. In brown dwarf clouds there is likely a range of particle sizes, ranging from very small, recently condensed grains, to larger grains that have grown by agglomeration. The mean particle size for silicate grains in L dwarf model atmospheres is typically computed to be in the range of several to several tens of microns (Ackerman & Marley 2001, Helling et al. 2008b) . Figure 5 compares the absorption efficiency of silicate grains of various sizes, composition, and crystal structures to the spectrum of 2MASS J2224-0158 (L4.5). For each species the quantity $Q_{\rm abs}/a$, or Mie absorption efficiency divided by particle radius, is shown; all else being equal, the total cloud optical depth is proportional to this quantity (Marley 2000). Large grain sizes tend to have a relatively flat absorption spectra (dashed lines) across the IRS spectral range. Only grains smaller than about $3\,\rm\mu m$ in radius show the classic $10\,\rm\mu m$ silicate feature (Hanner et al. 1994). Figure 5 suggests that the mismatch between the models and data may arise from a population of silicate grains that is not captured by the cloud model used to construct the figure (Ackerman & Marley 2001). The actual silicate cloud may contain both more small particles and a mixture of enstatite and forsterite grains (e.g., Helling & Woitke 2006), although detailed models for this particular dataset have not been attempted. Furthermore the model shown in the figure employs optical properties of amorphous silicate. It is possible, especially at the higher pressures found in brown dwarf atmospheres, that the grains are crystalline, not amorphous. Indeed laboratory solar-composition condensation experiments produce crystalline, not amorphous, silicates (Toppani et al. 2004). Crystalline grains (Figure 5) can have larger and spectrally richer absorption cross sections. ### 4.3 The Transition from L to T The evolutionary cooling behavior of a given substellar object can be inferred from the field brown dwarf near-infrared color-magnitude diagram (Fig. 6a). Over tens to hundreds of millions of years a given substellar object first moves to redder $J-K$ colors as it cools while falling to fainter J magnitudes. Around ${\rm M}_{J}\sim 14-15$ the $J-K$ color turns bluer and the $J$ magnitude slightly (and counter-intuitively) brightens (Dahn et al. 2002, Tinney et al. 2003, Vrba et al. 2004). With further cooling a given dwarf finally falls to fainter $J$ magnitudes and apparently continues to slightly turn somewhat bluer in $J-K$. The behavior with even greater cooling is as yet uncertain until many more objects with $T_{\rm eff}<700\,\rm K$ are found. The ‘L to T’ transition is the ‘horizontal branch’ of the color-magnitude diagram as objects move from red to blue in the diagram. From Figure 3 we know that this color (and underlying spectral) change from late-type L dwarfs with $J-K\sim 2.5$ to blue T dwarfs with $J-K\sim-1$ happens rapidly over a small range of effective temperature. No brown dwarf evolution model can currently reproduce the magnitude of the observed color change over such a small range of $T_{\rm eff}$. Various explanations have been suggested including holes forming in the condensate cloud decks (Ackerman & Marley 2001, Burgasser et al. 2002b), an increase in the efficiency of grain sedimentation (Knapp et al. 2004), or a change in particle size (Burrows et al. 2006). In a series of papers Tsuji (Tsuji 2002, Tsuji & Nakajima 2003, Tsuji et al. 2004) proposed that a physically very thin cloud could self-consistently explain the rapid L to T transition. These models indeed exhibit a somewhat faster L- to T-like transition, but are still not consistent with the observed rapidity of the color change. More recently Tsuji (2005) has favored a sudden collapse of the global cloud deck at the transition along the lines of the Knapp et al. (2004) suggestion. Support for a rapid increase in sedimentation efficiency at the L to T transition has come from the model analysis of the 0.8–$14.5\,\mu$m spectra of four transition dwarfs by Cushing et al. (2008), who find that the cloud sedimentation efficiency (Ackerman & Marley 2001) indeed increases across the transition. The Cushing et al. (2008) sample includes two pairs of mid- to late-L dwarfs with very different near-IR colors. The authors find that the redder L dwarfs have less efficient sedimentation and therefore thicker cloud decks consisting of smaller particles, although gravity also may play a role for one pair. For one of the red L dwarfs Cushing et al. (2006) identified a broad absorption feature at 9–$11\,\mu$m which may be due to the presence of small silicate grains (Figure 5). The difficulty in characterizing the L to T transition arises from our lack of understanding of the masses and effective temperatures of objects at various locations in the ultra-cool dwarf color-magnitude diagram. It is not clear, for example, if the reddest field L dwarfs are more or less massive than bluer objects or if the turn towards the blue in $J-K$ is mass dependent (although there are some indications that it may be (Metchev & Hillenbrand 2006)). There are two ways in which this shortcoming in current understanding could be addressed. First, observing the orbits of binary brown dwarfs allows the total system mass to be measured. Secondly, photometry leading to near-infrared color-magnitude diagrams for many clusters with a variety of ages and metalicities will constrain the nature of transition. To date, the cluster color magnitude diagram (CMD) has only reached the transition in the Pleiades and perhaps the Sigma Orionis clusters. Figure 6b shows the currently best available CMD for the Pleiades. Two objects in this figure can be seen to have turned towards the blue. Deeper searches to fainter magnitudes in this cluster should soon reveal the expected downward turn to the fainter J magnitudes apparent in the field CMD. By constructing evolution models at the age of the Pleiades, it should be possible to constrain the mass at the turnoff from the red L sequence. Given enough clusters of different ages, the turnoff effective temperature and gravity can be constrained, thus illuminating the dependence of the turnoff on gravity and perhaps metallicity. Figure 6: Near-infrared color-magnitude diagrams for field and cluster ultracool dwarfs. (a) Black dots show single field L & T dwarfs. Green dots are resolved components of binary systems. Dotted circles are suspected (but unresolved) binaries (figure courtesy M. Liu based on Liu et al. (2006)). (b) Candidate ultracool dwarfs in the Pleiades in the most sensitive current survey (Casewell et al. 2007). Faintest objects in this plot have masses of about $11\,\rm M_{Jup}$. Note that at a fixed magnitude the cluster members tend to be redder than the field objects, which is likely a signature of low gravity. JWST will obtain spectra of quality comparable to Figure 1 for the candidate objects shown on this panel which will help calibrate evolutionary models of the brown dwarf cooling sequence. The detection limit for NIRCam on JWST is at about $J=22$ for this cluster or $\sim 1\,\rm M_{Jup}$ . Model predictions for colors of objects with $J>15$ are shown in Figure 8. JWST will be able to obtain moderate resolution ($R\sim 1000$) spectra on the current Pleiades candidates in Figure 6 in about 2.5 hours at $S/N\sim 20$. This spectral resolution should be sufficient to identify, for example, FeH and $\rm CH_{4}$ bands as they vary through the spectral sequence. This combination of evolution models and spectra should tightly constrain the empirical cooling sequence. Although this cluster is likely too large on the sky for efficient searching by JWST, NIRCAM could in principle find objects with masses as low as about $1\,\rm M_{J}$. Surveys of more compact clusters would not reach to such low masses, but should nevertheless be deep enough to find many young T dwarfs that have already undergone the transition. ### 4.4 The Latest T Dwarfs At this time only 16 very cool ($T_{\rm eff}<900\,\rm K$) dwarfs with types T7 and later are known, and of these only four are T8 or later. New surveys that go fainter than 2MASS and SDSS have started or are planned, and several groups are attempting to push to later and cooler types (e.g. Warren et al. 2007). All but two of the very late T dwarfs are isolated (the exceptions are Gl 570 D, Burgasser et al. 2000, and HD 3651B, Mugrauer et al. 2007). Since age is unknown for field dwarfs and brown dwarf cool with time, observed spectra must be compared with models or spectra of fiducial objects, to constrain mass and age. Since there are only a few T dwarfs with highly constrained properties, accurate model analysis is crucial for secure determination of gravity and hence mass at the bottom of the T sequence. For field brown dwarfs with ages in the range $\sim$ 1-5 Gyr and masses of 20–50 Jupiter-masses, effective temperature will lie in the range of $\sim$600–800 K. To understand the physical parameters of these elusive, cold, and low-mass dwarfs requires observation of their full spectral energy distribution. ### 4.5 Vertical Mixing and Chemical Disequilibrium It has long been understood that the abundances of molecules in Jupiter’s atmosphere depart from the values predicted purely from equilibrium chemistry (Prinn & Barshay 1977; Barshay & Lewis 1978; Fegley & Prinn 1985; Noll et al. 1988; Fegley & Lodders 1994; Fegley & Lodders 1996)444In stellar atmospheres, departures from thermochemical equilibrium can arise from interactions of atoms and molecules with the non-thermal radiation field (Hauschildt et al. 1997; Schweitzer, Hauschildt & Baron 2000). In brown dwarf atmospheres this effect is negligible.. Rapid upwelling can carry compounds from the deep atmosphere up into the observable regions of the atmosphere on time scales of hours to days. When this convective timescale is shorter than the timescale for chemical reactions to reach equilibrium, then the atmospheric abundances will differ from those that would be found under pure equilibrium conditions. The canonical example of this situation is carbon monoxide in Jupiter’s atmosphere. In Jupiter’s cold and dense upper troposphere, carbon should almost entirely be found in the form of methane. Deeper into the atmosphere, where temperatures and pressures are higher, CO should be the principal carrier of carbon. Rapid vertical mixing, combined with the strong C-O molecular bond, means that CO molecules can be transported to the observed atmosphere faster than chemical reactions can reduce the CO into $\rm CH_{4}$. The observed enhancement of CO, combined with (uncertain) reaction rates places limits on the vigor of convective mixing in the atmosphere. ##### CO Analyses of the 4.5 – $5\,\rm\mu m$ spectra of the T dwarfs Gl 229B and Gl 570 D reveal an abundance of CO that is over 3 orders of magnitude larger than expected from chemical equilibrium calculations (Noll et al. 1997, Oppenheimer et al. 1998, Griffith & Yelle 1999, Saumon et al. 2000), as anticipated by Fegley & Lodders (1996). Photometry in the $M$ band, which overlaps the CO band at $\rm 4.6\,\rm\mu m$, shows that an excess of CO may be a common feature of T dwarfs. Golimowski et al. (2004) have found that the $M$ band flux is lower than equilibrium models predict based on the $K$ and $L^{\prime}$ fluxes in all of the T dwarfs in their sample. The low $M$ band flux certainly arises from an excess of CO above that expected by equilibrium models. Since CO is a strong absorber in the $M$ band, a brown dwarf can be much fainter in this band than would be predicted by equilibrium chemistry. Equilibrium models predict that brown dwarfs and cool extrasolar giant planets should be bright at $M$ band, hence any flux decrement at this wavelength would have implications for surveys for cool dwarfs and giant planets. The degree to which this is a concern depends upon how the vigor of mixing declines with following effective temperature. Hubeny & Burrows (2007) recently have argued that this will not be a concern at temperatures below about 500 K, however, because the vigor of mixing falls with effective temperature. ##### NH3 Ammonia forms from $\rm N_{2}$ by the reaction $\rm N_{2}+3\,H_{2}\Leftrightarrow 2\,NH_{3}$. $\rm N_{2}+3H_{2}$ is favored at low pressures and high temperatures because of higher entropy. $\rm NH_{3}$ is favored at low temperatures, but since molecular nitrogen is a strongly bound molecule, reactions involving this molecule typically have high reaction energies and proceed very slowly at low temperatures. Like CO, $\rm N_{2}$ is favored at the higher temperatures found deep in brown dwarfs atmospheres. Again, like CO, vigorous vertical transport can bring $\rm N_{2}$ in the upper atmosphere faster than it can be converted to $\rm NH_{3}$, resulting in an excess of $\rm N_{2}$ compared to the values expected from chemical equilibrium. Figure 7 illustrates the effect of mixing (Saumon et al. 2007) on the $10\,\rm\mu m$ ammonia band in the spectra of the T8 dwarf 2MASS0415-0935 (Burgasser et al. 2002a). Figure 7: Fits of the IRS spectrum of 2MASS J0415-0935 (Saumon et al. 2007) showing the difference between a model in chemical equilibrium and a model that includes vertical transport that drives the nitrogen and carbon chemistry out of equilibrium. The red thin curve is the best-fitting model in chemical equilibrium, and the blue thin curve is the best-fitting nonequilibrium model. The data and the noise spectrum are shown by the histograms (black). The uncertainty on the flux calibration of the IRS spectrum is $\pm 5\%$. The model fluxes, which have not been normalized to the data, are shown at the resolving power of the IRS spectrum. JWST will obtain higher resolution mid-infrared spectra than the Spitzer data analyzed by Saumon et al. (2007). Higher spectral resolution on more targets will allow more in depth studies of atmospheric mixing. Since the vertical profile of mixing also influences cloud particle sizes and optical depths (in L- and L to T transition dwarfs), mapping out the eddy diffusion coefficient (which parameterizes mixing) as a function of mass and effective temperature will help to shed light on cloud dynamics as well as atmospheric chemistry. ## 5 Opportunities for JWST As we have highlighted in the above sections, there are many unsolved problems in the study of ultracool dwarfs. In this section we will briefly summarize some of the most promising avenues for JWST. ### 5.1 Characterizing Rosetta Stone Dwarfs The characterization of most field brown dwarfs, particularly the L and early T dwarfs, is hampered by the dependency of model fits on the particulars of the cloud models used to generate model atmospheres and spectra for comparison to data. Brown dwarfs of known mass, metallicity, and age are thus of particular importance as calibrators for the entire brown dwarf cooling sequence. This section discusses some opportunities for unearthing and deciphering ‘Rosetta Stone’ ultracool dwarfs with known or easily deduced masses and effective temperatures. Such objects could turn the page to much greater understanding of our library of known L and T dwarfs. #### Color-magnitude diagram for clusters to low masses By providing cluster color-magnitude diagrams to very low masses (a Jupiter- mass or less) JWST will revolutionize our understanding of brown dwarf cooling in environments controlled for age and metallicity. Comparison of spectra of objects with known properties to models will finally provide insight into the variation in cloud properties with mass and effective temperature. In the Pleiades, Casewell et al. (2007) have detected objects with masses as low as 11 Jupiter masses (Figure 6). In each cluster a brown dwarf of a given mass will be found either earlier or later on its cooling track, depending on the cluster age. Since the evolutionary cooling of brown dwarfs is well understood, spectra of cluster objects with known masses will definitively connect spectral features with gravity for mid- to late L dwarfs. In the Pleiades such a project should be straightforward for JWST as NIRSPEC will be able to obtain $R\sim 1000$ JHK spectra of the known low-mass cluster members in as little as a few hours. A Jupiter-mass object will be about 6 magnitudes fainter in $J$ band. Assuming JWST could survey a sufficiently large area to find candidates, it would define the brown dwarf cooling curve to a degree still not reached in the disk population. Model photometric predictions (Burrows et al. 2003) for such objects are shown in Figure 8. The actual trajectory in color-magnitude space will ultimately depend upon the interplay of water clouds and atmospheric mixing with the emitted spectra. Comparison of models such as those in the figure with data will test our understanding through this as-yet unexplored range of parameter space. Figure 8: Predicted absolute J magnitude ($M_{J}$) vs. $J-K$ color for a range of brown dwarf masses and ages. The numbers by the symbols denote the masses of the objects in Jupiter mass units. In the Pleiades the JWST NIRCAM detection limit will be about 1 Jupiter mass. Figure and description from Burrows et al. (2003); see discussion therein for greater detail. #### Resolved Spectra of Close Binaries Binary stars have long served as astronomical workhorses, helping to reveal important details of stellar astrophysics. Likewise binary brown dwarfs are also exceptionally useful. The orbits of close L- and T-dwarf binaries allow the total system mass to be determined as Bouy et al. (2004) have done for the binary 2MASSW J0746425+2000321. Assuming co-evality and equal metallicity combined with the total system mass and resolved spectra of the individual dwarfs furthermore allows such systems to elucidate the interpretation of brown dwarf spectra. Many more tight binaries have been found by the combination of HST and ground based adaptive optics imaging (see summary in Bouy et al. 2008). Orbital periods for many of these systems appear to be less than twenty years, so dynamical masses will be available during the JWST mission lifetime. The combination of the ground-based astrometry and photometry and resolved NIRSPEC high $R$ spectra of many of the individual objects, particularly for the tightest binaries, in these systems should provide important constraints on models of brown dwarf evolution, atmospheric structure, and emergent spectra. #### Cloud Behavior from L to T to Y Perhaps the greatest single observational result that could drive improvements in understanding of the L to T transition would be a direct measurement of surface gravity (or almost equivalently, mass) and effective temperature of late L and early T dwarfs. This would elucidate the dependence of the initiation of the L to T transition on mass and $T_{\rm eff}$. Constraining the turnoff absolute magnitude in a variety of clusters of known ages and resolving the spectra of close binaries that have measurable dynamical masses would highly constrain the nature of the L to T transition. Likewise as brown dwarfs further cool through the T sequence a number of open issues remain. Although T dwarfs are generally modeled as being entirely cloud free, some models that include very thin cloud decks better reproduce the spectra and color of the T’s. As brown dwarfs cool through about 500 K, thin water clouds should appear high in their atmospheres. With falling effective temperature these clouds are expected to thicken and begin to substantially alter the spectra of the Y dwarfs. It is entirely possible that, like the departure of clouds in the late L dwarfs, the arrival of clouds in the early Y dwarfs will produce unexpected and perhaps rapid color and spectral changes. Although we cannot yet identify what these changes might be, the same type of observations noted above will also be invaluable in constraining the Y dwarfs. Since the optical and near-IR flux of the Y dwarfs is expected to rapidly decrease (Burrows et al. 2003), the signatures of the water clouds will be best obtained in the mid-IR by JWST. ### 5.2 Spectra of Very Cool Objects As of early 2008, the brown dwarf with the lowest estimated effective temperature is ULAS J0034-00 with $T_{\rm eff}\sim 650\,\rm K$ (Warren et al. 2007). A number of ongoing and future searches will certainly find cool objects in the solar neighborhood (e.g., UKIDSS, Pan-STARRS, and the WISE mission). In particular WISE will have sufficient sensitivity to detect a $T_{\rm eff}\sim 200\,\rm K$ brown dwarf ($2\,\rm M_{J}$ at 1 Gyr) at a distance of about 2 pc. Assuming such nearby targets are found, JWST will produce exquisite spectra that will be unobtainable from the ground. A survey with NIRSPEC in high resolution mode with the YJH and LM gratings would nicely sample the spectra of cool field dwarfs. For a 500K dwarf at 10 pc, we estimate an exposure of about a minute will provide a spectrum with S/N of about 100 in $M$ band. A one hour exposure would be required for the same dwarf at 25 pc. Detection limits and model spectra are shown in Figure 9. Figure 9: Spectra (flux in millijanskys) vs. wavelength (in microns) for a range of brown dwarf masses at an age of 1 Gyr and a distance of 10 pc. Superposed are the approximate point-source sensitivities for instruments on Spitzer (red) and JWST (blue). The JWST/NIRCam sensitivities are $5\,\sigma$ and assume an exposure time of $5\times 10^{4}\,\rm sec$. The JWST/MIRI sensitivity curve from 5.0 to $27\,\rm\mu m$ is $10\,\sigma$ and assumes an exposure time of $10^{4}\,\rm sec$. Figure and description from Burrows et al. (2003); see discussion therein for greater detail. High quality spectra of cool dwarfs will be important for a number of reasons. First, cold disk objects possess effective temperatures comparable to those of middle-aged to old extrasolar giant planets. The disk population of brown dwarfs will thus provide ground truth for the spectral features that such cold objects exhibit (for example $\rm NH_{3}$ should appear in the near-IR (Saumon et al. 2001, Burrows et al. 2003, Leggett et al. 2008)). These objects will also have water clouds. The experience with the challenge of modeling silicate and iron clouds in L dwarfs alluded to above implies that water clouds will be no more tractable. The cold disk population will thus provide a proving ground for exoplanet water cloud modeling, which is undoubtedly needed (see Marley et al. 2007 and references therein). Atmospheric mixing, long recognized in Jupiter’s atmosphere, is also important for brown dwarfs (see §4.5), yet the most diagnostic spectral region for this process (the CO band at $4.6\,\rm\mu m$) lies in a blind spot for Spitzer spectroscopy and most groundbased observatories, but not JWST. Thus the nearby disk brown dwarfs will elucidate the extent to which $M$ band flux of objects in this effective temperature range is impacted by excess atmospheric CO. Vertical mixing could be an important consideration for the direct detection of giant planets around nearby stars (Golimowski et al. 2004; Marley et al. 2006; Hinz et al. 2006). After the discovery of Gl 229B, Marley et al. (1996) suggested that a substantial 4 to $5\,\rm\mu m$ flux peak should be a universal feature of giant planets and brown dwarfs. This expectation, combined with a favorable planet/star flux ratio, has made the band a favorite for planet detection (Burrows et al. 2005). However, groundbased and IRAC photometry suggests that cool dwarfs are fainter in this region—and the $L$ band region is brighter—than predicted by equilibrium chemistry. Given these and other considerations, Leggett et al. (2007) suggested that the comparative advantage of ground-based searches for young, bright giant planets at $M$ band might, such as the searches planned with the JWST coronagraph, might be somewhat less than currently expected (see also Marley et al. 2007). Hubeny et al. (2007), however, recently predicted that the vigor of atmospheric mixing will decline with effective temperature. If this is indeed the case then $M$ band will remain a fruitful hunting ground for extrasolar giant planet coronagraphic imaging. NIRCAM photometry of cold brown dwarfs will certainly illuminate this issue. ### 5.3 Mid-IR Spectra Beyond Spitzer Examples of some of the best available Spitzer IRS mid-infrared spectra of L and T dwarfs are shown in Figures 5 and 7. The spectral region between about 6 and $15\,\rm\mu m$ is important for a number of reasons. First there are several strong molecular bands in this region, including water, methane, and ammonia. As recounted above methane and ammonia are particularly sensitive to atmospheric mixing. The Si-O vibrational band, seen in the opacity of small silicate grains (Figure 5), also may trace the arrival of silicate clouds. Finally a number of other molecules, not yet detected in brown dwarf spectra, have absorption features in this range (Mainzer et al. 2007). MIRI will produce much higher resolution and S/N spectra than the best data from IRS (see the sensitivity curve in Figure 9). Higher quality spectra will allow for more robust detection of silicate features, perhaps including the sort of fine structure in the grain opacity seen in the lower panel of Figure 5, as well as for fine detail on the molecular features (see model prediction in Burrows et al. (2003)). If silicates are indeed detected, high resolution spectra could in principle differentiate between the particular silicate species, including forsterite, enstatite, and even quartz ($\rm SiO_{2}$, Helling et al. 2006), and whether the grains are in crystalline or amorphous form (Figure 5). ## 6 Conclusions The next decade holds the potential to substantially improve our fundamental understanding of ultracool dwarfs. Ground and space based surveys for very cool dwarfs, including UKIDSS, Pan-STARRS, and the WISE mission as well as deep surveys of young clusters will provide a host of ultracool dwarf targets for JWST. For cold nearby dwarfs JWST will provide unparalleled near- and especially mid-infrared spectra. These observations will constrain the water clouds expected to be present in objects with $T_{\rm eff}<500\,\rm K$ and measure the degree of atmospheric mixing. Both types of observations are highly relevant to the ultimate direct detection and characterization of extrasolar giant planets by coronagraphy. In young clusters (e.g, the Pleiades and younger) JWST will provide exceptional quality spectra of many known cluster members and will have the capability of imaging objects down to about one Jupiter mass and below. Such observations will constrain the evolutionary cooling tracks for dwarfs with lower gravities than most field objects and will tightly constrain the nature of the L to T transition by revealing its dependence on gravity. Many other opportunities, including producing resolved spectra of tight binary dwarfs and searching for spectral signatures of condensates and low abundance gasses are also possible. Combined with the inevitable unexpected discoveries, there is no doubt that JWST will bring brown dwarf astrophysics into the same highly constrained realm as stellar astrophysics. Interpretting these expected datasets will undoubtedly require substantial improvements to atmosphere and evolution modeling, particularly cloud and chemical transport modeling of ultracool dwarf atmospheres. ###### Acknowledgements. The authors thank M. Cushing, M. Liu, & D. Saumon, for helpful conversations on the future of brown dwarf science and A. Burrows, Ch. Helling, X. Tielens and K. Zahnle for thoughtful comments on the manuscript. We thank M. Cushing, M. Liu, and K. Lodders for preparing Figures 1, 6a & 2, respectively and C. Nixon for kindly providing the Cassini CIRS spectrum of Jupiter for Figure 1. ## 7 References Ackerman, A. S., & Marley, M. S. 2001, ApJ, 556, 872 Barshay, S. S., & Lewis, J. S. 1978, Icarus, 33, 593 Bouy, H., et al. 2004, A&A, 423, 341 Bouy, H., et al. 2008, ArXiv e-prints, 801, arXiv:0801.4424 Burgasser, A. J., et al. 1999, ApJL, 522, L65 Burgasser, A. J., et al. 2002a, ApJ, 564, 421 Burgasser, A. J., Marley, M. S., Ackerman, A. S., Saumon, D., Lodders, K., Dahn, C. C., Harris, H. C., & Kirkpatrick, J. D. 2002b, ApJL, 571, L151 Burgasser, A. J., Geballe, T. R., Leggett, S. K., Kirkpatrick, J. D., & Golimowski, D. A. 2006, ApJ, 637, 1067 Burrows, A., et al. 1997, ApJ, 491, 856 Burrows, A., Marley, M. S., & Sharp, C. M. 2000, ApJ, 531, 438 Burrows, A., Hubbard, W. B., Lunine, J. I., & Liebert, J. 2001, Reviews of Modern Physics, 73, 719 Burrows, A., Sudarsky, D., & Lunine, J. I. 2003, ApJ, 596, 587 Burrows, A., Sudarsky, D., & Hubeny, I. 2006, ApJ, 640, 1063 Casewell, S. L., Dobbie, P. D., Hodgkin, S. T., Moraux, E., Jameson, R. F., Hambly, N. C., Irwin, J., & Lodieu, N. 2007, MNRAS, 378, 1131 Cooper, C. S., Sudarsky, D., Milsom, J. A., Lunine, J. I., & Burrows, A. 2003, ApJ, 586, 1320 Cushing, M. C., et al. 2006, ApJ, 648, 614 Cushing, M. C., et al. 2008, ApJ, in press, arXiv:0711.0801 Dahn, C. C., et al. 2002, AJ, 124, 1170 Fegley, B. J., & Lodders, K. 1994, Icarus, 110, 117 Fegley, B., Jr., & Prinn, R. G. 1985, ApJ, 299, 1067 Fegley, B. J., & Lodders, K. 1996, ApJL, 472, L37 Freedman, R. S., Marley, M. S., & Lodders, K. 2008, ApJSup, 174, 504 Geballe, T. R., et al. 2002, ApJ, 564, 466 Golimowski, D. A., et al. 2004, AJ, 127, 3516 Griffith, C. A., & Yelle, R. V. 1999, ApJL, 519, L85 Hanner, M. S., Lynch, D. K., & Russell, R. W. 1994, ApJ, 425, 274 Helling, Ch., Thi, W.-F., Woitke, P., & Fridlund, M. 2006, A&A, 451, L9 Helling, Ch., et al. 2008a, in prep. Helling, Ch., Dehn, M., Woitke, P., & Hauschildt, P. H. 2008b, ApJL, 675, L105 Hubeny, I., & Burrows, A. 2007, ApJ, 669, 1248 Kirkpatrick, J. D., et al. 1999, ApJ, 519, 802 Kirkpatrick, J. D. 2005, Ann. Rev. Astron. & Astrophys., 43, 195 Kirkpatrick, J. D. 2007, ArXiv e-prints, 704, arXiv:0704.1522 Knapp, G. R., et al. 2004, AJ, 127, 3553 Kunde, V. G., et al. 2004, Science, 305, 1582 Leggett, S. K., et al. 2000, ApJL, 536, L35 Leggett, S. K., Saumon, D., Marley, M. S., Geballe, T. R., Golimowski, D. A., Stephens, D., & Fan, X. 2007, ApJ, 655, 1079 Liu, M. C., Leggett, S. K., Golimowski, D. A., Chiu, K., Fan, X., Geballe, T. R., Schneider, D. P., & Brinkmann, J. 2006, ApJ, 647, 1393 Lodders, K., & Fegley, B. 2002, Icarus, 155, 393 Lodders, K., & Fegley, B., Jr. 2006, Astrophysics Update 2, 1 Lodieu, N., et al. 2007, MNRAS, 379, 1423 Mainzer, A. K., et al. 2007, ApJ, 662, 1245 Marley, M. S., Saumon, D., Guillot, T., Freedman, R. S., Hubbard, W. B., Burrows, A., & Lunine, J. I. 1996, Science, 272, 1919 Marley, M. 2000, From Giant Planets to Cool Stars, 212, 152 Marley, M. S., Fortney, J., Seager, S., & Barman, T. 2007, Protostars and Planets V, 733 Metchev, S. & Hillenbrand, L. A. 2006, ApJ, 651, 1166 Nakajima, T., Oppenheimer, B. R., Kulkarni, S. R., Golimowski, D. A., Matthews, K., & Durrance, S. T. 1995, Nature, 378, 463 Noll, K. S., Geballe, T. R., & Marley, M. S. 1997, ApJL, 489, L87 Noll, K. S., Geballe, T. R., Leggett, S. K., & Marley, M. S. 2000, ApJL, 541, L75 Oppenheimer, B. R., Kulkarni, S. R., Matthews, K., & van Kerkwijk, M. H. 1998, ApJ, 502, 932 Patten, B. M., et al. 2006, ApJ, 651, 502 Prinn, R. G., & Barshay, S. S. 1977, Science, 198, 1031 Roellig, T. L., et al. 2004, ApJS, 154, 418 Saumon, D., Geballe, T. R., Leggett, S. K., Marley, M. S., Freedman, R. S., Lodders, K., Fegley, B., Jr., & Sengupta, S. K. 2000, ApJ, 541, 374 Saumon, D., Marley, M. S., Cushing, M. C., Leggett, S. K., Roellig, T. L., Lodders, K., & Freedman, R. S. 2006, ApJ, 647, 552 Saumon, D., et al. 2007, ApJ, 656, 1136 Sharp, C. M., & Burrows, A. 2007, ApJSup, 168, 140 Skrutskie, M. F., et al. 2006, AJ, 131, 1163 Strauss, M. A., et al. 1999, ApJL, 522, L61 Tinney, C. G., Burgasser, A. J., & Kirkpatrick, J. D. 2003, AJ, 126, 975 Toppani, A., Libourel, G., Robert, F., Ghanbaja, J., & Zimmermann, L. 2004, Lunar and Planetary Institute Conference Abstracts, 35, 1726 Tsuji, T. 2002, ApJ, 575, 264 Tsuji, T., & Nakajima, T. 2003, ApJL, 585, L151 Tsuji, T., Nakajima, T., & Yanagisawa, K. 2004, ApJ, 607, 511 Tsuji, T. 2005, ApJ, 621, 1033 Vrba, F. J., et al. 2004, AJ, 127, 2948 Warren, S. J., et al. 2007, MNRAS, 381, 1400 Woitke, P., & Helling, Ch. 2003, A & A, 399, 297 York, D. G., et al. 2000, AJ, 120, 1579

arxiv-papers

2024-09-04T02:48:54.236070

0803.1513

# Proton to pion ratio at RHIC from dynamical quark recombination Alejandro Ayala Mauricio Martínez Guy Paić G. Toledo Sánchez ###### Abstract We propose an scenario to study, from a dynamical point of view, the thermal recombination of quarks in the midsts of a relativistic heavy-ion collision. We coin the term dynamical quark recombination to refer to the process of quark-antiquark and three-quark clustering, to form mesons and baryons, respectively, as a function of energy density. Using the string-flip model we show that the probabilities to form such clusters differ. We apply these ideas to the calculation of the proton and pion spectra in a Bjorken-like scenario that incorporates the evolution of these probabilities with proper time and compute the proton to pion ratio, comparing to recent RHIC data at the highest energy. We show that for a standard choice of parameters, this ratio reaches one, though the maximum is very sensitive to the initial evolution proper time. ###### Keywords: Relativistic heavy-ion collisions, dynamical quark recombination ###### : 25.75.-q ## 1 Introduction Recently, it has been recognized that thermal recombination of quarks plays an important role for hadron production at intermediate $p_{t}$ in relativistic heavy-ion collisions. This idea, first studied in Refs. recomb ; Fries , explains the formation of low to intermediate $p_{t}$ hadrons from the bounding of quarks in a densely populated phase space, assigning appropriate degeneracy factors for mesons and baryons An implicit assumption is that hadronization happens at a single temperature. However, it is known that hadronization is not an instantaneous process but rather that it spans a window of temperatures and densities. For instance lattice calculations Karsch show that the phase transition from a deconfined state of quarks and gluons to a hadron gas is, as a function of temperature, not sharp. Motivated by these shortcomings of the original recombination scenario, here we set out to explore to what extent the probability to recombine quarks into mesons and baryons depends on density and temperature and whether this probability differs for hadrons with two and three constituents, that is to say, whether the relative population of baryons and mesons can be attributed not only to the degeneracy factors but rather to the dynamical properties of quark clustering in a varying density environment. A detailed answer to the above question stemming from first principles can only be found by means of non-perturbative QCD. Nevertheless, in order to get a simpler but still quantitative answer, here we address such question by resorting to the so called string-flip model stringflip which has proven to be successful in the study of quark/hadron matter as a function of density string1 ; Genaro1 ; Genaro2 . In this proceedings contribution, we only outline the main features of the calculation and refer the interested reader to Ref. ampt for further details. Other approaches toward a dynamical description of recombination, in the context of fluctuations in heavy-ion collisions, have been recently formulated in terms of the qMD model 0702188 . ## 2 Thermal particle spectra In the recombination model, the phase space particle density is taken as the convolution of the product of Wigner functions for each hadron’s constituent quark at a given temperature and the constituent quark wave function inside the hadron. For instance, the meson phase space distribution is given by $F^{M}(x,P)=\sum_{a,b}\int_{0}^{1}dz|\Psi_{ab}^{M}(z)|^{2}w_{a}({\mathbf{x}},zP^{+})\bar{w}_{b}({\mathbf{x}},(1-z)P^{+})\,,$ (1) where $P^{+}$ is the light-cone momentum, $\Psi_{ab}^{M}(z)$ is the meson wave function and $a,\ b$ represent the quantum numbers (color, spin, flavor) of the constituent quark and antiquark in the meson, respectively. An analogous equation can also be written for baryons. When each constituent quark’s Wigner function is approximated as a Boltzmann distribution and momentum conservation is used, the product of Wigner functions is given by a Boltzmann-like factor that depends only on the light-cone momentum of the hadron Fries . For instance, in the case of mesons $w_{a}({\mathbf{x}},zP^{+})\bar{w}_{b}({\mathbf{x}},(1-z)P^{+})\sim e^{-zP^{+}/T}e^{-(1-z)P^{+}/T}=e^{-P^{+}/T}\,.$ (2) In this approximation, the product of parton distributions is independent of the parton momentum fraction and the integration of the wave function over $z$ is trivially found by normalization. There can be corrections from a dependence of each constituent quark Wigner function on momentum components that are not additive because energy is not conserved in this scenario Fries2 . An important feature to keep in mind is that in this formalism, the QCD dynamics between quarks inside the hadron is encoded in the wave function. In order to allow for a more realistic dynamical recombination scenario let us take the above description as a guide, modifying the ingredients that account for the QCD dynamics of parton recombination. Let us assume that the phase space occupation can be factorized into the product of a term containing the thermal occupation number, including the effects of a possible flow velocity, and another term containing the system energy density $\epsilon$ driven probability ${\mathcal{P}}(\epsilon)$ of the coalescence of partons into a given hadron. We thus write the analog of Eq. (1) as $F(x,P)=e^{-P\cdot v(x)/T}{\mathcal{P}}(\epsilon)\,,$ (3) where $v(x)$ is the flow velocity. In order to compute the probability ${\mathcal{P}}(\epsilon)$ we explicitly consider a model that is able to provide information about the likelihood of clustering of constituent quarks to form hadrons from an effective quark-quark interaction, the string-flip model, which we proceed to describe. ## 3 String Flip Model and Hadron Recombination Probability The String Flip Model is formulated incorporating a many-body quark potential able to confine quarks within color-singlet clusters stringflip . At low densities, the model describes a given system of quarks as isolated hadrons while at high densities, this system becomes a free Fermi gas of quarks. For our purposes, we consider up and down flavors and three colors (anticolors) quantum numbers. Our approach is very close to that described in Refs. string1 and Genaro1 , where we refer the reader for an extensive discussion of the model details. The many-body potential $V$ is defined as the optimal clustering of quarks into color-singlet objects, that is, the configuration that minimizes the potential energy. In our approach, the interaction between quarks is pair- wise. Therefore, the optimal clustering is achieved by finding the optimal pairing between two given sets of quarks of different color for all possible color charges. The minimization procedure is performed over all possible permutations of the quarks and the interaction between quarks is assumed to be harmonic with a spring constant $k$. Through this procedure, we can distinguish two types of hadrons: i) Meson-like. In this case the pairing is imposed to be between color and anticolors and the many-body potential of the system made up of mesons is given by: $V_{\pi}=V_{B\bar{B}}+V_{G\bar{G}}+V_{R\bar{R}}\,$ (4) where $R(\bar{R})$, $B(\bar{B})$ and $G(\bar{G})$ are the labels for red, blue and green color (anticolor) respectively. Note that this potential can only build pairs. ii) Baryon-like. In this case the pairing is imposed to be between the different colors in all the possible combinations. In this manner, the many- body potential is: $V_{p}=V_{RB}+V_{BG}+V_{RG}\,$ (5) which can build colorless clusters by linking 3(RBG), 6(RBGRBG),… etc., quarks. Since the interaction is pair-wise, the 3-quark clusters are of the delta (triangular) shape. The formed hadrons should interact weakly due to the short-range nature of the hadron-hadron interaction. This is partially accomplished by the possibility of a quark flipping from one cluster to another. At high energy density, asymptotic freedom demands that quarks must interact weakly. This behavior is obtained once the average inter-quark separation is smaller than the typical confining scale. We study the meson and baryon like hadrons independently. Therefore, $V=V_{\pi}$ or $V_{p}$, depending on the type of hadrons we wish to describe. We use a variational Monte Carlo approach to describe the evolution of a system of $N$ quarks as a function of the particle density. We consider the quarks moving in a three-dimensional box whose sides have length a and the system described by a variational wave function of the form: $\Psi_{\lambda}(\textbf{x}_{1},...,\textbf{x}_{N})=e^{-\lambda V(\textbf{x}_{1},...,\textbf{x}_{N})}\Phi_{FG}(\textbf{x}_{1},...,\textbf{x}_{N}),$ (6) where $\lambda$ is the single variational parameter, $V$(x1,…,xN) is the many- body potential either for mesons or baryons and $\Phi_{FG}$(x1,…,xN) is the Fermi-gas wave function given by a product of Slater determinants, one for each color-flavor combination of quarks. These are built up from single- particle wave functions describing a free particle in a box Genaro1 . The variational parameter has definite values for the extreme density cases. At very low density it must correspond to the wave function solution of an isolated hadron. For example, the non-relativistic quark model for a hadron consisting of 2 and 3 quarks, bound by a harmonic potential, predicts, in units where $k=m=1$ that $\lambda_{\pi}\to\lambda_{0\pi}=\sqrt{1/2}$ and $\lambda_{p}\to\lambda_{0p}=\sqrt{1/3}$ respectively; at very high densities the value of $\lambda$ must vanish for both cases. Since the simulation was performed taking $m=k=1$, to convert to physical units we consider each case separately. Baryons: To fix the the energy unit we first notice that in a 3-body system the energy per particle, including its mass, is given by (with $m=k=1$): $\frac{E}{3}=\sqrt{3}+1.$ (7) If we identify the state as the proton of mass $M_{p}=938$ MeV, then the correspondence is $\sqrt{3}+1\rightarrow 312.7\ {\mbox{MeV}}.$ (8) To fix the length unit we use the mean square radius, which for a 3-body system is: $\sqrt{<r^{2}>}=(3)^{1/4}$. The experimental value for the proton is $\sqrt{<r^{2}>}=0.880\pm 0.015\ {\mbox{fm}}.$ (9) Then the correspondence is: $(3)^{1/4}\rightarrow 0.88$ fm. Mesons: In a similar fashion we obtain for mesons (taking the pion as the representative 2-body particle): Energy: $\frac{3}{2\sqrt{2}}+1\rightarrow 70$ MeV, length: $2^{1/4}\rightarrow 0.764$ fm. Our results come from simulation done with 384 particles, 192 quarks and 192 antiquarks, corresponding to having 32 $u\ (\bar{u})$ plus 32 $d\ (\bar{d})$ quarks (antiquarks) in the three color charges (anti-charges). To determine the variational parameter as a function of density we first select the value of the particle density $\rho$ in the box, which, for a fixed number of particles, means changing the box size. Then we compute the energy of the system as a function of the variational parameter using a Monte Carlo Method. The minimum of the energy determines the optimal variational parameter. We repeat the procedure for a set of values of the particle densities in the region of interest. The information contained in the variational parameter is global, in the sense that it only gives an approximate idea about the average size of the inter- particle distance at a given density, which is not necessarily the same for quarks in a single cluster. This is reflected in the behavior of the variational parameter $\lambda_{p}$ for the case of baryons which goes above 1 for energies close to where the sudden drop in the parameter happens. We interpret this behavior as as a consequence of the procedure we employ to produce colorless clusters for baryons, which, as opposed to the case to form mesons, allows the formation of clusters with a number of quarks greater than 3. When including these latter clusters, the information on their size is also contained in $\lambda$. To correct for this, we compute the likelihood to find clusters of 3 quarks $P_{3}$. Recall that for $3N$ quarks in the system, the total number of clusters of 3 quarks that can be made is equal to $N$. However this is not always the case as the density changes, given that the potential allows the formation of clusters with a higher number of quarks. $P_{3}$ is defined as the ratio between the number of clusters of 3 quarks found at a given density, with respect to $N$. Therefore, within our approach, we can define the probability of forming a baryon as the product of the $\lambda/\lambda_{0p}$ parameter times $P_{3}$, namely ${\mathcal{P}}_{p}=\lambda/\lambda_{0p}\times P_{3}.$ (10) For the case of mesons, since the procedure only takes into account the formation of colorless quark-antiquark pairs, we simply define the probability of forming a meson as the value of the corresponding normalized variational parameter, namely ${\mathcal{P}}_{\pi}=\lambda/\lambda_{0\pi}.$ (11) The probabilities ${\mathcal{P}}_{p}$ and ${\mathcal{P}}_{\pi}$ as a function of the energy density are displayed in fig. 1. Notice the qualitative differences between these probabilities. In the case of baryons, the sudden drop found in the behavior of the variational parameter is preserved at an energy density around $\epsilon=0.7$ GeV/fm3 whereas in the case of mesons, this probability is smooth, indicating a difference in the production of baryons and mesons with energy density. Figure 1: Probabilities to form baryons and mesons as a function of energy density. ## 4 proton to pion ratio In order to quantify how the different probabilities to produce sets of three quarks (protons) as compared to sets of two quarks (pions) affect these particle’s yields as the energy density changes during hadronization, we need to resort to a model for the space-time evolution of the collision. For the present purposes, we will omit describing the effect of radial flow and take Bjorken’s scenario which incorporates the fact that initially, expansion is longitudinal, that is, along the beam direction which we take as the $\hat{z}$ axis. In this 1+1 expansion scenario, the relation between the temperature $T$ and the 1+1 proper-time $\tau$ is given by $T=T_{0}\left(\frac{\tau_{0}}{\tau}\right)^{v_{s}^{2}},$ (12) where $\tau=\sqrt{t^{2}-z^{2}}$. Equation (12) assumes that the speed of sound $v_{s}$ changes slowly with temperature. A lattice estimate of the speed of sound in quenched QCD Gupta shows that $v_{s}^{2}$ increases monotonically from about half the ideal gas limit for $T\sim 1.5T_{c}$ and approaches this limit only for $T>4T_{c}$, where $T_{c}$ is the critical temperature for the phase transition. No reliable lattice results exist for the value of the speed of sound in the hadronic phase though general arguments indicate that the equation of state might become stiffer below $T_{c}$ and eventually softens as the temperature approaches zero. For the ease of the argument, here we take $v_{s}$ as a constant equal to the ideal gas limit $v_{s}^{2}=1/3$. We also consider that hadronization takes place on hypersurfaces $\Sigma$ characterized by a constant value of $\tau$ and therefore $d\Sigma=\tau\rho\ d\rho\ d\phi\ d\eta,$ (13) where $\eta$ is the spatial rapidity and $\rho$, $\phi$ are the polar transverse coordinates. Thus, the transverse spectrum for a hadron species $H$ is given as the average over the hadronization interval, namely $E\frac{dN^{H}}{d^{3}P}=\frac{g}{\Delta\tau}\int_{\tau_{0}}^{\tau_{f}}d\tau\int_{\Sigma}d\Sigma\ \frac{P\cdot u(x)}{(2\pi)^{3}}F^{H}(x,P),$ (14) where $\Delta\tau=\tau_{f}-\tau_{0}$. To find the relation between the energy density $\epsilon$ –that the probability ${\mathcal{P}}$ depends upon– and $T$, we resort to lattice simulations. For the case of two flavors, a fair representation of the data Karsch is given by the analytic expression $\epsilon/T^{4}=a\left[1+\tanh\left(\frac{T-T_{c}}{bT_{c}}\right)\right],$ (15) with $a=4.82$ and $b=0.132$. We take $T_{c}=175$ MeV. For a purely longitudinal expansion, the flow four-velocity vector $v^{\mu}$ and the normal to the freeze-out hypersurfaces of constant $\tau$, $u^{\mu}$, coincide and are given by $v^{\mu}=u^{\mu}=(\cosh\eta,0,0,\sinh\eta)$, therefore, the products $P\cdot u$ and $P\cdot v$ appearing in Eq. (14) can be written as $P\cdot v=P\cdot u=m_{t}\cosh(\eta-y),$ (16) where $m_{t}=\sqrt{m_{H}^{2}+p_{t}^{2}}$ is the transverse mass of the hadron and $y$ is the rapidity. Considering the situation of central collisions and looking only at the case of central rapidity, $y=0$, the final expression for the hadron’s transverse distribution is given by $E\frac{dN^{H}}{d^{3}P}=\frac{g}{(2\pi)^{3}}\frac{2m_{t}A}{\Delta\tau}\int_{\tau_{0}}^{\tau_{f}}d\tau\tau K_{1}\left[\frac{m_{t}}{T(\tau)}\right]{\mathcal{P}}[\epsilon(\tau)].$ (17) To obtain the the pion and proton distributions, we use the values $\tau_{0}=0.75$ fm and $\tau_{f}=3.5$ fm and an initial temperature $T_{0}=200$ MeV. From Eq. (12), this corresponds to a final freeze-out temperature of $\sim 120$ MeV. For protons we take a degeneracy factor $g=2$ whereas for pions $g=1$, to account for the spin degrees of freedom. Figure 2 shows the proton to pion ratio for three different values of the initial evolution proper time $\tau_{0}=0.5,\ 0.75$ and $1$ fm and the same finial freeze-out proper-time $\tau_{f}=3.5$ fm, compared to data for this ratio for Au + Au collisions at $\sqrt{s_{NN}}=200$ GeV from PHENIX PHENIXBM . We notice that the maximum height reached by this ratio is sensitive to the choice of the initial evolution time. We also notice that the $p_{t}$ value for which the maximum is reached is displaced to larger values than what the experimental values indicate. This result is to be expected since the model assumptions leading to Eq. (17) do not include the effects of radial flow that, for a common flow velocity, are known to be larger for protons than for pions, and which will produce the displacement of the ratio toward lower $p_{t}$ values. Figure 2: Proton to pion ratio as a function of transverse momentum for three different values of the initial evolution proper-time $\tau_{0}=0.5,\ 0.75$ and $1$ fm and the same finial freeze-out proper-time $\tau_{f}=3.5$ fm, compared to data for Au + Au collisions at $\sqrt{s_{NN}}=200$ GeV from PHENIX. The height of this ratio is very sensitive to the choice of the initial evolution time. ## 5 Summary and Conclusions In conclusion, we have used the string-flip model to introduce a dynamical quark recombination scenario that accounts for the evolution of the probability to form a meson or a baryon as a function of the energy density during the collision of a heavy-ion system. We have used the model variational parameter as a measure of the probability to form colorless clusters of three quarks (baryons) or of quark-antiquark (mesons). We have shown that these probabilities differ; whereas the probability to form a pion transits smoothly from the high to the low energy density domains, the probability to form a baryon changes abruptly at a given critical energy density. We attribute this difference to the way the energy is distributed during the formation of clusters: whereas for mesons the clustering happens only for quark-antiquark pairs, for baryons the energy can be minimized by also forming sets of three, six, etc., quarks in (colorless) clusters. These produces competing minima in the energy that do not reach each other smoothly. We interpret this behavior as a signal for a qualitative difference in the probability to form mesons and a baryons during the collision evolution. We have incorporated these different probabilities to compute the proton and pion spectra in a thermal model for a Bjorken-like scenario. We use these spectra to compute the proton to pion ratio as a function of transverse momentum and compare to experimental data at the highest RHIC energies. We argue that the ratio computed from the model is able to reach a height similar to the one shown by data, although the maximum is displaced to larger $p_{t}$ values. This could be understood by recalling that the model does not include the effects of radial flow which is known to be stronger for protons (higher mass particles) than pions. The inclusion of these effects is the subject of current research that will be reported elsewhere. Support for this work has been received by PAPIIT-UNAM under grant number IN116008 and CONACyT under grant number 40025-F. M. Martinez was supported by DGEP-UNAM. ## References * (1) R. C. Hwa and C. B. Yang, _Phys. Rev. C_ 67, 034902 (2003); V. Greco, C. M. Ko, and P. Lévai, _Phys. Rev. Lett._ 90, 202302 (2003). * (2) R.J. Fries, B. Müller, C. Nonaka and S.A. Bass, _Phys. Rev. Lett._ 90, 202303 (2003). * (3) F. Karsch, E. Laermann and a Peikert, _Phys. Lett._ B478, 447 (2000); F. Karsch, _Lect. Notes in Phys._ 583, 209 (2002). * (4) C.J. Horowitz, E.J. Moniz and J.W. Negele, _Phys. Rev. D_ 31, 1689 (1985). * (5) C. Horowitz and J. Piekarewicz, _Nucl. Phys._ A536, 669-696 (1992). * (6) G. Toledo Sánchez and J. Piekarewicz, _Phys. Rev. C_ 65, 045208 (2002). * (7) G. Toledo Sánchez and J. Piekarewicz, _Phys. Rev. C_ 70, 035206 (2004). * (8) A. Ayala, M. Martínez, G. Paić and G. Toledo Sánchez, _Dynamical quark recombination in ultrarelativistic heavy-ion collisions and the proton to pion ratio_ , arXiv:0710.3629 [hep]. * (9) S. Haussler, S. Scherer and M. Bleicher, _The effect of dynamical parton recombination on event-by-event observables_ , hep-ph/0702188. * (10) R.J. Fries, B. Müller, C. Nonaka and S.A. Bass, _Phys. Rev. C_ 68, 044902 (2003). * (11) S. Gupta, Pramana 61, 877 (2003). * (12) S.S. Adler et al. (PHENIX Collaboration), Phys. Rev. C 69, 034909 (2004).

arxiv-papers

2024-09-04T02:48:54.242570

0803.1516

We have studied the adsorption of gas molecules (CO, NO, NO2, O2, N2, CO2, and NH3) on graphene nanoribbons (GNRs) using first principles methods. The adsorption geometries, adsorption energies, charge transfer, and electronic band structures are obtained. We find that the electronic and transport properties of the GNR with armchair-shaped edges are sensitive to the adsorption of NH3 and the system exhibits _n_ -type semiconducting behavior after NH3 adsorption. Other gas molecules have little effect on modifying the conductance of GNRs. Quantum transport calculations further indicate that NH3 molecules can be detected out of these gas molecules by GNR based sensor. # Adsorption of gas molecules on graphene nanoribbons and its implication for nano-scale molecule sensor Bing Huang1, Zuanyi Li1, Zhirong Liu2, Gang Zhou1, Shaogang Hao1, Jian Wu1, Bing-Lin Gu1 and Wenhui Duan1111Author to whom correspondence should be addressed. E-mail address: dwh@phys.tsinghua.edu.cn 1Department of Physics, Tsinghua University, Beijing 100084, People’s Republic of China 2College of Chemistry and Molecular Engineering, Peking University, Beijing 100871, People’s Republic of China ## I Introduction Sensing gas molecules is critical to environmental monitoring, control of chemical processes, space missions, and agricultural and medical applicationsMRS . Solid-state gas sensors are renowned for their high sensitivity which have made them ubiquitous in the worldMoseley ; Capone . In the past few years, a new generation of gas sensors have been demonstrated using carbon nanotubes (CNTs) and semiconductor nanowiresKong ; Collins ; Qi ; Valentini ; Novak ; Li ; Zhang ; Chopra . It was reported that semiconducting CNTs could be used to detect small concentration of NH3, NO2, and O2 with high sensitivity by measuring changes of the CNTs conductance upon exposure to the gases at room temperatureKong ; Collins ; Qi ; Valentini ; Chopra . Graphene, a single atomic layer of graphite, has been successfully produced in experimentsK. S. Novoselov ; C. Berger , which have resulted in intensive investigations on graphene-based structures because of fundamental physics interests and promising applicationsA. K. Geim (_e.g._ , gas sensorF.Schedin ). Importantly, graphene can be patterned via standard lithographic techniques into quasi-one-dimension materialsC. Berger ; Kim ; Chen , graphene nanoribbons (GNRs), which have many properties similar to CNTs, such as energy gap dependence of widths and crystallographic orientationQimin ; Louie . Different from CNTs, however, GNRs present long and reactive edges which make GNRs not only notably more accessible to dopingQimin ; Bing ; Bing2 and chemical modificationWang ; Hod ; Hao ; Ferrari , but also more susceptible to structural defects and impuritesBing2 . Theoretical studies of gas molecular adsorption on the graphene surface have been reported recentlyWehling ; O.Leenaerts , which showed that NO2, H2O, NH3, CO and NO molecules are physically adsorbed on the pristine graphene: NH3 and CO molecules will act as donors while H2O and NO2 will act as acceptors, which are consistent with previous experimentF.Schedin . Compared with graphene, GNRs are advantageous in small volume and free reactive edges. In experiments, the edges of GNRs are not well controlledChen ; Kim and it is hard to obtain fully saturated edges without any dangling bond (DB) defects. It is well known that DB defects around the vacancy sites or at the tips play a very important role in CNTs gas sensors because they are very chemically reactiveSnow ; Andzelm ; Vladimir . Similar to CNTs, when there are DB defects at GNRs edges, covalent attachment of chemical groups and molecules also significantly influences their electronic propertiesWang ; Hao ; Ferrari . Therefore, it is very interesting and important to study the feasibility of using GNRs as gas sensors. In this article, using density functional theory (DFT) calculations, we study the adsorption of gas molecules (CO, NO, NO2, O2, N2, CO2 and NH3) around the sites of DB defects on armchair GNRs (AGNRs, having armchair-shaped edges) and explore the feasibility of using AGNRs as gas sensors. Following conventional notationdefination , a GNR is specified by the numbers ($n$) of dimer lines and zigzag chains along the ribbon forming the width, for the AGNR and zigzag GNRs (ZGNRs, having zigzag-shaped edges), respectively. For example, the structure in Fig. 1a is referred as a 10-AGNR (i.e., $n=10$). Previous works show that all AGNRs are semiconductor while ZGNRs are metaldefination ; Qimin . We focus on semiconducting AGNRs instead of metallic ZGNRs, since it is expected that gas molecule adsorption will have a much smaller effect on modifying the electronic properties of (metallic) ZGNRs. It is found that although all gas molecules can influence the electronic structure of AGNRs, only NH3 molecule adsorption can modify the conductance of AGNRs remarkably by acting as donors while other gas molecules have little effect on conductance. This property can be utilized to detect NH3 out of other common gases, which is requisite and significant in industrial, medical, and living environmentsTakao . ## II Calculation Method and Model Our electronic structure calculations were performed using the DFT in the spin-polarized generalized gradient approximation (GGA) with PW91 functional for the exchange and correlation effects of the electrons, as implemented in Vienna _Ab initio_ Simulation Package (VASP)VASP . The electron-ion interaction was described by the ultrasoft pseudopentials and the energy cutoff was set to be 400 eV. Structural optimization was carried out on all systems until the residual forces on all ions were converged to 0.01 eV/Å. The quantum transport calculations were performed using the ATK 2.3 packageTaylor , which implements DFT-based real-space, nonequilibrium Green’s function formalism. The mesh cutoff is chosen as 150 Ry to achieve the balance between calculation efficiency and accuracy. Figure 1: (a) The optimized structure of 10-AGNR with edge dangling bond (DB) defects, where the arrow shows the periodic direction. The sites of DB defects are shown by yellow. C atoms and H atoms are denoted by large and small (white) spheres, respectively. (b) Density of states (DOS) of perfect 10-AGNR (top panel) and spin-polarized DOS of 10-AGNR with DB defects (middle and bottom panels). The Fermi level is set to zero (the top of valence band). The middle (blue) and bottom (red) panels in the figure correspond to minority spin and majority spin, respectively. The green area corresponds to local DOS of the two carbon atoms (yellow balls in (a)) with DB defects. ## III Results and discussion DB defects would exist at both edges of GNRs due to the fact that the two edges of a GNR are equivalent in nature. So we will focus our study on double- edge-defect case (_i. e._ , both edges have DB defects). What is more, we also preform test calculations for the single-edge-defect case (DB defects only exist at one edge), and find little difference in essential results compared with the double-edge-defect case. The structure with one DB defect per edge per five unit cells in the ribbon axis direction is adopted in our calculations, as shown in Fig. 1a, corresponding to a DB defect concentration of 0.04 Å-1. [Several different initial configurations with one DB defect per edge have been considered and we find that the one shown in Fig. 1a is most favorable (stable) in energy.] Due to the dangling $\sigma$-bonds at the edges, the ground state of this system is spin-polarized with the magnetic moment of 1 $\mu$B per DB, which is localized at the carbon atoms with DB defects. Spin-polarized density of states (DOS) of this system is shown in the middle and bottom panels of Fig. 1b. For better comparison, the DOS of prefect 10-AGNR with the same supercell is also presented (the top panel). It can be seen that the perfect 10-AGNR is paramagnetic semiconductor with a band gap of 1.2 eV, consistent with our previous studyQimin . When there are DB defects at the ribbon edges, two new peaks appear in the DOS (indicated by the arrows in Fig. 1b). The local density of states (LDOS) analysis (the green area in Fig. 1b) shows that the two peaks are mainly contributed by the edge carbon atoms with DB defects. These two DB states are localized, respectively, within the valence band for minority spin and at the bottom of conduction band for majority spin, different from the usual DB states which are localized within band gapChenchen . Figure 2: Optimized structure of 10-AGNRs with gas molecule adsorption: (a) CO, (b) NO, (c) NO2, (d) O2, (e) CO2, and (f) NH3. We only show the structure around the adsorbed molecule. Table 1: Calculated adsorption energies ($E_{\rm ads}$) and charge transfer (CT) from the gas molecules to the 10-AGNR. Molecules | CO | NO | NO2 | O2 | N2 | CO2 | NH3 ---|---|---|---|---|---|---|--- $E_{\rm ads}$ (eV) | -1.34 | -2.29 | -2.70 | -1.88 | 0.24 | -0.31 | -0.18 CT (e) | -0.30 | -0.55 | -0.53 | -0.78 | / | -0.41 | 0.27 We start by investigating the adsorption geometries of seven gas molecules on 10-AGNR with DB defects (Fig. 1a). Only the gas molecules adsorption around the sites of DB defects is considered. Several different initial orientations of gas molecules on 10-AGNRs are adopted in searching the most stable configurations. Fig. 2 shows the top views of 10-AGNR with adsorbed molecules. We find that different gas molecules prefer different geometries in the adsorption: (a) CO molecule lies in the AGNR plane with C-C distance of 1.35 Å and a C-C-O angle of 168.6∘. The bond length of the adsorbed CO is 1.18 Å, a little longer than that of an isolated molecule (1.13 Å), indicating that adsorption process will weaken the original C-O bond of gas molecule. (b) NO molecule sits out of the AGNR plane with a C-N-O angle of 118.2∘, and the C-N and N-O distances are 1.43 Å and 1.24 Å, respectively. Also, the bond length of the adsorbed NO is 0.07 Å longer than that of an isolated molecule. (c) The adsorbed NO2 has C-N and N-O distances of 1.47 Å and 1.25 Å and a O-N-O angle of 127∘, and the O-N-O plane is tilted 60∘ with respect to the ribbon plane. Our results are consistent with previous studiesHao ; Ferrari . (d) O2 sits in the ribbon plane with C-O and O-O distances of 1.37 Å and 1.38 Å respectively and a C-O-O angle of 116.2∘. The bond length of O2 molecule increases by 0.15 Å after adsorption. (e) The geometry of CO2 adsorbed on the ribbon edge is different from that of an isolated CO2 molecule: it is not linear structure anymore but similar to the NO2 configuration. This indicates that the bond of CO2 transfers from _sp_ into _sp 2_ hybridization in order to lower the total energy. The C-C and C-O distances are 1.51 Å and 1.26 Å respectively, and the C-O-C angle is 127∘. (f) The adsorbed NH3 molecule sits 1.49 Å away from the edge carbon atom, the H-N distance is 1.03 Å and the dihedral angle is about 21∘ between the C-N bond and the AGNR plane. Besides the above gas molecules, we also study the N2 molecule adsorption on the AGNR, and find that it is difficult for N2 to adsorb on the ribbon due to its inert nature. Figure 3: Band structure and density of states (DOS) of 10-AGNRs with gas molecule adsorption: (a) CO, (b) NO, (c) NO2, (d) O2, (e) CO2, and (f) NH3. The LDOS of gas molecules is also plotted (red filled area under DOS curve). The Fermi level is set to zero. The calculated adsorption energies of the gas molecules with the 10-AGNRs are shown in Table 1. Herein, The adsorption energy is defined as: $E_{\rm ads}=\\{E_{\rm tot}({\rm{ribbon}}+m{\rm{Molecule}})-E_{\rm tot}({\rm{ribbon}})-nE_{tot}({\rm{Molecule}})\\}/m$, where $E_{\rm tot}({\rm{ribbon}}+m{\rm{Molecule}})$, $E_{\rm tot}({\rm{ribbon}})$, and $E_{\rm tot}({\rm{Molecule}})$ are the total energies of the AGNR with molecule adsorption, the isolated AGNR (with DB defects) and the molecules, respectively. And $m$ is the number of molecules adsorption on the AGNR. The results reveal that all these adsorption configurations are energetically favorable except that the N2 adsorption process is endothermic reaction (note: the negative adsorption energy corresponds to the exothermic reaction). The adsorption energies of CO, NO, NO2, and O2 are all larger than 1 eV, corresponding to strong chemisorption. The adsorption energies for the NH3 and CO2 on AGNRs are -0.18 eV and -0.31 eV respectively, indicating that the adsorption are between weak chemisorption and strong physisorption. The above results illuminate that gas molecule adsorption at AGNR edges is quite different from the weak physisorption of gas molecules on the graphene surfaceF.Schedin ; O.Leenaerts . Furthermore, the Bader analysisBader of the charge distribution is used to understand the nature of the interaction between the gas molecules and the AGNRs, and to evaluate the induced effects on the molecules. The trend of calculated charge transfer (Table 1) can be understood on the basis of relative electron-withdrawing or -donating capability of the adsorbed molecular groups. From Table 1 (note: the positive value means a charge transfer from the adsorbed molecule to the AGNR), we can see that CO, NO, NO2, O2, and CO2 have electron-withdrawing capability, while NH3 is electron-donating functional molecule. It is well known that NO2 and O2 are relatively strong electron-withdrawing molecules while NH3 is relatively strong electron-donating molecule in other carbon-based materialsKong ; Collins ; F.Schedin ; O.Leenaerts ; Andzelm . Comparing with gas adsorption on the graphene (GNR) surfaceO.Leenaerts , the values of the charge transfer are much larger. This is consistent with our conclusion that the interaction between the gas molecules and the GNR edges is much stronger than that of surface. Figure 4: Iso-surface plots of the partial charge density at $\Gamma$ point of the band crossing the Fermi level for the 10-AGNRs with (a) CO2 adsorption and (b) NH3 adsorption. The iso-value is 0.01 e/Å3. The calculated band structures and DOS of 10-AGNRs with molecule adsorption are shown in Fig. 3. Comparing with the DOS of AGNR (Fig. 1b), the total DOS of the system and LDOS of the molecules show that these molecules modulate the electronic property of AGNRs in different manners: i) CO and NO molecules adsorption introduces impurity states in the band gap and the Fermi levels of two systems cross these states, as shown in Figs. 3a and 3b. Therefore, gas adsorption will decrease the original band gap, and probably have some influence on the optical properties of AGNRs. For CO adsorption, there are two half-occupied states in the band gap, but we do not expect these impurity states can enhance the conductance of the system because these states are very localized and deep in the original band gap. It would be very difficult for charge carriers to transit between the valence (or conduction) band and impurity states at finite temperature. ii) LDOS analysis (Fig. 3c) shows that NO2 adsorption will introduce fully-occupied states which are strongly hybridized with the original “bulk” states in the valence band and these states are nonlocalized. It suggests that the interaction between NO2 molecules and dangling bonds of the ribbon is very strong, consistent with the calculated adsorption energy. The Fermi level is pinned in the top of valence band, which is the same as the case of the ribbon without molecule adsorption, so the system is still semiconducting. iii) Figs. 3d and 3e show the cases of O2 and CO2 adsorptions, respectively. From LDOS analysis we can see that the states contributed by CO2 (or O2) molecules are localized around the top of valence band and hybridize with the original valence band. Partial charge density analysis (Fig. 4a) shows that the states near the Fermi level are quite localized and mainly contributed by CO2 molecule and the carbon atoms of ribbon around the CO2 molecule. This suggests that the conductance of this system can not be enhanced notably. When the molecular doping concentration is low enough, these impurity states will become more localized on the gas molecules. But due to these half-occupied impurity states being near the top of valence band, the electrons of the valence band can transit into these states and the system will exhibit _p_ -type semiconducting behavior at finite temperature. iv) NH3 molecule adsorption induces unoccupied local states in the conduction band, and more importantly, the Fermi level is shifted into original conduction bands, resulting in n-type semiconducting behavior (Fig. 3f). Furthermore, partial charge density analysis (see Fig. 4b) shows that the states near the Fermi level are mainly contributed by the carbon atoms of the ribbon rather than NH3 molecules. The above results indicate a transition from semiconducting to conducting behavior after NH3 molecule adsorption. The stabilization of the Fermi level in the conduction band by the impurity resonant levels was also reported in semiconductors doped by Cr and Tlpss448 ; prb3903 . Among all gas molecules considered, obviously NH3 molecule adsorption can greatly enhance the conductance of AGNRs, where the system will exhibit metallic behavior after NH3 adsorption. CO2 and O2, on the other hand, may enhance the conductance of GNRs to some extent at finite temperature since the CO2 and O2 adsorption will turn AGNRs to _p_ -type semiconductor. Based on the analysis, we can expect that AGNRs may act as effective sensor to detect NH3 out of other gas molecules discussed above by measuring the change of conductance after the gas adsorption. What is more interesting, the adsorption energy of NH3 molecule is only -0.18 eV (Table I), so NH3 molecules can be desorbed at higher temperature. This implies that GNR senors could be recycled more than once. Figure 5: The $I$-$V_{\rm bias}$ curves for the GNR sensor before and after the adsorption of NH3 and CO2. The inset shows the schematics of such a GNR sensor, consisting of one 10-AGNR (detection region) and two metallic 7-ZGNRs leads. The gas molecules can be adsorbed around the DB defects of the GNR sensor. A design of a GNR-based junction (GNR sensor) to detect NH3 is given in the inset of Fig. 5 as an example. It contains a 8.60 nm long 10-AGNR with one DB defect per edge as the detection region and two semi-infinite metallic 7-ZGNRs as the leads. Gas molecules can be adsorbed on the DB defect sites, and the conductance is measured by applying a bias voltage through the junction. The DB defect concentration in this model is about 0.011/Å per edge, which is practical in experiments. We calculate a series of current versus bias voltage ($I$-$V_{\rm bias}$) curves for such GNR junction with different gas molecule adsorption on the edges. For the sake of clarity, we only show $I-V_{\rm bias}$ curves for the AGNRs before and after NH3 and CO2 adsorption due to the fact that the currents induced by other gas molecules adsorption are almost zero (much smaller than the currents induced by CO2 or NH3 adsorption). As shown in Fig. 5, without gas molecule adsorption, the channel (GNR sensor) exhibits a semiconducting behavior, and the current is always zero even under a bias of 0.5 V. After NH3 adsorption, however, the current increases notably and $I$-$V_{bias}$ curve is nearly linear, corresponding to a metallic behavior of ohmic contact. CO2 adsorption can also increase the current, but its value is much smaller than that induced by NH3 adsorption. This phenomenon indicates that NH3 can be detected out of other gases by applying a bias voltage upon the GNR junction, which is consistent with our analysis based on electronic properties. ## IV Summary In summary, we have performed first-principles calculation to study the adsorption geometries and electronic structure of graphene nanoribbons with gas molecule adsorption. We find that NH3 molecule adsorption can significantly influence the electronic and transport properties of AGNRs, while other gas molecules have little effect. Based on this characteristic, we demonstrate that an AGNR can be used to detect NH3 molecules out of many familiar gas molecules. Furthermore, our work also suggests an effective way to fabricate _n_ -type (_p_ -type) transistors by NH3 (CO2 or O2) adsorption on graphene nanoribbons. ###### Acknowledgements. This work was supported by the Ministry of Science and Technology of China (Grant Nos. 2006CB605105 and 2006CB0L0601), the Natural Science Foundation of China (Grant Nos. 10674077 and 10774084) and the Ministry of Education of China. ## References * (1) Special issue on Gas-Sensing Materials, _MRS Bull_. 1999, _24_. * (2) Moseley, P. T. _Meas. Sci. Technol._ 1997, _8_ , 223. * (3) Capone, S.; Forleo, A.; Francioso, L.; Rella, R.; Siciliano,P.; Spadavecchia, J.; Presicce, D. S.; Taurino, A. M. _J. Optoelect. Adv. Mater._ 2003, _5_ , 1335. * (4) Kong, J.; Franklin, N. R.; Zhou, C.; Chapline, M. G.; Peng, S.; Cho, K.; Dai, H. _Science_ 2000, _287_ , 622. * (5) Collins, P. G.; Bradley, K.; Ishigami, M.; Zettl, A. _Science_ 2000, _287_ , 1801. * (6) Qi, P.; Vermesh, O.; Grecu, M.; Javey, A.; Wang, Q.; Dai, H., _Nano Lett._ 2003, _3_ , 347\. * (7) Valentini, L.; Armentano, I.; Kenny, J. M.; Cantalini, C.; Lozzi, L.; Santuccia, S. _Appl. Phys. Lett._ 2003, _82_ , 961. * (8) Novak, J. P.; Snow, E. S.; Houser, E. J.; Park, D.; Stepnowski, J. L.; McGill, R. A. _Appl. Phys. Lett._ 2003, _83_ , 4026. * (9) Li, J.; Lu, Y.; Ye, Q.; Cinke, M.; Han, J.; Meyyappan, M. _Nano Lett._ 2003, _3_ , 929\. * (10) Zhang, D.; Liu, Z.; Li, C.; Tang, T.; Liu, X.; Han, S.; Lei, B.; Zhou, C. _Nano Lett._ 2004, _4_ , 1919. * (11) Chopra, S.; McGuire, K.; Gothard, N.; Rao, A. M. _Appl. Phys. Lett._ 2003, _83_ , 2280. * (12) Novoselov, K. S.; Geim, A. K., Morozov, S. V.; Jiang, D.; Zhang, Y.; Dubonos, S. V.; Grigorieva, I. V.; Firsov, A. A. _Science_ 2004, _306_ , 666. * (13) Berger, C.; Song, Z.; Li, X.; Wu, X.; Brown, N.; Naud, C.; Mayou, D.; Ji, T.; Hass, J.; Marchenkov, A. N.; Conrad, E. H.; First, P. N.; de Heer, W. A. _Science_ 312, 1191 (2006). * (14) Geim, A. K.; Novoselov, K. S. _Nat. Mater._ 2007, _6_ , 183. * (15) Schedin, F.; Geim, A. K.; Morozov, S. V.; Hill, E. H.; Blake, P.; Katsnelson, M. I.; Novoselov, K. S. _Nat. Mater._ 2007, _8_ , 652. * (16) Han, M. Y.; Öyilmaz, B.; Zhang, Y.; Kim, P. _Phys. Rev. Lett._ 2007, _98_ , 206805. * (17) Chen, Z.; Lin, Y.; Rooks, M. J.; Avouris, Ph. _Physica E_ 2007, _40_ , 228. * (18) Yan, Q. M.; Huang, B.; Yu, J.; Zheng, F. W.; Zang, J.; Wu, J.; Gu, B. L.; Liu, F.; Duan, W. H. _Nano Lett._ 2007, _7_ , 1459. * (19) Li, Y.; Park, C. H.; Son, Y. W., Cohen, M. L.; Louie, S. G. _Phys. Rev. Lett._ 2007, _99_ , 186801. * (20) Huang, B.; Yan, Q. M.; Zhou, G.; Wu, J.; Gu, B. L., Duan, W. H.; Liu, F. _Appl. Phys. Lett._ 2007, _91_ , 253122. * (21) Huang, B.; Liu, F.; Wu, J.; Gu, B. L.; Duan, W. H. 2007, cond-mat/0708.1795. * (22) Wang, Z. F.; Li, Q.; Zheng, H.; Ren, H.; Su, H.; Shi, Q. W.; Chen, J. _Phys. Rev. B_ 2007, _75_ , 113406. * (23) Hod, O.; Barone, V.; Peralta, J. E.; Scuseria, G. E. _Nano Lett._ 2007, _7_ , 1459. * (24) Ren, H.; Li, Q.; Su, H.; Shi, Q. W.; Chen, J.; Yang, J. L. 2007, cond-mat/0711.1700. * (25) Cervantes-Sodi, F.; Csanyi, G.; Piscanec, S.; Ferrari, A. C. 2007, cond-mat/0711.2340. * (26) Wehling, T. O.; Novoselov, K. S.; Morozov, S. V.; Vdovin, E. E.; Katsnelson, M. I.; Geim, A. K.; Lichtenstein, A. I. _Nano Lett._ 2008, _8_ , 173. * (27) Leenaerts, O.; Partoens, B.; Peeters, F. M. 2007, cond-mat/0710.1757. * (28) Robinson, J. A.; Snow, E. S.; Badescu, S. C.; Reinecke, T. L.; Perkins, F. K. _Nano Lett._ 2006, _6_ , 1747. * (29) Andzelm, J.; Govind, N.; Maiti, A. _J. Chem. Phys._ 2006, _421_ , 58. * (30) Basiuk, V. A. _Nano Lett._ 2002, _2_ , 835. * (31) Nakada, K.; Fujita, M.; Dresselhaus, G.; Dresselhaus, M. S. _Phys. Rev. B_ 1996, _54_ , 17954. * (32) Takao, Y.; Miyazaki, K.; Shimizu, Y.; Egashira, M.; _J. Electrochem. Soc._ 1994, _141_ , 1028. * (33) Kresse, G.; Furthmüller, J.; _Comput. Mater. Sci._ 1996, _6_ , 15. * (34) Taylor, J.; Guo, H.; Wang, J. _Phys. Rev. B_ 2001, _63_ , 245407. * (35) Wang, C. C.; Zhou, G.; Wu, J.; Gu, B. L.; Duan, W. H. _Appl. Phys. Lett._ 2006, _89_ , 173130. * (36) Henkelman, G.; Arnaldsson, A.; Jonsson, H. _Comput. Mater. Sci._ 2006, _36_ , 354. * (37) Skipetrov, E.; Golubev, A.; Pichugin, N.; Plastun, A.; Dmitriev, N.; Slyn’ko, V. _Phys. Stat. Sol. B_ 2007, _244_ , 448. * (38) Feit, Z.; Eger, D.; Zemel, Z. _Phys. Rev. B_ 1985, _31_ , 3903.

arxiv-papers

2024-09-04T02:48:54.247443

0803.1533

# Quantum gravity as the way from spacetime to space quantum states thermodynamics Lukasz A. Glinka1,2111E-mail to: laglinka@gmail.com , glinka@theor.jinr.ru 1 _Bogoliubov Laboratory of Theoretical Physics_ , _Joint Institute for Nuclear Research_ , _Joliot–Curie 6, 141980 Dubna, Moscow Region, Russia_ 2 _Universit $\grave{\mathit{a}}$ degli Studi di Udine,_ _Dipartimento di Matematica e Informatica,_ _via delle Scienze, 206 33100 UDINE (UD) -Italy_ ###### Abstract Physical spacetime geometry follows from some effective thermodynamics of quantum states of all fields and particles described in frames of General Relativity. In the sense of pure field theoretical Einstein’s point of view on gravitation the thermodynamic information is actually quantum gravity. We propose new realization of this old idea by studying the canonical $3+1$ Dirac–ADM approach to pseudo–Riemannian (Lorentzian) manifold of General Relativity. We derive the Wheeler–DeWitt theory as the Global One-Dimensional classical field theory of the Bose field associated with embedded 3-space, where Wheeler’s superspace metric is absent. The classical theory is discussed, some deductions on tachyon state, Dark Energy density and cosmological constant are included. Reduction to 1st order evolution is carried out, and quantum theory by the second quantization in the Fock space of creators and annihilators is constructed by employing the Heisenberg equation and the bosonic Bogoliubov transformation for diagonalization. In result we find the static reper with stable vacuum, where quantum states of 3-space can be considered, and finally space quantum states thermodynamics is formulated. ###### Contents 1. 1 Einstein’s Thermodynamic Legacy 2. 2 Hamiltonian $3+1$ Quantum Gravity 1. 2.1 The Einstein–Hilbert field equations 2. 2.2 $3+1$ General Relativity 3. 2.3 The Wheeler–DeWitt equation 3. 3 Bosonization. Global One-Dimensionality. 4. 4 Space quantum states 1. 4.1 Canonical reduction 2. 4.2 Second quantization in the Fock space 3. 4.3 Quantum bosonic field. One–point correlations. 5. 5 Classical bosonic embedded space. Tachyon. 1. 5.1 Field mass by its energy 2. 5.2 Perturbations, cosmological constant, Dark Energy 3. 5.3 Tachyon 6. 6 Quantum gravity by thermodynamics 1. 6.1 Density matrix 2. 6.2 The Bogoliubov coefficients 3. 6.3 Thermodynamics of the Bose gas 4. 6.4 Classically stable phase. Cold Big Bang. 5. 6.5 Entropy 7. 7 Summary ## 1 Einstein’s Thermodynamic Legacy Thermodynamics is the only physical theory of universal content concerning which I am convinced that, within the framework of the applicability of its basic concepts, it will never be overthrown. These words of Dr. Albert Einstein written in his autobiographical notes [1] are the testament of his life in Science. One hundred years after Einstein’s discoveries, in the centaury of microcomputers, the testament sounds strange. Today theoretical as well as mathematical physics treats thermodynamical investigations with some very subtle kind of contempt. In common established conviction thermodynamics is clear and well understood branch of physics, and is need only for engineering sciences, but not for theoreticians and mathematicians – this branch is categorized by more technologic than scientific debatable level. However, from methodological point of view, it is interesting that the General Relativity founder was the well defined thermodynamic physicist, and in spite of this crucial fact he successfully formulated physical as well as mathematical fundamentals on the modern view on gravitation, so that General Relativity today has a status of a physical theory well confirmed by experimental data from the nearest regions of Cosmos. As it is commonly known, towards the end of his life Einstein did general field theoretical considerations about physics. The reason of the general field theory dream seems to be very simple as well as very complex. Namely, from the historical point of view, the man who gave contributions to theory of diffusion processes and explained unexpected difference between theoretical and experimental predictions in the photoelectrical effect of falling down light quanta on a metal, surprisingly gave simultaneously and practically in the same time the crucial investigation to contemporary thinking about gravitational phenomena on classical level as, _i.e._ he computed theoretical values of precession of the perihelion of Mercury and binding of light rays around the sun, and generalizes the Newton law of universal gravitation. These regions of scientific activity at first glance lie in the most far conceptual points, and are not connected by straight intellectual line. Can it be only a coincidence that these outermost points was present in the Einstein considerations? Maybe it is not obvious, but really it seems that Dr. Einstein discovered only the one universal true, namely that _thermodynamics is an essence of all physics_ , because really thermodynamic effects are present on _experimental level_. In this manner also gravitation, by its obvious presence in Nature and deep physical consequences can be also considered as an thermodynamic effect. Really, the main goal of the Einstein vision was understanding that _it should be possible to describe geometry by thermodynamics_. By this reason, thermodynamics realizes the concept of the _general field theory_ searched by Einstein. In this paper we will derive some new realization of the Einstein idea. This is a mathematical way between a pseudo-Riemannian manifold given by spacetime of General Relativity, and thermodynamics of quantum states of this spacetime. Existence of this way is not obvious, and by this we propose to take into considerations the following point of view ###### Conjecture. Gravitation determined by General Relativistic spacetime can be described as the effective quantum states thermodynamics. The realization of this conjecture is essentially contained in the idea called in this paper _Thermodynamical Einstein’s Dream_. This idea is comprehend as some hypothetical way from the classical object that is a pseudo-Riemannian manifold defined by four dimensional spacetime metrics being a solution of the Einstein–Hilbert field equations of General Relativity, to generalized thermodynamics of the Bose gas of quantum states of 3-dimensional space. The conception of space quantum states arises naturally from the canonical $3+1$ Dirac–ADM approach to General Relativity, that determines theory of gravitation as an time evolution of 3-dimensional geometry of embedded space. Conceptually the thermodynamics realizes the quantum theory of general gravitational fields in the strict sense of general field theory. In this manner, this paper is devoted to give the new proposal for quantum gravity realization. Contents of this paper is as follows. First, we recall very briefly basic classical $3+1$ approach to the Einstein–Hilbert General Relativity [2, 3] called geometrodynamics, that is studying of the splitting of 4-dimensional pseudo-Riemannian geometry into 1-dimensional time and 3-dimensional space, and interprets of General Relativity as time evolution of the embedded 3-space geometry. This canonical $3+1$ Arnowitt–Deser–Misner approach to General Relativity [4, 5] leads immediately to the Hamiltonian secondary constraint of General Relativity given by some generalized the Einstein–Hamilton–Jacobi equation, that after application of the Dirac first quantization method [6] gives in the result the Wheeler–DeWitt equation and leads to the conception of the Wheeler’s superspace [7, 8, 9] as the configurational space of General Relativity. The superspatial quantum equation of motion mathematically is the second order functional differential equation with respect to metrics of 3-dimensional space, and is commonly interpreted [9, 10, 11] as a kind of the nonrelativistic Schrödinger quantum mechanics in superspace for the Wheeler–DeWitt wave function that is a functional of embedded 3-space metric field. However, by the relativistic character of General Relativity, in this paper it is proposed to reinterpret this equation not as nonrelativistic equation but as relativistic one, that is the Klein–Gordon–Fock equation for some abstract Bose field. It is demonstrated explicitly that in frames of $3+1$ metric field decomposition one can get rid of the superspace metric from the Wheeler–DeWitt equation and to change differentiation from metrics of 3-dimensional space onto determinant of 3-metrics, and in result treat the Wheeler–DeWitt wave function as the one dimensional Bose field associated with a pseudo–Riemannian manifold of General Relativity by three dimensional geometry of embedded space. Some aspects of classical theory of the boson are developed and discussed in this paper, deductions for the tachyon state, Dark Energy and cosmological constant value are included. The classical field theory is quantized by employing the second quantization in form of generalized canonical commutation relations agreed with general Von Neumann–Araki–Woods algebraic approach [12, 13] and the correct Fock space is builded. By nonlinear character of equations of motion in the Fock space, diagonalization procedure based on the Heisenberg operator evolution with using of the bosonic Bogoliubov transformation [14] is proposed, and in the result the fundamental functional operator Fock reper associated to initial data is obtained, where stable quantum vacuum state is naturally present. Quantum states of 3-dimensional space, called here Space Quantum States (SQS), are defined with respect to this static reper, and thermodynamics of the Bose gas of space quantum states is formulated by application of the one-particle density matrix method in this static basis. The SQS system is analyzed from point of view of one-point correlations of the Bose field, and thermodynamically stable phase of the system in the limit of huge number of quantum states produced from stable Bogoliubov vacuum is chosen as the correct in thermodynamical equilibrium sense. Fundamental thermodynamic characteristics are computed. The equipartition law is used to obtain a number of degrees of freedom in the classical limit of the Bose gas, and spacetime coordinates are interpreted as four thermodynamical degrees of freedom. Entropy of the Bose gas of space quantum states is analyzed, and roles of initial data vacuum quantum states, and the Cold Big Bang of SQS from vacuum are discussed. In this manner the quantum theory of gravitation is realized as generalized thermodynamics of the Bose gas of space quantum states. ## 2 Hamiltonian $3+1$ Quantum Gravity In this section we present some standard results which have a basic status for General Relativity and quantum gravity, and are need for further developments of this paper. This is the Arnowitt–Deser–Misner canonical $3+1$ approach to a Lorentzian manifold given by a solution of the Einstein–Hilbert field equations of General Relativity and its Dirac’s primary quantization that leads to the Wheeler–DeWitt equation and the concept of Superspace. ### 2.1 The Einstein–Hilbert field equations General Relativity can be obtained from the four-geometry action with fixed on a boundary three-geometry [9, 15] 222In this paper the units $8\pi G/3=c=\hbar=k_{B}=1$ are used. $S[g]=\int_{M}d^{4}x\sqrt{-g}\left\\{-\dfrac{1}{6}R[g]+\dfrac{\Lambda}{3}+\mathcal{L}\right\\}-\dfrac{1}{3}\int_{\partial M}d^{3}x\sqrt{h}K[h],$ (1) where $(M,g)$ is a pseudo–Riemannian manifold [16] with a boundary $(\partial M,h)$, $h=\det{h_{ij}}$ is 3-volume form, $K[h]=\mathrm{Tr}K_{ij}$ is traced the second fundamental form, related to unit normal vector $n^{i}$ by $K_{ij}=-\nabla_{(i}n_{j)}$ and called the extrinsic Gauss–Codazzi curvature (see _e.g._ [17]) of a boundary, $g=\det{g_{\mu\nu}}$ is 4-volume form, $R[g]$ is the Ricci scalar curvature, $\Lambda$ is cosmological constant, and $\mathcal{L}$ is a lagrangian of all physical fields considered on a manifold, called Matter Lagrangian. By application of the Hilbert–Palatini variational principle [3, 18] with respect to the fundamental field $g_{\mu\nu}$ to the action (1) $\dfrac{\delta S[g]}{\delta g_{\mu\nu}}=0,$ (2) with boundary condition $\delta S[g]\left|{}_{\partial M}\right.=0,$ (3) one can obtain the Einstein–Hilbert field equations of General Relativity $R_{\mu\nu}-\dfrac{R}{2}g_{\mu\nu}+\Lambda g_{\mu\nu}=3T_{\mu\nu},$ (4) where $T_{\mu\nu}$ is the stress–energy tensor $T_{\mu\nu}=\frac{2}{\sqrt{-g}}\frac{\delta\left(\sqrt{-g}\mathcal{L}\right)}{\delta g^{\mu\nu}},$ (5) $R_{\mu\nu}$ is the Ricci curvature tensor that is contracted the Riemann–Christoffel curvature tensor $R^{\lambda}_{\mu\alpha\nu}$, and is dependent on the Christoffel affine connections $\Gamma^{\rho}_{\mu\nu}$ and their coordinate derivatives $\displaystyle R^{\lambda}_{\mu\alpha\nu}$ $\displaystyle=$ $\displaystyle\Gamma^{\lambda}_{\mu\nu,\alpha}-\Gamma^{\lambda}_{\mu\alpha,\nu}+\Gamma^{\lambda}_{\sigma\alpha}\Gamma^{\sigma}_{\mu\nu}-\Gamma^{\lambda}_{\sigma\nu}\Gamma^{\sigma}_{\mu\alpha},$ (6) $\displaystyle R_{\mu\nu}$ $\displaystyle=$ $\displaystyle R^{\lambda}_{\mu\lambda\nu}=\Gamma^{\lambda}_{\mu\lambda\alpha}-\Gamma^{\lambda}_{\mu\alpha,\lambda}+\Gamma^{\lambda}_{\sigma\alpha}\Gamma^{\sigma}_{\mu\lambda}-\Gamma^{\lambda}_{\sigma\lambda}\Gamma^{\sigma}_{\mu\alpha},\leavevmode\nobreak\ \leavevmode\nobreak\ R=g^{\mu\nu}R_{\mu\nu}$ (7) $\displaystyle\Gamma^{\rho}_{\mu\nu}$ $\displaystyle=$ $\displaystyle\dfrac{1}{2}g^{\rho\sigma}\left(g_{\mu\sigma,\nu}+g_{\sigma\nu,\mu}-g_{\mu\nu,\sigma}\right),$ (8) where holonomic basis [15] was chosen. ### 2.2 $3+1$ General Relativity Let us introduce coordinate system chosen by the condition so that boundary space is a constant time $t$ surface and write the spacetime metric field being a solution of the Einstein–Hilbert field equations (4) in the following way $\displaystyle ds^{2}$ $\displaystyle=$ $\displaystyle g_{\mu\nu}dx^{\mu}dx^{\nu}=-N^{2}dt^{2}+h_{ij}\left(dx^{i}+N^{i}dt\right)\left(dx^{j}+N^{j}dt\right)=$ (9) $\displaystyle=$ $\displaystyle-\left(N^{2}-N_{i}N^{i}\right)dt^{2}+N_{i}dx^{i}dt+N_{j}dx^{j}dt+h_{ij}dx^{i}dx^{j},$ that actually is the Pythagoras’ theorem between two points lying on two distinguish constant time (spacelike) hypersurfaces, and was firstly investigated by Arnowitt, Deser and Misner (ADM) [4]. By this four-dimensional metrics $g_{\mu\nu}$ of the Einstein–Hilbert General Relativity Riemannian manifold in the canonical $3+1$ ADM approach has the following form $\displaystyle g_{\mu\nu}=\left[\begin{array}[]{cc}-N^{2}+N_{i}N^{i}&N_{j}\\\ N_{i}&h_{ij}\end{array}\right],$ (12) $\displaystyle g^{\mu\nu}=\left[\begin{array}[]{cc}-\dfrac{1}{N^{2}}&\dfrac{N^{j}}{N^{2}}\vspace*{5pt}\\\ \dfrac{N^{i}}{N^{2}}&h^{ij}-\dfrac{N^{i}N^{j}}{N^{2}}\end{array}\right],$ (15) $\displaystyle h_{ik}h^{kj}=\delta_{i}^{j},\leavevmode\nobreak\ \leavevmode\nobreak\ N^{i}=h^{ij}N_{j},\leavevmode\nobreak\ \leavevmode\nobreak\ g=N^{2}h,$ (16) In this case the action (1) becomes $\displaystyle S[g]=\int dt\left[\int_{\partial M}d^{3}x\left\\{\pi\dot{N}+\pi^{i}\dot{N_{i}}+\pi^{ij}\dot{h}_{ij}-NH- N_{i}H^{i}\right\\}\right],$ (17) where $\displaystyle H$ $\displaystyle=$ $\displaystyle\sqrt{h}\left\\{(K^{i}_{i}[h])^{2}-(K^{2}[h])^{i}_{i}+R[h]-2\Lambda-6\varrho\right\\},$ (18) $\displaystyle H^{i}$ $\displaystyle=$ $\displaystyle-2\pi^{ij}_{\leavevmode\nobreak\ ;j}=-2\pi^{ij}_{\leavevmode\nobreak\ ,j}-h^{il}\left(2h_{jl,k}-h_{jk,l}\right)\pi^{jk},$ (19) $\displaystyle K_{ij}[h]$ $\displaystyle=$ $\displaystyle\dfrac{1}{2N}\left(N_{i|j}+N_{j|i}-\dot{h}_{ij}\right),$ (20) where (20) follows from the Gauss-Codazzi equations for embedded space. Here $K_{ij}$ is the extrinsic-curvature tensor, $\varrho$ is the stress-energy tensor projected onto unit timelike normal vector field $n^{\mu}=\left[-\dfrac{1}{N},-\dfrac{N^{i}}{N}\right],\leavevmode\nobreak\ \leavevmode\nobreak\ n_{\mu}=\left[-N,0\right],\leavevmode\nobreak\ \leavevmode\nobreak\ n^{\mu}n_{\mu}=1,$ (21) to induced embedded 3-space $\varrho=n^{\mu}n^{\nu}T_{\mu\nu}=\dfrac{1}{N^{2}}T_{00}-\dfrac{N^{i}}{N^{2}}(T_{0i}+T_{i0})+\dfrac{N^{i}N^{j}}{N^{2}}T_{ij},$ (22) and $\pi^{ij}$ is the canonical conjugate momentum field to the field $h_{ij}$ $\pi^{ij}=\dfrac{\delta L}{\delta\dot{h}_{ij}}=\sqrt{h}\left(h^{ij}K^{i}_{i}[h]-K^{ij}[h]\right).$ (23) Time-preservation requirement [6] of the primary constraints [8] for (17) $\displaystyle\pi$ $\displaystyle=$ $\displaystyle\dfrac{\delta L}{\delta\dot{N}}\approx 0,$ (24) $\displaystyle\pi^{i}$ $\displaystyle=$ $\displaystyle\dfrac{\delta L}{\delta\dot{N_{i}}}\approx 0,$ (25) leads to the secondary constraints $\displaystyle H$ $\displaystyle\approx$ $\displaystyle 0,$ (26) $\displaystyle H^{i}$ $\displaystyle\approx$ $\displaystyle 0,$ (27) called the Hamiltonian constraint and the diffeomorphism constraint, respectively. The diffeomorphism constraint (27) merely reflects spatial diffeoinvariance of the theory, and dynamics is given by the Hamiltonian constraint (26). By using of the conjugate momentum field (23), the Hamiltonian constraint (26) can be written in the equivalent form $H=G_{ijkl}\pi^{ij}\pi^{kl}+\sqrt{h}\left(R[h]-2\Lambda-6\varrho\right)=0,$ (28) called the Einstein–Hamilton–Jacobi equation [ham1]–[ham33]. Here $G_{ijkl}=\dfrac{1}{2}h^{-1/2}\left(h_{ik}h_{jl}+h_{il}h_{jk}-h_{ij}h_{kl}\right)$ (29) is called the Wheeler superspace metric. ### 2.3 The Wheeler–DeWitt equation The classical geometrodynamics is given by the Dirac–ADM Hamiltonian constraint (28) and can be quantized by direct application of the Dirac primary quantization [6] $\displaystyle i\left[\pi^{ij}(x),h_{kl}(y)\right]$ $\displaystyle=$ $\displaystyle\dfrac{1}{2}\left(\delta_{k}^{i}\delta_{l}^{j}+\delta_{l}^{i}\delta_{k}^{j}\right)\delta^{(3)}(x,y),$ (30) $\displaystyle i\left[\pi^{i}(x),N_{j}(y)\right]$ $\displaystyle=$ $\displaystyle\delta^{i}_{j}\delta^{(3)}(x,y),$ (31) $\displaystyle i\left[\pi(x),N(y)\right]$ $\displaystyle=$ $\displaystyle\delta^{(3)}(x,y),$ (32) that in result demands to introduce the canonical conjugate momentum operator in the form ${\pi}^{ij}=-i\dfrac{\delta}{\delta h_{ij}},\leavevmode\nobreak\ \leavevmode\nobreak\ {\pi}^{j}=-i\dfrac{\delta}{\delta N_{j}},\leavevmode\nobreak\ \leavevmode\nobreak\ {\pi}=-i\dfrac{\delta}{\delta N},$ (33) and leads to the Wheeler–DeWitt equation [8] ${H}\Psi[h_{ij}]=\left\\{-G_{ijkl}\dfrac{\delta^{2}}{\delta h_{ij}\delta h_{kl}}+h^{1/2}\left(R[h]-2\Lambda-6\varrho\right)\right\\}\Psi[h_{ij}]=0.$ (34) Other first class constraints are conditions on the wave function $\Psi[h]$ ${\pi}\Psi[h_{ij}]=0,\leavevmode\nobreak\ \leavevmode\nobreak\ {\pi}^{i}\Psi[h_{ij}]=0,\leavevmode\nobreak\ \leavevmode\nobreak\ {H}^{i}\Psi[h_{ij}]=0,$ (35) and the canonical commutation relations hold $\left[{\pi}(x),{\pi}^{i}(y)\right]=\left[{\pi}(x),{H}^{i}(y)\right]=\left[{\pi}^{i}(x),{H}^{j}(y)\right]=\left[{\pi}^{i}(x),{H}(y)\right]=0.$ (36) In result $H_{i}$ are generators of diffeomorphisms $\widetilde{x}^{i}=x^{i}+\delta x^{i}$ [8] $\displaystyle i\left[h_{ij},\int_{\partial M}H_{a}\delta x^{a}d^{3}x\right]$ $\displaystyle=$ $\displaystyle-h_{ij,k}\delta x^{k}-h_{kj}\delta x^{k}_{\leavevmode\nobreak\ ,i}-h_{ik}\delta x^{k}_{\leavevmode\nobreak\ ,j}\leavevmode\nobreak\ \leavevmode\nobreak\ ,$ (37) $\displaystyle i\left[\pi_{ij},\int_{\partial M}H_{a}\delta x^{a}d^{3}x\right]$ $\displaystyle=$ $\displaystyle-\left(\pi_{ij}\delta x^{k}\right)_{,k}+\pi_{kj}\delta x^{i}_{\leavevmode\nobreak\ ,k}+\pi_{ik}\delta x^{j}_{\leavevmode\nobreak\ ,k}\leavevmode\nobreak\ \leavevmode\nobreak\ ,$ (38) which can be expressed also as constraints commutators $\displaystyle i\left[H_{i}(x),H_{j}(y)\right]\\!\\!$ $\displaystyle=$ $\displaystyle\\!\\!\int_{\partial M}H_{a}c^{a}_{ij}d^{3}z,$ (39) $\displaystyle i\left[H(x),H_{i}(y)\right]\\!\\!$ $\displaystyle=$ $\displaystyle\\!\\!H\delta^{(3)}_{,i}(x,y),$ (40) $\displaystyle i\left[\int_{\partial M}H\delta x_{1}d^{3}x,\int_{\partial M}H\delta x_{2}d^{3}x\right]\\!\\!$ $\displaystyle=$ $\displaystyle\\!\\!\int_{\partial M}H^{a}\left(\delta x_{1,a}\delta x_{2}-\delta x_{1}\delta x_{2,a}\right)d^{3}x,$ (41) where $H_{i}=h_{ij}H^{j}$, and $c^{a}_{ij}$’s are structure constants of diffeomorphism group $c^{a}_{ij}=\delta^{a}_{i}\delta^{b}_{j}\delta^{(3)}_{,b}(x,z)\delta^{(3)}(y,z)-\delta^{a}_{j}\delta^{b}_{i}\delta^{(3)}_{,b}(y,z)\delta^{(3)}(x,z)$ (42) Commutators (39-41) show the first-class constrained system property. ## 3 Bosonization. Global One-Dimensionality. Commonly the Wheeler–DeWitt theory (34) is interpreted in terms of the nonrelativistic Schrödinger quantum mechanics in configuration space of General Relativity. This point of view seems to be misleading, the conception of superspace is rather mysterious mathematical creation than real physical existence, and in this interpretation the Wheeler–Dewitt equation becomes physically senseless. Indeed one can ask: _Why primary quantization of relativistic classical field theory, that is General Relativity, must be nonrelativistic Schrödinger quantum mechanics?_ From conceptual point of view it is completely unnatural to interpret quantization of relativistic theory as nonrelativistic one. Really this question is old and seems to have answer in Dirac’s considerations – the result of classical field theory primary quantization should be relativistic quantum mechanics that is also a classical field theory. This is unique correct conceptual way on classical field theory level. However, in spite of this famous fact previous investigations of authors was concentrated on studying the Wheeler–DeWitt equation (34) as a kind of nonrelativistic stationary wave mechanics. This quantum mechanical logics applied to quantization of the Einstein–Hilbert theory of gravitation is the most popular approach in the present state of quantum cosmology and quantum gravity (See, _e.g._ , [9, 10, 11]). For example so called Loop Quantum Gravity and Loop Quantum Cosmology develop also quantum mechanics point of view. In result, in spite of beautiful philosophical as well as sophisticated mathematical constructions and many promises to gravitational physics, description of quantum gravity in terms of nonrelativistic quantum mechanics did not give any phenomenological results that can be confronted with experimental data. By this reason in this section we will study the Wheeler–DeWitt equation (34) from some new point of view, that is relativistic quantum mechanics as well as classical field theory. Recall that we have begun our considerations of quantum gravity by studying of $3+1$ decomposition of Lorentzian metric field of General Relativity, which means that actually we have considered some relativistic classical field theory after the Dirac primary quantization. From the famous candidates for the relativistic quantum mechanics equation it seems to be the best choice for this role the stationary the bosonic evolution, that is the Klein–Gordon–Fock wave equation. Indeed, the Wheeler–DeWitt theory is based on the second order differential equation in the superspace coordinate $h_{ij}$. However, it is some conceptual and formal problem to consider this equation with explicit presence of the superspace metrics $G_{ijkl}$. One can try to eliminate this refined tensor from our considerations and reduce $3+1$ Quantum Gravity to global one-dimensional classical field theory. Let us consider the standard relation between functional differentials of 4-metric field and 4-volume form (See, _e.g._ , [15]) $\delta g=gg^{\mu\nu}\delta g_{\mu\nu},$ (43) where summation convention is assumed. By employing the $3+1$ decomposition (12) one can determine the variations of contravariant metric field components $\displaystyle\delta g_{00}$ $\displaystyle=$ $\displaystyle-\delta N^{2}+N^{i}N^{j}\delta h_{ij}+h_{ij}N^{i}\delta N^{j}+h_{ij}N^{j}\delta N^{i},$ (44) $\displaystyle\delta g_{ij}$ $\displaystyle=$ $\displaystyle\delta h_{ij},$ (45) $\displaystyle\delta g_{0j}$ $\displaystyle=$ $\displaystyle h_{ij}\delta N^{i}+N^{i}\delta h_{ij},$ (46) $\displaystyle\delta g_{i0}$ $\displaystyle=$ $\displaystyle h_{ij}\delta N^{j}+N^{j}\delta h_{ij},$ (47) as well as the variation of 4-volume form $\displaystyle\delta g=N^{2}\delta h+h\delta N^{2}.$ (48) So by using of covariant metric field components we obtain finally in result the formula $\displaystyle N^{2}\delta h=N^{2}hh^{ij}\delta h_{ij},$ (49) which establishes the global relation between 3-volume form and 3-metric field contravariant components. However, the relation (49) simultaneously allows to determine the functional derivative with respect to contravariant 3-metric field as an object proportional to covariant space metrics with functional differentiation with respect to the scalar field that is the space metrics determinant (3-volume form) $\dfrac{\delta}{\delta h_{ij}}=hh^{ij}\dfrac{\delta}{\delta h}.$ (50) Double using of this functional differential operator to the wave function of the Wheeler–DeWitt equation leads to $\displaystyle\dfrac{\delta}{\delta h_{ij}}\dfrac{\delta}{\delta h_{kl}}\Psi[h]$ $\displaystyle=$ $\displaystyle hh^{ij}\dfrac{\delta}{\delta h}\left(hh^{kl}\dfrac{\delta}{\delta h}\right)\Psi[h]=$ (51) $\displaystyle=$ $\displaystyle hh^{ij}\left(h^{kl}\dfrac{\delta}{\delta h}+h\dfrac{\delta h^{kl}}{\delta h}\dfrac{\delta}{\delta h}+hh^{kl}\dfrac{\delta^{2}}{\delta h^{2}}\right)\Psi[h],$ and by direct computation of the variation of covariant space metrics with using by the relation (43) $\delta h^{ij}=\delta\dfrac{1}{h_{ij}}=-\dfrac{\delta h_{ij}}{\left(h_{ij}\right)^{2}},$ (52) we obtain as result $\delta h^{ij}=-\dfrac{h^{ij}}{h}\delta h\longrightarrow\dfrac{\delta h^{ij}}{\delta h}=-\dfrac{h^{ij}}{h},$ (53) that after application in the second term of (51) causes that two first terms vanishes and in result we obtain important for further development conclusion $\dfrac{\delta}{\delta h_{ij}}\dfrac{\delta}{\delta h_{kl}}\Psi[h]=h^{2}h^{ij}h^{kl}\dfrac{\delta^{2}\Psi[h]}{\delta h^{2}}.$ (54) One can see now that really we have not to deal with the superspace metrics $G_{ijkl}$ explicitly, namely the key relation (54) in result leads to the scalar beeing double contraction of the superspace metrics $\displaystyle G_{ijkl}h^{ij}h^{kl}=\dfrac{1}{2}h^{-1/2}\left(h_{ik}h_{jl}+h_{il}h_{jk}-h_{ij}h_{kl}\right)h^{ij}h^{kl}=-\dfrac{3}{2}h^{-1/2},$ (55) and by this the functional differentiation with respect to space metrics being an origin of the Wheeler–DeWitt equation transits into the functional differentiation with respect to space metrics determinant with some scalar coefficient $\displaystyle-G_{ijkl}\dfrac{\delta^{2}\Psi[h]}{\delta h_{ij}\delta h_{kl}}=\dfrac{3}{2}h^{3/2}\dfrac{\delta^{2}\Psi[h]}{\delta h^{2}}.$ (56) In consequence we conclude that the superspatial Wheeler–DeWitt equation (34) transforms by the following way $\left\\{\dfrac{3}{2}h^{3/2}\dfrac{\delta^{2}}{\delta h^{2}}+h^{1/2}\left(R[h]-2\Lambda-6\varrho\right)\right\\}\Psi[h]=0.$ (57) This equation can be rewritten in form of the functional 1–dimensional Klein–Gordon–Fock equation for the classical massive Bose field $\Psi[h]$ $\left\\{\dfrac{\delta^{2}}{\delta{h^{2}}}+m^{2}[h]\right\\}\Psi[h]=0,$ (58) that lies in accordance with general relativistic character of the classical Einstein–Hilbert theory of gravitation. Formally one can understand the quantity $\dfrac{2}{3h}\left(R[h]-2\Lambda-6\varrho\right)\equiv m^{2}[h],$ (59) as square of mass for the classical bosonic field $\Psi[h]$, and build a quantum theory of the Einstein–Hilbert general gravitational fields as quantum field theory, where classical Riemannian manifold is an effect of the Bose gas. ## 4 Space quantum states The previous section was devoted to presentation of results for the quantum geometrodynamics treated in terms of the relativistic quantum mechanics defined by the Klein–Gordon–Fock equation (173) for the classical Bose field $\Psi[h]$ associated with the Einstein–Hilbert Riemannian manifold of General Relativity. In the present section we will construct quantum field theory of the considered Bose field by language of the Fock space of annihilation and creation operators. ### 4.1 Canonical reduction Let us consider again the Klein–Gordon–Fock equation (173). Formally one can consider this Euler–Lagrange equation of motion as combination of two equations: the equation of motion for the canonical conjugate momentum field given by (177) and the constraint for the canonical conjugate momentum field that is (178). This system of equations can be rewritten in the reduced form, that is the first order functional differential equation as follows $\dfrac{\delta}{\delta{h}}\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right]=\left[\begin{array}[]{cc}0&1\\\ -m^{2}[h]&0\end{array}\right]\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right].$ (60) With using of the following abbreviated notation $\Phi_{\mu}=\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right],\leavevmode\nobreak\ \leavevmode\nobreak\ \partial_{\nu}=\left[\begin{array}[]{c}\dfrac{\delta}{\delta h}\\\ 0\end{array}\right],$ (61) the reduced equation (60) can be presented in the form that is looks like formally to the Dirac equation $\left(i\Gamma^{\mu}\partial_{\nu}-\mathrm{M}^{\mu}_{\nu}\right)\Phi_{\mu}=0,$ (62) where the positively defined mass matrix $\mathrm{M}^{\mu}_{\nu}$ is determined by $\mathrm{M}^{\mu}_{\nu}=\left[\begin{array}[]{cc}0&-1\\\ -m^{2}&0\end{array}\right]\geq 0.$ (63) However, in the considered case the matrices $\Gamma^{\mu}=\left[-i\mathbf{I}_{2},\mathbf{0}_{2}\right]$ create different the Clifford algebra than in the case the Dirac algebra $\left\\{\Gamma^{\mu},\Gamma^{\nu}\right\\}=2\eta^{\mu\nu}\mathbf{I}_{2},$ (64) where $\\{,\\}$ are anticommutator brackets, $\mathbf{I}_{2}$ is 2-dimensional unit matrix, and $\mathbf{0}_{2}$ is 2-dimensional null matrix, and the metrics $\eta_{\mu\nu}$ in this case is given by $\eta^{\mu\nu}=\left[\begin{array}[]{cc}-1&0\\\ 0&0\end{array}\right],$ (65) that is agreed with 1-dimensional equation (173). Let us investigate the generalized second quantization of the reduced relativistic quantum mechanics (62) by application of the Fock space of creation and annihilation functional operators. Note that the classical field theory Hamiltonian of considered system given by the relation (179) $H[h]=\dfrac{\Pi_{\Psi}^{2}[h]+m^{2}[h]\Psi^{2}[h]}{2},$ (66) can be presented in the matrix form $H[h]=[\Psi,\Pi_{\Psi}]\left[\begin{array}[]{cc}\alpha&\beta\\\ \gamma&\delta\end{array}\right]\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right]=\Phi_{\mu}^{\dagger}H^{\mu\nu}\Phi_{\nu},$ (67) where $\alpha,\leavevmode\nobreak\ \beta,\leavevmode\nobreak\ \gamma,\leavevmode\nobreak\ \delta$ are some functionals of $h$, generally. From classical point of view the Hamiltonian (67) can be rewritten as $H[h]=\alpha\Psi^{2}+\delta\Pi_{\Psi}^{2}+\gamma\Pi_{\Psi}\Psi+\beta\Psi\Pi_{\Psi},$ (68) with natural identification concluded from the form of the Hamiltonian (66) $\alpha=\dfrac{1}{2},\leavevmode\nobreak\ \leavevmode\nobreak\ \delta=\dfrac{1}{2}m^{2}[h],\leavevmode\nobreak\ \leavevmode\nobreak\ \beta=\gamma=0.$ (69) Let us build the second quantization of the reduced equation (62) based on quantization of the classical field theory Hamiltonian. ### 4.2 Second quantization in the Fock space The quantization of the considered classical Bose field theory will understand in this paper in terms of the fundamental operator quantization of the reduced Klein–Gordon–Fock field equation (62). This quantization, called the second quantization, that can be presented formally as the transition between classical fields and quantum fields operators as follows $\displaystyle\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right]\longrightarrow\left[\begin{array}[]{c}\mathbf{\Psi}\\\ \mathbf{\Pi}_{\Psi}\end{array}\right],$ (74) and applied to the reduced equation (62) gives as the result the following functional operator equation $\left(i\Gamma^{\mu}\partial_{\nu}-\mathrm{M}^{\mu}_{\nu}\right)\mathbf{\Phi}_{\mu}=0.$ (75) According to standard rules of quantum field theory, the quantization must be constructed by application of canonical commutation relations that are agreed with quantum statistics that have the entrance relativistic quantum mechanical equation given in the considered case by the Klein–Gordon–Fock equation (173). Obviously, this is the Bose statistics, and in this manner we should apply the standard rules of bosonic quantum field theory thats are [22] $\displaystyle\left[\mathbf{\Pi}_{\Psi}[h^{\prime}],\mathbf{\Psi}[h]\right]$ $\displaystyle=$ $\displaystyle-i\delta(h^{\prime}-h),$ (76) $\displaystyle\left[\mathbf{\Pi}_{\Psi}[h^{\prime}],\mathbf{\Pi}_{\Psi}[h]\right]$ $\displaystyle=$ $\displaystyle 0,$ (77) $\displaystyle\left[\mathbf{\Psi}[h^{\prime}],\mathbf{\Psi}[h]\right]$ $\displaystyle=$ $\displaystyle 0.$ (78) where $[,]$ are commutator brackets. From the quantum field theory point of view, that we want to construct here, the classical field theory Hamiltonian (66) must be quantized in terms of field operators, and in this case we should consider instead the classical Hamiltonian (67) more general quadratic form $\displaystyle\mathbf{H}[h^{\prime},h]$ $\displaystyle=$ $\displaystyle[\mathbf{\Psi}[h^{\prime}],\mathbf{\Pi}_{\Psi}[h^{\prime}]]\left[\begin{array}[]{cc}\alpha[h^{\prime},h]&\beta[h^{\prime},h]\\\ \gamma[h^{\prime},h]&\delta[h^{\prime},h]\end{array}\right]\left[\begin{array}[]{c}\mathbf{\Psi}[h]\\\ \mathbf{\Pi}_{\Psi}[h]\end{array}\right]\equiv$ (83) $\displaystyle\equiv$ $\displaystyle\mathbf{\Phi}_{\mu}^{\dagger}[h^{\prime}]H^{\mu\nu}[h^{\prime},h]\mathbf{\Phi}_{\nu}[h],$ (84) and the formula (68) in this case has a form $\displaystyle\mathbf{H}[h^{\prime},h]$ $\displaystyle=$ $\displaystyle\alpha[h^{\prime},h]\mathbf{\Psi}[h^{\prime}]\mathbf{\Psi}[h]+\delta[h^{\prime},h]\mathbf{\Pi}_{\Psi}[h^{\prime}]\mathbf{\Pi}_{\Psi}[h]+$ (85) $\displaystyle+$ $\displaystyle\gamma[h^{\prime},h]\mathbf{\Pi}_{\Psi}[h^{\prime}]\mathbf{\Psi}[h]+\beta[h^{\prime},h]\mathbf{\Psi}[h^{\prime}]\mathbf{\Pi}_{\Psi}[h],$ that by existence of the canonical commutation relations (76), (77), (78) is nonequivalent to the classical field theory Hamiltonian (67). Now we want to see directly that if we want to preserve in quantum field theory the classical form of the field Hamiltonian (67), _i.e._ $\mathbf{H}[h]=\dfrac{\mathbf{\Pi}_{\Psi}^{2}[h]+m^{2}[h]\mathbf{\Psi}^{2}[h]}{2},$ (86) where $\mathbf{H}[h]\equiv\mathbf{H}[h,h]$, then we must take into consideration the following identification $\alpha[h^{\prime},h]=\dfrac{1}{2},\leavevmode\nobreak\ \leavevmode\nobreak\ \delta[h^{\prime},h]=\dfrac{1}{2}m[h^{\prime}]m[h],\leavevmode\nobreak\ \leavevmode\nobreak\ \gamma[h^{\prime},h]=C,\leavevmode\nobreak\ \leavevmode\nobreak\ \beta[h^{\prime},h]=-C,$ (87) where $C$ is some constant c-number independent on $h$. Then from (85) we obtain directly $\mathbf{H}[h]=\dfrac{\mathbf{\Pi}_{\Psi}^{2}[h]+m^{2}[h]\mathbf{\Psi}^{2}[h]}{2}-iC\delta(0),$ (88) where the last term can be omitted by c-number character. However, generally the quantum field theory of considered boson has the following field Hamiltonian $\mathbf{H}[h]=\mathbf{\Phi}^{\dagger}[h]\left[\begin{array}[]{cc}\dfrac{1}{2}&-C\\\ C&\dfrac{m^{2}[h]}{2}\end{array}\right]\mathbf{\Phi}[h],$ (89) that is obviously nondiagonal. Because of, as it was seen in the relation (88), the constant c-number $C$ does not play role in physics - it is only a choice of reference Hamiltonian value - one can put into computations the simplest case $C\equiv 0$. Then the quantum field theory Hamiltonian (89) is diagonal, and moreover the demanded classical form of the quantum field Hamiltonian is preserved $\mathbf{H}[h]=\dfrac{\mathbf{\Pi}_{\Psi}^{2}[h]+m^{2}[h]\mathbf{\Psi}^{2}[h]}{2}.$ (90) In this manner we must not search for special diagonalizable basis, and we can directly apply to the system the Fock space quantization. We propose apply to the reduced Klein–Gordon–Fock equation (75) the following generalized fundamental operator quantization in the Fock space of creation and annihilation functional operators $\mathsf{G}^{\dagger}[h]$ and $\mathsf{G}[h]$ $\displaystyle\left[\begin{array}[]{c}\mathbf{\Psi}[h]\\\ \mathbf{\Pi}_{\Psi}[h]\end{array}\right]=\left[\begin{array}[]{cc}\dfrac{1}{\sqrt{2|m[h]|}}&\dfrac{1}{\sqrt{2|m[h]|}}\\\ -i\sqrt{\dfrac{|m[h]|}{2}}&i\sqrt{\dfrac{|m[h]|}{2}}\end{array}\right]\left[\begin{array}[]{c}\mathsf{G}[h]\\\ \mathsf{G}^{\dagger}[h]\end{array}\right],$ (97) Let us note that this second quantization lies in strict accordance with the bosonic character of the equation (173), and also with the generalized algebraic approach investigated in papers of Von Neumann, Araki and Woods [12, 13]. The principal canonical commutation relations (76), (77), and (78) are automatically fulfilled if the considered Bose system is described by the dynamical basis $\mathfrak{B}[h]$ in the proposed Fock space construction $\mathfrak{B}[h]=\left\\{\left[\begin{array}[]{c}\mathsf{G}[h]\\\ \mathsf{G}^{\dagger}[h]\end{array}\right]:\left[\mathsf{G}[h^{\prime}],\mathsf{G}^{\dagger}[h]\right]=\delta\left(h^{\prime}-h\right),\left[\mathsf{G}[h^{\prime}],\mathsf{G}[h]\right]=0\right\\},$ (98) so that the Fock space quantization (97) can be rewritten briefly as action of the second quantization matrix on the dynamical basis $\mathbf{\Phi}[h]=\mathbb{Q}[h]\mathfrak{B}[h],$ (99) where the second quantization matrix can be determined directly as $\mathbb{Q}[h]=\left[\begin{array}[]{cc}\dfrac{1}{\sqrt{2|m[h]|}}&\dfrac{1}{\sqrt{2|m[h]|}}\\\ -i\sqrt{\dfrac{|m[h]|}{2}}&i\sqrt{\dfrac{|m[h]|}{2}}\end{array}\right].$ (100) The quantum field dynamics considered from the point of view of the basis $\mathfrak{B}[h]$ is described by the Klein–Gordon–Fock equation (173) with application of the second quantization (99). By elementary calculations one can obtain that the $h$-evolution of the basis $\mathfrak{B}[h]$ is governed by the following equation of motion $\dfrac{\delta\mathfrak{B}[h]}{\delta h}=\left[\begin{array}[]{cc}-im[h]&\dfrac{1}{2m[h]}\dfrac{\delta m[h]}{\delta h}\\\ \dfrac{1}{2m[h]}\dfrac{\delta m[h]}{\delta h}&im[h]\end{array}\right]\mathfrak{B}[h].$ (101) Formally, this is nonlinear first order functional differential equation. This system of equations for the creation and annihilation functional operators can not be solved directly by application of the standard path integrals method, here the coupling between annihilation and creation operators is present in form of nondiagonal elements. Moreover, in the dynamical basis (98) we have to deal with some kind of global situation – by reference frame dependence of the quantum vacuum state, vacuum expectation values of the quantum field theory Hamiltonian (90) that in the dynamical basis has a following form $\displaystyle\mathbf{H}[h]$ $\displaystyle=$ $\displaystyle\mathfrak{B}^{\dagger}[h]\left[\begin{array}[]{cc}\dfrac{m[h]}{2}&0\\\ 0&\dfrac{m[h]}{2}\end{array}\right]\mathfrak{B}[h]=$ (104) $\displaystyle=$ $\displaystyle m[h]\left(\mathsf{G}^{\dagger}[h]\mathsf{G}[h]+\dfrac{1}{2}\delta(0)\right),$ (105) can not be treated as correctly defined, by the dynamical character of the basis. For correctness we must build some static functional operator basis – in this type of basis, the vacuum expectation values can be determined by local status of the basis, and by this quantum field theory is no senseless. The recept to the similar evolutions is only one, unique, and unambiguous – this is diagonalization of the operator evolution to the Heisenberg canonical form together with using of the Bogoliubov transformation agreed with canonical commutation relations. Let us apply the diagonalization procedure. Firstly, we take into considerations the supposition that sounds that in the Fock space exists some local basis where exactly the same canonical commutation relations between creation and annihilation functional operators are preserved $\mathfrak{B}^{\prime}[h]=\left\\{\left[\begin{array}[]{c}\mathsf{G}^{\prime}[h]\\\ \mathsf{G}^{\prime\dagger}[h]\end{array}\right]:\left[\mathsf{G}^{\prime}[h^{\prime}],\mathsf{G}^{\prime\dagger}[h]\right]=\delta\left(h^{\prime}-h\right),\left[\mathsf{G}^{\prime}[h^{\prime}],\mathsf{G}^{\prime}[h]\right]=0\right\\}.$ (106) This basis in our studies has the fundamental status, namely we suppose that in this basis the functional operator evolution (101) diagonalizes directly to the canonical Heisenberg operator evolution $\dfrac{\delta\mathfrak{B}^{\prime}[h]}{\delta h}=\left[\begin{array}[]{cc}-i\lambda[h]&0\\\ 0&i\lambda[h]\end{array}\right]\mathfrak{B}^{\prime}[h],$ (107) where $\lambda[h]$ is generally some functional of the evolution parameter. By using of the supposition that in the local basis are preserved the bosonic canonical commutation relations, one can deduce directly that the fundamental basis $\mathfrak{B}^{\prime}[h]$ should be obtained from the dynamical basis $\mathfrak{B}[h]$ by some generalized canonical transformation in the considered Fock space, that is a rotation of basis determined by standardly defined the Bogoliubov transformation, which in the case of systems with Bose statistics has the following form $\mathfrak{B}^{\prime}[h]=\left[\begin{array}[]{cc}u[h]&v[h]\\\ v^{\ast}[h]&u^{\ast}[h]\end{array}\right]\mathfrak{B}[h],$ (108) where the Bogoliubov coefficients $u$ and $v$ are functionals of $h$, and by rotational character of the bosonic Bogoliubov transformation these coefficients obey the Lobachevskiy–Gauss–Bolyai hyperbolic space condition $|u[h]|^{2}-|v[h]|^{2}=1.$ (109) This can be checked by direct elementary computation that the proposed two- step diagonalization operator evolution equations procedure in result leads to demanding of vanishing of the functional $\lambda[h]$, but simultaneously transits a whole dynamical evolution from the operator basis onto the system of the bosonic Bogoliubov coefficients $\dfrac{\delta}{\delta h}\left[\begin{array}[]{c}u[h]\\\ v[h]\end{array}\right]=\left[\begin{array}[]{cc}-im[h]&\dfrac{1}{2m[h]}\dfrac{\delta m[h]}{\delta h}\\\ \dfrac{1}{2m[h]}\dfrac{\delta m[h]}{\delta h}&im[h]\end{array}\right]\left[\begin{array}[]{c}u[h]\\\ v[h]\end{array}\right].$ (110) By this the procedure leads to realization of the main aim of this construction - namely this gives definition of the static operator basis that has the fundamental character; the static basis is completely determined by initial data problem in the Fock space, and is given by usual ladder operators $\mathfrak{B}^{\prime}[h]=\mathfrak{B}_{I}=\left\\{\left[\begin{array}[]{c}\mathsf{G}_{I}\\\ \mathsf{G}^{\dagger}_{I}\end{array}\right]:\left[\mathsf{G}_{I},\mathsf{G}^{\dagger}_{I}\right]=1,\left[\mathsf{G}_{I},\mathsf{G}_{I}\right]=0\right\\}.$ (111) Furthermore, by the static character the fundamental operator basis, this basis defines static quantum vacuum state given as $|0\rangle_{I}=\left\\{|0\rangle_{I}:\mathsf{G}_{I}|0\rangle_{I}=0,\leavevmode\nobreak\ 0={{}_{I}}\langle 0|\mathsf{G}_{I}^{\dagger}\right\\},$ (112) and vacuum expectation values computed on this _initial data_ vacuum state are well-defined by local status of the fundamental basis $\mathfrak{B}_{I}$. The functional differential equations for the Bogoliubov coefficients (110) can be solved directly by famous methods of linear analytical algebra based on the Cayley–Hamilton theorem. However, in the present situation we have very special evolution - actually the bosonic Bogoliubov coefficients can not be chosen arbitrary, by the fact that they are constrained by the rotational condition (109), and in possible solving method we should construct firstly some parametrization that lies in accordance with this hyperbolic constraint, and then to try solve the coefficients evolution equation (110) in this concretely chosen parametrization. It can be checked by direct algebraic manipulation that reverse conduct leads to bad-defined algebraical problem. By hyperbolic view of the rotational condition (109) we suggest to use the very special parametrization of the bosonic Bogoliubov coefficients, so called the superfluid coordinate system defined by the following transformation $\displaystyle u[h]$ $\displaystyle=$ $\displaystyle\exp\left\\{i\theta[h]\right\\}\cosh\phi[h],$ (113) $\displaystyle v[h]$ $\displaystyle=$ $\displaystyle\exp\left\\{i\theta[h]\right\\}\sinh\phi[h].$ (114) Elementary algebraic manipulations lead to the system of functional differential equations for the parameters $\theta[h]$ and $\phi[h]$, that can be solved directly and the solutions can be written in the following form $\displaystyle\theta[h]$ $\displaystyle=$ $\displaystyle\pm m_{I}\int_{h_{I}}^{h}\sqrt{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}\delta h,$ (115) $\displaystyle\phi[h]$ $\displaystyle=$ $\displaystyle\ln{\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}},\leavevmode\nobreak\ \leavevmode\nobreak\ m_{I}=m[h_{I}].$ (116) The interpretation of these solutions is obvious – the quantity $\theta[h]$ is integrated mass of the considered boson, and the solution $\phi[h]$ is logarithmic field associated with mass of the boson. By this the bosonic Bogoliubov coefficients (110) can be determined as follows $\displaystyle u[h]$ $\displaystyle=$ $\displaystyle\dfrac{1}{2}\exp\left\\{\pm im_{I}\int_{h_{I}}^{h}\dfrac{m[h]}{m_{I}}\delta h\right\\}\left(\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}+\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\right),$ (117) $\displaystyle v[h]$ $\displaystyle=$ $\displaystyle\dfrac{1}{2}\exp\left\\{\pm im_{I}\int_{h_{I}}^{h}\dfrac{m[h]}{m_{I}}\delta h\right\\}\left(\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}-\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\right).$ (118) We have a freedom in sign choosing of the phases, but we decide to choose the positive phases. Actually by definition of the coefficients $u$ and $v$, we determinate the monodromy matrix $\mathbb{G}[h]$ that transits the fundamental basis $\mathfrak{B}_{I}$ into the dynamical one $\mathfrak{B}[h]$ defined as $\mathfrak{B}[h]=\mathbb{G}[h]\mathfrak{B}_{I}$ (119) that has the following form $\displaystyle\mathbb{G}[h]\\!=\\!\left[\\!\\!\\!\begin{array}[]{cc}\left(\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\\!+\\!\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}\right)\dfrac{e^{-i\theta[h]}}{2}\vspace*{10pt}\\!\\!\\!\\!&\left(\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\\!-\\!\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}\right)\dfrac{e^{i\theta[h]}}{2}\\\ \left(\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\\!-\\!\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}\right)\dfrac{e^{-i\theta[h]}}{2}\\!\\!\\!\\!&\left(\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}[h]}\right|}\\!+\\!\sqrt[4]{\left|\dfrac{m^{2}[h]}{m_{I}^{2}}\right|}\right)\dfrac{e^{i\theta[h]}}{2}\end{array}\\!\\!\\!\right]\\!\\!.$ (122) In this manner one can conclude directly that in the presented approach the quantum theory of gravitation is completely determined by the correct choose of the monodromy matrix between dynamic and static bases in the Fock space of creation and annihilation functional operators. By this reason in the Fock space formulation, quantum gravitation one can immediately understood as the phenomena that is an effect of the choice of operator basis. The initial data basis $\mathfrak{B}_{I}$ is directly related with intial data of the creation and annihilation operators, and in this manner this has the fundamental status for description of physical phenomena – the monodromy matrix (122) consists whole information about dynamics of a space geometry of the Riemannian manifold given by a solution of the Einstein–Hilbert field equations of General Relativity (LABEL:gr1). Moreover, one can see directly from the form of the monodromy matrix (122), that this fundamental quantity is completely determined by a quotient of two squares of mass for the Bose field - one taken in the initial point, and the second taken in the current evolution point. By this reason, actually this quotient of squares of mass has the fundamental physical meaning for the quantum theory of the considered Bose field. ### 4.3 Quantum bosonic field. One–point correlations. The field operator $\mathbf{\Phi}[h]$ which is directly associated with a 3-dimensional spatial part of the Einstein–Hilbert Riemannian manifold of General Relativity, and represents a spacetime in terms of bosonic quantum field theory can be concluded immediately as an effect of transformation of the fundamental static initial data basis by directed action of the monodromy matrix $\mathbb{G}[h]$ and the second quantization matrix $\mathbb{Q}[h]$ as follows $\mathbf{\Phi}[h]=\mathbb{Q}[h]\mathbb{G}[h]\mathfrak{B}_{I}.$ (123) In this manner by multiplication of matrices $\mathbb{Q}$ and $\mathbb{G}$ given by the formulas (99) and (122) the bosonic field operator associated with a spatial geometry of spacetime can be concluded in the form $\displaystyle\mathbf{\Psi}[h]=\frac{1}{2\sqrt{2m_{I}}}\sqrt{\dfrac{m_{I}^{2}}{m^{2}[h]}}\left(e^{-i\theta[h]}\mathsf{G}_{I}+e^{i\theta[h]}\mathsf{G}^{\dagger}_{I}\right).$ (124) This field operator is formally hermitian operator $\displaystyle\mathbf{\Psi}^{\dagger}[h]=\mathbf{\Psi}[h],$ (125) and acts on the static vacuum state by the following way $\displaystyle\mathbf{\Psi}[h]|0\rangle_{I}$ $\displaystyle=$ $\displaystyle\frac{1}{2\sqrt{2m_{I}}}\sqrt{\dfrac{m_{I}^{2}}{m^{2}[h]}}e^{i\theta[h]}\mathsf{G}^{\dagger}_{I}|0\rangle_{I},$ (126) $\displaystyle{{}_{I}}\langle 0|\mathbf{\Psi}^{\dagger}[h]$ $\displaystyle=$ $\displaystyle{{}_{I}}\langle 0|\mathsf{G}_{I}\frac{1}{2\sqrt{2m_{I}}}\sqrt{\dfrac{m_{I}^{2}}{m^{2}[h]}}e^{-i\theta[h]}.$ (127) Linear algebra gives the theorem that states that eigenvalues of operator function are given by functions of the operator eigenvalues, and in considered case it particularly allows to define the multifield quantum states $\displaystyle\left(\mathbf{\Psi}[h]\right)^{n}|0\rangle_{I}$ $\displaystyle=$ $\displaystyle\left(\frac{1}{2\sqrt{2m_{I}}}\sqrt{\dfrac{m_{I}^{2}}{m^{2}[h]}}e^{i\theta[h]}\right)^{n}\mathsf{G}^{\dagger n}_{I}|0\rangle_{I},$ (128) $\displaystyle{{}_{I}}\langle 0|\left(\mathbf{\Psi}^{\dagger}[h^{\prime}]\right)^{n^{\prime}}$ $\displaystyle=$ $\displaystyle{{}_{I}}\langle 0|\mathsf{G}_{I}^{n^{\prime}}\left(\frac{1}{2\sqrt{2m_{I}}}\sqrt{\dfrac{m_{I}^{2}}{m^{2}[h]}}e^{-i\theta[h^{\prime}]}\right)^{n^{\prime}},$ (129) where $n^{\prime}$ and $n$ are natural numbers, and $h^{\prime}$ and $h$ are determinants of space metrics and characterize quantum state of spacetime given by a metrics with space part described respectively by $h_{\mu\nu}^{\prime}$ and $h_{\mu\nu}$. We will these states as _space quantum states of a spacetime_ and for shortness we will note these states as $\displaystyle\left(\mathbf{\Psi}[h]\right)^{n}|0\rangle_{I}$ $\displaystyle\equiv$ $\displaystyle|h,n\rangle,$ (130) $\displaystyle{{}_{I}}\langle 0|\left(\mathbf{\Psi}^{\dagger}[h^{\prime}]\right)^{n^{\prime}}$ $\displaystyle\equiv$ $\displaystyle\langle n^{\prime},h^{\prime}|.$ (131) Generalized two-point correlation functions of two space quantum states can be determined immediately as $\displaystyle\langle n^{\prime},h^{\prime}|h,n\rangle=\dfrac{m_{I}^{(n^{\prime}+n)/2}}{2^{3(n^{\prime}+n)/2}}\frac{e^{-i(n^{\prime}\theta[h^{\prime}]-n\theta[h])}}{(m[h^{\prime}])^{n^{\prime}}(m[h])^{n}}{{}_{I}}\langle 0|\mathsf{G}_{I}^{n^{\prime}}\mathsf{G}^{\dagger n}_{I}|0\rangle_{I}.$ (132) By normalization to unity the initial data correlator $\langle 1,h_{I}|h_{I},1\rangle=\dfrac{1}{8m_{I}}{{}_{I}}\langle 0|0\rangle_{I}\equiv 1,$ (133) one can determinate the vacuum-vacuum amplitude as ${{}_{I}}\langle 0|0\rangle_{I}=8m_{I},$ (134) and simultaneously it can be treated as the definition of initial data mass $m_{I}$. Especially interesting for further developments correlators are $\displaystyle\langle 1,h|h,1\rangle$ $\displaystyle=$ $\displaystyle\dfrac{m_{I}^{2}}{m^{2}[h]},$ (135) $\displaystyle\langle n^{\prime},h|h,n\rangle$ $\displaystyle=$ $\displaystyle\left(\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}}\right)^{(n^{\prime}+n)/2}e^{-i(n^{\prime}-n)\theta[h]}{{}_{I}}\langle 0|\mathsf{G}_{I}^{n^{\prime}}\mathsf{G}^{\dagger n}_{I}|0\rangle_{I},$ (136) $\displaystyle\langle 1,h^{\prime}|h,1\rangle$ $\displaystyle=$ $\displaystyle\dfrac{m_{I}^{2}}{m[h^{\prime}]m[h]}\exp\left\\{i\int_{h^{\prime}}^{h}m[h^{\prime\prime}]\delta h^{\prime\prime}\right\\},$ (137) $\displaystyle\dfrac{\langle n,h^{\prime}|h,n\rangle}{{{}_{I}}\langle 0|0\rangle_{I}}$ $\displaystyle=$ $\displaystyle\left(\dfrac{\langle 1,h^{\prime}|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}}\right)^{n},$ (138) where in calculations of vacuum expectation values was used the identity $\displaystyle{{}_{I}}\langle 0|\mathsf{G}_{I}^{n}\mathsf{G}^{\dagger n}_{I}|0\rangle_{I}$ $\displaystyle=$ $\displaystyle{{}_{I}}\langle 0|\left(\mathsf{G}_{I}\mathsf{G}^{\dagger}_{I}\right)^{n}|0\rangle_{I}={{}_{I}}\langle 0|\left(1+\mathsf{G}^{\dagger}_{I}\mathsf{G}_{I}\right)^{n}|0\rangle_{I}=$ (139) $\displaystyle=$ $\displaystyle{{}_{I}}\langle 0|\sum_{k=0}^{n}C^{n}_{k}\left(\mathsf{G}^{\dagger}_{I}\mathsf{G}_{I}\right)^{k}|0\rangle_{I}=\sum_{k=0}^{n}C^{n}_{k}{{}_{I}}\langle 0|\left(\mathsf{G}^{\dagger}_{I}\mathsf{G}_{I}\right)^{k}|0\rangle_{I}=$ $\displaystyle=$ $\displaystyle C^{n}_{0}{{}_{I}}\langle 0|0\rangle_{I}=8m_{I},$ with $C^{n}_{k}=\dfrac{n!}{k!(n-k)!}$ being the Newton binomial coefficient. The correlator (135) is basic, naturally one can find from (132) that $\dfrac{\langle n^{\prime},h^{\prime}|h,n\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{(n^{\prime}+n)/2}}=\sqrt{\langle 1,h^{\prime}|h^{\prime},1\rangle^{n^{\prime}}\langle 1,h|h,1\rangle^{n}}e^{-im_{I}\theta_{n^{\prime},n}[h^{\prime},h]}{{}_{I}}\langle 0|\mathsf{G}_{I}^{n^{\prime}}\mathsf{G}^{\dagger n}_{I}|0\rangle_{I},$ (140) where $\theta_{n^{\prime},n}[h^{\prime},h]=n^{\prime}\int_{h_{I}}^{h^{\prime}}\dfrac{\delta h^{\prime\prime}}{\sqrt{\langle 1,h^{\prime\prime}|h^{\prime\prime},1\rangle}}-n\int_{h_{I}}^{h}\dfrac{\delta h^{\prime\prime}}{\sqrt{\langle 1,h^{\prime\prime}|h^{\prime\prime},1\rangle}}.$ (141) For example one can directly define the two-point correlator (137) in the following way $\displaystyle\langle 1,h^{\prime}|h,1\rangle=\sqrt{\langle 1,h^{\prime}|h^{\prime},1\rangle\langle 1,h|h,1\rangle}\exp\left\\{im_{I}\int_{h^{\prime}}^{h}\dfrac{\delta h^{\prime\prime}}{\sqrt{\langle 1,h^{\prime\prime}|h^{\prime\prime},1\rangle}}\right\\},$ (142) or by application of the functional Taylor series expansion of the integral in exponent of the last relation $\int_{h^{\prime}}^{h}\dfrac{\delta h^{\prime\prime}}{\sqrt{\langle 1,h^{\prime\prime}|h^{\prime\prime},1\rangle}}=\sum_{n=0}^{\infty}\kappa_{n}[h,h^{\prime}|h_{I}]{\dfrac{\delta^{n}}{\delta h^{n}}\langle 1,h|h,1\rangle\Biggr{|}_{h_{I}}},$ (143) with the functional coefficients $\kappa_{n}[h,h^{\prime}|h_{I}]=\dfrac{(2n-3)!}{2^{2n-1}(n-1)!}\sum_{k=0}^{n+1}\dfrac{(-1)^{k}}{k!(n-k+1)!}(h_{I})^{n-k+1}\left(h^{k}-h^{\prime k}\right),$ (144) the considered two-point correlator becomes $\langle 1,h^{\prime}|h,1\rangle=\sqrt{\langle 1,h^{\prime}|h^{\prime},1\rangle\langle 1,h|h,1\rangle}\prod_{n=0}^{\infty}\sum_{p=0}^{\infty}\dfrac{\left(\kappa_{n}^{\prime}\right)^{p}}{p!}\left(\dfrac{\delta^{n}}{\delta h^{n}}\langle 1,h|h,1\rangle\Biggr{|}_{h_{I}}\right)^{p},$ (145) where for shortness $\kappa_{n}^{\prime}\equiv im_{I}\kappa_{n}[h,h^{\prime}|h_{I}]$. Let us consider in detail the one point correlator (135). Firstly let us note that by application of the correlator (135) the bosonic field (124) is immediately determined by this one-point correlator in the following way $\displaystyle\mathbf{\Psi}[h]=\frac{1}{2\sqrt{2m_{I}}}\sqrt{\langle 1,h|h,1\rangle}\left(e^{-i\theta[h]}\mathsf{G}_{I}+e^{i\theta[h]}\mathsf{G}^{\dagger}_{I}\right),$ (146) and by this it gives also the self-interaction $\mathbf{\Psi}^{\dagger}[h]\mathbf{\Psi}[h]=\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}}\left[\left(e^{-2i\theta[h]}\mathsf{G}_{I}+\mathsf{G}_{I}^{\dagger}\right)\mathsf{G}_{I}+h.c.\right],$ (147) where $h.c.$ means hermitean conjugation. From the other side by using of the square of mass definition (59) one can obtain the relation $\dfrac{2}{3h}\left(R[h]-2\Lambda-6\varrho\right)=m^{2}[h]=\dfrac{m_{I}^{2}}{\langle 1,h|h,1\rangle}=\dfrac{1}{64}\dfrac{{{}_{I}}\langle 0|0\rangle_{I}^{2}}{\langle 1,h|h,1\rangle},$ (148) that can be interpreted as the other definition of the one-point correlator by basic geometrical quantities associated with 3-dimensional space metrics $\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}}=\dfrac{3}{128}\dfrac{h}{R[h]-2\Lambda-6\varrho}{{}_{I}}\langle 0|0\rangle_{I}.$ (149) Also the Dark Matter energy density (228) can be expressed immediately by this correlator $\rho_{DM}=\dfrac{3h}{788}\dfrac{{{}_{I}}\langle 0|0\rangle_{I}^{2}}{\langle 1,h|h,1\rangle}.$ (150) Moreover, as it was noted in the previous subsection the monodromy matrix (122) is immediately determined by the quotient of squares of mass, that is really the one-point correlator by the formula (135). In this manner the monodromy matrix (122) is really dependent on the one-point correlator in the following way $\displaystyle\mathbb{G}[h]\\!=\\!\left[\\!\\!\\!\begin{array}[]{cc}\dfrac{\sqrt{\langle 1,h|h,1\rangle}+1}{\sqrt[4]{\langle 1,h|h,1\rangle}}\dfrac{e^{-i\theta[h]}}{2}\vspace*{10pt}\\!\\!\\!\\!&\dfrac{\sqrt{\langle 1,h|h,1\rangle}-1}{\sqrt[4]{\langle 1,h|h,1\rangle}}\dfrac{e^{i\theta[h]}}{2}\\\ \dfrac{\sqrt{\langle 1,h|h,1\rangle}-1}{\sqrt[4]{\langle 1,h|h,1\rangle}}\dfrac{e^{-i\theta[h]}}{2}\\!\\!\\!\\!&\dfrac{\sqrt{\langle 1,h|h,1\rangle}+1}{\sqrt[4]{\langle 1,h|h,1\rangle}}\dfrac{e^{i\theta[h]}}{2}\end{array}\\!\\!\\!\right]\\!\\!,$ (153) where according to the definition (115) and with accepted sign, the phase is equal to $\theta[h]=m_{I}\int_{h_{I}}^{h}\sqrt{\langle 1,h|h,1\rangle}\delta h.$ (154) Similarly the second quantization matrix (100) can be completely determined by the considered one-point correlator in the following way $\mathbb{Q}[h]=\left[\begin{array}[]{cc}\dfrac{1}{\sqrt{2m_{I}}}\sqrt[4]{\langle 1,h|h,1\rangle}&\dfrac{1}{\sqrt{2m_{I}}}\sqrt[4]{\langle 1,h|h,1\rangle}\\\ -i\sqrt{\dfrac{m_{I}}{2}}\dfrac{1}{\sqrt[4]{\langle 1,h|h,1\rangle}}&i\sqrt{\dfrac{m_{I}}{2}}\dfrac{1}{\sqrt[4]{\langle 1,h|h,1\rangle}}\end{array}\right].$ (155) Note that in the previous section we obtained the relation for square of mass (208) by the coefficients $\alpha$’s $\displaystyle m^{2}[h]$ $\displaystyle=$ $\displaystyle-\dfrac{1}{(h-h_{I})^{2}}+\left(\dfrac{\dfrac{\alpha_{2}}{h-h_{I}}}{1+\dfrac{\alpha_{2}}{h-h_{I}}}\right)^{2}\left[\dfrac{1}{(h-h_{I})^{2}}-\dfrac{\alpha_{1}^{2}+2\alpha_{1}}{\alpha_{2}}\right]+$ (156) $\displaystyle+$ $\displaystyle 2\dfrac{\dfrac{\alpha_{2}}{h-h_{I}}}{\left(1+\dfrac{\alpha_{2}}{h-h_{I}}\right)^{2}}\left[\dfrac{1}{(h-h_{I})^{2}}+\dfrac{\alpha_{0}}{\alpha_{2}}+\dfrac{\alpha_{0}}{h-h_{I}}\right],$ and by this the correlator (135) can be determined now by the following way $\displaystyle\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}\\!\\!\\!$ $\displaystyle=$ $\displaystyle\\!\\!\\!\dfrac{1}{64}\Biggr{\\{}-\dfrac{1}{(h-h_{I})^{2}}+\left(\dfrac{\dfrac{\alpha_{2}}{h-h_{I}}}{1+\dfrac{\alpha_{2}}{h-h_{I}}}\right)^{2}\left[\dfrac{1}{(h-h_{I})^{2}}-\dfrac{\alpha_{1}^{2}+2\alpha_{1}}{\alpha_{2}}\right]+$ (157) $\displaystyle+$ $\displaystyle\\!\\!\\!2\dfrac{\dfrac{\alpha_{2}}{h-h_{I}}}{\left(1+\dfrac{\alpha_{2}}{h-h_{I}}\right)^{2}}\left[\dfrac{1}{(h-h_{I})^{2}}+\dfrac{\alpha_{0}}{\alpha_{2}}+\dfrac{\alpha_{0}}{h-h_{I}}\right]\Biggr{\\}}^{-1},$ that for case of constant energies of the rang $\varepsilon$ according to (218) becomes $\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}=\dfrac{(h-h_{I})^{2}}{64}\dfrac{\varepsilon^{2}(h-h_{I})^{3}+4\varepsilon(h-h_{I})^{2}+4(h-h_{I})}{3\varepsilon^{2}(h-h_{I})^{3}+8\varepsilon(h-h_{I})^{2}-4(h-h_{I})-4\varepsilon},$ (158) and in the tachyon limit we obtain $\lim_{\varepsilon\rightarrow 0}\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}=-\dfrac{1}{64}(h-h_{I})^{2}.$ (159) The one-point correlation function (158) can be interpreted as the information source on quantum stable states of the considered Bose system. Namely, as it is common accepted in research on similar situation in particle physics, one can consider the poles of the correlator with respect to the variable $h-h_{I}$. The poles are given by zeros of the correlator (158) denominator $3\varepsilon^{2}(h-h_{I})^{3}+8\varepsilon(h-h_{I})^{2}-4(h-h_{I})-4\varepsilon=0.$ (160) This is the third order polynomial equation, and generally this kind of equations has three roots – two complex and one real. Let us consider only the real root, because it is the stable state only. By elementary algebraic methods of the Galois group, one can obtain that the real solution of the equation (158) is given by $\displaystyle h-h_{I}$ $\displaystyle=$ $\displaystyle-\dfrac{8}{9\varepsilon}+\dfrac{50}{9\varepsilon\sqrt[3]{\dfrac{243}{4}\varepsilon^{2}-118+9\sqrt{3}\sqrt{-7-59\varepsilon^{2}+\dfrac{243}{16}\varepsilon^{4}}}}+$ (161) $\displaystyle+$ $\displaystyle\dfrac{\sqrt[3]{-944+486\varepsilon^{2}+18\sqrt{3}\sqrt{-112-944\varepsilon^{2}+243\varepsilon^{4}}}}{9\varepsilon^{2}}.$ This solution has a little bit complicated form, but we can exchange these some complex solution by the following recept. Namely, we know that the real solution is only one, and by this one can use the parametrization of the constant $\varepsilon$, related with $h-h_{I}$ by the inequality (219), by the following way $\varepsilon=\theta\dfrac{2}{h-h_{I}},$ (162) where $\theta$ is a number, that by the formula (219) must fulfills $|\theta|<1\Rightarrow-1<\theta<1.$ (163) With this supposition the formula (158) can be expressed as follows $\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}=\left(\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}\right)_{\theta=0}\dfrac{(\theta+1)^{2}(h-h_{I})^{2}}{2\theta-(3\theta^{2}+4\theta-1)(h-h_{I})^{2}},$ (164) where the correlator $\left(\dfrac{\langle 1,h|h,1\rangle}{{{}_{I}}\langle 0|0\rangle_{I}^{2}}\right)_{\theta=0}=-\dfrac{(h-h_{I})^{2}}{64},$ (165) is identified with tachyon state. The real and positive pole of (164) is determined by very simple relation $h-h_{I}=\dfrac{2\theta}{3\theta^{2}+4\theta-1},$ (166) In this manner, by using of the regularization (162), actually one can consider the relative correlator $\dfrac{\langle 1,h|h,1\rangle}{\left(\langle 1,h|h,1\rangle\right)_{\theta=0}}=\dfrac{(\theta+1)^{2}(h-h_{I})^{2}}{2\theta-(3\theta^{2}+4\theta-1)(h-h_{I})^{2}},$ (167) where the vacuum-vacuum amplitude was absorbed by reduction. This correlator expressed in units for which the square of the tachyon mass is equal to minus unity, _i.e._ for $m_{0}^{2}=-\dfrac{1}{(h-h_{I})^{2}}\equiv-1,$ (168) becomes very simple $\left(\dfrac{\langle 1,h|h,1\rangle}{\left(\langle 1,h|h,1\rangle\right)_{\theta=0}}\right)_{m_{0}^{2}=-1}=\dfrac{1+\theta}{1-3\theta},$ (169) and we see that this reduced one-point correlator has the real pole for $\theta=\dfrac{1}{3}$. The Figure (1) presents graphics of this one-point correlator as a function of $\theta$, and the one-point correlator (167) as function of the argument $h-h_{I}$ for some values of $\theta\in(-1,1)$. Figure 1: Graphics of (a) the normalized relative correlator (169) (b) the correlator (167) (vertical axis) as a function of the argument $h-h_{I}$ (horizontal axis) for some values of $\theta\in(-1,1)$. Because we have to deal with one-point correlation function of the quantum bosonic field $\mathbf{\Psi}$ determined on the configurational space of General Relativity that is the superspace, where the point means some concrete compact 3-geometry, the poles of the one-point correlator (169) have an interpretation of free stable states of the considered quantum field theory. By this the point $\theta=\dfrac{1}{3}$ localizes the quantum stable state, and in this manner the quantum stable state can be identified with the quanta of gravity, _i.e._ with the graviton. Generally the graviton is detected in the superspace points $h$ that fulfill the relation $h-h_{I}=\dfrac{2\theta}{3\theta^{2}+4\theta-1},$ (170) and are presented on the Figure (1) in the part (b) by points where the correlator has singularity. In the tachyon limit $\theta=0$ the pole value of $h-h_{I}$ (170) becomes $(h-h_{I})_{\theta=0}=0,$ (171) and by this in the point $h=h_{I}$ is localized graviton in the tachyon limit. It is interesting that in the point $\theta=\theta_{\infty}=\dfrac{\sqrt{7}-2}{3}\approx 0.2152504370,$ (172) stable quantum state can not be detected, as it is presented on the Figure (2). Figure 2: Dependence of the argument $h-h_{I}$ from the parameter $\theta$ for that the graviton is localized. ## 5 Classical bosonic embedded space. Tachyon. Let us consider now some aspects of the classical Bose field $\Psi[h]$ introduced in the previous section. We are not going to resolve the equation (58) again, but analyze some structural elements of the relativistic quantum mechanics described by this equation, especially the square of mass (59) by its direct connection with 3-geometry of spacetime. ### 5.1 Field mass by its energy The relativistic wave equation obtained in the previous section $\left\\{\dfrac{\delta^{2}}{\delta h^{2}}+m^{2}[h]\right\\}\Psi[h]=0,$ (173) from classical point of view describes some one-dimensional classical particle with mass dependent on the point $h_{ij}$ in superspace characterized by its determinant $h$. In the case of the mass independent on the superspace point, this equation has a very well known solution in terms of plane waves, but in the general case, _i.e._ for nonconstant mass, this equation is not very simple for direct solving. Plane waves are not a solution in this case. However, we are not going to concentrate our considerations on search for general classical solutions of the equation (173), but in the next sections we will try to construct a second quantization of this equation, that is independent on classical field theory solution. In this section we will discuss classical field theory that gives the equation (173) as the classical Euler–Lagrange equation of motion. Obviously, because of we have to deal with the Bose field, in this section we will consider some theory of the bosonic string. Firstly, let us consider the equation (173) as the classical Euler–Lagrange equation of motion obtained by some field theory lagrangian $L\left[\Psi[h],\dfrac{\delta\Psi[h]}{\delta h}\right]$ according to the system of equations $\displaystyle\dfrac{\delta\Pi_{\Psi}[h]}{\delta h}-\dfrac{\partial}{\partial\Psi[h]}L\left[\Psi[h],\dfrac{\delta\Psi[h]}{\delta h}\right]$ $\displaystyle=$ $\displaystyle 0,$ (174) $\displaystyle\Pi_{\Psi}[h]-\dfrac{\partial}{\partial\left(\dfrac{\delta\Psi[h]}{\delta h}\right)}L\left[\Psi[h],\dfrac{\delta\Psi[h]}{\delta h}\right]$ $\displaystyle=$ $\displaystyle 0,$ (175) where $\Pi_{\Psi}[h]$ is canonical momentum conjugate to the classical field $\Psi[h]$. Standardly, one can construct the Lagrangian by field theory action functional $S[\Psi]$ directly by using of the equation of motion (173) in the following way $\displaystyle S[\Psi]$ $\displaystyle=$ $\displaystyle-\dfrac{1}{2}\int\delta{h}\Psi[h]\left\\{\dfrac{\delta^{2}}{\delta{h}^{2}}+m^{2}[h]\right\\}\Psi[h]=$ (176) $\displaystyle=$ $\displaystyle-\dfrac{1}{2}\int\delta{h}\left\\{\dfrac{\delta}{\delta{h}}\left(\Psi[h]\dfrac{\delta\Psi[h]}{\delta{h}}\right)-\left(\dfrac{\delta\Psi[h]}{\delta{h}}\right)^{2}+m^{2}[h]\Psi^{2}[h]\right\\}=$ $\displaystyle=$ $\displaystyle\dfrac{1}{2}\int\delta{h}\left\\{\left(\dfrac{\delta\Psi[h]}{\delta{h}}\right)^{2}-m^{2}[h]\Psi^{2}[h]\right\\}\equiv\int\delta hL\left[\Psi[h],\dfrac{\delta\Psi[h]}{\delta h}\right]\\!\\!,$ where we have applied integration of full divergence, and $\Psi^{\dagger}[h]=\Psi[h]$. This means we have to deal with the classical field theory given by the Euler–Lagrange system of equations in the form $\displaystyle\dfrac{\delta\Pi_{\Psi}[h]}{\delta h}+m^{2}[h]\Psi[h]$ $\displaystyle=$ $\displaystyle 0,$ (177) $\displaystyle\Pi_{\Psi}[h]-\dfrac{\delta\Psi[h]}{\delta h}$ $\displaystyle=$ $\displaystyle 0.$ (178) The classical field theory Hamiltonian can be constructed immediately from the Lagrangian (176) by application of the standard Legendre transformation [20] between Hamiltonian and Lagrangian of the classical dynamical system $H[\Pi_{\Psi},\Psi]=\Pi_{\Psi}[h]\dfrac{\delta\Psi[h]}{\delta h}-L\left[\Psi[h],\dfrac{\delta\Psi[h]}{\delta h}\right]=\dfrac{\Pi_{\Psi}^{2}[h]+m^{2}[h]\Psi^{2}[h]}{2},$ (179) that for fixed mass $m[h]$ describes ellipse in space $(H[\Pi_{\Psi},\Psi],\Pi_{\Psi},\Pi)$. The set of these all ellipses lies on paraboloid parameterized by continue parameter $h$. Direct using of the momentum constraint (178) and the fact that really the Hamiltonian (179) can be treated as functional with respect to the evolution parameter $h$ of the equation (173), _i.e._ $H[\Pi_{\Psi}[h],\Psi[h]]=H[h]$, allows to rewrite the classical field theory Hamiltonian (179) as the definition of mass $m^{2}[h]\Psi^{2}[h]=2H[h]-\left(\dfrac{\delta\Psi[h]}{\delta h}\right)^{2},$ (180) and after simple elimination of the square of mass by using of the equation of motion (173) this leads to the functional differential equation for the classical field $\Psi[h]$ $\dfrac{\delta^{2}\Psi[h]}{\delta h^{2}}\Psi[h]=2H[h]-\left(\dfrac{\delta\Psi[h]}{\delta h}\right)^{2},$ (181) which after collecting terms leads to the following relation between the field $\Psi[h]$ and the Hamiltonian $H[h]$ $\dfrac{\delta}{\delta h}\left(\dfrac{\delta\Psi[h]}{\delta h}\Psi[h]\right)=2H[h].$ (182) So, presently one can be integrate the last equation directly with initial value of $h$ taken as $h_{I}$. In result we obtain $\Psi^{2}[h]=4\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}],$ (183) and by this the solution of the classical wave equation (173) can be formally accepted as the functional $\Psi[h]=\left(\Psi^{2}[h]\right)^{1/2}$. From the other side the equation (181) can be integrated into the form $\Pi_{\Psi}[h]\Psi[h]=2\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}],$ (184) which combined together with the solution (183) fixes values of the canonical conjugate momentum $\Pi_{\Psi}[h]$ with respect to values of the classical Hamiltonian $H[h]$ in the following way $\Pi_{\Psi}[h]=\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\left(\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]\right)^{1/2}}.$ (185) Taking into account the basic relation for the classical field theory Hamiltonian (179) one can obtain by direct algebraic manipulation $m^{2}[h]=\dfrac{2H[h]}{\Psi^{2}[h]}-\left(\dfrac{\Pi_{\Psi}[h]}{\Psi[h]}\right)^{2},$ (186) or by employing the relation (184) $m^{2}[h]=\dfrac{2H[h]}{\Psi^{2}[h]}-\left(\dfrac{2\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\Psi^{2}[h]}\right)^{2}.$ (187) By treating this relation as a constraint that fixes mass value and by application of the solution (183) one can determine easily the dependence between field mass and its energy values $m^{2}[h]=\dfrac{1}{4}\left[\dfrac{2H[h]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}-\left(\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right)^{2}\right].$ (188) In this manner, the classical field theory of the Bose field $\Psi[h]$ can be studied in terms of values of its square mass in dependence on values of the classical field theory Hamiltonian. ### 5.2 Perturbations, cosmological constant, Dark Energy The last formula (188) determines fundamental relation between the square of mass of the considered boson and the classical field energy distribution. By using of the definition (59) one can consider this relation in terms of the constraint $\dfrac{8}{3h}\left(R[h]-2\Lambda-6\varrho\right)=\dfrac{2H[h]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}-\left(\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right)^{2},$ (189) that fixes values of the normal to the boundary space component of the stress–energy tensor $\varrho$ on $\displaystyle\varrho\\!\\!$ $\displaystyle=$ $\displaystyle\\!\\!\dfrac{h}{16}\left[\left(\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right)^{2}-\dfrac{2H[h]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right]+$ (190) $\displaystyle+$ $\displaystyle\\!\\!\dfrac{R[h]}{6}-\dfrac{\Lambda}{3},$ and by positive definiteness of the classical energy density can be used to determine an upper limit for the cosmological constant $\Lambda\leq\dfrac{R[h]}{2}+\dfrac{3h}{8}\left[\left(\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right)^{2}-\dfrac{2H[h]}{\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}H[h^{\prime\prime}]}\right]$ (191) Let us define the mass groundstate of the classical field theory presented above by the following condition $H[h^{\prime}]=C\delta(h^{\prime}-h),\leavevmode\nobreak\ \leavevmode\nobreak\ H[h]=0,$ (192) so that the constant $C\rightarrow 0$ formally, here $\delta(h^{\prime}-h)$ is the Dirac delta function. For so defined groundstate the first term in the square of mass (188) vanishes automatically, and the second term gives finite contribution to the square of mass $m^{2}_{0}[h]\equiv m^{2}[h]\Big{|}_{\mathrm{groundstate}}=-\left(\dfrac{C}{C(h-h_{I})}\right)^{2}\Bigg{|}_{C\rightarrow 0}=-\dfrac{1}{4\left(h-h_{I}\right)^{2}}.$ (193) This number is negative for all values of $h$ and by this relation describes formally the tachyon, that is the fundamental excitation of the bosonic string [21]. For the considered groundstate are fulfilled the following relations $\displaystyle\varrho^{(0)}$ $\displaystyle=$ $\displaystyle\dfrac{1}{16}\dfrac{h}{\left(h-h_{I}\right)^{2}}+\dfrac{R[h]}{6}-\dfrac{\Lambda}{3},$ (194) $\displaystyle\Lambda$ $\displaystyle\leq$ $\displaystyle\dfrac{R[h]}{2}+\dfrac{1}{16}\dfrac{h}{\left(h-h_{I}\right)^{2}}.$ (195) If we demand additionally that for the initial metric $h_{I}$ the classical boson field $\Psi[h]$ should have a some finite mass $m_{I}$ then the formula (199) should be renormalized as follows $m^{2}_{0}[h]=-\dfrac{1}{\left(h-h_{I}-i\sqrt{\dfrac{1}{m_{I}^{2}}}\right)^{2}},$ (196) so the initial square of mass should be huge for correctness $\eta=\sqrt{\dfrac{1}{m_{I}^{2}}}\rightarrow 0$. For the field $\Psi[h]$, we conclude from the basic relation (183) that in so defined mass groundstate the classical field $\Psi[h]$ is $\Psi[h]=2\sqrt{C(h-h_{I})},$ (197) and in this case the phase space $(\Pi_{\Psi},\Psi)$ determined by the relation (184) is given by a family of hyperbolas $\Pi_{\Psi}[h]=\dfrac{2C}{\Psi[h]}\equiv\sqrt{\dfrac{C}{h-h_{I}}},$ (198) or simply by the condition that the product of phase space variables is the first integral of the considered classical field theory $\Pi_{\Psi}[h]\Psi[h]=constans$. For all constant, but nonzero values of the classical field theory Hamiltonian $H[h]=H_{0}\neq 0$, the square of mass vanishes identically $m^{2}[h]\Big{|}_{H[h]=H_{0}}=0,$ (199) and these states are massless excitations of the bosonic string, by fact that here $H_{0}$ is arbitrary constant, number of massless states is continuum. For the massless states we have simplified relations for normal stress–energy tensor and cosmological constant $\varrho=\dfrac{R[h]}{6}-\dfrac{\Lambda}{3},\leavevmode\nobreak\ \leavevmode\nobreak\ \Lambda\leq\dfrac{R[h]}{2}.$ (200) However, presence of massless states in the theory means that $\mu^{2}=0$, what is unphysical mass scale value by $\mu\geq 1$. From the string theory point of view the tachyon state is treated as mass groundstate of the considered theory of bosonic string. One can generate the process of symmetry breaking in frames of the perturbational calculus with respect to the classical field theory Hamiltonian $H[h]$. Namely, in the most general case, one can imagine that an arbitrary mass state of the considered bosonic string, and as the context suggests arbitrary metrics of General Relativity, is generated by small deviation from the tachyon state. Let the deviation is an arbitrary functional so that $\delta H[h]\ll 1$, then deviation from the groundstate of the classical Hamiltonian given by $H[h^{\prime}]=C\delta(h^{\prime}-h)+\delta H[h^{\prime}],\leavevmode\nobreak\ \leavevmode\nobreak\ H[h]=\delta H[h],$ (201) leads to perturbations from the mass groundstate in the form $m^{2}[h]=m^{2}_{0}[h]+\delta m^{2}[h],$ (202) where the term $\delta m^{2}[h]$ describes the full contribution to the square of mass from the perturbation and breaks mass groundstate directly. Let us assume that the term has a form of the series $\delta m^{2}[h]=\sum_{n=1}^{\infty}\delta^{(n)}m^{2}[h]=\delta^{(1)}m^{2}[h]+\delta^{(2)}m^{2}[h]+\delta^{(3)}m^{2}[h]+\ldots,$ (203) where the partial terms $\delta^{(k)}m^{2}[h]$ consist all corrections taken up to the $k$-th order in the perturbation $\delta H[h]$ of the classical Hamiltonian. By introduce of shorten notation $\displaystyle\alpha_{0}\equiv\alpha_{0}[h]$ $\displaystyle=$ $\displaystyle\dfrac{\delta H[h]}{C},$ (204) $\displaystyle\alpha_{1}\equiv\alpha_{1}[h]$ $\displaystyle=$ $\displaystyle\int_{h_{I}}^{h}\delta h^{\prime}\alpha_{0}[h^{\prime}],$ (205) $\displaystyle\alpha_{2}\equiv\alpha_{2}[h]$ $\displaystyle=$ $\displaystyle\int_{h_{I}}^{h}\delta h^{\prime}\alpha_{1}[h^{\prime}]=\int_{h_{I}}^{h}\delta h^{\prime}\int_{h_{I}}^{h^{\prime}}\delta h^{\prime\prime}\alpha_{0}[h^{\prime\prime}],$ (206) one can check easily by elementary computation that the $k$-th contribution to the series (203) has a following form $\displaystyle\delta^{(k)}m^{2}[h]$ $\displaystyle=$ $\displaystyle(-1)^{k+1}\dfrac{k+1}{4(h-h_{I})^{k+2}}\alpha_{2}^{k}+(-1)^{k}\dfrac{2k}{4(h-h_{I})^{k+1}}\alpha_{1}\alpha_{2}^{k-1}+$ (207) $\displaystyle+$ $\displaystyle(-1)^{k-1}\dfrac{2\alpha_{0}\alpha_{2}^{k-1}+(k-1)\alpha_{1}^{2}\alpha_{2}^{k-2}}{4(h-h_{I})^{k}},$ and by this the series (203) can be summed immediately, so that the full result for the square of mass (202) can be determined by dependence from the parameters $\alpha^{\prime}s$ $\displaystyle m^{2}[h]$ $\displaystyle=$ $\displaystyle-\dfrac{1}{4(h-h_{I})^{2}}+\left(\dfrac{\dfrac{1}{2}\dfrac{\alpha_{2}}{h-h_{I}}}{1+\dfrac{\alpha_{2}}{h-h_{I}}}\right)^{2}\left[\dfrac{1}{(h-h_{I})^{2}}-\dfrac{\alpha_{1}^{2}+2\alpha_{1}}{\alpha_{2}}\right]+$ (208) $\displaystyle+$ $\displaystyle\dfrac{\dfrac{1}{2}\dfrac{\alpha_{2}}{h-h_{I}}}{\left(1+\dfrac{\alpha_{2}}{h-h_{I}}\right)^{2}}\left[\dfrac{1}{(h-h_{I})^{2}}+\dfrac{\alpha_{0}}{\alpha_{2}}+\dfrac{\alpha_{0}}{h-h_{I}}\right],$ and the parameters can be treated as free parameters of the theory. ### 5.3 Tachyon From the relation (208) we see explicitly that if $\alpha$’s are constrained by the following system of equations $\left\\{\begin{array}[]{cc}\dfrac{1}{(h-h_{I})^{2}}-\dfrac{\alpha_{1}^{2}+2\alpha_{1}}{\alpha_{2}}&=0\\\ \dfrac{1}{(h-h_{I})^{2}}+\dfrac{\alpha_{0}}{\alpha_{2}}+\dfrac{\alpha_{0}}{h-h_{I}}&=0\end{array}\right.$ (209) then we have to deal with the tachyon state – in this case the square of mass is negative and equal to the first term of this formula. The system of equations (209) can be solved directly, in result we obtain the relations between $\alpha$’s $\displaystyle\alpha_{1}$ $\displaystyle=$ $\displaystyle\pm\sqrt{1-\dfrac{\alpha_{0}}{1+\alpha_{0}(h-h_{I})}}-1,$ (210) $\displaystyle\alpha_{2}$ $\displaystyle=$ $\displaystyle-\dfrac{\alpha_{0}(h-h_{I})^{2}}{1+\alpha_{0}(h-h_{I})}\leavevmode\nobreak\ \leavevmode\nobreak\ \mathrm{for}\leavevmode\nobreak\ \leavevmode\nobreak\ \mathrm{both}\leavevmode\nobreak\ \leavevmode\nobreak\ \alpha_{1}.$ (211) By this the tachyon, which is the mass groundstate of the considered bosonic theory, can be completely determined by arbitrary value of $\alpha_{0}$ and connected with this value the functions $\alpha_{1}$ and $\alpha_{2}$ determined by relations (210) and (211). One can see easily that this system of equations leads to the surface $T$ in space of parameters $(\alpha_{0},\alpha_{1},\alpha_{2})$ given by the set of points $T=\left\\{(\alpha_{0},\alpha_{1},\alpha_{2})\in{\mathbb{R}^{3}}:\alpha_{2}(\alpha_{0},\alpha_{1})=\dfrac{\left(\alpha_{0}+2\alpha_{1}+\alpha_{1}^{2}\right)^{2}}{\alpha_{0}^{2}\left(2\alpha_{1}+\alpha_{1}^{2}\right)}\right\\},$ (212) that describes tachyon state in this space, see Figure 3. Figure 3: Tachyon state in space of parameters $(\alpha_{0},\alpha_{1},\alpha_{2})$: the part (a) presents large scale view of the surface (212); the part (b) presents the surface in neighborhood of the point $(0,0,0)$. Consider the case of the constant perturbation $\epsilon$ that is very small in comparison with $C$ $\dfrac{\int_{h_{I}}^{h}\delta h^{\prime}\delta H[h^{\prime}]}{\int_{h_{I}}^{h}\delta h^{\prime}H[h^{\prime}]}=\dfrac{\epsilon}{C}=\varepsilon\ll 1.$ (213) Then by direct combination of the relations (204), (205), and (206) we obtain $\displaystyle\alpha_{0}$ $\displaystyle=$ $\displaystyle\varepsilon,$ (214) $\displaystyle\alpha_{1}$ $\displaystyle=$ $\displaystyle(h-h_{I})\varepsilon,$ (215) $\displaystyle\alpha_{2}$ $\displaystyle=$ $\displaystyle\dfrac{(h-h_{I})^{2}}{2}\varepsilon=\dfrac{\alpha_{1}^{2}}{2\alpha_{0}},$ (216) and in this case $\displaystyle\delta^{(k)}m^{2}[h]=(-1)^{k-1}\dfrac{(h-h_{I})^{k-2}}{2^{k}}\left[2k+4-\dfrac{2k}{(h-h_{I})^{2}}\right]\varepsilon^{k},$ (217) so, the sum (202) can be calculated directly $m^{2}[h]=-\dfrac{1}{(h-h_{I})^{2}}+\dfrac{4\varepsilon}{(h-h_{I})^{3}}\dfrac{\varepsilon(h-h_{I})^{3}+3(h-h_{I})^{2}-1}{\varepsilon^{2}(h-h_{I})^{2}+4\varepsilon(h-h_{I})+4},$ (218) where the small constant $\varepsilon$ is chosen according to the condition $|\varepsilon|<\dfrac{2}{|h-h_{I}|}.$ (219) It is clear now that tachyon state is obtained by the limit $\varepsilon\rightarrow 0$. Equivalently one can treat the square of mass (202) in terms of power series in the function $\dfrac{1}{h-h_{I}}$ $m^{2}[h]=\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}},$ (220) where the coefficients $a_{n}$ as functions of (204), (205), and (206) are functionals of $h$ and initial data described by $h_{I}$, and they can be directly written in the compact form $\displaystyle a_{n}[h;h_{I}]=(-1)^{n}\left\\{\left[2\alpha_{0}\alpha_{2}-\left(1+\alpha_{1}\right)^{2}\right]G[n]-2n\alpha_{0}\alpha_{2}\right\\}\alpha_{2}^{n-2},$ (221) where $G[n]$ is the step function defined as $G[n]=\left\\{\begin{array}[]{cc}0,&\mathrm{for}\leavevmode\nobreak\ n<1\\\ n-1,&\mathrm{for}\leavevmode\nobreak\ n\geq 1\end{array}\right.$ (222) By using of the main relation for the square of mass (59) $m^{2}[h]=\dfrac{2}{3h}\left(R[h]-2\Lambda-6\varrho\right),$ one can obtain by direct comparison with the power series (220) the relation $\dfrac{2}{3h}\left(R[h]-2\Lambda-6\varrho\right)=\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}},$ (223) that can be treated as a definition of the stress–energy tensor $\varrho$ projected onto normal vector field to boundary 3-dimensional surface as $\varrho[h]=\dfrac{R[h]}{6}-\dfrac{\Lambda}{3}-\dfrac{h}{4}\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}}.$ (224) This energy density is positive iff $\Lambda\leq\dfrac{R[h]}{2}-\dfrac{3h}{4}\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}},$ (225) and this actually defines an upper limit for the cosmological constant $\Lambda$. One can view on the relation (223) by different point of view. Namely, when we rewrite this formula in the following form $\dfrac{2}{3h}\left(R[h]-2\Lambda-6\varrho\right)-\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}}=0,$ (226) then we see that this suggests redefinition of the energy density by the way $T^{DM}=\varrho+\rho_{DM}[h],$ (227) where $\displaystyle\rho_{DM}[h]$ $\displaystyle=$ $\displaystyle\dfrac{h}{4}\sum_{n=0}^{\infty}\dfrac{a_{n}[h;h_{I}]}{\left(h-h_{I}\right)^{n}},$ (228) can be interpreted as a density energy from Dark Matter fields. Equivalently one can determine the Dark Matter density energy (228) by application the relation (208) as follows $\displaystyle\rho_{DM}[h]$ $\displaystyle=$ $\displaystyle\dfrac{h}{16}\Biggr{\\{}-\dfrac{1}{(h-h_{I})^{2}}+\left(\dfrac{\dfrac{\alpha_{2}}{h-h_{I}}}{1+\dfrac{\alpha_{2}}{h-h_{I}}}\right)^{2}\left[\dfrac{1}{(h-h_{I})^{2}}-\dfrac{\alpha_{1}^{2}+2\alpha_{1}}{\alpha_{2}}\right]+$ (229) $\displaystyle+$ $\displaystyle\dfrac{2\dfrac{\alpha_{2}}{h-h_{I}}}{\left(1+\dfrac{\alpha_{2}}{h-h_{I}}\right)^{2}}\left[\dfrac{1}{(h-h_{I})^{2}}+\dfrac{\alpha_{0}}{\alpha_{2}}+\dfrac{\alpha_{0}}{h-h_{I}}\right]\Biggr{\\}},$ that for constant energies becomes $\displaystyle\rho_{DM}[h]=\dfrac{h}{4}\dfrac{\varepsilon}{(h-h_{I})^{3}}\dfrac{\varepsilon(h-h_{I})^{3}+3(h-h_{I})^{2}-1}{\varepsilon^{2}(h-h_{I})^{2}+4\varepsilon(h-h_{I})+4}.$ (230) By direct resolving of the equation (226) with respect to the cosmological constant $\Lambda$ one can determine the cosmological constant as the quantity dependent only on scalar curvature of boundary 3-geometry, and summarized energy density of normal Matter fields and Dark Matter $\Lambda=\dfrac{R[h]-3T^{DM}}{2}.$ (231) In this manner, the Einstein–Hilbert action that is the second integral of (1) takes the form $S_{EH}[g]\longrightarrow S_{M+DM}[g]=\int_{M}d^{4}x\sqrt{-g}\left\\{-\dfrac{1}{6}R[g]+\mathcal{L}_{M+DM}\right\\},$ (232) where $\mathcal{L}_{M+DM}$ is the total lagrangian of Matter fields and Dark Matter $\mathcal{L}_{M+DM}=\mathcal{L}+\dfrac{R[h]}{6}-\dfrac{T^{DM}}{2},$ (233) with $\mathcal{L}$ as the Lagrangian of Matter fields. Moreover, by constant value of $\Lambda$, with Dark Matter contribution, the General Relativity field equations (LABEL:gr1) presently are $R_{\mu\nu}=\left[\dfrac{1}{2}R[h]-\dfrac{3}{2}\left(\varrho+\rho_{DM}[h]\right)\right]g_{\mu\nu}+3\left(T_{\mu\nu}-\dfrac{T}{2}g_{\mu\nu}\right),$ (234) where $T_{\mu\nu}$ is the stress-energy tensor of Matter fields. One can consider the case when we have to deal with vanishing cosmological constant $\Lambda\equiv 0$. In this case, from the relations (231) and (227) we directly obtain that stress-energy tensor projected onto normal field vector has a value $\lim_{\Lambda\rightarrow 0}\varrho[h]=\dfrac{R[h]}{3}-\rho_{DM}[h],$ (235) that for small energies becomes $\lim_{\Lambda\rightarrow 0}\varrho[h]=\dfrac{R[h]}{3}-\dfrac{h}{4(h-h_{I})^{3}}\dfrac{\varepsilon^{2}(h-h_{I})^{3}+\varepsilon[3(h-h_{I})^{2}-1]}{\varepsilon^{2}(h-h_{I})^{2}+4\varepsilon(h-h_{I})+4},$ (236) and in the tachyon limit takes the value $\lim_{\varepsilon\rightarrow 0}\lim_{\Lambda\rightarrow 0}\varrho[h]=\dfrac{R[h]}{3}.$ (237) By this the Einstein–Hilbert field equations of General Relativity with Dark Matter existence (234) in the case of vanishing cosmological constant within the tachyon limit are simply $R_{\mu\nu}=3\left(T_{\mu\nu}-\dfrac{T}{2}g_{\mu\nu}\right),$ (238) but the tachyon state in the neighborhood of zero in the space of parameters $(\alpha_{0},\alpha_{1},\alpha_{2})$ with nonzero cosmological constant is described by completely other field equations $R_{\mu\nu}=\left(\dfrac{1}{2}R[h]-\dfrac{3}{2}\varrho\right)g_{\mu\nu}+3\left(T_{\mu\nu}-\dfrac{T}{2}g_{\mu\nu}\right).$ (239) Perturbation calculus ideas presented above, completely describe the classical field theory (173) in terms of the spontaneously breaking of mass groundstate of the bosonic string with respect to the field theory Hamiltonian (179). ## 6 Quantum gravity by thermodynamics The last section was devoted to presentation of the quantum theory of the Bose field $\mathbf{\Psi}[h]$ based on the quantization in the Fock space of creation and annihilation functional operators, and proper choice of the initial data basis. This approach led us to notion of space quantum states associated with three-dimensional spatial part of a Riemannian spacetime classically treated as a solution of the Einstein–Hilbert field equations of General Relativity. Furthermore, as the main result of our studies of the one- point two-field correlator real poles of the bosonic field $\mathbf{\Psi}[h]$ we have obtained a localization of the stable quantum states of the quantum field theory that can be interpreted as the quantum particle of generalized gravitational fields – the graviton. In this section we will investigate thermodynamical description of the considered bosonic statistical system. We will use the density functional method in order to formulation of equilibrium statistical thermodynamics of many space quantum states. Actually, it is the last step of the Thermodynamical Einstein’s Dream, that is the main motivation to this paper. Let us try to create thermodynamical picture that arises from the presented quantum field theory. When we build thermodynamics, we should use the simplest rules of statistical physics, that in some sense give the general information about the considered physical system. In usual thermal situations in physics, we have to deal with some concrete set of possible physical states, and we try to construct statistical description of the system by using of ensemble that given the prescription for averaging procedure. Generally in real physical systems we have to deal with the only one classical statistics, _i.e._ the Boltzmann distribution, and in case of quantum states of the Bose systems with the Bose–Einstein statistics, and in case with Fermi particles with Fermi–Dirac statistics. Furthermore, the real systems are no isolated and open, so interaction with environment is inevitable. Let us consider the situation of the concrete system as is the system of space quantum states of a spacetime. By quantum character of the set we have to deal with quantum statistics, in the considered case the quantum mechanics, that is classical field theory, is described by the Wheeler–DeWitt equation in form of the Klein–Gordon–Fock evolution equation (173). Naturally, this is the Bose system, and we should describe statistical properties of the system in frames of the Bose–Einstein statistics. Moreover, by its Nature the system is open and no isolated, but we have proposed the diagonalization procedure and this framework generates the fundamental static operator basis in the Fock space associated with initial data of the system. Actually, this initial data basis also defines the thermal equilibrium state, and generalized thermodynamics of the set of space quantum states can be investigated from this point of view. Let us consider the thermodynamics of space quantum states as quantum theory of general gravitational fields. The initial data basis (111) gives an opportunity to introduce a notion of the thermodynamical equilibrium state in the statistical ensemble of many space quantum states that are some generalized quantum particles of the classical Einstein–Hilbert Riemannian manifold of General Relativity. Essentially, the fundamental static operator basis $\mathfrak{B}_{I}$ is given by creation and annihilation operators in the Fock space of the quantum field theory. It means that initial data are directly jointed with static description of the ensemble, and from the point of view of the fundamental basis the set of space quantum states is isolated and no open system, and can be characterized by usual thermodynamical description. By this from conceptual side of thermodynamics as the only theory between quantum field theory and statistical mechanics, and from as logical well as ontological points should be possible to obtain the statistical characterizations of the space quantum states system. In the context of this paper the following supposition seems to be the most natural Thermodynamics of space quantum states is quantum gravity. --- Let us study this generalized thermodynamics and its physical aspects. ### 6.1 Density matrix We will investigate here thermodynamical description treated as one-particle approximation of density operator. In real physical systems, as for example for photon gas or the system of free electrons, this is sufficient approximation to obtain satisfactory accordance with experimental data. The one-particle density operator is standardly given by occupation number operator of quantum states. For the considered case the quantum states are described by the dynamical operator basis (98), and by this in demanded approximation the density operator has a form $\mathsf{D}[h]={\mathsf{G}}^{\dagger}[h]{\mathsf{G}}[h].$ (240) This dynamical density operator has the following matrix representation in the dynamical basis $\mathsf{D}[h]=\mathfrak{B}^{\dagger}[h]\left[\begin{array}[]{cc}1&0\\\ 0&0\end{array}\right]\mathfrak{B}^{\dagger}[h]=\mathfrak{B}^{\dagger}[h]\mathbb{D}\mathfrak{B}^{\dagger}[h],$ (241) and by direct application of the Bogoliubov transformation can be immediately expressed in the static initial data basis as follows $\mathsf{D}[h]=\mathfrak{B}_{I}^{\dagger}\left[\begin{array}[]{cc}|u[h]|^{2}&-u[h]v[h]\\\ -u^{\ast}[h]v^{\ast}[h]&|v[h]|^{2}\end{array}\right]\mathfrak{B}_{I}\equiv{\mathfrak{B}}_{I}^{\dagger}\mathbb{D}[h]{\mathfrak{B}}_{I},$ (242) where $\mathbb{D}[h]$ has an interpretation of the matrix representation of the density operator (240) in the initial data operator basis, and actually describes the system of space quantum states in thermodynamical equilibrium with respect to the fundamental basis. The explicit form of the functional matrix $\mathbb{D}[h]$ is $\mathbb{D}[h]=\left[\begin{array}[]{cc}\dfrac{1}{4}\left(\sqrt[4]{\left|\dfrac{m^{2}}{m_{I}^{2}}\right|}+\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}}\right|}\right)^{2}&\dfrac{e^{2i\theta}}{4}\left(\sqrt{\left|\dfrac{m_{I}^{2}}{m^{2}}\right|}-\sqrt{\left|\dfrac{m^{2}}{m_{I}^{2}}\right|}\right)\vspace*{10pt}\\\ \dfrac{e^{-2i\theta}}{4}\left(\sqrt{\left|\dfrac{m_{I}^{2}}{m^{2}}\right|}-\sqrt{\left|\dfrac{m^{2}}{m_{I}^{2}}\right|}\right)&\dfrac{1}{4}\left(\sqrt[4]{\left|\dfrac{m^{2}}{m_{I}^{2}}\right|}-\sqrt[4]{\left|\dfrac{m_{I}^{2}}{m^{2}}\right|}\right)^{2}\end{array}\right],$ (243) and has the natural properties $\mathbb{D}^{\dagger}[h]=\mathbb{D}[h],\leavevmode\nobreak\ \leavevmode\nobreak\ \det\mathbb{D}[h]=0,$ (244) where for compact notation $m=m[h]$, and $\theta=\theta[h]$. This type of reasoning is some kind of the Heisenberg picture. It is natural to assume that the set of space quantum states is described in the Grand Canonical Ensemble [23]. The grand partition function is standardly defined as $\Omega(z,V,T)=\mathrm{Tr}z\exp\left(-\dfrac{U}{T}\right)=\mathrm{Tr}\exp\left\\{-\dfrac{U-\mu N}{T}\right\\},$ (245) where $z=\exp\dfrac{\mu N}{T}$ is called activity, $U$ is internal energy, $\mu$ is chemical potential, $N$ is averaged occupation number, $V$ is volume, and $T$ is temperature of the system. The ensemble average of quantity $A$ in the Grand Canonical Ensemble is $\langle A\rangle=\dfrac{\mathrm{Tr}\left(A\exp\left\\{-\dfrac{U-\mu N}{T}\right\\}\right)}{\mathrm{Tr}\exp\left\\{-\dfrac{U-\mu N}{T}\right\\}}.$ (246) Thermodynamical equation of state for the Bose system can be calculated as $\dfrac{PV}{T}=\ln\Omega(z,V,T),$ (247) where $P$ is pressure. The famous grand partition function for the Bose statistics in the case associated with our problem is $\Omega(z,V,T)=\dfrac{1}{1-z\exp\left(-\dfrac{U}{T}\right)},$ (248) and by this the equation of state (247) becomes $\dfrac{PV}{T}=-\ln\left(1-z\exp\left(-\dfrac{U}{T}\right)\right)=\ln\dfrac{z^{-1}\exp\dfrac{U}{T}}{z^{-1}\exp\dfrac{U}{T}-1}.$ (249) Moreover, the averaged occupation number can be determined $N=z\dfrac{\partial}{\partial z}\ln\Omega(z,V,T)=\dfrac{z\exp\left(-\dfrac{U}{T}\right)}{1-z\exp\left(-\dfrac{U}{T}\right)}=\dfrac{1}{z^{-1}\exp\dfrac{U}{T}-1}.$ (250) Entropy of the Bose gas is then determined by the following relation $S=\left(\dfrac{U}{T}-\ln z\right)\dfrac{z^{-1}\exp\dfrac{U}{T}}{z^{-1}\exp\dfrac{U}{T}-1}-\ln\left(z^{-1}\exp\dfrac{U}{T}-1\right).$ (251) ### 6.2 The Bogoliubov coefficients Let us consider the space quantum states system in Grand Canonical Ensemble. The basic quantity of statistical mechanics is an entropy, that for an arbitrary quantum system described is defined by the standard Gibbs–Von Neumann formula $S[h]=-\dfrac{\mathrm{Tr}\left(\mathbb{D}[h]\ln\mathbb{D}[h]\right)}{\mathrm{Tr}\mathbb{D}[h]},$ (252) and in considered case can be immediately computed from the density matrix (243). Using of linear algebra methods, especially the Cayley–Hamilton characteristic polynomial and its properties, one can compute directly the logarithm of the density matrix as $\ln\mathbb{D}=\left[\begin{array}[]{cc}-\dfrac{3}{2}\dfrac{|v|^{2}}{|u|^{2}+|v|^{2}}+\ln\left(|u|^{2}+|v|^{2}\right)&\dfrac{5}{2}\dfrac{uv}{|u|^{2}+|v|^{2}}\\\ \dfrac{5}{2}\dfrac{u^{\ast}v^{\ast}}{|u|^{2}+|v|^{2}}&-\dfrac{3}{2}\dfrac{|u|^{2}}{|u|^{2}+|v|^{2}}+\ln\left(|u|^{2}+|v|^{2}\right)\end{array}\right],$ (253) where $u=u[h]$ and $v=v[h]$ are the Bogoliubov coefficients given by (113). Taking the proper traces according to the definition (252) one can directly obtain the compact relation for entropy $S[h]=\dfrac{8|u[h]|^{2}|v[h]|^{2}}{(|u[h]|^{2}+|v[h]|^{2})^{2}}-\ln\left(|u[h]|^{2}+|v[h]|^{2}\right),$ (254) that can be immediately compared with the entropy of the Bose gas (251), and in result leads to the following identification $\displaystyle|u[h]|^{2}+|v[h]|^{2}$ $\displaystyle=$ $\displaystyle z^{-1}[h]\exp\dfrac{U[h]}{T[h]}-1,$ (255) $\displaystyle\dfrac{8|u[h]|^{2}|v[h]|^{2}}{(|u[h]|^{2}+|v[h]|^{2})^{2}}$ $\displaystyle=$ $\displaystyle\left(\dfrac{U[h]}{T[h]}-\ln z[h]\right)\dfrac{z^{-1}[h]\exp\dfrac{U[h]}{T[h]}}{z^{-1}[h]\exp\dfrac{U[h]}{T[h]}-1},$ (256) where the activity $z[h]$ is now $z[h]=\exp\dfrac{\mu[h]N[h]}{T[h]}.$ (257) After using of the hyperbolic property of the Bogoliubov coefficients the first identification leads to the following result $z^{-1}[h]\exp\dfrac{U[h]}{T[h]}=2|u[h]|^{2}=2|v[h]|^{2}+2,$ (258) and by this the second identification gives simply $\dfrac{U[h]-\mu[h]N[h]}{T[h]}=\dfrac{4|v[h]|^{2}}{2|v[h]|^{2}+1}.$ (259) Similarly the equation of state (249) for the Bose gas of space quantum states becomes $\dfrac{P[h]V[h]}{T[h]}=\ln\dfrac{2|u[h]|^{2}}{2|u[h]|^{2}-1}=\ln\left(1+\dfrac{1}{2|v[h]|^{2}+1}\right).$ (260) The formula determined averaged occupation number (250) expressed by the Bogoliubov coefficients becomes $N[h]=\dfrac{1}{2|u[h]|^{2}-1}=\dfrac{1}{2|v[h]|^{2}+1}.$ (261) The presented relations give an opportunity to determine the Helmholtz free energy $F[h]=U[h]-T[h]S[h],$ (262) as well as the Gibbs free energy $G[h]=U[h]-T[h]S[h]+P[h]V[h],$ (263) and the enthalpy of the system defined as $H[h]=U[h]+P[h]V[h],$ (264) iff the free energy $U[h]$ is understood as the ensemble average of the matrix representation of the Hamiltonian of the system expressed in the stable Bogoliubov vacuum $U[h]=\dfrac{\mathrm{Tr}\mathbb{D}[h]\mathbb{H}[h]}{\mathrm{Tr}\mathbb{D}[h]},$ (265) and the thermodynamical chemical potential is simply the functional derivative of the internal energy with respect to the averaged occupation number $\mu[h]=\dfrac{\delta U[h]}{\delta N[h]}.$ (266) ### 6.3 Thermodynamics of the Bose gas In this part of the paper we will construct the space quantum states thermodynamics, that according to the conjecture presented in the first section of this text is the quantum theory of gravitation. Let us start from derivation of the thermodynamical quantities for the Bose gas of space quantum states that give crucial information about this many-body statistical system. Firstly, we will consider the internal energy of the gas. In order to derivation this characteristics, let us consider the matrix representation $\mathbb{H}$ of the quantum field theory Hamiltonian (104) of the space quantum states with respect to the initial data fundamental operator basis $\mathfrak{B}_{I}$, that is $\mathbb{H}[h]=\left[\begin{array}[]{cc}\dfrac{m[h]}{2}\left(|v[h]|^{2}+|u[h]|^{2}\right)&-m[h]u[h]v[h]\\\ -m[h]u^{\ast}[h]v^{\ast}[h]&\dfrac{m[h]}{2}\left(|v[h]|^{2}+|u[h]|^{2}\right)\end{array}\right].$ (267) As it can be checked directly, this Hamiltonian matrix for fixed space metrics has the discrete spectrum that consists two different type eigenvalues $\mathrm{Spec}\mathbb{H}=\left\\{\dfrac{m[h]}{2}\left(|v[h]|+\sqrt{1+|v[h]|^{2}}\right)^{2},\dfrac{m[h]}{2}\left(|v[h]|-\sqrt{1+|v[h]|^{2}}\right)^{2}\right\\}.$ (268) By using of the definition (265) and some elementary algebraic computations one can obtain directly the internal energy of the Bose gas, that is equal to $U[h]=m[h]\left(|v[h]|^{2}+\dfrac{1}{2}+\dfrac{|v[h]|^{2}\left(1+|v[h]|^{2}\right)}{|v[h]|^{2}+\dfrac{1}{2}}\right).$ (269) Let us concentrate our attention on the occupation number of quantum states for the considered Bose gas of space quantum states. The number of space quantum states generated from the stable Bogoliubov vacuum related to initial data fundamental operator basis can be derived by standard method, as the vacuum expectation value of the one–particle density operator (240), namely by the following way $\xi=\dfrac{{{}_{I}}\langle 0\left|\mathsf{D}[h]\right|0\rangle_{I}}{{{}_{I}}\langle 0|0\rangle_{I}}=\dfrac{{{}_{I}}\langle 0\left|{\mathsf{G}}^{\dagger}[h]{\mathsf{G}}[h]\right|0\rangle_{I}}{{{}_{I}}\langle 0|0\rangle_{I}}.$ (270) After direct application of the bosonic Bogoliubov transformation and by using of the canonical commutation relations related to the fundamental initial data operator basis in the Fock space, one can simply derive the number of vacuum quantum states as follows $\displaystyle\xi=\dfrac{{{}_{I}}\left\langle 0\left|\left(u[h]\mathsf{G}_{I}^{\dagger}-v[h]\mathsf{G}_{I}\right)\left(-v^{\ast}[h]\mathsf{G}_{I}^{\dagger}+u^{\ast}[h]\mathsf{G}_{I}\right)\right|0\right\rangle_{I}}{{{}_{I}}\langle 0|0\rangle_{I}}=$ $\displaystyle=\dfrac{{{}_{I}}\left\langle 0\left||v[h]|^{2}\mathsf{G}_{I}\mathsf{G}_{I}^{\dagger}+|u[h]|^{2}\mathsf{G}_{I}^{\dagger}\mathsf{G}_{I}-v^{\ast}[h]u[h]\mathsf{G}_{I}^{\dagger}\mathsf{G}_{I}^{\dagger}-v[h]u^{\ast}[h]\mathsf{G}_{I}\mathsf{G}_{I}\right|0\right\rangle_{I}}{{{}_{I}}\langle 0|0\rangle_{I}}=$ $\displaystyle=|v[h]|^{2}\dfrac{{{}_{I}}\left\langle 0\left|\mathsf{G}_{I}\mathsf{G}_{I}^{\dagger}\right|0\right\rangle_{I}}{{{}_{I}}\langle 0|0\rangle_{I}}=|v[h]|^{2}.$ (271) From the other point of view one can calculate the number of all possible states that can be occupied by the ensemble. This quantity can be determined by grand canonical ensemble average of the matrix representation of one–particle density operator according to the definition $\langle N\rangle[h]=\dfrac{\mathrm{Tr}\left(\mathbb{D}[h]\mathbb{N}[h]\right)}{\mathrm{Tr}\mathbb{D}[h]}.$ (272) However, in the considered case we have the identification $\mathbb{N}[h]\equiv\mathbb{D}[h]$, and by this reason the number can be computed immediately with the following result $\langle N\rangle[h]=\dfrac{\mathrm{Tr}\left(\mathbb{D}^{2}[h]\right)}{\mathrm{Tr}\mathbb{D}[h]}=\dfrac{\mathrm{Tr}\left((\mathrm{Tr}\mathbb{D}[h])\mathbb{D}[h]\right)}{\mathrm{Tr}\mathbb{D}[h]}=\mathrm{Tr}\mathbb{D}[h],$ (273) that after application of the matrix representation (243) and the relation (6.3) leads finally to $\langle N\rangle=2\xi+1.$ (274) By this the grand canonical ensemble average of an occupation number of the Bose gas of space quantum states determined firstly by the relation $(\ref{occu})$ really equals $N=\dfrac{1}{2\xi+1},$ (275) and gives an information that statistically the volume of the Bose gas of space quantum states is occupied by one space quantum state. The relation between the number of states generated from the stable Bogoliubov vacuum $\xi$ and the mass $m[h]$ arises directly from the formula (118) as $m_{\pm}=m_{I}\left(\sqrt{\xi}\pm\sqrt{\xi+1}\right)^{2}.$ (276) By this reason the spectrum of the Hamiltonian eigenvalues (277) actually is determined by $\mathrm{Spec}\mathbb{H}=\left\\{\dfrac{m_{I}}{2}\left(\sqrt{\xi}+\sqrt{\xi+1}\right)^{4},\dfrac{m_{I}}{2}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{4}\right\\}.$ (277) This result can be interpreted as follows – the Bose gas of space quantum states consists two physically independent phases, associated with the sign $+$ and $-$ respectively. However, these two possible phases have no the same physical status. For demystify of this fact let us consider the basic quantity (135) of the quantum field theory formulated in previous parts of this paper – the correlation function, that carries an information about one-point bosonic field configuration and is the key quantity by this fact $\langle 1h|h1\rangle_{\pm}=\left(\dfrac{m_{I}}{m_{\pm}}\right)^{2}=\dfrac{1}{\left(\sqrt{\xi}\pm\sqrt{\xi+1}\right)^{4}}.$ (278) The character of changeability of this one-point correlator strongly depends on the choice of the sign in the denominator, and has completely different physical meaning for the case of the sign $+$ and for the case of the sign $-$. Namely, in the case the positive sign this correlator goes to zero for huge values of particles generated from the initial data vacuum, but for the case of negative sign the one-point correlations become asymptotically infinite for huge number of vacuum quantum states $\lim_{\xi\rightarrow\infty}\langle 1h|h1\rangle_{\pm}=\left\\{\begin{array}[]{cc}0&\leavevmode\nobreak\ ,\leavevmode\nobreak\ \leavevmode\nobreak\ \mathrm{for}\leavevmode\nobreak\ \leavevmode\nobreak\ $+$\\\ \infty&,\leavevmode\nobreak\ \leavevmode\nobreak\ \mathrm{for}\leavevmode\nobreak\ \leavevmode\nobreak\ $--$\end{array}\right.$ (279) The physical meaning of this situation can be explained in the following way. In the case of the positive sign, the one-point correlations in the limit of huge number of quantum states generated from the stable Bogoliubov vacuum asymptotically vanish, that physically means we have to deal with unstable situation in the classical limit, and by this reason the classical object associated with the positive sign in the one-point correlator (278) is the unstable object. However, in the second case, that is for the negative sign, the one-point correlations asymptotically arise to infinity with arise to infinity of the number of vacuum quantum states, and by this reason in this case the one-point correlator (278) describes stable configuration of space quantum states in the classical limit, it is stable physical object, see Figure (4). In this manner, at the present text we will discuss only the case of the negative sign. Figure 4: The basic one-point correlation function for stable configuration of space quantum states. For the classical limit, _i.e._ for huge number of vacuum quantum states, the one-point correlations arises infinitely. ### 6.4 Classically stable phase. Cold Big Bang. The internal energy of the Bose gas of space quantum states (269) for the stable fields configuration reads $U=m_{I}\dfrac{3\xi^{2}+3\xi+1}{2\xi+1}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2},$ (280) and is monotonic function of the argument $\xi$ (see Figure (5)) that in classical limit of the huge argument values goes asymptotically to the constant value that is $\lim_{\xi\rightarrow\infty}U=\dfrac{3}{8}m_{I}.$ (281) Figure 5: Internal energy for the Bose gas of space quantum states primordially is given by initial data, and asymptotically goes to constant value determined also by initial data. Figure 6: Chemical potential for the Bose gas of space quantum states asymptotically describes open system, but primordially is associated with a point object ( The Big Bang point). Figure 7: Temperature for the Bose gas of space quantum states asymptotically goes to constant value determined by initial data, but primordially in the point of Big Bang is characterized by minus infinite value of temperature (Cold Big Bang). One can characterize some statistical properties of the Bose gas of space quantum states by derivation of the chemical potential for this system. This quantity can be calculated by direct using of the standard thermodynamical relation $\mu=\dfrac{\delta U}{\delta N},$ (282) that by using of the fact that the internal energy $U$ as well as the averaged occupation number $N$ are functions of the number of vacuum quantum states $\xi$ leads to the definition $\mu=\dfrac{\delta U}{\delta\xi}\dfrac{\delta\xi}{\delta N}.$ (283) By using of the relations (280) and (275) one can compute some elementary derivatives that are need for derivation of the chemical potential $\displaystyle\dfrac{\delta U}{\delta\xi}$ $\displaystyle=$ $\displaystyle m_{I}\left[\dfrac{6\xi^{2}+6\xi+1}{2\xi+1}-\dfrac{3\xi^{2}+3\xi+1}{\sqrt{\xi(\xi+1)}}\right]\dfrac{\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}}{2\xi+1},$ (284) $\displaystyle\dfrac{\delta\xi}{\delta N}$ $\displaystyle=$ $\displaystyle\left(\dfrac{\delta N}{\delta\xi}\right)^{-1}=-\dfrac{1}{2}(2\xi+1)^{2},$ (285) so that actually the chemical potential (283) depends from number of vacuum quantum states by the following formula $\displaystyle\mu=-m_{I}\left[3\xi^{2}+3\xi+\dfrac{1}{2}-\dfrac{3\xi^{2}+3\xi+1}{2\sqrt{\xi(\xi+1)}}\left(2\xi+1\right)\right]\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}.$ (286) The chemical potential (286) is also monotonic function of the argument $\xi$ and asymptotically decreases to zero for huge number of vacuum particles. This fact physically means that we actually consider the system with nonconserved number of quantum states (see Figure (6)). Obviously, it is not new fact for our considerations, we have considered this type system in this paper from the beginning. Now temperature of the Bose gas of space quantum states can be determined by direct application of the relation (259) as follows $\displaystyle T=\dfrac{2\xi+1}{4\xi}\left(U-\mu N\right),$ (287) that after application of the relations (280), (275), and (286) leads to the relation between temperature and number of the space quantum states produced from initial data vacuum $T=m_{I}\left[4\xi^{2}+4\xi+1-\dfrac{3\xi^{2}+3\xi+1}{\sqrt{\xi(\xi+1)}}(2\xi+1)\right]\dfrac{3\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}}{8\xi}.$ (288) This temperature globally is not monotonic function, but has stable value in the classical limit (see Figure (7)) $\lim_{\xi\rightarrow\infty}T=\dfrac{3}{16}m_{I}.$ (289) One can see now that the following relation between internal energy and temperature of space quantum states holds $\dfrac{U}{T}=\dfrac{\dfrac{8}{3}\dfrac{\xi}{2\xi+1}}{\dfrac{4\xi^{2}+4\xi+1}{3\xi^{2}+3\xi+1}-\dfrac{2\xi+1}{3\sqrt{\xi(\xi+1)}}},$ (290) and in the limit of huge number of vacuum quantum states one can suppose that the principle of energy equipartition should be fulfilled – in a sense of the classical thermal equilibrium, the energy is shared equally among on all degrees of freedom $f$ of the system $\xi\rightarrow\infty\Longrightarrow U=\dfrac{f}{2}T.$ (291) One can calculate immediately the classical limit of the relation (290). The result exactly accords with the equipartition law (291), for this case the number of degrees of freedom equals $f=4.$ (292) For huge number of vacuum quantum states we have to deal with classical thermal equilibrium state of the system of space quantum states, that is a Riemannian manifold given by a solution of the Einstein–Hilbert field equations of General Relativity. Simultaneously out of the presented way looks into view the following fact: _classical thermal equilibrium state of the system of space quantum states is associated with an object described by 4 thermodynamical degrees of freedom_ (see Figure (8)). These degrees of freedom have the natural interpretation – they can be identified with four spacetime coordinates - one time and three space coordinates. Figure 8: Relation between quotient of internal energy and temperature (the blue line), and number of space quantum states generated from the initial data Bogoliubov vacuum. For the limit of huge value of this quantum number (the red line), _i.e._ for the classical equilibrium state of the system of space quantum states described by a Riemannian manifold given by a solution of the Einstein–Hilbert field equations of General Relativity, according to the law of equipartition this quotient asymptotically is related to 4 degrees of freedom, which have an interpretation of four spacetime coordinates. By direct using the equation of state (260) one can determinate the product of pressure and volume as $\displaystyle PV$ $\displaystyle=$ $\displaystyle m_{I}\left[\dfrac{4\xi^{2}+4\xi+1}{2\xi+1}-\dfrac{3\xi^{2}+3\xi+1}{\sqrt{\xi(\xi+1)}}\right]\times$ (293) $\displaystyle\times$ $\displaystyle\dfrac{3(2\xi+1)}{8\xi}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}\ln\left(\dfrac{2\xi+2}{2\xi+1}\right),$ and similarly the product of temperature and entropy as $\displaystyle TS$ $\displaystyle=$ $\displaystyle 3m_{I}\left[\dfrac{4\xi^{2}+4\xi+1}{2\xi+1}-\dfrac{3\xi^{2}+3\xi+1}{\sqrt{\xi(\xi+1)}}\right]\times$ (294) $\displaystyle\times$ $\displaystyle\left[\dfrac{\xi+1}{2\xi+1}-\dfrac{2\xi+1}{8\xi}\ln(2\xi+1)\right]\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}.$ The product of pressure and volume (293) goes to zero for huge number of vacuum space quantum states, what physically means that in this limit the pressure goes to zero for arbitrary big volume. The product of entropy and temperature (294) goes to minus infinity in this limit. Now the Helmholtz free energy $F$ given by general relation (262) can be determined directly as follows $\displaystyle F$ $\displaystyle=$ $\displaystyle m_{I}\Biggr{\\{}1+3\left[\dfrac{2\xi+1}{8\xi}\ln(2\xi+1)-\dfrac{\xi+1}{2\xi+1}\right]\dfrac{3\xi^{2}+3\xi+1}{2\xi+1}\times$ (295) $\displaystyle\times$ $\displaystyle\left[\dfrac{4\xi^{2}+4\xi+1}{3\xi^{2}+3\xi+1}-\dfrac{2\xi+1}{\sqrt{\xi(\xi+1)}}\right]\Biggr{\\}}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2},$ similarly the Gibbs free energy $G$ determined standardly by the relation (263) now is equal to $\displaystyle G$ $\displaystyle=$ $\displaystyle m_{I}\Biggr{\\{}1+3\left[\dfrac{2\xi+1}{8\xi}\ln(2\xi+2)-\dfrac{\xi+1}{2\xi+1}\right]\dfrac{3\xi^{2}+3\xi+1}{2\xi+1}\times$ (296) $\displaystyle\times$ $\displaystyle\left[\dfrac{4\xi^{2}+4\xi+1}{3\xi^{2}+3\xi+1}-\dfrac{2\xi+1}{\sqrt{\xi(\xi+1)}}\right]\Biggr{\\}}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2},$ and the enthalpy $H$ defined by the formula (264) now reads $\displaystyle H$ $\displaystyle=$ $\displaystyle m_{I}\Biggr{\\{}1+\dfrac{3(2\xi+1)}{8\xi}\left[\dfrac{4\xi^{2}+4\xi+1}{3\xi^{2}+3\xi+1}-\dfrac{2\xi+1}{\sqrt{\xi(\xi+1)}}\right]\times$ (297) $\displaystyle\times$ $\displaystyle\ln\left(\dfrac{2\xi+2}{2\xi+1}\right)\Biggr{\\}}\dfrac{3\xi^{2}+3\xi+1}{2\xi+1}\left(\sqrt{\xi}-\sqrt{\xi+1}\right)^{2}.$ These thermodynamical potentials have the following asymptotical values for huge number of vacuum quantum states (see Figures (9), (10), and (11)) $\displaystyle\lim_{\xi\rightarrow\infty}F$ $\displaystyle=$ $\displaystyle\infty,$ (298) $\displaystyle\lim_{\xi\rightarrow\infty}G$ $\displaystyle=$ $\displaystyle\infty,$ (299) $\displaystyle\lim_{\xi\rightarrow\infty}H$ $\displaystyle=$ $\displaystyle\dfrac{3}{8}m_{I}.$ (300) Figure 9: Helmholtz free energy for the Bose gas of space quantum states asymptotically arises to infinity. Figure 10: Gibbs free energy for the Bose gas of space quantum states asymptotically arises to infinity. Figure 11: Enthalpy for the Bose gas of space quantum states asymptotically goes to constant value determined by initial data. Above thermodynamical characteristics determine complete physical information about the Bose gas of space quantum states related to the initial data stable Bogoliubov vacuum state. The variable $\xi$, that really is a number of space quantum states generated from the stable vacuum and simultaneously the square of one of the Bogoliubov coefficients, can be treated as the fundamental quantity directly related with the basic one-point correlator $\langle 1h|h1\rangle$ by the following way $\xi=\dfrac{1}{4}\left(\dfrac{1}{\sqrt{\left|\langle 1h|h1\rangle\right|}}+\sqrt{\left|\langle 1h|h1\rangle\right|}\right)-\dfrac{1}{2},$ (301) and allows to study the presented relations between thermodynamics of space quantum states and classical equilibrium states determined only by the one- point correlator. ### 6.5 Entropy Let us consider the entropy of the Bose gas of space quantum states (254), graphically presented on the Figure (12). Figure 12: Entropy of the Bose gas of space quantum states as a function of the parameter $\xi$ has nontrivial maximum, that is identified with the Bose condensation in the system. The relation (254) actually establishes the nontrivial connection between disorder in the Bose gas of space quantum states with respect to the initial data fundamental operator basis, and the number of space quantum states generated from the stable initial data vacuum state. Let us take into our considerations the set of initial data space quantum states. According to the relation for the mass (276) this group of space quantum states is described by the initial data mass $m=m_{I}\Longleftrightarrow\xi=0,$ (302) that really is the initial tachyon mass. It can be seen directly that the entropy (254) for these quantum states vanishes $S_{I}=0.$ (303) The complete thermodynamical characterizations of the group of initial data space quantum states can be computed by taking the limit $\displaystyle\lim_{\xi\rightarrow 0}N$ $\displaystyle=$ $\displaystyle 1,$ (304) $\displaystyle\lim_{\xi\rightarrow 0}U$ $\displaystyle=$ $\displaystyle m_{I},$ (305) $\displaystyle\lim_{\xi\rightarrow 0}\mu$ $\displaystyle=$ $\displaystyle\infty,$ (306) $\displaystyle\lim_{\xi\rightarrow 0}T$ $\displaystyle=$ $\displaystyle-\infty,$ (307) $\displaystyle\lim_{\xi\rightarrow 0}\dfrac{PV}{T}$ $\displaystyle=$ $\displaystyle\ln 2,$ (308) $\displaystyle\lim_{\xi\rightarrow 0}F$ $\displaystyle=$ $\displaystyle\infty,$ (309) $\displaystyle\lim_{\xi\rightarrow 0}G$ $\displaystyle=$ $\displaystyle-\infty,$ (310) $\displaystyle\lim_{\xi\rightarrow 0}H$ $\displaystyle=$ $\displaystyle-\infty.$ (311) We see that initially the Bose gas of space quantum states has finite internal energy and unit averaged occupation number, but all other characterizations are $\pm$ infinite. The chemical potential is also infinite, that physically means that the system is no open and is compact point object with huge negative temperature. In this manner the initial data point can be interpreted as the Big Bang point, where objects that bangs are space quantum states spontaneously generated from the stable quantum vacuum. The temperature in the Big Bang limit is negative infinite; this phenomena can be called the Cold Big Bang. On the Figure (12) we see that the next interesting group of states are the space quantum states associated with the maximal value of the entropy. Let us consider now the maximally entropy point of the system of space quantum states. This especial point is determined by the number of quantum states generated from the initial data vacuum given by $\xi=\dfrac{1}{2},$ (312) and by maximal value of entropy and chemical equilibrium character, this point has the natural interpretation of the point of the condensation in the Bose gas of space quantum states. The averaged occupation number for the condensate state equals $N_{cond}=\dfrac{1}{2}=\xi,$ (313) and the mass of the condensate is $m_{cond}\approx 0.26795m_{I}.$ (314) The Bose condensate point has the entropy $S_{cond}=\dfrac{3}{2}-\ln 2,$ (315) and by values of thermodynamical characteristics are as follows $\displaystyle U_{cond}$ $\displaystyle\approx$ $\displaystyle 0.43542m_{I},$ (316) $\displaystyle\mu_{cond}$ $\displaystyle\approx$ $\displaystyle 0.26869m_{I},$ (317) $\displaystyle T_{cond}$ $\displaystyle\approx$ $\displaystyle 0.30107m_{I},$ (318) $\displaystyle F_{cond}$ $\displaystyle\approx$ $\displaystyle 0.19250m_{I},$ (319) $\displaystyle G_{cond}$ $\displaystyle\approx$ $\displaystyle 0.31457m_{I},$ (320) $\displaystyle H_{cond}$ $\displaystyle\approx$ $\displaystyle 0.55749m_{I},$ (321) with the following equation of state $\left(\dfrac{PV}{T}\right)_{cond}=\ln\dfrac{3}{2}.$ (322) The group of space quantum states with maximal value of entropy are formally associated with chemical equilibrium of the Bose system of space quantum states. The fact that the maximal value of entropy is not localized in the initial data point $\xi=0$ is the typical characteristic property of systems with the Bose condensate presence. However, as it was presented the group of initial data space quantum states play the crucial role in context of the Einstein–Hilbert General Relativity Riemannian manifold. These states have the fundamental status, they have the primordial states meaning. From the Figure (12) we see that in the region between the Big Bang and the Bose condensation of space quantum states, _i.e._ $0\leq\xi\leq\dfrac{1}{2}$, we observe entropy arising, and from the Bose condensation point up to classical equilibrium state, _i.e._ in the region $\xi\geq\dfrac{1}{2}$, entropy decreases to minus infinity, and system goes to thermodynamical disorder. It is interesting that in this region exists the point when entropy again vanishes, _i.e_ the point where the system of space quantum states has the initial value of entropy, but other characteristics are not the same as initial ones. ## 7 Summary In this paper we have presented is details the new realization of the old problem, that is formulation of quantum gravity by effective thermodynamics of quantum states. This realization was based on the fundamental fact – the Wheeler–DeWitt theory following from $3+1$ decomposition of a Lorentzian manifold metric field of General Relativity actually is not nonrelativistic Schrödinger quantum mechanics for wave function of Universe, but is the Global One–Dimensional Klein–Gordon–Fock equation of classical field theory of the Bose field associated with embedded 3-dimensional space. Moreover, we have proved directly that the Wheeler-DeWitt equation with presence superspace metrics can be represented in the form without explicitly using of the superspace metrics, that is the idea of Global One-Dimensionality, where the dimension is 3-volume form of a space. This simplified equation that further exists in the configurational space of General Relativity, was treated in this paper as the equation of classical field theory of the Bose field associated with spatial geometry of the Einstein–Hilbert pseudo–Riemannian manifold. The tachyon state of the classical field theory and more important from physical point of view the Dark Energy and the cosmological constant problems were described and discussed. Some limits for value of cosmological constant was derived. This little interpretational and cosmetic changes in the form of the Wheeler–DeWitt theory completely changed essence of the geometrodynamics that in the presented form is a classical field theory of some relativistic system. Quantization of this classical field theory is natural by application of the language of the Fock space functional annihilation and creation operators, and as it was presented in this paper gives beautiful and elegant results on physical nature of quantum gravitation. The quantum field theory was used as the main link between the Einstein–Hilbert General Relativity and thermodynamical description of a Lorentzian manifold as an effect of the many- body quantum field theory of the Bose gas of space quantum states. It is the general field theory according to depictions of Dr. Albert Einstein. The quantum theory of gravitation presented in this paper essentially describes the classical spacetime given by a solution of General Relativity field equations as an effect of asymptotical equilibrium of the generalized thermodynamics of the Bose gas of many quantum states of three-dimensional space that classically evolves in 1-dimensional time. These quantum states of $3+1$ splitted pseudo–Riemannian spacetime was called in this paper by name of space quantum states, and it was shown here that these quantum states can be considered in terms of gravity quanta. These are gravitons, in the sense of quantum field theory formulated in the Fock space the stable Bogoliubov vacuum state. As it was seen, the quantization of classical field theory, given by one-dimensional Klein–Gordon–Fock equation, can be constructed correctly only by using of the Fock space operator basis that is the Heisenberg type, _i.e._ has static character. This fact caused using of the bosonic Bogoliubov transformation, and leads to the fundamental operator basis associated with initial data. However, still we have to deal with open system, where number of quantum states is not conserved. The description related to the stable Bogoliubov vacuum state gave an opportunity to understand the quantum field theory in terms of the Bose gas of space quantum states and construct proper statistical mechanics of this system. As it was shown explicitly, this amazing Bose gas has some state of chemical equilibrium that is related to non-zero number of quantum states generated from the initial data vacuum, and has an interpretation of the Bose condensation in the Bose gas of space quantum states with respect to the initial data vacuum state. Really, as it was mentioned the Bose gas of space quantum consists two physical phases – one phase is characterized by some kind of "condensation" of one-point correlations in the case of arising of number of quantum states generated from the fundamental initial data vacuum, the second phase has vanishing correlations of the Bose gas in this classical limit. Physically this fact means that classical solution given by a Lorentizan manifold of General Relativity is described by the stable classical equilibrium in the first case, and by completely unstable state in the case of the second phase. By this physical reasoning we have chosen to further consideration the phase with condensing one-point correlations. This solution is thermodynamically stable in the classical limit, _i.e._ in the limit of huge value of number of vacuum quantum states gives a classical object that can be identified with the pseudo–Riemannian spacetime. Furthermore, as it was computed by using of the equipartition law, this solution gives a number of thermodynamical degrees of freedom which accords to number of classical spacetime coordinates - it is equal to four. By this reason, from the point of view of the thermodynamics of the Bose gas of space quantum states, spacetime coordinates have a status of thermodynamical degrees of freedom in the presented approach. In this classical limit the considered phase has asymptotically constant values of temperature and internal energy, and chemical potential vanishes classically. In this manner the stable solution describes the classical open quantum system in constant temperature. In the case of the second solution, by vanishing of the one-point correlations in classical limit, also the temperature of the space quantum states system arises to infinity for huge number of vacuum quantum states. For this solution exists Hot Big Bang of the initial data quantum states from the Bogoliubov vacuum opportunity. As it was presented in details, the analysis of the stable phase of the Bose gas of space quantum states leads to the Cold Big Bang phenomena, and to the interpretation of the initial state of the Bose gas as the primordial compact point object with negative infinite temperature. Finally, it was shown that the condensed state of the Bose gas of space quantum states presented for nontrivial value of number of vacuum quantum states, has a natural physical interpretation of chemical equilibrium state. Actually, by application of an analogy famous from condensed matter physics, the Bose condensate has a nature of the quantum object with macroscopic dimensions, and this should be observed in experiments as in the case of the Bose–Einstein condensation of photons. The condensed state of the Bose gas of space quantum states can be responsible for Dark Matter effects. ## Acknowledgements Special thanks are directed to Dr. B. G. Sidharth and Prof. F. Honsell for full hospitality 3 - 25 June 2008 at Dipartimento di Matematica e Informatica of Universit$\grave{\mathrm{a}}$ degli Studi di Udine and discussions during the stay. The author benefitted many valuable discussions from Prof. G. ’t Hooft and Dr. B. G. Sidharth, and is grateful to Profs. A. B. Arbuzov, I. Ya. Aref’eva, B. M. Barbashov, K. A. Bronnikov, I. L. Buchbinder, V. N. Pervushin, and V. B. Priezzhev for the critical remarks and motivation. ## References * [1] A. Einstein, autobiographical notes in Albert Einstein: Philosopher-Scientist, Vol. 1, p. 1–94, ed. by P.A. Schilpp, Open Court, (1969). * [2] A. Einstein, Sitzungsber. Preuss. Akad. Wiss. Berlin 44, N2, 778, (1915); Sitzungsber. Preuss. Akad. Wiss. Berlin 46, N2, 799, (1915); Sitzungsber. Preuss. Akad. Wiss. Berlin 48, N2, 844, (1915). * [3] D. Hilbert, Konigl. Gesell. d. Wiss. Göttinger, Nachr., Math.-Phys. Kl. 27, 395, (1915). * [4] R. Arnowitt, S. Deser and Ch.W. Misner, in Gravitation: an introduction to current research, ed. by L. Witten, p. 227, John Wiley and Sons, (1961), (arXiv:gr-qc/0405109v1) * [5] A. Peres , _Nuovo Cimento_ 26, 53, (1962). * [6] P.A.M. Dirac, Proc. Roy. Soc. Lond. A 246, 333, (1958); Phys. Rev. 114, 924, (1959); Proc. Roy. Soc. Lond. A 246, 326, (1958); Can. J. Math. 2, 129, (1950). * [7] J.A. Wheeler, in Battelle Rencontres: 1967 Lectures in Mathematics and Physics, Editors C.M. DeWitt and J.A. Wheeler, New York, 1968, p. 242 * [8] B.S. DeWitt, Phys. Rev. 160, 1113, (1967). * [9] J.B. Hartle, S.W. Hawking, _Phys. Rev. D_ 28, 2960, (1983). * [10] A. Ashtekar, M. Bojowald and J. Lewandowski, Adv. Theor. Math. Phys. 7, 233, (2003). * [11] C. Rovelli, Quantum gravity, Cambridge University Press, (2004). * [12] J. von Neumann, Math. Ann. 104, 570, (1931). * [13] H. Araki and E.J. Woods, J. Math. Phys. 4, 637, (1963). * [14] J.-P. Blaizot and G. Ripka, Quantum theory of finite systems, Massachusetts Institute of Technology Press, (1986). * [15] Ch.W. Misner, K.S. Thorne, J.A. Wheeler, Gravitation, W. H. Freeman, (1973). * [16] B. Riemann, Nachr. Ges. Wiss. Göttingen 13, 133, (1920). * [17] P. Petersen, Riemannian Geometry, 2nd ed., Grad. Texts Math. 171, (2006). * [18] A. Palatini, Rend. Pal. 43, 203, (1919). * [19] G. ’t Hooft, _Private communication_. * [20] H. Goldstein, Ch. Poole, J. Safko, Classical Mechanics, 3rd ed., Addison-Wesley, (2000). * [21] D. Lüst and S. Theisen, Lect. Notes Phys. 346, (1989). * [22] N.N. Bogoliubov, A.A. Logunov, A.I. Oksak, and I.T. Todorov, General Principles of Quantum Field Theory, Nauka, Moscow, (1991) * [23] K. Huang, Statistical Mechanics, 2nd ed., John Wiley & Sons, Inc, (1987).

arxiv-papers

2024-09-04T02:48:54.255068

0803.1548

# Adiabatic Loading of Cold Bosons in Three-Dimensional Optical Lattices and Superfluid-Normal Phase Transition S. Yoshimura1,2, S. Konabe3, and T. Nikuni3 1Department of Physics, Graduate School of Science, The University of Tokyo, 7-3-1 Hongo, Bunkyo-Ku, Tokyo 113-8656, Japan 2CREST, JST, 4-1-8 Honcho Kawaguchi, Saitama 332-0012, Japan 3Department of Physics, Faculty of Science, Tokyo University of Science, 1-3 Kagurazaka, Shinjuku-ku, Tokyo 162-8601, Japan ###### Abstract We investigate the effects of the adiabatic loading of optical lattices to the temperature by applying the mean-field approximation to the three-dimensional Bose-Hubbard model at finite temperatures. We compute the lattice-height dependence of the isentropic curves for the given initial temperatures in case of the homogeneous system i.e., neglecting the trapping potential. Taking the unit of temperatures as the recoil energy, the adiabatic cooling/heating through superfluid (SF) - normal (N) phase transition is clearly understood. It is found that the cooling occurs in SF phase while the heating occurs in N phase and the efficiency of adibatic cooling/heating is higher at higher temperatures. We also explain how its behavior can be understood from the lattice-hight dependence of dispersion relation in each phase. Furthermore, the connection of the adiabatic heating/cooling between the cases with/without the trapping potential is discussed. ###### pacs: Recently ultracold atoms in optical lattices have been studied intensively both theoretically and experimentally (for the reviews, see Jaksch and Zoller (2005); Morsch and Oberthaler (2006); M.Lewenstein et al. (2007); Bloch et al. (2007)). Not only atomic, molecule, and optical (AMO) physics party, but also quantum information and condensed matter physics party come into this field and it has the possibilities of producing new kind of physics. From the viewpoint of quantum information, ultracold atoms in optical lattices can be used as one-way quantum computing Clark et al. (2005); Kay et al. (2006); Christandl et al. (2005) and dynamical controlling of entanglement Jaksch et al. (1999); Mandel et al. (2003) under well-controlled conditions. Considering strongly correlated physics, this system offers the possibility of realizing various quantum lattice models like the Bose-Hubbard model, the Fermi-Hubbard model, and the Bose-Fermi Hubbard model, which have various rich quantum phases M.Lewenstein et al. (2007). In order to investigate the above subjects, it is crucial to understand the lattice-height dependence of the temperatures. In the experiments, the temperature of Bose gases is measured before inserting optical lattices. However, the experimental method to investigate the temperature of Bose gases in optical lattices has not been established. Usually the loading process can be treated as adiabatic since the loading speed is very low. The behavior of the temperature of this system during the adiabatic loading of optical lattices is therefore of great interests, which has been studied in several papers Blakie and Porto (2004); Rey et al. (2006); Ho and Zhou (2007); Gerbier (2007a). For the non-interacting Bose gases, the adiabatic cooling only occurs in the tight-binding regime (on the other hand, the adiabatic heating occurs when the thermal energy lies in the first excited band). The mechanism of the adiabatic heating/cooling in this case can be understood in terms of the change of density of states Blakie and Porto (2004). The adiabatic loading including the interaction between atoms and the effect of the trapping potential has been studied for deep lattices. In that case, it is found that the adiabatic heating occurs due to increasing the Mott gap and the trapping effect induces the adiabatic compression and expansion which cause the adiabatic heating and cooling Rey et al. (2006); Ho and Zhou (2007); Gerbier (2007a). On the other hand, the mechanism of the adiabatic cooling/heating through SF-N phase transition is not fully understood. In this paper, we examine the three-dimensional Bose-Hubbard model at finite temperatures within the mean-field approximation in order to investigate the effect of the adiabatic loading to the temperature of the system through SF phase to N phase. We show that the adiabatic cooling occurs in SF phase while the adiabatic heating occurs in N phase. The number fluctuation conserves during the adiabatic loading after crossing SF-N phase transition point, which yields that it is necessary to have ultracold temperatures at the phase transition point in order to obtain the system with very low number fluctuation at deep optical lattices. We argue that the mechanism of the adiabatic cooling/heating is due to the dispersion relation in each phase. Finally, we will mention that in the case with the trapping potential, the mechanism of the adiabatic heating/cooling is essentially same as in the case without the trapping potential. The Hamiltonian for Bose atoms in optical lattices can be written as $\displaystyle\hat{H}$ $\displaystyle=$ $\displaystyle\int d^{3}x\hat{\psi}^{\dagger}(\mbox{\bm{$x$}})\left[-\frac{\hbar^{2}}{2m}\nabla^{2}+V_{o}(\mbox{\bm{$x$}})\right]\hat{\psi}(\mbox{\bm{$x$}})$ (1) $\displaystyle+$ $\displaystyle\frac{1}{2}\frac{4\pi a_{s}\hbar^{2}}{m}\int d^{3}x\hat{\psi}^{\dagger}(\mbox{\bm{$x$}})\hat{\psi}^{\dagger}(\mbox{\bm{$x$}})\hat{\psi}(\mbox{\bm{$x$}})\hat{\psi}(\mbox{\bm{$x$}})$ where $\hat{\psi}(\mbox{\bm{$x$}})$ is a field operator for Bose atoms and $V_{o}(\mbox{\bm{$x$}})$ is the optical lattice potential. We consider three- dimensional optical lattices where $V_{o}(\mbox{\bm{$x$}})$ has the form $V_{o}(\mbox{\bm{$x$}})=V(\sin^{2}kx+\sin^{2}ky+\sin^{2}kz).$ (2) Here $k=2\pi/\lambda$ where $\lambda$ is the wave length of standing wave laser forming optical lattices. The lattice constant is determined by $a=\lambda/2$. The lattice height of optical lattices $V$ is measured by the recoil energy $E_{R}=\hbar^{2}k^{2}/2m$ where $m$ is mass of the atom. We use the dimensionless lattice height $s=V/E_{R}$. The binary interaction between atoms is approximated by s-wave scattering, which is charactrized by the scattering length $a_{s}$. The Bose-Hubbard Hamiltonian can be derived by applying the tight-binding approximation to Eq. (1) Jaksch et al. (1998). We expand the field operator by the Wannier function $w_{0}(\mbox{\bm{$x$}}-\mbox{\bm{$x$}}_{i})$ as $\hat{\psi}^{\dagger}(\mbox{\bm{$x$}})=\sum_{i}\hat{b_{i}}^{\dagger}w_{0}(\mbox{\bm{$x$}}-\mbox{\bm{$x$}}_{i})$, where $\hat{b_{i}}$ is the destruction operator for a boson at a lattice site $\mbox{\bm{$x$}}_{i}$. We can rewrite Eq. (1) as $\hat{H}=-t\sum_{\langle ij\rangle}(\hat{b}_{i}^{\dagger}\hat{b_{j}}+h.c.)+\frac{U}{2}\sum_{i}\hat{n}_{i}(\hat{n}_{i}-1)-\sum_{i}\mu\hat{n}_{i}.$ (3) As usual, we consider the nearest-neighbor hopping and the on-site interaction. Note that we added the chemical potential $\mu$ in order to treat the grand-canonical ensemble. The relation between $U,t$ and $s$ under approximating the Wannier function as the Gaussian functioin can be found as $\frac{U}{E_{R}}=5.97\frac{a_{s}}{\lambda}s^{0.88}$ and $\frac{t}{E_{R}}=1.43s^{0.98}e^{-2.07\sqrt{s}}$ Gerbier et al. (2005). In this paper, we consider 87Rb and set $m=1.44\times 10^{-25}$ [kg], $a_{s}=545$ [nm], and $\lambda=852$ [nm] as in the Greiner’s experiment Greiner et al. (2002). The validity of using the Bose-Hubbard model was examined in Ref. Jaksch et al. (1998). The tight-binding approximation at finite temperatures implies that the energy scale including the thermal energy must be less than the gap energy between the lowest band and the second band. We calculated the thermal energy $k_{B}T$ normalized by the gap energy $\Delta$ and confirmed $k_{B}T/\Delta\ll 1$ for the parameters we will use in this paper. Let us apply the mean-field approximation van Oosten et al. (2001); Lu and Yu (2006) as $\hat{b}^{\dagger}_{i}\hat{b}_{j}\approx\phi(\hat{b}^{\dagger}_{i}+\hat{b}_{j})-\phi^{2}$ where $\phi=\langle\hat{b}^{\dagger}\rangle=\langle\hat{b}\rangle$ is taken to be real. Then we can rewrite the Hamiltonian Eq.(3) as sum of on-site Hamiltonians as $\hat{H}=\sum_{i}\hat{H_{i}}$, and the Hamiltonian of the $i$-th site is $\hat{H}_{i}=\frac{U}{2}\hat{n_{i}}(\hat{n_{i}}-1)-zt\phi(\hat{b}^{\dagger}_{i}+\hat{b}_{i})+zt\phi^{2}-\mu\hat{n}_{i}.$ (4) Here $z$ denotes the coordinate number. In the actual conditions, we truncate the size of the Hilbert space of $\hat{H}_{i}$ by assuming the maximum number of particles which can be at one site is $n_{t}$. We take large $n_{t}$ so that the truncation effect on the calculated physical quantities is neglegible. Diagonalizing this Hamiltonian $\hat{H}_{i}$ under given $U$, $t$, and $\mu$, we obtain the eigenstates and the corresponding eigenenergies. Then we can calculate the partition function $Z$ and the Helmholtz free energy $F$ as functions of $\phi$ for given temperatures. $\phi$ is determined by the self-consistent equation $\partial F/\partial\phi=0$. At first, we assume that the system is homogeneous i.e., neglecting the trapping potential. In order to investigate the behavior of this system during the adiabatic loading of optical lattices, we calculate the entropy $S/k_{B}$, the condensate density $\rho_{s}\equiv\phi^{2}$, and the number fluctuation $\sigma\equiv\sqrt{\langle\hat{n}^{2}\rangle-\langle\hat{n}\rangle^{2}}$ under the fixed chemical potential $\mu$ as the occupation number $\rho=1$. Fig. 1 (a) shows the relation between the isentropic curves and the condensate density. Loading optical lattices adiabatically, the system goes along the isentropic curve which corresponds to the initial entropy. One finds that there exists two areas: the adiabatic cooling region and the adiabatic heating one. Former is the region where the system is cooled as the lattice height increases while the latter is that of where the system is heated as the lattice height increases. Notice that the efficiency of the adiabatic cooling/heating is higher for higher initial entropy. Seeing the condensate density, one finds that SF phase, which is charcterized by $\rho_{s}>0$, exists in shallow lattices and at low temperatures. The cooling region lies in SF phase, whereas the heating region lies in N phase where $\rho_{s}=0$. We also find that the efficiency of the adiabatic cooling/heating is not related with the amount of the condensate density. We plot the number fluctuation and the isentropic curves in Fig. 1 (b). For $k_{B}T/E_{R}<0.04$, there exists a rectangle region with the very low number fluctuation, which is called as the thermal insulator in Ref. Gerbier (2007b). In that region, the critical lattice height $s_{c}$ does not change much. It is necessary to satisfy $S/k_{B}<0.01$ in order to reach the thermal insulator. Note, however, that the number fluctuation is constant on the same isentropic curve in N phase. Thus, although the system is heated up, the number fluctuation does not change with loading optical lattices adiabatically after crossing SF-N phase transition point. | ---|--- Figure 1: (Color online) (a) The isentropic curves and the condensate density $\rho_{s}$ for the given lattice height $s$ and temperatures $k_{B}T/E_{R}$. We choose the chemical potential $\mu$ which satisfies $\rho=1$. The number inside box on the isentropic curve indicates the value of the entropy $S/k_{B}$. The system goes along the isentropic curve which corresponds to the initial entropy. SF phase stands for the cooling region, whereas N phase stands for the heating region. (b) The isentropic curves added the density plot of the number fluctuation $\sigma$. The isentropic curves is the same one as in (a). In SF phase, the number fluctuation is enhanced due to the quantum fluctuation. As loading optical lattices adiabatically, the number fluctuation decreases and it takes the minimum value at the critical lattice height $s_{c}$. The temperature increases after crosing $s_{c}$ whereas the number fluctuation is same. One can understand the reason why the adiabatic cooling occurs in SF phase while the adiabatic heating does in N phase by investigating the dispersion relation in each phase (see Fig. 2). The adiabatic loading means that the number of state is conserved during this process. As shown in Fig. 2 (a), the dispersion relation in SF phase exhibits the phonon-dispersion, where the sound velocity decreases with increasing the lattice height from the initial state. Then the number of state which can be occupied by the thermal energy increases if the temperature does not change. Thus the temperature must decrease in the final state in order to keep the number of state (i.e. the entropy) being constant. Fig. 2 (b) shows that, in N phase, the energy gap increases as the lattice height increases from the initial state. Then the number of state which can be accesjsible by the thermal energy decreases if the temperature is same. Therefore the temperature rises to maintain the number of state being constant in the final state. In this way, the behavior of the adiabatic heating/cooling is determined by whether the dispersion relation increases or decreases from the initial state to the final state. | ---|--- Figure 2: (Color online) Schematic picture of the dispersion relation and the thermal energy. Arrows indicate the direction of loading optical lattices. Its tail (head) corresponds to the initial (finial) state. (Dot: the thermal energy in the initial state $(k_{B}T)_{\rm ini}$, Short Dashed: the thermal energy in the final state $(k_{B}T)_{\rm fin}$, Long Dashed: The excitation spectrum in the initial state $\epsilon_{\rm ini}$, Dot Dashed: The excitation spectrum in the final state $\epsilon_{\rm fin}$) (a) SF phase. The sound velocity decreases with increasing the lattice height. Thus the temperature must decrease in order to keep the number of state. (b) N phase. The energy gap opens with loading optical lattices. Hence the temperature increase to maintain the entropy being a constant. Let us consider the system with the harmonic trapping potential $V_{t}(r)=\frac{m}{2}\omega^{2}r^{2}$, which usually exists in experiments. Note that $r$ is the distance from the center of the trapping potential. In the presence of the trapping potential, the system becomes inhomogenous and has the mixture of SF and N phases. In order to investigate the behavior of the adiabatic heating/cooling in this case, we perform the same calculation as the homogeneous case except inserting the local chemical potential $\mu(r)=\mu_{0}-V_{t}(r)$. We take the trapping frequency $\omega=2\pi\times 24$, the lattice size to be $65^{3}$, and the total number density $N=2\times 10^{5}$ as in the Greiner’s work Greiner et al. (2002). Figs. 3 (a) and (b) show qualitativley same results as homogeneous case. The adiabatic cooling occurs in the presence of condensate while the adiabatic heating occurs when there is no condensate. Therefore we can say that the behavior of adiabatic heating/cooling does not change essentially whether there is the trapping potential or not. We note, however, that the effects of the trapping potential is important quantitatively for realizing the strong correlated system as shown in Refs. Ho and Zhou (2007); Gerbier (2007b). In Figs. 3 (c) and (d), we plot the spatial distributions of the occupation number, the condensate density, the entropy, and the number fluctuation along the adiabatic line $S_{total}/Nk_{B}=0.3$ indicated by two white dots in Figs. 3 (a) and (b). We see the high condensate density in Fig. 3 (c), while the wedding-cake structure is exhibited in Fig. 3 (d). Since we use local density approximation, further detailed studies are needed for the case with the trapping potential. | ---|--- | Figure 3: (Color online) (a) The isentropic curves and the density plot of the condensate density per particle. (b) The isoentropic curves and the density plot of the number fluctuation per particle. (c) and (d) Spatial distributions of $\rho$ (circle, blue), $\rho_{s}$ (square, purple), $S/k_{B}$ (diamond, blown), and $\sigma$ (triangle, green) at $(s,k_{B}T/E_{R})=(8.9,0.085),(17.4,0.045)$ respectively. This two $(s,k_{B}T/E_{R})$ are on the adiabatic line $S_{total}/Nk_{B}=0.3$, which are indicated by two white dots in (a) and (b). The condensate density exists in (c) while the wedding-cake structure is exbited in (d). In conclusion, we calculated the three-dimensional Bose-Hubbard model at finite temperatures within the mean-field approximation and investigated the effects of the adiabatic loading of optical lattices to the temperature. The lattice-height dependence of the isentropic curves for given initial temperatures in case of the homogeneous system was computed. We found that the cooling occurs in SF phase while the heating occurs in N phase and the efficiency of the adibatic cooling/heating is higher for higher temperatures. Its behavior is determined by whether the dispersion relation of the system increases or decreases as loading optical lattices. Finally, we showed that the case with the trapping potential is essentially same as the homogeneous one. Note added. Recently we have become aware of a related paper by Pollet et al. Pollet et al. (2008). They studied the adiabatic loading in one-dimensional and two-dimensional optical lattices by quantum monte carlo method. The effect of the trapping potential to the adiabatic loading was investigated with high accuracy in these lower dimensional cases. ###### Acknowledgements. S.Y. thanks to S. Miyashita, N. Kawashima, and I. Danshita for fruitful discussions. S.Y. is supported by NAREGI Nanoscience Project from Ministry of Education Culture, Sports, Science, and Technology, Japan. S.K. is supproted by JSPS (Japan Society for the Promotion of Science) Research Fellowship for Young Scientists. ## References * Jaksch and Zoller (2005) D. Jaksch and P. Zoller, Annals of Physics 315, 52 (2005). * Morsch and Oberthaler (2006) O. Morsch and M. Oberthaler, Rev. Mod. Phys. 78, 179 (2006). * M.Lewenstein et al. (2007) M.Lewenstein, A.Sanpera, V.Ahufinger, B.Damski, A.S.De, and U.Sen, Adv. Phys. 56, 243 (2007). * Bloch et al. (2007) I. Bloch, J. Dalibard, and W. Zwerger, cond-mat/0704.3011 (2007), eprint 0704.3011. * Clark et al. (2005) S. Clark, C. Moura-Alves, and D. Jaksch, New J. Phys. 7, 124 (2005). * Kay et al. (2006) A. Kay, J. Pachos, and C. Adams, Phys. Rev. A 73, 022310 (2006). * Christandl et al. (2005) M. Christandl, N. Datta, T. Dorlas, A. Ekert, A. Kay, and A. Labdahl, Phys. Rev. A 71, 032312 (2005). * Jaksch et al. (1999) D. Jaksch, H.-J. Briegel, J. I. Cirac, C. W. Gardiner, and P. Zoller, Phys. Rev. Lett. 82, 1975 (1999). * Mandel et al. (2003) O. Mandel, M. Greiner, A. Widera, T. Rom, and T. Hänsch, Nature 425, 937 (2003). * Blakie and Porto (2004) P. Blakie and J. Porto, Phys. Rev. A 69, 13603 (2004). * Rey et al. (2006) A. Rey, G. Pupillo, and J. Porto, Phys. Rev. A 73, 23608 (2006). * Ho and Zhou (2007) T.-L. Ho and Q. Zhou, Phys. Rev. Lett. 99 (2007). * Gerbier (2007a) F. Gerbier, Phys. Rev. Lett. 99 (2007a). * Jaksch et al. (1998) D. Jaksch, C. Bruder, J. Cirac, C. Gardiner, and P. Zoller, Phys. Rev. Lett. 81, 3108 (1998). * Gerbier et al. (2005) F. Gerbier, A. Widera, S. Fölling, O. Mandel, T. Gericke, and I. Bloch, Phys. Rev. A 72, 053606 (2005). * Greiner et al. (2002) M. Greiner, O. Mandel, T. Esslinger, T. W. Hänsch, and I. Bloch, Nature (London) 45, 39 (2002). * van Oosten et al. (2001) D. van Oosten, P. van der Straten, and H. Stoof, Phys. Rev. A 63, 053601 (2001). * Lu and Yu (2006) X. Lu and Y. Yu, Phys. Rev. A 74, 063615 (2006). * Gerbier (2007b) F. Gerbier, Phys. Rev. Lett. 99, 12045 (2007b). * Pollet et al. (2008) L. Pollet, C. Kollath, K. Van Houcke, and M. Troyer, cond-mat/0801.1887 (2008).

arxiv-papers

2024-09-04T02:48:54.264673

0803.1606

# Symmetric Numerical Semigroups Generated by Fibonacci and Lucas Triples Leonid G. Fel Department of Civil Engineering, Technion, Haifa 3200, Israel e-mail: lfel@tx.technion.ac.il ###### Abstract The symmetric numerical semigroups ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ and ${\sf S}\left(L_{k},L_{m},L_{n}\right)$ generated by three Fibonacci $(F_{a},F_{b},F_{c})$ and Lucas $(L_{k},L_{m},L_{n})$ numbers are considered. Based on divisibility properties of the Fibonacci and Lucas numbers we establish necessary and sufficient conditions for both semigroups to be symmetric and calculate their Hilbert generating series, Frobenius numbers and genera. Keywords: Symmetric numerical semigroups, Fibonacci and Lucas numbers. 2000 Mathematics Subject Classification: Primary – 20M14, Secondary – 11N37. ## 1 Introduction Recently the numerical semigroups ${\sf S}\left(F_{i},F_{i+2},F_{i+k}\right)$, $i,k\geq 3$, generated by three Fibonacci numbers $F_{j}$ were discussed in [8]. It turns out that the remarkable properties of $F_{j}$ in these triples suffice to calculate the Frobenius number ${\cal F}\left({\sf S}\right)$ and genus $G\left({\sf S}\right)$ of semigroup. In this article we show that a nature of Fibonacci and Lucas numbers is sufficient not only to calculate the specific parameters of semigroups, but also to describe completely the structure of symmetric numerical semigroups ${\sf S}\left(F_{a},F_{b},F_{c}\right)$, $3\leq a<b<c$, and ${\sf S}\left(L_{k},L_{m},L_{n}\right)$, $2\leq k<m<n$, generated by Fibonacci 111We avoid to use the term ”Fibonacci semigroup” because it has been already reserved for another algebraic structure [10]. and Lucas numbers, respectively. Based on divisibility properties of these numbers we establish necessary and sufficient conditions for both semigroups to be symmetric and calculate their Hilbert generating series, Frobenius numbers and genera. ## 2 Basic properties of the 3D symmetric numerical semigroups Recall basic definitions and known facts about 3D numerical semigroups mostly focusing on their symmetric type. Let ${\sf S}\left(d_{1},d_{2},d_{3}\right)\subset{\mathbb{Z}}_{+}\cup\\{0\\}$ be the additive numerical semigroup with zero finitely generated by a minimal set of positive integers $\\{d_{1},d_{2},d_{3}\\}$ such that $3\leq d_{1}<d_{2}<d_{3}$, $\gcd(d_{1},d_{2},d_{3})=1$. Semigroup ${\sf S}(d_{1},d_{2},d_{3})$ is said to be generated by the minimal set of three natural numbers if there are no nonnegative integers $b_{i,j}$ for which the following dependence holds: $d_{i}=\sum_{j\neq i}^{m}b_{i,j}d_{j}\;,\;\;\;b_{i,j}\in\\{0,1,\ldots\\}\;\;\;\mbox{for any}\;\;i\leq m\;.$ (1) For short we denote the vector $(d_{1},d_{2},d_{3})$ by ${\bf d}^{3}$. Following Johnson [6] define the minimal relation ${\cal R}_{3}$ for given ${\bf d}^{3}$ as follows $\displaystyle{\cal R}_{3}\left(\begin{array}[]{r}d_{1}\\\ d_{2}\\\ d_{3}\end{array}\right)=\left(\begin{array}[]{r}0\\\ 0\\\ 0\end{array}\right)\;,\;\;\;{\cal R}_{3}=\left(\begin{array}[]{rrr}a_{11}&-a_{12}&-a_{13}\\\ -a_{21}&a_{22}&-a_{23}\\\ -a_{31}&-a_{32}&a_{33}\end{array}\right)\;,\;\;\;\left\\{\begin{array}[]{r}\gcd(a_{11},a_{12},a_{13})=1\\\ \gcd(a_{21},a_{22},a_{23})=1\\\ \gcd(a_{31},a_{32},a_{33})=1\end{array}\right.\;,$ (14) where $\displaystyle a_{11}$ $\displaystyle=$ $\displaystyle\min\left\\{v_{11}\;|\;v_{11}\geq 2,\;v_{11}d_{1}=v_{12}d_{2}+v_{13}d_{3},\;v_{12},v_{13}\in{\mathbb{N}}\cup\\{0\\}\right\\}\;,$ $\displaystyle a_{22}$ $\displaystyle=$ $\displaystyle\min\left\\{v_{22}\;|\;v_{22}\geq 2,\;v_{22}d_{2}=v_{21}d_{1}+v_{23}d_{3},\;v_{21},v_{23}\in{\mathbb{N}}\cup\\{0\\}\right\\}\;,$ (15) $\displaystyle a_{33}$ $\displaystyle=$ $\displaystyle\min\left\\{v_{33}\;|\;v_{33}\geq 2,\;v_{33}d_{3}=v_{31}d_{1}+v_{32}d_{2},\;v_{31},v_{32}\in{\mathbb{N}}\cup\\{0\\}\right\\}\;.$ The uniquely defined values of $v_{ij},i\neq j$ which give $a_{ii}$ will be denoted by $a_{ij},i\neq j$. Note that due to minimality of the set $(d_{1},d_{2},d_{3})$ the elements $a_{ij},i,j\leq 3$ satisfy $\displaystyle a_{11}=a_{21}+a_{31}\;,\;\;\;a_{22}=a_{12}+a_{32}\;,\;\;\;a_{33}=a_{13}+a_{23}\;,$ $\displaystyle d_{1}=a_{22}a_{33}-a_{23}a_{32}\;,\;\;\;d_{2}=a_{11}a_{33}-a_{13}a_{31}\;,\;\;\;d_{3}=a_{11}a_{22}-a_{12}a_{21}\;.$ (16) The smallest integer $C\left({\bf d}^{3}\right)$ such that all integers $s,\;s\geq C\left({\bf d}^{3}\right)$, belong to ${\sf S}\left({\bf d}^{3}\right)$ is called the conductor of ${\sf S}\left({\bf d}^{3}\right)$, $\displaystyle C\left({\bf d}^{3}\right):=\min\left\\{s\in{\sf S}\left({\bf d}^{3}\right)\;|\;s+{\mathbb{Z}}_{+}\cup\\{0\\}\subset{\sf S}\left({\bf d}^{3}\right)\right\\}\;.$ The number ${\cal F}\left({\bf d}^{3}\right)=C\left({\bf d}^{3}\right)-1$ is referred to as the Frobenius number. Denote by $\Delta\left({\bf d}^{3}\right)$ the complement of ${\sf S}\left({\bf d}^{3}\right)$ in ${\mathbb{Z}}_{+}\cup\\{0\\}$, i.e. $\Delta\left({\bf d}^{3}\right)={\mathbb{Z}}_{+}\cup\\{0\\}\setminus{\sf S}\left({\bf d}^{3}\right)$. The cardinality ($\\#$) of the set $\Delta\left({\bf d}^{3}\right)$ is called the number of gaps, $G\left({\bf d}^{3}\right):=\\#\left\\{\Delta\left({\bf d}^{3}\right)\right\\}$, or genus of ${\sf S}\left({\bf d}^{3}\right)$. The semigroup ring ${\sf k}\left[X_{1},X_{2},X_{3}\right]$ over a field ${\sf k}$ of characteristic 0 associated with ${\sf S}\left({\bf d}^{3}\right)$ is a polynomial subring graded by $\deg X_{i}=d_{i}$, $i=1,2,3$ and generated by all monomials $z^{d_{i}}$. The Hilbert series $H({\bf d}^{3};z)$ of a graded subring ${\sf k}\left[z^{d_{1}},z^{d_{2}},z^{d_{3}}\right]$ is defined [11] by $H({\bf d}^{3};z)=\sum_{s\;\in\;{\sf S}\left({\bf d}^{3}\right)}z^{s}=\frac{Q({\bf d}^{3};z)}{\left(1-z^{d_{1}}\right)\left(1-z^{d_{2}}\right)\left(1-z^{d_{3}}\right)}\;,$ (17) where $Q({\bf d}^{3};z)$ is a polynomial in $z$. The semigroup ${\sf S}\left({\bf d}^{3}\right)$ is called symmetric iff for any integer $s$ holds $\displaystyle s\in{\sf S}\left({\bf d}^{3}\right)\;\;\;\Longleftrightarrow\;\;\;{\cal F}\left({\bf d}^{3}\right)-s\not\in{\sf S}\left({\bf d}^{3}\right)\;.$ (18) Otherwise ${\sf S}\left({\bf d}^{3}\right)$ is called non–symmetric. The integers $G\left({\bf d}^{3}\right)$ and $C\left({\bf d}^{3}\right)$ are related [5] as, $\displaystyle 2G\left({\bf d}^{3}\right)=C\left({\bf d}^{3}\right)\;\;\mbox{if}\;\;{\sf S}\left({\bf d}^{3}\right)\;\;\mbox{is symmetric semigroup, and}\;\;2G\left({\bf d}^{3}\right)>C\left({\bf d}^{3}\right)\;\;\mbox{otherwise}.$ (19) Notice that ${\sf S}\left({\bf d}^{2}\right)$ is always symmetric semigroup [1]. The number of independent entries $a_{ij}$ in (14) can be reduced if ${\sf S}\left({\bf d}^{3}\right)$ is symmetric: at least one off-diagonal element of $\widehat{\cal R}_{3}$ vanishes, e.g. $a_{13}=0$ and therefore $a_{11}d_{1}=a_{12}d_{2}$. Due to minimality of the last relation we have by (14) the following equalities and consequently the matrix representation as well [4] (see also [3], Section 6.2) $\displaystyle\left.\begin{array}[]{l}a_{11}=a_{21}={\sf lcm}(d_{1},d_{2})/d_{1},\;\;\;a_{12}=a_{22}={\sf lcm}(d_{1},d_{2})/d_{2}\;,\\\ a_{33}=d_{1}/a_{22}=d_{2}/a_{11}\;,\;\;a_{23}=0\;,\end{array}\right.\;\;\widehat{\cal R}_{3s}=\left(\begin{array}[]{rrr}a_{11}&-a_{22}&0\\\ -a_{11}&a_{22}&0\\\ -a_{31}&-a_{32}&a_{33}\end{array}\right),$ (25) where subscript ”$s$” stands for symmetric semigroup. Combining (25) with formula for the Frobenius number of symmetric semigroup [4], ${\cal F}\left({\bf d}^{3}_{s}\right)=a_{22}d_{2}+a_{33}d_{3}-\sum_{i=1}^{3}d_{i}$, we get finally, $\displaystyle{\cal F}\left({\bf d}^{3}_{s}\right)=e_{1}+e_{2}-\sum_{i=1}^{3}d_{i}\;,\;\;\;\;e_{1}={\sf lcm}(d_{1},d_{2})\;,\;\;\;e_{2}=d_{3}\;{\sf gcd}(d_{1},d_{2})\;.$ (26) If ${\sf S}\left({\bf d}^{3}\right)$ is symmetric semigroup then ${\sf k}\left[{\sf S}\left({\bf d}^{3}\right)\right]$ is a complete intersection [4] and the numerator $Q({\bf d}^{3};z)$ in the Hilbert series (17) reads [11] $\displaystyle Q({\bf d}^{3};z)=(1-z^{e_{1}})(1-z^{e_{2}})\;.$ (27) ### 2.1 Structure of generating triples of symmetric numerical semigroups Two following statements, Theorem 1 and Corollary 1, give necessary and sufficient conditions for ${\sf S}\left({\bf d}^{3}\right)$ to be symmetric. ###### Theorem 1 ([4] and Proposition 3, [14]) If a semigroup ${\sf S}\left(d_{1},d_{2},d_{3}\right)$ is symmetric then its minimal generating set has the following presentation with two relatively not prime elements: $\displaystyle\gcd(d_{1},d_{2})=\lambda\;,\;\;\gcd(d_{3},\lambda)=1\;,\;\;d_{3}\in{\sf S}\left(\frac{d_{1}}{\lambda},\frac{d_{2}}{\lambda}\right)\;.$ (28) It turns out that (28) gives also sufficient conditions for ${\sf S}\left({\bf d}^{3}\right)$ to be symmetric. This follows by Corollary 1 of the old Lemma of Watanabe [14] for semigroup ${\sf S}\left({\bf d}^{m}\right)$ ###### Lemma 1 (Lemma 1, [14]) Let ${\sf S}\left(d_{1},\ldots,d_{m}\right)$ be a numerical semigroup, $a$ and $b$ be positive integers such that: (i) $c\in{\sf S}\left(d_{1},\ldots,d_{m}\right)$ and $c\neq d_{i}$, (ii) $\gcd(c,\lambda)=1$. Then semigroup ${\sf S}\left(\lambda d_{1},\ldots,\lambda d_{m},c\right)$ is symmetric iff ${\sf S}\left(d_{1},\ldots,d_{m}\right)$ is symmetric. Combining Lemma 1 with the fact that every semigroup ${\sf S}\left({\bf d}^{2}\right)$ is symmetric we arrive at Corollary. ###### Corollary 1 Let ${\sf S}\left(d_{1},d_{2}\right)$ be a numerical semigroup, $c$ and $\lambda$ be positive integers, $\gcd(c,\lambda)=1$. If $c\in{\sf S}\left(d_{1},d_{2}\right)$, then the semigroup ${\sf S}\left(\lambda d_{1},\lambda d_{2},c\right)$ is symmetric. In Corollary 1 the requirement $c\neq d_{1},d_{2}$ can be omitted since both semigroups ${\sf S}\left(\lambda d_{1},\lambda d_{2},d_{1}\right)$ and ${\sf S}\left(\lambda d_{1},\lambda d_{2},d_{2}\right)$ are generated by two elements ($d_{1},\lambda d_{2}$) and are also symmetric. Finish this Section with important proposition adapted to the 3D numerical semigroups. ###### Theorem 2 ([5], Proposition 1.14) The numerical semigroup ${\sf S}\left(3,d_{2},d_{3}\right)$, $\gcd(3,d_{2},d_{3})=1$, $3\nmid d_{2}$ and $d_{3}\not\in{\sf S}\left(3,d_{2}\right)$, is never symmetric. ## 3 Divisibility of Fibonacci and Lucas numbers We recall a remarkable divisibility properties of Fibonacci and Lucas numbers which are necessary for further consideration. Theorem 3 dates back to E. Lucas [7] (Section 11, p. 206), ###### Theorem 3 Let $F_{m}$ and $F_{n}$, $m>n$, be the Fibonacci numbers. Then $\displaystyle\gcd\left(F_{m},F_{n}\right)=F_{\gcd(m,n)}\;.$ (29) As for Theorem 4, its weak version was given by Carmichael [2] 222Carmichael [2] (Theorem 7, p. 40) has proven only the most hard part of Theorem 4, namely, the 1st equality in (33).. We present here its modern form proved by Ribenboim [12] and McDaniel [9]. ###### Theorem 4 Let $L_{m}$ and $L_{n}$ be the Lucas numbers, and let $m=2^{a}m^{\prime}$, $n=2^{b}n^{\prime}$, where $m^{\prime}$ and $n^{\prime}$ are odd positive integers and $a,b\geq 0$. Then $\displaystyle\gcd\left(L_{m},L_{n}\right)=\left\\{\begin{array}[]{lll}L_{\gcd(m,n)}&\mbox{if}&a=b\;,\\\ 2&\mbox{if}&a\neq b\;,\;\;3\mid\gcd(m,n)\;,\\\ 1&\mbox{if}&a\neq b\;,\;\;3\nmid\gcd(m,n)\;.\end{array}\right.$ (33) We also recall another basic divisibility property of Lucas numbers, $\displaystyle L_{m}=0\pmod{2}\;,\;\;\;\;\mbox{iff}\;\;\;\;m=0\pmod{3}\;.$ (34) We’ll need a technical Corollary which follows by consequence of Theorem 4. ###### Corollary 2 Let $L_{m}$ and $L_{n}$ be the Lucas numbers, and let $m=2^{a}m^{\prime}$, $n=2^{b}n^{\prime}$, where $m^{\prime}$ and $n^{\prime}$ are odd positive integers and $a,b\geq 0$. Then $\displaystyle\gcd\left(L_{m},L_{n}\right)=1\;,\;\;\;\;\mbox{iff}\;\;\;\;\left\\{\begin{array}[]{ll}a=b=0\;,&\gcd\left(m^{\prime},n^{\prime}\right)=1\;,\\\ a\neq b\;,&\gcd\left(3,\gcd(m,n)\right)=1\;.\end{array}\right.$ (37) ## 4 Symmetric numerical semigroups generated by Fibonacci triple In this Section we consider symmetric numerical semigroups generated by three Fibonacci numbers $F_{c}$, $F_{b}$ and $F_{a}$, $c>b>a\geq 3$. The two first values $a=3,4$ are of special interest because of Fibonacci numbers $F_{3}=2$ and $F_{4}=3$. First, the semigroup ${\sf S}\left(F_{3},F_{b},F_{c}\right)$, $\gcd(2,F_{b},F_{c})=1$, is always symmetric and has actually 2 generators. Next, according to Theorem 2 the semigroup ${\sf S}\left(F_{4},F_{b},F_{c}\right)$ is symmetric iff at least one of two requirements, $3\nmid F_{b}$ and $F_{c}\not\in{\sf S}\left(3,F_{b}\right)$, is broken. Avoiding those trivial cases we state ###### Theorem 5 Let $F_{c}$, $F_{b}$ and $F_{a}$ be the Fibonacci numbers where $c>b>a\geq 5$. Then a numerical semigroup ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ is symmetric iff $\displaystyle\lambda=\gcd(a,b)\geq 3\;,\;\;\;\gcd(\lambda,c)=1,2\;,\;\;\;F_{c}\in{\sf S}\left(\frac{F_{a}}{F_{\lambda}},\frac{F_{b}}{F_{\lambda}}\right)\;,$ (38) Proof By Theorem 1 and Corollary 1 a numerical semigroup ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ is symmetric iff $\displaystyle g=\gcd\left(F_{a},F_{b}\right)>1\;,\;\;\;\gcd(g,F_{c})=1\;,\;\;\;F_{c}\in{\sf S}\left(\frac{F_{a}}{g},\frac{F_{b}}{g}\right)\;.$ (39) By consequence of Theorem 3 and definition of Fibonacci numbers we get $\displaystyle\left\\{\begin{array}[]{lll}g=F_{\lambda}>1&\rightarrow&\gcd(a,b)\geq 3\;,\\\ \gcd(F_{\lambda},F_{c})=F_{\gcd(\lambda,c)}=1&\rightarrow&\gcd(\lambda,c)=1,2\;.\end{array}\right.$ (42) The last containment in (39) gives $\displaystyle F_{c}=A\frac{F_{a}}{g}+B\frac{F_{b}}{g}=A\frac{F_{a}}{F_{\lambda}}+B\frac{F_{b}}{F_{\lambda}}\;,\;\;\;\;A,B\in{\mathbb{Z}}_{+}\;,$ that finishes the proof of Theorem.$\;\;\;\;\;\;\Box$ Theorem 5 remains true for any permutation of indices in triple $(F_{a},F_{b},F_{c})$. By (26), (27) and (38) we get ###### Corollary 3 Let $F_{c}$, $F_{b}$ and $F_{a}$ be the Fibonacci numbers and numerical semigroup ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ be symmetric. Then its Hilbert series and Frobenius number are given by $\displaystyle H\left(F_{a},F_{b},F_{c}\right)$ $\displaystyle=$ $\displaystyle\frac{(1-z^{f_{1}})(1-z^{f_{2}})}{\left(1-z^{F_{a}}\right)\left(1-z^{F_{b}}\right)\left(1-z^{F_{c}}\right)}\;,\;\;\;f_{1}=\frac{F_{a}F_{b}}{F_{\gcd(a,b)}}\;,\;\;\;f_{2}=F_{c}\cdot F_{\gcd(a,b)}\;,$ (43) $\displaystyle{\cal F}\left(F_{a},F_{b},F_{c}\right)$ $\displaystyle=$ $\displaystyle f_{1}+f_{2}-(F_{a}+F_{b}+F_{c})\;.$ The next Corollary 4 gives only the sufficient condition for ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ to be symmetric and is less strong than Theorem 5. However, instead of containment (38) it sets an inequality which is easy to check out. ###### Corollary 4 Let $F_{c}$, $F_{b}$ and $F_{a}$ be the Fibonacci numbers where $c>b>a\geq 5$. Then a numerical semigroup ${\sf S}\left(F_{a},F_{b},F_{c}\right)$ is symmetric if $\displaystyle\lambda=\gcd(a,b)\geq 3\;,\;\;\;\gcd(\lambda,c)=1,2\;,\;\;\;F_{c}F_{\lambda}>{\sf lcm}(F_{a},F_{b})-F_{a}-F_{b}\;.$ (44) The Hilbert series and Frobenius number are given by (43). Proof The two first relations in (44) are taken from Theorem 5 and were proven in (42). We have to use also the containment (38). For this purpose take $F_{c}$ exceeding the Frobenius number of semigroup generated by two numbers $F_{a}/F_{\lambda}$ and $F_{b}/F_{\lambda}$. This number ${\cal F}\left(F_{a}/F_{\lambda},F_{b}/F_{\lambda}\right)$ is classically known due to Sylvester [13]. So, we get $\displaystyle F_{c}>\frac{F_{a}}{F_{\lambda}}\frac{F_{b}}{F_{\lambda}}-\frac{F_{a}}{F_{\lambda}}-\frac{F_{b}}{F_{\lambda}}=\frac{{\sf lcm}(F_{a},F_{b})-F_{a}-F_{b}}{F_{\lambda}}\;,$ where the Hilbert series $H\left(F_{a},F_{b},F_{c}\right)$ and Frobenius number ${\cal F}\left(F_{a},F_{b},F_{c}\right)$ are given by (43). Thus, Corollary is proven.$\;\;\;\;\;\;\Box$ We finish this Section by Example 1 where the Fibonacci triple does satisfy the containment in (38) but does not satisfy inequality in (44). ###### Example 1 $\\{d_{1},d_{2},d_{3}\\}=\\{F_{6}=8,F_{8}=21,F_{9}=34\\}$ $\displaystyle\gcd(F_{6},F_{9})=F_{3}\;,\;\;\;\;\gcd(F_{3},F_{8})=1\;,\;\;\;\;F_{8}\in{\sf S}\left(\frac{F_{6}}{F_{3}},\frac{F_{9}}{F_{3}}\right)={\sf S}\left(4,17\right)\;,$ $\displaystyle f_{1}={\sf lcm}(F_{6},F_{9})=136\;,\;\;\;\;f_{2}=F_{8}\cdot F_{3}=42\;,\;\;\;\;F_{8}\cdot F_{3}<{\sf lcm}(F_{6},F_{9})-F_{6}-F_{9}\;,$ $\displaystyle H\left(F_{6},F_{8},F_{9}\right)=\frac{(1-z^{136})(1-z^{42})}{\left(1-z^{8}\right)\left(1-z^{21}\right)\left(1-z^{34}\right)}\;,\;\;\;{\cal F}\left(F_{6},F_{8},F_{9}\right)=115\;,\;\;\;\;G\left(F_{6},F_{8},F_{9}\right)=58\;.$ ## 5 Symmetric numerical semigroups generated by Lucas triple In this Section we consider symmetric numerical semigroups generated by three Lucas numbers $L_{n}$, $L_{m}$ and $L_{k}$, $n>m>k\geq 2$. Note that the case $k=2$ is trivial because of Lucas number $L_{2}=3$ and Theorem 2. The semigroup ${\sf S}\left(L_{2},L_{m},L_{n}\right)$ is symmetric iff at least one of two requirements, $3\nmid L_{m}$ and $L_{n}\not\in{\sf S}\left(3,L_{m}\right)$, is broken. ###### Theorem 6 Let $L_{k}$, $L_{m}$ and $L_{n}$, $\;n,m,k\geq 3$, be the Lucas numbers and let $\displaystyle m=2^{a}m^{\prime}\;,\;\;n=2^{b}n^{\prime}\;,\;\;k=2^{c}k^{\prime}\;,\;\;\;\mbox{where}\;\;\;m^{\prime}=n^{\prime}=k^{\prime}=1\pmod{2}\;,\;\;\;a,b,c\geq 0\;,\;\;\;\;\;\;$ (45) $\displaystyle l=\gcd(m,n)=2^{d}l^{\prime}\;,\;\;\;\mbox{where}\;\;\;l^{\prime}=\gcd(m^{\prime},n^{\prime})=1\pmod{2}\;,\;\;\;d=\min\\{a,b\\}\;.$ Then a numerical semigroup generated by these numbers is symmetric iff $L_{k}$, $L_{m}$ and $L_{n}$ satisfy $\displaystyle L_{k}\in{\sf S}\left(\frac{L_{m}}{L_{l}},\frac{L_{n}}{L_{l}}\right)\;,\;\;\;\mbox{if}\;\;a=b\;,\;\;\;\mbox{or}\;\;\;L_{k}\in{\sf S}\left(\frac{L_{m}}{2},\frac{L_{n}}{2}\right)\;,\;\;\;\mbox{if}\;\;a\neq b\;,$ (46) and one of three following relations: $\displaystyle\begin{array}[]{ll}1)\;\;\;\;a=b\neq 0\;,&\;\;a=b\neq c\;,\;\;\;3\nmid\gcd(k,l)\;,\\\ 2)\;\;\;\;a=b=0\;,\;\;\gcd\left(m^{\prime},n^{\prime}\right)>1\;\;,&\;\left\\{\begin{array}[]{ll}c=0\;,&\gcd\left(k^{\prime},l^{\prime}\right)=1\;,\\\ c\neq 0\;,&3\nmid\gcd(k,l)\;,\end{array}\right.\\\ 3)\;\;\;\;a\neq b\;,\;\;\;3\mid\gcd(m,n)\;,&\;\;3\nmid k\;.\end{array}$ (52) Proof By Theorem 1 and Corollary 1 a numerical semigroup ${\sf S}\left(L_{k},L_{m},L_{n}\right)$ is symmetric iff there exist two relatively not prime elements of its minimal generating set such that $\displaystyle\eta=\gcd(L_{n},L_{m})>1\;,\;\;\gcd(L_{k},\eta)=1\;,\;\;L_{k}\in{\sf S}\left(\frac{L_{n}}{\eta},\frac{L_{m}}{\eta}\right)\;.$ (53) Represent $n$ and $m$ as in (45) and substitute them into the 1st relation in (53). By consequence of Theorem 4 it holds iff $\displaystyle 1)\;\;a=b\;,\;\;\gcd(m,n)>1\;\;\;\;\;\mbox{or}\;\;\;\;\;2)\;\;a\neq b\;,\;\;3\mid\gcd(m,n)\;.$ (54) First, assume that the 1st requirement in (54) holds that results by Theorem 4 in $\eta=L_{l}$. Making use of notations (45) for $k$ move on to the 2nd requirement in (53) and apply Corollary (2). Here we have to consider two cases $a=b\neq 0$ and $a=b=0$ separately. $\displaystyle a=b\neq 0\;,\;\;\;a=b\neq c\;,\;\;\;3\nmid\gcd(k,l)=1\;,$ (55) $\displaystyle a=b=0\;,\;\;\;\gcd\left(m^{\prime},n^{\prime}\right)>1\;,\;\;\;\left\\{\begin{array}[]{ll}c=0\;,&\gcd\left(k^{\prime},l^{\prime}\right)=1\;,\\\ c\neq 0\;,&3\nmid\gcd(k,l)\;.\end{array}\right.$ (58) Now, assume that the 2nd requirement in (54) holds that results by Theorem 4 in $\eta=2$. Making use of the 2nd requirement in (53) and applying (34) we get, $\displaystyle a\neq b\;,\;\;3\mid\gcd(m,n)\;,\;\;\;3\nmid k\;.$ (59) Combining (55), (58) and (59) we arrive at (52). The last requirement in (53) together with Theorem 4 gives $\displaystyle L_{k}=A\frac{L_{m}}{\eta}+B\frac{L_{n}}{\eta}=\left\\{\begin{array}[]{lll}A\cdot L_{m}/L_{l}+B\cdot L_{n}/L_{l}&\mbox{if}&\;a=b\\\ A\cdot L_{m}/2+B\cdot L_{n}/2&\mbox{if}&a\;\neq b\end{array}\right.\;,\;\;\;A,B\in{\mathbb{Z}}_{+}\;,$ (62) that proves (46) and finishes proof of Theorem.$\;\;\;\;\;\;\Box$ By consequence of Theorem 6 the following Corollary holds for the most simple Lucas triples. ###### Corollary 5 Let $L_{k^{\prime}}$, $L_{m^{\prime}}$ and $L_{n^{\prime}}$ be the Lucas numbers with odd indices such that $\displaystyle\gcd(m^{\prime},n^{\prime})>1\;,\;\;\;\;\gcd(m^{\prime},n^{\prime},k^{\prime})=1\;.$ (63) Then a numerical semigroup generated by these numbers is symmetric iff $\displaystyle L_{k^{\prime}}\in{\sf S}\left(\frac{L_{m^{\prime}}}{L_{\gcd(m^{\prime},n^{\prime})}},\frac{L_{n^{\prime}}}{L_{\gcd(m^{\prime},n^{\prime})}}\right)\;.$ (64) Proof follows if we apply Theorem 6 in the case $a=b=c=0$, see (58). We give without derivation the Hilbert series and Frobenius number for symmetric semigroup ${\sf S}\left(L_{k^{\prime}},L_{m^{\prime}},L_{n^{\prime}}\right)$. $\displaystyle H\left(L_{n^{\prime}},L_{m^{\prime}},L_{k^{\prime}}\right)=\frac{(1-z^{l_{1}})(1-z^{l_{2}})}{\left(1-z^{L_{n^{\prime}}}\right)\left(1-z^{m^{\prime}}\right)\left(1-z^{L_{k^{\prime}}}\right)}\;,\;\;\;l_{1}=\frac{L_{n^{\prime}}\cdot L_{m^{\prime}}}{L_{\gcd(m^{\prime},n^{\prime})}}\;,$ $\displaystyle{\cal F}\left(L_{n^{\prime}},L_{m^{\prime}},L_{k^{\prime}}\right)=l_{1}+l_{2}-(L_{n^{\prime}}+L_{m^{\prime}}+L_{k^{\prime}})\;,\;\;\;l_{2}=L_{k^{\prime}}\cdot L_{\gcd(m^{\prime},n^{\prime})}\;.$ (65) In general, the containment (64) is hardly to verify because it presumes algorithmic procedure. Instead, one can formulate a simple inequality which provide only the sufficient condition for semigroup ${\sf S}\left(L_{n^{\prime}},L_{m^{\prime}},L_{k^{\prime}}\right)$ to be symmetric. ###### Corollary 6 Let $L_{n^{\prime}}$, $L_{m^{\prime}}$ and $L_{k^{\prime}}$ be the Lucas numbers with odd indices such that (63) is satisfied and the following inequality holds, $\displaystyle L_{k^{\prime}}\;L_{\gcd(m^{\prime},n^{\prime})}>\frac{L_{n^{\prime}}\;L_{m^{\prime}}}{L_{\gcd(m^{\prime},n^{\prime})}}-L_{n^{\prime}}-L_{m^{\prime}}\;.$ (66) Then a numerical semigroup ${\sf S}\left(L_{n^{\prime}},L_{m^{\prime}},L_{k^{\prime}}\right)$ is symmetric and its Hilbert series and Frobenius number are given by (65). Its proof is completely similar to the proof of Corollary 4 for symmetric semigroup generated by three Fibonacci numbers. We finish this Section by Example 2 where the Lucas triple does satisfy the containment in (64) but does not satisfy inequality (66). ###### Example 2 $\\{d_{1},d_{2},d_{3}\\}=\\{L_{9}=76,L_{15}=1364,L_{17}=3571\\}$ $\displaystyle\gcd(L_{9},L_{15})=L_{3}\;,\;\;\;\;\gcd(L_{3},L_{17})=1\;,\;\;\;\;L_{17}\in{\sf S}\left(\frac{L_{9}}{L_{3}},\frac{L_{15}}{L_{3}}\right)={\sf S}\left(19,341\right)\;,$ $\displaystyle l_{1}={\sf lcm}(L_{9},L_{15})=25916\;,\;\;\;\;l_{2}=L_{17}\cdot L_{3}=14264\;,\;\;\;\;L_{17}\cdot L_{3}<{\sf lcm}(L_{9},L_{15})-L_{9}-L_{15}\;,$ $\displaystyle H\left(L_{9},L_{15},L_{17}\right)=\frac{(1-z^{25916})(1-z^{14264})}{\left(1-z^{76}\right)\left(1-z^{1364}\right)\left(1-z^{3571}\right)}\;,$ $\displaystyle{\cal F}\left(L_{9},L_{15},L_{17}\right)=35189\;,\;\;\;\;G\left(L_{9},L_{15},L_{17}\right)=17595\;.$ ## Acknowledgement I thank C. Cooper for bringing the paper [9] to my attention. ## References * [1] R. Apéry, Sur les Branches superlinéaires des Courbes Algébriques, C. R. Acad. Sci. Paris, 222, 1198 (1946). MR 8, 221 * [2] R. D. Carmichael, On the Numerical Factors of the Arithmetic Forms $\alpha^{n}\pm\beta^{n}$, Annals of Math., 15, 30-70 (1913) * [3] L. G. Fel, Frobenius Problem for Semigroups ${\sl S}\left(d_{1},d_{2},d_{3}\right)$, Funct. Analysis and Other Math., 1, # 2, 119-157 (2006) * [4] J. Herzog, Generators and Relations of Abelian Semigroups and Semigroup Rings, Manuscripta Math., 3, 175 (1970) * [5] J. Herzog and E. Kunz, Die Werthalbgruppe Eines Lokalen Rings der Dimension 1, Sitzungsberichte der Heidelberger Akademie der Wissenschaften, Springer, Berlin (1971) * [6] S. M. Johnson, A Linear Diophantine Problem, Canad. J. Math., 12, 390 (1960) * [7] E. Lucas, Theorie des Fonctions Numeriques Simplement Periodiques, Amer. J. Math., 1, 184-240, 289-321 (1878) * [8] J. M. Marin, J. Ramirez Alfonsin and M. P. Revuelta, On the Frobenius Number of Fibonacci Numerical Semigroups, Integers: Electron. J. Comb. Number Theory, 7, # A14 (2007) * [9] W. L. McDaniel, The G.C.D in Lucas Sequences and Lehmer Number Sequences, Fibonacci Quarterly, 29, 24-29 (1991) * [10] A. Restivo, Permutation property and the Fibonacci semigroup, Semigroup Forum, 38, 337-345 (1989) * [11] R. P. Stanley, Combinatorics and Commutative Algebra, Birkhäuser Boston, 2nd ed, (1996) * [12] P. Ribenboim, Square Classes of Fibonacci and Lucas Numbers, Port. Math., 46, 159-175 (1989) * [13] J. J. Sylvester, Problems from the Theory of Numbers, with Solutions, Educational Times, 4, 171 (1884) * [14] K. Watanabe, Some Examples of 1–dim Gorenstein Domains, Nagoya Math. J., 49, 101 (1973)

arxiv-papers

2024-09-04T02:48:54.269357

0803.1608

# Spectrum of diffractively produced gluons in onium–nucleus collisions Yang Li$\,{}^{a}$ and Kirill Tuchin$\,{}^{a,b}$ ${}^{a}\,$Department of Physics and Astronomy, Iowa State University, Ames, IA 50011 ${}^{b}\,$RIKEN BNL Research Center, Upton, NY 11973-5000 ###### Abstract We calculate spectrum of diffractively produced gluons in onium–heavy nucleus collisions at high energies. We show that it exhibits a characteristic dependence on nucleus atomic number $A$ and energy/rapidity. We argue that this dependence offers a unique possibility for determining the low-$x$ structure of nuclear matter. Applications to RHIC, LHC and EIC experimental programs are discussed. ††preprint: RBRC-731 ## I Introduction Diffractive dissociation is one of the most interesting processes in high energy QCD. It played a pivotal role in identifying early signatures of the gluon saturation Gribov:1983tu ; Mueller:1986wy ; McLerran:1993ni ; Jalilian- Marian:1997jx ; Jalilian-Marian:1997gr ; Jalilian-Marian:1998cb ; Jalilian- Marian:1997dw ; Kovner:2000pt ; Iancu:2000hn ; Iancu:2001ad ; Iancu:2001md ; Ferreiro:2001qy in deep inelastic scattering (DIS) at HERA GolecBiernat:1998js ; GolecBiernat:1999qd ; Gotsman:1999vt ; Gotsman:2000gb ; Gotsman:2002zi ; Levin:2002fj ; Levin:2001pr ; Kovchegov:1999kx ; Bartels:2002cj . While the measurements at HERA revealed the first indication that the gluon saturation has become an important effect, the possible measurements of diffractive dissociation in p(d)A collisions at RHIC and LHC as well as in DIS in the proposed EIC collider will be able to probe gluon densities deeply in the saturation region. This statement is supported by the recent phenomenological success of models based on gluon saturation that accurately describe the experimental data on total hadron multiplicities Kharzeev:2000ph ; Kharzeev:2001gp ; Kharzeev:2001yq ; Kharzeev:2002ei , inclusive gluon production Kovchegov:1998bi ; Kovchegov:2001sc ; Braun:2000bh ; Dumitru:2001ux ; Blaizot:2004wu ; Kharzeev:2002pc ; Kharzeev:2003wz ; Kharzeev:2004yx ; Baier:2003hr ; Iancu:2004bx and heavy quark production Kharzeev:2003sk ; Gelis:2003vh ; Tuchin:2004rb ; Blaizot:2004wv ; Kovchegov:2006qn ; Tuchin:2007pf ; Kharzeev:2005zr ; Tuchin:2006hz . This motivated us to calculate multiplicity of diffractively produced gluons in coherent diffraction of onium on a heavy nucleus in a recent publication Li:2008bm . We observed that the diffractive gluon multiplicity is very sensitive to the low-$x$ dynamics in onium. On the other hand, it showed only a weak dependence on the gluon density in the nucleus. The reason is that the total cross section is dominated by soft gluon momenta which are not sensitive to the short-distance structure on the nuclear color field. Consequently, in the present paper we set to calculate the diffractive gluon spectrum. Diffractive gluon production in DIS has been discussed in many publications Wusthoff:1997fz ; GolecBiernat:1999qd ; Bartels:1999tn ; Kopeliovich:1999am ; Gotsman:1999vt ; Kovchegov:2001ni ; Munier:2003zb ; Marquet:2004xa ; GolecBiernat:2005fe ; Marquet:2007nf ; Kovner:2006ge ; Kovner:2001vi . The goal of our calculation is to calculate, for the first time, the diffractive gluon spectrum taking into account the low-$x$ gluon evolution in all rapidity intervals, i.e. in the rapidity interval between the onium and the emitted gluon and between the emitted gluon and the nucleus. We believe that this calculation opens a new avenue towards the phenomenological applications in pA and eA collisions. The paper is organized as follows. In Sec. II we briefly review the result for the gluon spectrum given in our previous paper Li:2008bm . Eq. (1) represents the cross section in terms of the dipole density in onium $n_{p}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{b},y)$ and the dipole–nucleus forward scattering amplitude $N(\mathbf{r},\mathbf{b},y)$. In Sec. III we deliberate about the behavior of $n_{p}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{b},y)$ and $N(\mathbf{r},\mathbf{b},y)$ in various kinematic regions. We turn to analysis of the gluon spectrum for large dipoles $r>1/Q_{s}$ in Sec. IV and small ones $r<1/Q_{s}$ in Sec. V. The results are summarized in Sec. VI where we also discuss possible phenomenological applications. ## II Diffractive gluon production in onium–nucleus collisions Figure 1: Fan diagram for the diffractive gluon production in onium–nucleus collisions with maximal rapidity gap $Y_{0}=y$. The source of gluon multiplicity is the cut pomeron hanging out the onium P Li:2008bm . Consider a process of diffractive gluon production in onium–nucleus scattering such that the rapidity gap equals the produced gluon rapidity $y$. The corresponding fan diagram is displayed in Fig. 1. In Ref. Li:2008bm we used the Mueller’s dipole model dip to generalize the quasi-classical result of Kovchegov Kovchegov:2001sc (derived independently in Kovner:2006ge ) by including the quantum evolution effects. The method is based on the principal idea of the dipole model that, due to the large difference between the coherence length of the low-$x$ gluons in the onium light-cone “wave-function” and the nuclear size, we can split in the light–cone time the process of the low-$x$ evolution in onium and the instantaneous interaction. Indeed, the coherence length of the low-$x$ gluons is inversely proportional to $x$, whereas the size of the interaction region (in the nucleus rest frame) is $2R_{A}$. In the large $N_{c}$ approximation the onium wave function decomposes into a system of independent color dipoles. Up to terms suppressed at low $x$, dipole transverse size does not change in a course of interaction with the nucleus. We therefore, are able to write the cross section for the gluon production as a convolution of the onium dipole density and dipole forward scattering amplitude. Let us introduce the following notations, see Fig. 2: transverse coordinates of quark, anti-quark, gluon in the amplitude and gluon in the c.c. amplitude are denoted by $\mathbf{x}$, $\mathbf{y}$, $\mathbf{z}_{1}$, $\mathbf{z}_{2}$ respectively; gluon transverse momentum is denoted by $\mathbf{k}$. Figure 2: One of the diagrams contributing to the diffractive gluon production at the quasi-classical level. Notations are detailed in text. In this notation the cross section takes the following form $\displaystyle\frac{d\sigma(k,y)}{d^{2}k\,dy}$ $\displaystyle=$ $\displaystyle\frac{\alpha_{s}C_{F}}{\pi^{2}}\frac{1}{(2\pi)^{2}}\,\int d^{2}b\,d^{2}B\,\int d^{2}r^{\prime}\,n_{1}(\mathbf{x}-\mathbf{y},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{B}-\mathbf{b},Y-y)\,$ (1) $\displaystyle\times\bigg{|}\int d^{2}z_{1}\,e^{-i\mathbf{k}\cdot\mathbf{z}_{1}}\left(\frac{\mathbf{z}_{1}-\mathbf{x}^{\prime}}{|\mathbf{z}_{1}-\mathbf{x}^{\prime}|^{2}}-\frac{\mathbf{z}_{1}-\mathbf{y}^{\prime}}{|\mathbf{z}_{1}-\mathbf{y}^{\prime}|^{2}}\right)\,\left[N(\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{b},y)\right.$ $\displaystyle\left.-N(\mathbf{x}^{\prime}-\mathbf{z}_{1},\mathbf{b},y)-N(\mathbf{y}^{\prime}-\mathbf{z}_{1},\mathbf{b},y)+N(\mathbf{x}^{\prime}-\mathbf{z}_{1},\mathbf{b},y)N(\mathbf{y}^{\prime}-\mathbf{z}_{1},\mathbf{b},y)\right]\bigg{|}^{2}\,,$ where $n_{1}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{B}-\mathbf{b},Y-y)$ is the dipole density and $N(\mathbf{r}^{\prime},\mathbf{b},y)$ is the forward dipole–nucleus scattering amplitude. Function $n_{1}(\mathbf{x}-\mathbf{y},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{B}-\mathbf{b},Y-y)$ has the meaning of the number of dipoles of size $\mathbf{x}^{\prime}-\mathbf{y}^{\prime}$ at rapidity $Y-y$ and impact parameter $\mathbf{b}$ generated by evolution from the original dipole $\mathbf{x}-\mathbf{y}$ having rapidity $Y$ and impact parameter $\mathbf{B}$ dip . It satisfies the BFKL equation Kuraev:1977fs ; Balitsky:1978ic $\displaystyle\frac{\partial n_{1}(\mathbf{x}-\mathbf{y},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{b},y)}{\partial y}=\frac{\alpha_{s}N_{c}}{2\pi^{2}}\int d^{2}z\,\frac{(\mathbf{x}-\mathbf{y})^{2}}{(\mathbf{x}-\mathbf{z})^{2}(\mathbf{y}-\mathbf{z})^{2}}$ $\displaystyle\big{[}n_{1}(\mathbf{x}-\mathbf{z},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{b},y)+n_{1}(\mathbf{y}-\mathbf{z},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{b},y)-n_{1}(\mathbf{x}-\mathbf{y},\mathbf{x}^{\prime}-\mathbf{y}^{\prime},\mathbf{b},y)\big{]}\,,$ (2) with the initial condition $n_{1}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{b},0)=\delta(\mathbf{r}-\mathbf{r}^{\prime})\,\delta(\mathbf{b})\,,$ (3) where we denoted $\mathbf{r}=\mathbf{x}-\mathbf{y}$ and $\mathbf{r}^{\prime}=\mathbf{x}^{\prime}-\mathbf{y}^{\prime}$. The forward elastic dipole–nucleus scattering amplitude satisfies the nonlinear BK equation Balitsky:1995ub ; Kovchegov:1999yj $\displaystyle\frac{\partial N(\mathbf{x}-\mathbf{y},\mathbf{b},y)}{\partial y}=\frac{\alpha_{s}N_{c}}{2\pi^{2}}\int d^{2}z\,\frac{(\mathbf{x}-\mathbf{y})^{2}}{(\mathbf{x}-\mathbf{z})^{2}(\mathbf{y}-\mathbf{z})^{2}}\big{[}N(\mathbf{x}-\mathbf{z},\mathbf{b},y)$ $\displaystyle+N(\mathbf{y}-\mathbf{z},\mathbf{b},y)-N(\mathbf{x}-\mathbf{y},\mathbf{b},y)-N(\mathbf{x}-\mathbf{z},\mathbf{b},y)N(\mathbf{y}-\mathbf{z},\mathbf{b},y)\big{]}\,,$ (4) with the initial condition given by Mue $N(\mathbf{r},\mathbf{b},0)=1-e^{-\frac{1}{8}\mathbf{r}^{2}\,Q_{s0}^{2}}\,.$ (5) The _gluon_ saturation scale is given by $Q_{s0}^{2}=\frac{4\pi^{2}\alpha_{s}N_{c}}{N_{c}^{2}-1}\,\rho\,T(\mathbf{b})\,xG(x,1/\mathbf{r}^{2})\,,$ (6) where $\rho$ is the nuclear density, $T(\mathbf{b})$ is the nuclear thickness function as a function of the impact parameter $\mathbf{b}$. In the following we will assume for notational brevity that the nuclear profile is cylindrical. An explicit impact parameter dependence, which are required in the numerical analysis, can be easily restored in the final expressions. Accordingly, it is convenient to proceed by defining the quantity $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)=\int d^{2}b\,n_{p}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{b},y)$ (7) which satisfies the BFKL equation (II) with the initial condition $n_{p}(\mathbf{r},\mathbf{r}^{\prime},0)=\delta(\mathbf{r}-\mathbf{r}^{\prime})\,.$ (8) For the following calculations it is convenient to cast (1) in a different form extracting the explicit dependence on $\mathbf{r}^{\prime}$. First, we change the integration variable $\mathbf{w}=\mathbf{z}_{1}-\mathbf{y}^{\prime}$. Then, introduce the following transverse vector $\displaystyle\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)$ $\displaystyle=$ $\displaystyle\int d^{2}w\,e^{-i\mathbf{k}\cdot\mathbf{w}}\left(\frac{\mathbf{w}-\mathbf{r}^{\prime}}{|\mathbf{w}-\mathbf{r}^{\prime}|^{2}}-\frac{\mathbf{w}}{\mathbf{w}^{2}}\right)$ (9) $\displaystyle\times\left[N(\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w},\mathbf{b},y)+N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)N(\mathbf{w},\mathbf{b},y)\right]\,.$ Using (9), (1) can be rendered as $\frac{d\sigma(k,y)}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi^{2}}\frac{1}{(2\pi)^{2}}\,\int d^{2}b\,d^{2}B\,\int d^{2}r^{\prime}\,n_{1}(\mathbf{r},\mathbf{r}^{\prime},\mathbf{B}-\mathbf{b},Y-y)\,|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}\,,$ (10) Now, contribution of the first term in the round brackets of (9) can be written as $\displaystyle\int d^{2}w\,e^{-i\mathbf{k}\cdot w}\frac{\mathbf{w}-\mathbf{r}^{\prime}}{|\mathbf{w}-\mathbf{r}^{\prime}|^{2}}\left[N(\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w},\mathbf{b},y)+N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)N(\mathbf{w},\mathbf{b},y)\right]=\,$ $\displaystyle-\int d^{2}w\,e^{i\mathbf{k}\cdot(\mathbf{w}-\mathbf{r}^{\prime})}\frac{\mathbf{w}}{\mathbf{w}^{2}}\left[N(\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w},\mathbf{b},y)+N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)N(\mathbf{w},\mathbf{b},y)\right]\,$ (11) where we changed the integration variable $\mathbf{w}-\mathbf{r}^{\prime}\to-\mathbf{w}$ and used the fact that the amplitude depends only on the dipole size (and not on direction). Defining a new scalar function $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ as $\displaystyle Q(\mathbf{r}^{\prime},\mathbf{k},y)=$ $\displaystyle-\int d^{2}w\,e^{i\mathbf{k}\cdot\mathbf{w}}\frac{1}{w^{2}}\left[N(\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)-N(\mathbf{w},\mathbf{b},y)+N(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y)N(\mathbf{w},\mathbf{b},y)\right]\,.$ (12) and using (II) we write (9) in the following form $\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)=-e^{-i\mathbf{k}\cdot\mathbf{r}^{\prime}}\,i\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y)+i\nabla_{\mathbf{k}}Q^{*}(\mathbf{r}^{\prime},\mathbf{k},y)\,.$ (13) Consequently, $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=2|\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}-e^{-i\mathbf{k}\cdot\mathbf{r}^{\prime}}(\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y))^{2}-e^{i\mathbf{k}\cdot\mathbf{r}^{\prime}}(\nabla_{\mathbf{k}}Q^{*}(\mathbf{r}^{\prime},\mathbf{k},y))^{2}\,.$ (14) In the region where $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ is a real function, we can render (14) as $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=4\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,(\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y))^{2}\,,\quad\mathrm{when}\,\,\,Q(\mathbf{r}^{\prime},\mathbf{k},y)\in\Re\,.$ (15) In terms of $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)$, (10) reads $\frac{d\sigma(k,y)}{d^{2}kdy}=\frac{\alpha_{s}C_{F}}{\pi^{2}}\frac{1}{(2\pi)^{2}}\,S_{A}\int d^{2}r^{\prime}\,n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}\,,$ (16) where $S_{A}$ is the cross sectional area of the interaction region. At $y=0$ this expression reduces to the quasi-classical formula derived in Kovchegov:2001ni ; Kovner:2001vi ; Kovner:2006ge . So far we have been concentrating on a case in which the rapidity of the produced gluon $y$ coincides with the rapidity gap $Y_{0}$ in a diffractive event. In this case the diffractive scattering amplitude $N_{D}(\mathbf{r},\mathbf{b},y,Y_{0})$ coincides with the square of the forward elastic scattering amplitude $N(\mathbf{r},\mathbf{b},y)$. This case has the most phenomenological interest (since the invariant mass $M$ of the produced system is dominated by “slow” gluons $M^{2}\approx k^{2}/x$). Still, at high enough luminosity a single hadron spectrum measurements should become possible. Therefore, a question may arise about the diffractive production of a gluon with $y>Y_{0}$. Such process is shown in Fig. 3. Figure 3: Fan diagram for the diffractive gluon production in pA collisions with rapidity gap $Y_{0}$ smaller than the gluon rapidity $y$. In this case, the amplitude $N(\mathbf{r},\mathbf{b},y)$ must be replaced by the off-forward diffractive dipole amplitude which explicitly depends on a coordinate of quark or anti-quark of parent dipole and the coordinates of the emitted gluon in the amplitude and in the c.c. one.111In the case of inclusive gluon production, an off-forward amplitude was discussed in JalilianMarian:2004da . Investigation of properties of the off-forward diffractive amplitude would lead us astray of the main subject of this paper and hence will be discussed elsewhere. Let us only note here that in those cases when the coordinate of the gluon is the same on both sides of the cut, the off-forward diffractive amplitude reduces to the more familiar forward diffractive amplitude $N_{D}(\mathbf{r},\mathbf{b},y;Y_{0})$. Since $N_{D}(\mathbf{r},\mathbf{b},y;Y_{0})$ contains information about all possible pomeron cuts shown in Fig. 3 it can serve as a _phenomenological model_ for the yet unknown off-forward diffractive amplitude. In this case the cross section for the diffractive gluon production takes form: $\frac{d\sigma^{pA}(k,y)}{d^{2}kdy}=\frac{\alpha_{s}C_{F}}{\pi^{2}}\frac{1}{(2\pi)^{2}}S_{A}\int d^{2}r^{\prime}\,n_{p}(\mathbf{r}^{\prime},Y-y)\,|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y;Y_{0})|^{2}\,,$ (17) where now in place of (13) and (II) we write $\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y;Y_{0})=-e^{-i\mathbf{k}\cdot\mathbf{r}^{\prime}}\,i\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y;Y_{0})+i\nabla_{\mathbf{k}}Q^{*}(\mathbf{r}^{\prime},\mathbf{k},y;Y_{0})\,,$ (18) and $\displaystyle Q(\mathbf{r}^{\prime},\mathbf{k},y;Y_{0})$ $\displaystyle=$ $\displaystyle-\int d^{2}w\,e^{i\mathbf{k}\cdot\mathbf{w}}\frac{1}{w^{2}}\left[N^{\frac{1}{2}}_{D}(\mathbf{r}^{\prime},\mathbf{b},y;Y_{0})-N^{\frac{1}{2}}_{D}(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y;Y_{0})\right.$ (19) $\displaystyle\left.\,-N^{\frac{1}{2}}_{D}(\mathbf{w},\mathbf{b},y;Y_{0})+N^{\frac{1}{2}}_{D}(\mathbf{w}-\mathbf{r}^{\prime},\mathbf{b},y;Y_{0})N^{\frac{1}{2}}_{D}(\mathbf{w},\mathbf{b},y;Y_{0})\right]\,.$ Amplitude $N_{D}(\mathbf{r},\mathbf{b},y;Y_{0})$ equals the cross section of single diffractive dissociation of a dipole of transverse size $\mathbf{r}$, rapidity $y$ and impact parameter $\mathbf{b}$ on a target nucleus. It satisfies the Kovchegov–Levin evolution equation Kovchegov:1999ji $\displaystyle\frac{\partial N_{D}(\mathbf{x}-\mathbf{y},\mathbf{b},y;Y_{0})}{\partial y}$ (20) $\displaystyle=\frac{2\alpha_{s}C_{F}}{\pi^{2}}\int d^{2}z\,\left[\frac{(\mathbf{x}-\mathbf{y})^{2}}{(\mathbf{x}-\mathbf{z})^{2}(\mathbf{y}-\mathbf{z})^{2}}-2\pi\delta(\mathbf{y}-\mathbf{z})\ln(|\mathbf{x}-\mathbf{y}|\Lambda)\right]N_{D}(\mathbf{x}-\mathbf{z},\mathbf{b},y;Y_{0})$ $\displaystyle+$ $\displaystyle\frac{\alpha_{s}C_{F}}{\pi^{2}}\int d^{2}z\frac{(\mathbf{x}-\mathbf{y})^{2}}{(\mathbf{x}-\mathbf{z})^{2}(\mathbf{y}-\mathbf{z})^{2}}\left[N_{D}(\mathbf{x}-\mathbf{z},\mathbf{b},y;Y_{0})N_{D}(\mathbf{y}-\mathbf{z},\mathbf{b},y;Y_{0})\right.$ $\displaystyle-$ $\displaystyle\left.4N_{D}(\mathbf{x}-\mathbf{z},\mathbf{b},y;Y_{0})N(\mathbf{y}-\mathbf{z},\mathbf{b},y)+2N(\mathbf{x}-\mathbf{z},\mathbf{b},y)N(\mathbf{y}-\mathbf{z},\mathbf{b},y)\right]\,,$ with the initial condition $N_{D}(\mathbf{r},\mathbf{b},y=Y_{0};Y_{0})=N^{2}(\mathbf{r},\mathbf{b},Y_{0})\,.$ (21) Diffractive gluon production of the kind shown in Fig. 3 requires a dedicated study while in this paper we concentrate on the case $y=Y_{0}$. ## III Dipole evolution in onium and nucleus ### III.1 Dipole evolution in onium Dipole evolution in onium is encoded in the function $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)$ and is determined by solving the BFKL equation (II) with the initial condition (3). The result reads $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)=\frac{1}{2\pi^{2}r^{\prime 2}}\int_{-\infty}^{\infty}d\nu\,e^{2\bar{\alpha}_{s}\chi(\nu)y}\,(r/r^{\prime})^{1+2i\nu}\,,$ (22) where $\bar{\alpha}_{s}=\alpha_{s}N_{c}/\pi$ and $\chi(\nu)=\psi(1)-\frac{1}{2}\psi(\frac{1}{2}-i\nu)-\frac{1}{2}\psi(\frac{1}{2}+i\nu)\,,$ (23) with $\psi(\nu)$ being the digamma function $\psi(\nu)=\frac{\Gamma^{\prime}(\nu)}{\Gamma(\nu)}\,.$ (24) There are several cases when the integral (22) can be done analytically. Expansion near the maximum of $\chi(\nu)$ corresponds to the leading- logarithmic approximation. In this case we have $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)_{LLA}\approx\frac{1}{2\pi^{2}rr^{\prime}}\sqrt{\frac{\pi}{14\zeta(3)\bar{\alpha}_{s}y}}e^{(\alpha_{P}-1)y}\,e^{-\frac{\ln^{2}(r^{\prime}/r)}{14\zeta(3)\bar{\alpha}_{s}y}}\,,\quad\alpha_{s}y\gg\ln^{2}(r/r^{\prime})\,,$ (25) where $\alpha_{P}-1=4\bar{\alpha}_{s}\ln 2$. Alternatively, we can expand $\chi(\nu)$ near one of its two symmetric poles at $2i\nu=\pm 1$. This corresponds to the double logarithmic approximation depending on the relation between $r$ and $r^{\prime}$. The results for the dipole density read as follows $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)_{DLA}\approx\frac{r^{2}}{4\pi^{3/2}r^{\prime 4}}\frac{(2\bar{\alpha}_{s}y)^{1/4}}{\ln^{3/4}(r^{\prime}/r)}\,e^{2\sqrt{2\bar{\alpha}_{s}y\ln(r^{\prime}/r)}}\,,\quad r<r^{\prime}\,,\quad\ln(r^{\prime}/r)\gg\alpha_{s}y\,.$ (26) and $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)_{DLA}\approx\frac{1}{4\pi^{3/2}r^{\prime 2}}\frac{(2\bar{\alpha}_{s}y)^{1/4}}{\ln^{3/4}(r/r^{\prime})}\,e^{2\sqrt{2\bar{\alpha}_{s}y\ln(r/r^{\prime})}}\,,\quad r>r^{\prime}\,,\quad\ln(r/r^{\prime})\gg\alpha_{s}y\,.$ (27) ### III.2 Dipole evolution in a heavy nucleus In the region $rQ_{s}\ll 1$ where the forward elastic dipole–nucleus scattering amplitude $N(\mathbf{r},\mathbf{b},Y)$ satisfies the BFKL equation it can be calculated similarly to the dipole density of the previous subsection. The initial condition in this case is specified by (5) expanded at small dipole sizes to the leading order. The result is $N(\mathbf{r},\mathbf{b},y)_{LT}=\frac{1}{8\pi}\int_{-\infty}^{\infty}d\nu\,e^{2\bar{\alpha}_{s}\chi(\nu)y}\,(rQ_{s0})^{1+2i\nu}\,\frac{1+(1-2i\nu)\ln\frac{Q_{s0}}{\Lambda}}{(1-2i\nu)^{2}}\,.$ (28) Analogously to the derivation of (25) we obtain in the leading logarithmic approximation $N(\mathbf{r},\mathbf{b},y)_{LLA}=\frac{rQ_{s0}}{8\pi}\sqrt{\frac{\pi}{14\zeta(3)\bar{\alpha}_{s}y}}\ln\left(\frac{Q_{s0}}{\Lambda}\right)\,e^{(\alpha_{P}-1)y}\,e^{-\frac{\ln^{2}(rQ_{s0})}{14\zeta(3)\bar{\alpha}_{s}y}}\,,\quad\alpha_{s}y\gg\ln^{2}\left(\frac{1}{rQ_{s0}}\right)\,,$ (29) and in the double logarithmic approximation $\displaystyle N(\mathbf{r},\mathbf{b},y)_{DLA}=\frac{\sqrt{\pi}}{16\pi}\frac{\ln^{1/4}\left(\frac{1}{rQ_{s0}}\right)}{(2\bar{\alpha}_{s}y)^{3/4}}\,r^{2}Q_{s0}^{2}\,\left(1+\sqrt{\frac{2\alpha_{s}y}{\ln\frac{1}{rQ_{s0}}}}\,\ln\frac{Q_{s0}}{\Lambda}\right)e^{2\sqrt{2\bar{\alpha}_{s}y\ln\frac{1}{rQ_{s0}}}}\,,$ $\displaystyle r<1/Q_{s0}\,,\quad\ln\frac{1}{rQ_{s0}}\gg\alpha_{s}y\,.\qquad$ (30) Behavior of the scattering amplitude deeply in the saturation region $rQ_{s}\gg 1$ can be found by noting, that with the logarithmic accuracy, the parent dipole $\mathbf{r}$ tends to split into two daughter dipoles $\mathbf{w}$ and $\mathbf{r}-\mathbf{w}$ of different sizes: either $w\ll r\approx|\mathbf{w}-\mathbf{r}|$ or, symmetrically, $|\mathbf{w}-\mathbf{r}|\ll r\approx w$. Both give equal contribution to the integral over $\mathbf{w}$. Restricting ourself to the case $w\ll r$ and doubling the integral we write the BK equation as follows: $\frac{\partial N(\mathbf{r},\mathbf{b},y)}{\partial y}\approx\frac{\alpha_{s}C_{F}}{\pi}\,2\,\int_{1/Q_{s}^{2}}^{r^{2}}\frac{dw^{2}}{w^{2}}\,[N(\mathbf{w},\mathbf{b},y)-N(\mathbf{w},\mathbf{b},y)N(\mathbf{r},\mathbf{b},y)]\,.$ (31) Now, for the reason that in the saturation region, the amplitude $N(\mathbf{r},\mathbf{b},y)$ is close to unity we render (31) as $-\frac{\partial\\{1-N(\mathbf{r},\mathbf{b},y)\\}}{\partial y}\approx\frac{\alpha_{s}C_{F}}{\pi}\,2\,\int_{1/Q_{s}^{2}}^{r^{2}}\frac{dw^{2}}{w^{2}}\\{1-N(\mathbf{r},\mathbf{b},y)\\}=\frac{2\,\alpha_{s}C_{F}}{\pi}\ln(r^{2}Q_{s}^{2})\,\\{1-N(\mathbf{r},\mathbf{b},y)\\}\,.$ (32) The saturation scale $Q_{s}(y)$ can be found by equating the argument of the exponent in (III.2) to a constant which yields Levin:1999mw ; Bartels:1992ix $Q_{s}(y)\approx Q_{s0}e^{2\bar{\alpha}_{s}y}\,.$ (33) Introducing a new scaling variable $\tau=\ln(r^{2}Q_{s}^{2})$ we solve (32) and find the high energy limit of the forward scattering amplitude (32) Bartels:1992ix ; Levin:1999mw ; Levin:2000mv ; Levin:2001cv (in the fixed coupling approximation). It reads $N(\mathbf{r},\mathbf{b},y)=1-S_{0}\,e^{-\tau^{2}/8}=1-S_{0}\,e^{-\frac{1}{8}\ln^{2}(r^{2}Q_{s}^{2})}\,,\quad r\gg\frac{1}{Q_{s}}\,.$ (34) where we approximated $C_{F}\approx N_{c}/2$ in the large $N_{c}$ limit. $S_{0}$ is the integration constant. It determines the value of the amplitude at the critical line $r(y)=1/Q_{s}(y)$. ## IV Diffractive gluon spectrum: large onium $r>1/Q_{s}$ Now the stage is set for calculation of the diffractive gluon spectrum in various kinematic regions. Let us first analyze the differential gluon production cross section in the case of scattering of large onium $r>1/Q_{s}$. There are two interesting kinematic regions in this case depending on the relation between the gluon transverse momentum $k$ and the saturation scale $Q_{s}$. We consider these two cases separately. ### IV.1 Hard gluons $k>Q_{s}$ To begin we need to calculate the function $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ given by (II). Note, that in the region $w>1/k$ the integrand is a rapidly fluctuation function. Therefore, the dominant contribution to $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ arises from dipole sizes $w<1/k$. Consider now three possible cases. (i) $r^{\prime}<\frac{1}{k}<\frac{1}{Q_{s}}$. In this case splitting the integration region into two parts we write (II) as $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx\int_{0}^{r^{\prime}}\frac{d^{2}w}{w^{2}}\,[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)\,+\,\int_{r^{\prime}}^{1/k}\frac{d^{2}w}{w^{2}}\,2N(\mathbf{w},\mathbf{b},y)\,.$ (35) In the second integral on the r.h.s. we neglected $N(\mathbf{r}^{\prime},\mathbf{b},0)$ as compared to $N(\mathbf{w},\mathbf{b},0)$ since the amplitude is an increasing function of the dipole size and in most of the integration region $w\gg r^{\prime}$. To determine the kinematic region that gives the largest contribution we note that when $w\ll\frac{1}{Q_{s}}$ the amplitude scales as $N(\mathbf{w},\mathbf{b},y)\sim w^{2}$. It follows, that the first integral in the r.h.s. of (35) is of order $r^{\prime 2}Q_{s}^{2}$ whereas the second one is of order $Q_{s}^{2}/k^{2}$, i.e. the former is parametrically smaller than the latter. Thus, $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx\int_{r^{\prime}}^{1/k}\frac{d^{2}w}{w^{2}}\,2N(\mathbf{w},\mathbf{b},y)\,,\quad r^{\prime}<\frac{1}{k}<\frac{1}{Q_{s}}\,.$ (36) We obtain for the gradient $\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y)=-4\pi\frac{\hat{\mathbf{k}}}{k}\,N(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,,\quad r^{\prime}<\frac{1}{k}<\frac{1}{Q_{s}}\,.$ (37) Eq. (37) holds in the logarithmic approximation. Using (15) we get $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=4\,\frac{(4\pi)^{2}}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,,\quad r^{\prime}<\frac{1}{k}<\frac{1}{Q_{s}}\,.$ (38) In the second case (ii) $\frac{1}{k}<r^{\prime}<\frac{1}{Q_{s}}$ and the third case (iii) $\frac{1}{k}<\frac{1}{Q_{s}}<r^{\prime}$ there is only one significant integration region yielding $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,\int_{0}^{1/k}\frac{d^{2}w}{w^{2}}\,N(\mathbf{w},\mathbf{b},y)\,,\quad\frac{1}{k}<r^{\prime}<\frac{1}{Q_{s}}\,\,\mathrm{and}\,\,\frac{1}{k}<\frac{1}{Q_{s}}<r^{\prime}\,.$ (39) Therefore, $\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y)=-2\pi\frac{\hat{\mathbf{k}}}{k}\,N(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,\,,\quad\frac{1}{k}<r^{\prime}<\frac{1}{Q_{s}}\,\,\mathrm{and}\,\,\frac{1}{k}<\frac{1}{Q_{s}}<r^{\prime}\,.$ (40) Substitution into (15) yields $\displaystyle|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=4\,\frac{(2\pi)^{2}}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]^{2}\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,,$ $\displaystyle\quad\frac{1}{k}<r^{\prime}<\frac{1}{Q_{s}}\,\,\mathrm{and}\,\,\frac{1}{k}<\frac{1}{Q_{s}}<r^{\prime}\,.$ (41) To calculate the differential cross section (16) we now need to integrate over all possible dipole sizes $r^{\prime}$ and orientations. Integration over the dipole orientations produces $\displaystyle\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi^{3}}\,S_{A}\,\frac{(4\pi)^{2}}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\left\\{\int_{0}^{1/k}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,(1-J_{0}(k\,r^{\prime}))\right.$ $\displaystyle\left.+\frac{1}{4}\int_{1/k}^{1/Q_{s}}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,(1-J_{0}(k\,r^{\prime}))\right\\}\,.$ (42) We restricted integration over $r^{\prime}$ to the region $r^{\prime}<1/Q_{s}$ since otherwise the integrand is strongly suppressed by $[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]^{2}\to 0$, see (IV.1) with (34) or (5). To determine the largest contribution to the integral on the r.h.s. of (IV.1) we use the fact that the Bessel function $J_{0}(x)\sim x^{-1/2}$ at $x\gg 1$ and $J_{0}(x)\approx 1-x^{2}/4$ at $x\ll 1$ and write the expression in the curly brackets as $\int_{0}^{1/k}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,\frac{1}{4}k^{2}r^{\prime 2}+\frac{1}{4}\int_{1/k}^{1/Q_{s}}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,.$ (43) It follows from (26) and (27) that $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)\sim r^{2}/r^{\prime 4}$ at $r<r^{\prime}$ and $n_{p}(\mathbf{r},\mathbf{r}^{\prime},y)\sim 1/r^{\prime 2}$ at $r>r^{\prime}$. On that account, we determine that the first integral in (43) is of order unity, whereas the second one is logarithmically enhanced by $\ln(k/Q_{s})\gg 1$. A more accurate estimate is gained by substitution of (27) and explicit integration over $r^{\prime}$. Introducing a new integration variable $\zeta=2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln\frac{r}{r^{\prime}}}$ we have $\int_{1/k}^{1/Q_{s}}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\approx\frac{\sqrt{2}}{4\pi^{3/2}}\,\int_{\zeta_{*}}^{\zeta_{0}}\,\frac{d\zeta}{\sqrt{\zeta}}\,e^{\zeta}=\frac{\sqrt{2\pi}}{4\pi^{3/2}}\,\left[\mathrm{erfi}\left(\sqrt{\zeta_{0}}\right)-\mathrm{erfi}\left(\sqrt{\zeta_{*}}\right)\right]\,,$ (44) where $\mathrm{erfi}(z)$ is the imaginary error function defined as $\mathrm{erfi}(z)=-i\,\mathrm{erf}(iz)\,,$ (45) $\zeta_{0}=2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rk)}$ and $\zeta_{*}=2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rQ_{s})}$. In compliance with the double logarithmic approximation we must replace the imaginary error function by its asymptotic form at $\zeta_{0}\gg 1$ given by $\mathrm{erfi}(z)\approx\frac{1}{\sqrt{\pi}\,z}\,e^{z^{2}}\,,\quad z\gg 1\,.$ (46) Hence, keeping in mind that $\zeta_{0}\gg\zeta_{*}$ we get $\int_{1/k}^{1/Q_{s}}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\approx\frac{1}{4\pi^{3/2}}\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln(rk)\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rk)}}\,.$ (47) As expected, this integral is independent of $Q_{s}$ since the integrand is a steeply increasing function of $\frac{1}{r^{\prime}}$. Finally, the cross section is procured by plugging (47) into (IV.1) $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi^{5/2}}\,\frac{1}{k^{2}}\,S_{A}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln(rk)\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rk)}}\,,\quad r>\frac{1}{Q_{s}}>\frac{1}{k}\,.$ (48) ### IV.2 Soft gluons $k<Q_{s}$ We are now turning to analysis of soft gluon production by large onium. As in the case of hard gluons we wish to calculate $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ in three different cases. First case corresponds to (i) $r^{\prime}<\frac{1}{Q_{s}}<\frac{1}{k}$, i.e. size of dipole emitting the triggered gluon is smaller than any other scale in the problem. We have $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx\int_{0}^{r^{\prime}}\frac{d^{2}w}{w^{2}}[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)+\int_{r^{\prime}}^{1/Q_{s}}\frac{d^{2}w}{w^{2}}\,2\,N(\mathbf{w},\mathbf{b},y)+\int_{1/Q_{s}}^{1/k}\frac{d^{2}w}{w^{2}}\,,$ (49) where we used the properties of the amplitude $N(\mathbf{w},\mathbf{b},y)$ as discussed after (35). The three integrals on the r.h.s. of (49) is of order $r^{\prime 2}Q_{s}^{2}\ll 1$, 1 and $\ln\frac{Q_{s}}{k}\gg 1$ respectively. Evidently, the third one is dominating. Thus, $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx 2\pi\ln\frac{Q_{s}}{k}\,$ (50) implying that $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=\frac{4(2\pi)^{2}}{k^{2}}\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,,\quad r^{\prime}<\frac{1}{Q_{s}}<\frac{1}{k}\,.$ (51) In the second case (ii) $\frac{1}{Q_{s}}<r^{\prime}<\frac{1}{k}$ there are also three relevant regions of integration $\displaystyle Q(\mathbf{r}^{\prime},\mathbf{k},y)$ $\displaystyle\approx$ $\displaystyle\int_{0}^{1/Q_{s}}\frac{d^{2}w}{w^{2}}[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)+\int_{1/Q_{s}}^{r^{\prime}}\frac{d^{2}w}{w^{2}}[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)$ (52) $\displaystyle+\int_{r^{\prime}}^{1/k}\frac{d^{2}w}{w^{2}}[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)\,.$ In the second and the third integral $N(\mathbf{w},\mathbf{b},y)\approx 1$. The third integral is enhanced by $\ln\frac{1}{r^{\prime}k}$ and anyway it is the only integral that depends on $k$. Therefore, using (34) $\nabla_{\mathbf{k}}Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx-2\pi\frac{\hat{\mathbf{k}}}{k}\,[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]=-2\pi\frac{\hat{\mathbf{k}}}{k}\,S_{0}\,e^{-\frac{1}{8}\ln^{2}(Q_{s}^{2}r^{\prime 2})}\,.$ (53) Consequently, $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=\frac{4(2\pi)^{2}}{k^{2}}\,S_{0}^{2}\,e^{-\frac{1}{4}\ln^{2}(Q_{s}^{2}r^{\prime 2})}\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,,\quad\frac{1}{Q_{s}}<r^{\prime}<\frac{1}{k}\,.$ (54) The third case corresponds to (iii) $\frac{1}{Q_{s}}<\frac{1}{k}<r^{\prime}$. There are now two relevant regions $Q(\mathbf{r}^{\prime},\mathbf{k},y)\approx\int_{0}^{1/Q_{s}}\frac{d^{2}w}{w^{2}}[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)+\int_{1/Q_{s}}^{1/k}\frac{d^{2}w}{w^{2}}\,[1-N(\mathbf{r}^{\prime},\mathbf{b},y)]\,N(\mathbf{w},\mathbf{b},y)\,.$ (55) The second integral is enhanced by $\ln\frac{Q_{s}}{k}$ and, apart from the lower limit of integration, is the same as the third integral in (52). Evidently, the $k$ dependence of $Q(\mathbf{r}^{\prime},\mathbf{k},y)$ is the same as in the case (ii), implying that (54) holds in the case (iii) as well $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}=\frac{4(2\pi)^{2}}{k^{2}}\,S_{0}^{2}\,e^{-\frac{1}{4}\ln^{2}(Q_{s}^{2}r^{\prime 2})}\,\sin^{2}\left(\frac{\mathbf{k}\cdot\mathbf{r}^{\prime}}{2}\right)\,,\quad\frac{1}{Q_{s}}<\frac{1}{k}<r^{\prime}\,.$ (56) Essentially, what (54) and (56) tell us is that the region $r^{\prime}>\frac{1}{Q_{s}}$ does not contribute to the cross section for diffractive production of soft gluon by large onium. Thus, the only contribution to the cross section stems from $r^{\prime}<\frac{1}{Q_{s}}$. There are now two possibilities depending on the size $r$ of the incident onium: (a) $r>\frac{1}{k}>\frac{1}{Q_{s}}$ and (b) $\frac{1}{k}>r>\frac{1}{Q_{s}}$. However, in both cases $\frac{1}{Q_{s}}$ is the smallest size implying that the leading contribution to the cross section is the same in both cases. Expanding the argument of sinus in (51) and substituting to (16) we have $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi^{2}}\frac{1}{(2\pi)^{2}}\,S_{A}\,\int_{0}^{1/Q_{s}}d^{2}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,\frac{4(2\pi)^{2}}{k^{2}}\frac{1}{4}(\mathbf{k}\cdot\mathbf{r}^{\prime})^{2}\,.$ (57) The dipole density is given by (27). Notice that since the largest contribution to the integral arises from dipoles of size $r^{\prime}\sim\frac{1}{Q_{s}}$ (the integrand increase rapidly with $r^{\prime}$) we can approximate $\ln\frac{r}{r^{\prime}}\approx\ln(rQ_{s})$, neglecting contribution of very small dipole sizes $r^{\prime}$. Thus $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{8\pi^{5/2}}\,\frac{S_{A}}{Q_{s}^{2}}\,\frac{(2\bar{\alpha}_{s}(Y-y))^{1/4}}{\ln^{3/4}(rQ_{s})}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rQ_{s})}}\,,\quad r,\frac{1}{k}>\frac{1}{Q_{s}}\,,$ (58) which holds for any relation between $r$ and $1/k$. If we now wish to calculate the total cross section for diffractive gluon production at given rapidity $y$ we have to integrate (48) and (58) over $d^{2}k$. Clearly, the leading contribution stems from the integral over soft gluons given by (58). We attain $\frac{d\sigma}{dy}=\frac{\alpha_{s}C_{F}}{8\pi^{3/2}}\,S_{A}\,\frac{(2\bar{\alpha}_{s}(Y-y))^{1/4}}{\ln^{3/4}(rQ_{s})}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rQ_{s})}}\,,\quad r>1/Q_{s}\,,$ (59) in complete agreement with the result obtained in our previous paper Li:2008bm . ## V Diffractive gluon spectrum: Small onium $r<1/Q_{s}$ We now consider scattering of small onium on a heavy nucleus. We will again consider separately the two cases of hard and soft gluons. Calculation are facilitated a lot since we have already derived the function $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}$, which embodies information about the gluon emission and subsequent elastic scattering of the two intermediate dipoles $\mathbf{w}$ and $\mathbf{r}^{\prime}-\mathbf{w}$ off the nucleus. ### V.1 Hard gluons $k>Q_{s}$ Using (38) we obtain $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi}\,\frac{1}{(2\pi)^{2}}\,S_{A}\,\frac{4\,(4\pi)^{2}}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\int_{0}^{\infty}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,(1-J_{0}(kr^{\prime}))\,,$ (60) where we integrated over orientation of the dipole $\mathbf{r}^{\prime}$. To proceed we have to specify the relationship between the onium size $r$ and the inverse gluon transverse momentum $\frac{1}{k}$. Assume that (a) $r<\frac{1}{k}<\frac{1}{Q_{s}}$. Then, integral over $r^{\prime}$ can be divided into the following four regions: (i) $0<r^{\prime}<r$, (ii) $r<r^{\prime}<\frac{1}{k}$, (iii) $\frac{1}{k}<r^{\prime}<\frac{1}{Q_{s}}$ and (iv) $\frac{1}{Q_{s}}<r^{\prime}$. To estimate the integral in each of this regions we use the same procedure as before (it is explained after (43)). We find the following parametric dependence of the integral in these four regions: (i) $k^{2}r^{2}$, (ii) $k^{2}r^{2}\ln\frac{1}{kr}$, (iii) $k^{2}r^{2}$ and (iv) $r^{2}Q_{s}^{2}$. Region (ii) gives the largest contribution. We have $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi}\,\frac{1}{(2\pi)^{2}}\,S_{A}\,\frac{4\,(4\pi)^{2}}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\frac{k^{2}}{4}\,\int_{0}^{\infty}dr^{\prime}r^{\prime 3}\,n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,,$ (61) Upon substitution of (26) and changing to a new integration variable $\tilde{\zeta}=2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln\frac{r^{\prime}}{r}}$ we reduce the integral over $r^{\prime}$ to the imaginary error function as in (44). Following the same steps as those that led us to (47) we derive $\frac{d\sigma}{d^{2}k\,dy}=\frac{\alpha_{s}C_{F}}{\pi^{5/2}}\,S_{A}\,r^{2}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{kr}\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{kr}}}\,,\quad r<\frac{1}{k}<\frac{1}{Q_{s}}\,.$ (62) Consider region (b) $\frac{1}{k}<r<\frac{1}{Q_{s}}$. Repeating the same analysis as above we conclude that the dominant logarithmic contribution originates from the region $\frac{1}{k}<r^{\prime}<r$. Accordingly, we use (27) for the dipole density and neglect the Bessel function in (60). Doing the integral as explained in (44)–(47) we write $\frac{d\sigma}{d^{2}k\,dy}=\frac{\,\alpha_{s}C_{F}}{\pi^{5/2}}\,S_{A}\,\frac{1}{k^{2}}\,N^{2}(k^{-1}\hat{\mathbf{k}},\mathbf{b},y)\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln(rk)\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln(rk)}}\,,\quad\frac{1}{k}<r<\frac{1}{Q_{s}}\,.$ (63) ### V.2 Soft gluons $k<Q_{s}$ In the case $r<\frac{1}{Q_{s}}<\frac{1}{k}$ formulas for $|\mathbf{I}(\mathbf{r}^{\prime},\mathbf{k},y)|^{2}$ are given by (51),(54),(56). As was already mentioned, only small dipoles $r^{\prime}<1/Q_{s}$ contribute in the $r^{\prime}$ integral. We thus have two regions of integration: (i) $0<r^{\prime}<r$ and (ii) $r<r^{\prime}<\frac{1}{Q_{s}}$. The integral over the former is of order $r^{2}$ whereas over the latter it is of order $r^{2}\ln\frac{1}{rQ_{s}}$. That being the case we derive $\displaystyle\frac{d\sigma}{d^{2}k\,dy}$ $\displaystyle=$ $\displaystyle\frac{\alpha_{s}C_{F}}{\pi}\frac{1}{(2\pi)^{2}}\,S_{A}\,\int_{r}^{1/Q_{s}}dr^{\prime}r^{\prime}n_{p}(\mathbf{r},\mathbf{r}^{\prime},Y-y)\,\frac{4\,(2\pi)^{2}}{k^{2}}\,\frac{k^{2}\,r^{\prime 2}}{4}$ (64) $\displaystyle=$ $\displaystyle\frac{\alpha_{s}C_{F}}{4\pi^{5/2}}\,S_{A}\,r^{2}\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{rQ_{s}}\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{rQ_{s}}}}\,,\quad r<\frac{1}{Q_{s}}<\frac{1}{k}\,.$ (65) The total cross section is again dominated by soft gluons. Integrating (64) over $d^{2}k$ such that $k<Q_{s}$ we find $\frac{d\sigma}{dy}=\frac{\alpha_{s}C_{F}}{4\pi^{3/2}}\,S_{A}\,Q_{s}^{2}\,r^{2}\,\frac{1}{\left(2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{rQ_{s}}\right)^{1/4}}\,e^{2\sqrt{2\bar{\alpha}_{s}(Y-y)\ln\frac{1}{rQ_{s}}}}\,,\quad r<\frac{1}{Q_{s}}\,,$ (66) again in agreement with our previous result Li:2008bm . ## VI Summary The differential cross section for diffractive gluon production is given by formulas (48), (58), (62), (63) and (65). We can see that there are five distinct kinematic regions, which are really six. The behavior of gluon spectrum in these regions is sketched in Fig. 4. | ---|--- Figure 4: Sketch of diffractive gluon spectrum as a function of transverse momentum $k$ in two cases $r>Q_{s}^{-1}$ and $r<Q_{s}^{-1}$. Qualitative behavior in both regions is also indicated. $xG(x=e^{y-Y})$ is a “gluon distribution function” in onium, $\gamma$ is the anomalous dimension of the nuclear gluon distribution function and $N_{\infty}$ is a normalization constant. To make the figure self-contained we indicated an approximate transverse momentum $k$ and the saturation scale $Q_{s}$ dependence. $\gamma$ denotes the anomalous dimension of the nuclear gluon distribution. It varies from about unity at $k\gg Q_{s}^{2}/Q_{s0}$ to $\gamma\approx 1/2$ at $Q_{s}\lesssim k<Q_{s}^{2}/Q_{s0}$. Fig. 4 teaches us that by varying the incident onium size with respect to the saturation scale we obtain different behavior of the gluon spectrum as a function of transverse momentum. In DIS the typical onium size can be varied by means of triggering on the events with different photon virtuality. Depending on the relation between $k$, $Q_{s}$ and $r$ the gluon spectrum exhibits different pattern that allows a more direct measurement of the saturation scale $Q_{s}(y)$ (and hence the nuclear gluon density) than it is possible nowadays. The $k$-dependence of hadron spectra is of course significantly modified by the fragmentation process. On the other hand, dependence of hadron spectra on atomic number $A$ is the same as for the gluon spectrum (it arises from the $A$-dependence of $Q_{s}$, see (6), (33)). The reason is that, as we explained in Introduction, the coherence length for the gluon production is much larger than the nucleus size, implying that the fragmentation process is independent of $A$. Consequently, $A$-dependence is a powerful tool in studying the nuclear gluon distribution. Likewise, energy/rapidity dependence is independent of details of fragmentation (see however Li:2007zzc ) and has been successfully used along with $A$-dependence for analysis of inclusive hadron production at RHIC. Therefore, energy/rapidity and atomic number dependence at different values of hadron transverse momenta allows access to information about the anomalous dimension $\gamma$, which is of crucial importance for understanding the transition region between the region of gluon saturation and the hard perturbative QCD. Similar arguments apply to the diffractive gluon production in pA collisions. In this case, however, there is a substantial uncertainty regarding the structure of the proton wave function. Diffractive gluon production in the case when the distance between the three pairs of valence quarks is about the same is strongly suppressed as compared to the case when the distance between one pair of quarks is much smaller than the distance between the other two pairs (quark - diquark configuration), see Li:2008bm . In either case the $A$ and energy dependence are given by Fig. 4 (right or left panel). Since calculation of diffractive gluon production in pA collisions requires a substantial modeling of the proton wave function we intend to address it in a separate publication. To summarize, we calculated the spectrum of diffractively produced gluons at low-$x$ in onium–heavy nucleus collisions. In the forthcoming publications we are going to apply our results for calculation of the diffractive gluon production in DIS and pA collisions. ###### Acknowledgements. We would like to thank Yuri Kovchegov, Genya Levin and Jianwei Qiu for many informative discussions. The work of K.T. was supported in part by the U.S. Department of Energy under Grant No. DE-FG02-87ER40371. He would like to thank RIKEN, BNL, and the U.S. Department of Energy (Contract No. DE-AC02-98CH10886) for providing facilities essential for the completion of this work. ## References * (1) L. V. Gribov, E. M. Levin, and M. G. Ryskin, Phys. Rept. 100, 1 (1983). * (2) A. H. Mueller and J. w. Qiu, Nucl. Phys. B 268, 427 (1986). * (3) L. D. McLerran and R. Venugopalan, Phys. Rev. D 49, 2233 (1994) [arXiv:hep-ph/9309289]; Phys. Rev. D 49, 3352 (1994) [arXiv:hep-ph/9311205]; Phys. Rev. D 50, 2225 (1994) [arXiv:hep-ph/9402335]. * (4) J. Jalilian-Marian, A. Kovner, A. Leonidov, and H. Weigert, Nucl. Phys. B 504, 415 (1997) [arXiv:hep-ph/9701284]; * (5) J. Jalilian-Marian, A. Kovner, A. Leonidov, and H. Weigert, Phys. Rev. D 59, 014014 (1999) [arXiv:hep-ph/9706377]; * (6) J. Jalilian-Marian, A. Kovner, A. Leonidov, and H. Weigert, Phys. Rev. D 59, 034007 (1999) [Erratum-ibid. D 59, 099903 (1999)] [arXiv:hep-ph/9807462]; * (7) J. Jalilian-Marian, A. Kovner, and H. Weigert, Phys. Rev. D 59, 014015 (1999) [arXiv:hep-ph/9709432]; * (8) A. Kovner, J. G. Milhano, and H. Weigert, Phys. Rev. D 62, 114005 (2000) [arXiv:hep-ph/0004014]; H. Weigert, Nucl. Phys. A 703, 823 (2002) [arXiv:hep-ph/0004044]. * (9) E. Iancu, A. Leonidov, and L. D. McLerran, Nucl. Phys. A 692, 583 (2001) [arXiv:hep-ph/0011241]; * (10) E. Iancu, A. Leonidov, and L. D. McLerran, Phys. Lett. B 510, 133 (2001) [arXiv:hep-ph/0102009]; * (11) E. Iancu and L. D. McLerran, Phys. Lett. B 510, 145 (2001) [arXiv:hep-ph/0103032]; * (12) E. Ferreiro, E. Iancu, A. Leonidov, and L. McLerran, Nucl. Phys. A 703, 489 (2002) [arXiv:hep-ph/0109115]. * (13) K. J. Golec-Biernat and M. Wusthoff, Phys. Rev. D 59, 014017 (1999) [arXiv:hep-ph/9807513]. * (14) K. J. Golec-Biernat and M. Wusthoff, Phys. Rev. D 60, 114023 (1999) [arXiv:hep-ph/9903358]. * (15) E. Gotsman, E. Levin, M. Lublinsky, U. Maor, and K. Tuchin, arXiv:hep-ph/0007261. * (16) E. Gotsman, E. Levin, M. Lublinsky, U. Maor, and K. Tuchin, Nucl. Phys. A697, 521 (2002). * (17) Y. V. Kovchegov and L. D. McLerran, Phys. Rev. D 60, 054025 (1999) [Erratum-ibid. D 62, 019901 (2000)] [arXiv:hep-ph/9903246]. * (18) E. Levin and M. Lublinsky, Nucl. Phys. A 712, 95 (2002) [arXiv:hep-ph/0207374]. * (19) E. Levin and M. Lublinsky, Eur. Phys. J. C 22, 647 (2002) [arXiv:hep-ph/0108239]. * (20) J. Bartels, K. J. Golec-Biernat, and H. Kowalski, Phys. Rev. D 66, 014001 (2002) [arXiv:hep-ph/0203258]. * (21) E. Gotsman, E. Levin, M. Lublinsky, U. Maor and K. Tuchin, Phys. Lett. B 492, 47 (2000) [arXiv:hep-ph/9911270]. * (22) D. Kharzeev and M. Nardi, Phys. Lett. B 507, 121 (2001) [arXiv:nucl-th/0012025]; * (23) D. Kharzeev and E. Levin, Phys. Lett. B 523, 79 (2001) [arXiv:nucl-th/0108006]; * (24) D. Kharzeev, E. Levin, and M. Nardi, arXiv:hep-ph/0111315. * (25) D. Kharzeev, E. Levin, and M. Nardi, Nucl. Phys. A 730, 448 (2004) [Erratum-ibid. A 743, 329 (2004)] [arXiv:hep-ph/0212316]. * (26) Y. V. Kovchegov and A. H. Mueller, Nucl. Phys. B 529, 451 (1998). * (27) Y. V. Kovchegov and K. Tuchin, Phys. Rev. D 65, 074026 (2002). * (28) M. A. Braun, Phys. Lett. B 483, 105 (2000) [arXiv:hep-ph/0003003]. * (29) A. Dumitru and L. D. McLerran, Nucl. Phys. A 700, 492 (2002) [arXiv:hep-ph/0105268]. * (30) J. P. Blaizot, F. Gelis, and R. Venugopalan, Nucl. Phys. A 743, 13 (2004) [arXiv:hep-ph/0402256]. * (31) D. Kharzeev, E. Levin, and L. McLerran, Phys. Lett. B 561, 93 (2003) [arXiv:hep-ph/0210332]. * (32) D. Kharzeev, Y. V. Kovchegov, and K. Tuchin, Phys. Rev. D 68, 094013 (2003) [arXiv:hep-ph/0307037]. * (33) D. Kharzeev, Y. V. Kovchegov, and K. Tuchin, Phys. Lett. B 599, 23 (2004) [arXiv:hep-ph/0405045]. * (34) R. Baier, A. Kovner, and U. A. Wiedemann, Phys. Rev. D 68, 054009 (2003) [arXiv:hep-ph/0305265]. * (35) E. Iancu, K. Itakura, and D. N. Triantafyllopoulos, Nucl. Phys. A 742, 182 (2004) [arXiv:hep-ph/0403103]. * (36) F. Gelis and R. Venugopalan, Phys. Rev. D 69, 014019 (2004) [arXiv:hep-ph/0310090]. * (37) K. Tuchin, Phys. Lett. B 593, 66 (2004) [arXiv:hep-ph/0401022]. * (38) J. P. Blaizot, F. Gelis, and R. Venugopalan, Nucl. Phys. A 743, 57 (2004) [arXiv:hep-ph/0402257]. * (39) Y. V. Kovchegov and K. Tuchin, Phys. Rev. D 74, 054014 (2006) [arXiv:hep-ph/0603055]. * (40) K. Tuchin, Nucl. Phys. A 798, 61 (2008) [arXiv:0705.2193 [hep-ph]]. * (41) D. Kharzeev and K. Tuchin, Nucl. Phys. A 735, 248 (2004) [arXiv:hep-ph/0310358]. * (42) K. Tuchin, Nucl. Phys. A783, 173 (2007) [arXiv:hep-ph/0609258]. * (43) D. Kharzeev and K. Tuchin, Nucl. Phys. A 770, 40 (2006) [arXiv:hep-ph/0510358]. * (44) Y. Li and K. Tuchin, arXiv:0802.2954 [hep-ph]. * (45) M. Wusthoff, Phys. Rev. D 56, 4311 (1997) [arXiv:hep-ph/9702201]. * (46) J. Bartels, H. Jung, and M. Wusthoff, Eur. Phys. J. C 11, 111 (1999) [arXiv:hep-ph/9903265]. * (47) B. Z. Kopeliovich, A. Schafer, and A. V. Tarasov, Phys. Rev. D 62, 054022 (2000) [arXiv:hep-ph/9908245]. * (48) Y. V. Kovchegov, Phys. Rev. D 64, 114016 (2001) [Erratum-ibid. D 68, 039901 (2003)] [arXiv:hep-ph/0107256]. * (49) K. J. Golec-Biernat and C. Marquet, Phys. Rev. D 71, 114005 (2005) [arXiv:hep-ph/0504214]. * (50) C. Marquet, Nucl. Phys. B 705, 319 (2005) [arXiv:hep-ph/0409023]. * (51) C. Marquet, Phys. Rev. D 76, 094017 (2007) [arXiv:0706.2682 [hep-ph]]. * (52) S. Munier and A. Shoshi, Phys. Rev. D 69, 074022 (2004) [arXiv:hep-ph/0312022]. * (53) A. Kovner, M. Lublinsky, and H. Weigert, Phys. Rev. D 74, 114023 (2006) [arXiv:hep-ph/0608258]. * (54) A. Kovner and U. A. Wiedemann, Phys. Rev. D 64, 114002 (2001) [arXiv:hep-ph/0106240]. * (55) A.H. Mueller, Nucl. Phys. B415, 373 (1994); A.H. Mueller and B. Patel, Nucl. Phys. B425, 471 (1994); A.H. Mueller, Nucl. Phys. B437, 107 (1995). * (56) E. A. Kuraev, L. N. Lipatov, and V. S. Fadin, Sov. Phys. JETP 45, 199 (1977) [Zh. Eksp. Teor. Fiz. 72, 377 (1977)]. * (57) I. I. Balitsky and L. N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822 [Yad. Fiz. 28 (1978) 1597]. * (58) I. Balitsky, Nucl. Phys. B 463, 99 (1996) [arXiv:hep-ph/9509348]. * (59) Y. V. Kovchegov, Phys. Rev. D 60, 034008 (1999) [arXiv:hep-ph/9901281]. * (60) A.H. Mueller, Nucl. Phys. B335, 115 (1990). * (61) J. Jalilian-Marian and Y. V. Kovchegov, Phys. Rev. D 70, 114017 (2004) [Erratum-ibid. D 71, 079901 (2005)] [arXiv:hep-ph/0405266]. * (62) Y. V. Kovchegov and E. Levin, Nucl. Phys. B 577, 221 (2000) [arXiv:hep-ph/9911523]. * (63) J. Bartels and E. Levin, Nucl. Phys. B 387, 617 (1992). * (64) E. Levin and K. Tuchin, Nucl. Phys. B 573, 833 (2000) [arXiv:hep-ph/9908317]. * (65) E. Levin and K. Tuchin, Nucl. Phys. A 691, 779 (2001) [arXiv:hep-ph/0012167]. * (66) E. Levin and K. Tuchin, Nucl. Phys. A 693, 787 (2001) [arXiv:hep-ph/0101275]. * (67) Y. Li and K. Tuchin, Phys. Rev. D 75, 074022 (2007) [arXiv:hep-ph/0702208].

arxiv-papers

2024-09-04T02:48:54.274593

0803.1619

eurm10 msam10 1 # Massive Star Formation in the Galactic Center Don F.Figer Rochester Institute of Technology, Rochester, NY, USA (2006) ###### Abstract The Galactic center is a hotbed of star formation activity, containing the most massive star formation site and three of the most massive young star clusters in the Galaxy. Given such a rich environment, it contains more stars with initial masses above 100 M⊙ than anywhere else in the Galaxy. This review concerns the young stellar population in the Galactic center, as it relates to massive star formation in the region. The sample includes stars in the three massive stellar clusters, the population of younger stars in the present sites of star formation, the stars surrounding the central black hole, and the bulk of the stars in the field population. The fossil record in the Galactic center suggests that the recently formed massive stars there are present-day examples of similar populations that must have been formed through star formation episodes stretching back to the time period when the Galaxy was forming. ††volume: 1 ## 1 Introduction The Galactic center (GC) is an exceptional region for testing massive star formation and evolution models. It contains 10% of the present star formation activity in the Galaxy, yet fills only a tiny fraction of a percent of the volume in the Galactic disk111For the purposes of this review, the GC refers to a cylindrical volume with radius of $\approx$500 pc and thickness of $\approx$60 pc that is centered on the Galactic nucleus and is coincident with a region of increased dust and gas density, often referred to as the “Central Molecular Zone” (Serabyn & Morris, 1996).. The initial conditions for star formation in the GC are unique in the Galaxy. The molecular clouds in the region are extraordinarly dense, under high thermal pressure, and are subject to a strong gravitational tidal field. Morris (1993) argue that these conditions may favor the preferential formation of high mass stars. Being the closest galactic nucleus, the GC gives us an opportunity to observe processes that potentially have wide applicability in other galaxies, both in their centers and in the interaction regions of merging galaxies. Finally, the GC may be the richest site of certain exotic processes and objects in the Galaxy, i.e. runaway stellar mergers leading to intermediate mass black holes and stellar rejuvination through atmospheric stripping, to name a few. This review is primarily concerned with massive star formation in the region. For thorough reviews on a variety of topics concerning the Galactic center, see Genzel & Townes (1987), Genzel et al. (1994), Morris & Serabyn (1996), and Eckart et al. (2005). ## 2 The Galactic center environment and star formation The star formation efficiency in the GC appears to be high. Plotting the surface star formation rate ($\Sigma_{\rm SFR}\sim$5 M⊙ yr-1 pc-2) versus surface gas density ($\Sigma_{H_{2}}\sim$400 M⊙ pc-2) in a “Schmidt plot” suggests an efficiency of nearly 100%, comparable to that of the most intense infrared circumnuclear starbursts in other galaxies and a factor of twenty higher than in typical galaxies (see Figure 7 in Kennicutt, 1998). It is also higher than that elsewhere in the Galaxy; commensurately, stars in the GC emit about 5-10% of the Galaxy’s ionizing radiation and infrared luminosity. Morris & Serabyn (1996) review the content and conditions of the interstellar medium in the “Central Molecular Zone” (CMZ), noting that the molecular clouds in the region are extraordinarily dense ($n>10^{4}~{}cm^{-3}$) and warm ($T\sim 70~{}K$) with respect to those found in the disk of the Galaxy. Stark et al. (1989) argue that the density and internal velocities of clouds in the GC are a direct result of the strong tidal fields in the region, i.e. only the dense survive. Serabyn & Morris (1996) argue that the inexorable inflow of molecular material from further out in the Galaxy powers continuous and robust star formation activity in the region. It is still unclear how magnetic field strength affects star formation. If it does matter, then the GC might be expected to reveal such effects. The strength of the magnetic field in the GC has been estimated through far infrared polarized light from aligned dust grains (Hildebrand et al., 1993; Chuss et al., 2003) and Zeeman splitting of the OH molecule (Plante, Lo, & Crutcher, 1995). In both cases, the field is inferred to be of milliGuass strength. However, Uchida & Guesten (1995) argue strongly that these strengths are localized to bundles that delineate the extraordinary non-thermal filaments in the region (Yusef-Zadeh & Morris, 1987), and are not representative of the field strength that is pervasive in the region. If this is correct, then the fields inside GC molecular clouds may not be so strong versus those inside disk clouds ($B\sim 3~{}\mu$G). Metals in molecular clouds can provide cooling that aids protostellar collapse, but they also create opacity to the UV flux, winds, and bipolar outflows that emanate from newly formed stars. Measurements of metallicty in the Galactic center span a range of solar, observed in stars (Ramírez et al., 2000; Carr, Sellgren, & Balachandran, 2000; Najarro et al., 2004), to twice solar, observed in the gas phase (Shields & Ferland, 1994), to four times solar, observed through x-ray emission near the very center (Maeda et al., 2002). The errors from the stellar measurements are the smallest and suggest that stars in the GC are formed from material with roughly solar abundances. ## 3 Present-day star formation in the GC Present-day star formation in the GC is somewhat subdued compared to the episodes that produced the massive clusters we now see. A dozen or so ultra- compact HII regions are distributed throughout the central 50 pc, each containing one or a few O-stars still embedded in their natal environs. Yusef- Zadeh & Morris (1987) identify most of these sources in radio continuum observations (see Figure 1). Zhao et al. (1993) and Goss et al. (1985) infer lyman-continuum fluxes that are comparable to that expected from a single O7V star in each of the H1-H5 and A-D UCHII regions. Cotera et al. (1999) find that several of the recently formed stars in these regions have broken out of their dust shroud, revealing spectra of young massive stars; see also Figer et al. (1994) and Muno et al. (2006) for additional examples. Figure 1: Radio emission from the GC region at 6 cm, adapted in Figure 1 by Cotera et al. (1999) from Yusef-Zadeh & Morris (1987). The star symbols represent the three massive clusters. Hot stars in the Quintuplet and Arches clusters ionize gas on the surfaces of nearby molecular clouds to produce the radio emission in the “Sickle” and “G0.10+0.02/E1/E2 Filaments,” respectively. The radio emission near the Galactic center is due to a combination of thermal and non-thermal emission. The “H1-8” and “A-D” regions are ultra-compact HII regions surrounding recently formed stars. A bit further from the GC, the Sgr B2 molecular cloud harbors a massive star cluster in the making and is home to the most intense present-day star formation site in the Galaxy (Gaume et al., 1995; de Pree et al., 1995; McGrath, Goss, & De Pree, 2004; Takagi, Murakami, & Koyama, 2002; de Vicente et al., 2000; Liu & Snyder, 1999; Garay & Lizano, 1999; de Pree et al., 1996). Within the next few Myr, this activity should produce a star cluster that is comparable in mass to the Arches cluster (see Figure 2). Sato et al. (2000) note evidence in support of a cloud-cloud collision as the origin for the intense star formation in Sgr B2; these include velocity gradients, magnetic field morphology, shock-enhanced molecular emission, shock-induced molecular evaporation from dust grains, and distinctly different densities of certain molecular species throughout the cloud. Figure 2: Figure 3 from McGrath, Goss, & De Pree (2004) showing H2O and OH masers overplotted on 7 mm contours for a small portion of the Sgr B2 cloud. The activity in this region is typical of that found near the fifty or so ultra-compact HII regions in Sgr B2. ## 4 Continuous star formation in the GC There is ample evidence for persistent star formation in the GC in the form of upper-tip asymptotic giant branch stars distributed throughout the region (Lebofsky & Rieke, 1987; Narayanan, Gould, & Depoy, 1996; Frogel, Tiede, & Kuchinski, 1999; Sjouwerman et al., 1999). Figure 3 shows a plot for some of these stars, based on spectroscopic data, overlaid with intermediate age model isochrones (Blum et al., 2003). Note that the giants and supergiants in this plot require ages that span a few Myr to a few Gyr. One comes to similar conclusions by analyzing photometry of the field population in the GC. Figer et al. (2004) use observed luminosity functions to determine that the star formation rate has been roughly constant for the lifetime of the Galaxy in the GC, similar to the suggestion in Serabyn & Morris (1996) based on the sharp increase in unresolved infrared light towards the center and a mass-budget argument. Figure 4 shows model and observed luminosity functions (right) for various star formation scenarios (left) over the lifetime of the Galaxy, assuming a Salpeter IMF (Salpeter, 1955) for masses above 10 M⊙, and a flat slope below this mass. The observations were obtained with HST/NICMOS and have been corrected for incompleteness. The “burst” models (panels 1, 2, 4, and 5) produce unrealistic ratios of bright to faint stars in the luminosity functions, especially for the red clump near a dereddened K-band magnitude of 12. The continuous star formation model (panel 3) best fits the data. Figure 3: Estimates of absolute magnitude versus temperature for stars in the GC from Blum et al. (2003). The lines correspond to model isochrones having ages of 10 Myr, 100 Myr, 1 Gyr, 5 Gyr, and 12 Gyr. The supergiants (above the horizontal line) are descendant from stars having M$\approx$15-25 M⊙, whereas fainter stars are descendant from lower mass main sequence stars having a few to 15 M⊙. The presence of these stars in the GC demonstrates intermediate age star formation of massive stars. Figure 4: A figure adapted from Figer et al. (2004) showing various star formation scenarios (left), and resultant model luminosity functions (right, thick) compared to observed luminosity functions (right, thin) in the GC. The models assume a Salpeter IMF slope, an elevated lower-mass turnover of 10 M⊙, and are additionally constrained to produce 2(108) M⊙ in stars within the region. The observations have been corrected for incompleteness. The third panels from the top, i.e. continuous star formation, best fit the data. The observed turn-down at the faint end appears to be real and is only well fit only by assuming a very high lower mass turnover. ## 5 Properties of the Three Massive Clusters The majority of recent star formation activity in the GC over the past 10 Myr produced three massive clusters: the Central cluster, the Arches cluster, and the Quintuplet cluster. The following sections describe the stellar content in the clusters and the resultant implications for star formation in the region. They closely follow recent reviews (Figer et al., 1999a; Figer, 2003, 2004), with updates, as summarized in Table 1. Table 1: Properties of massive clusters in the Galactic Center 000“M1” is the total cluster mass in observed stars. “M2” is the total cluster mass in all stars extrapolated down to a lower-mass cutoff of 1 M⊙, assuming a Salpeter IMF slope and an upper mass cutoff of 120 M⊙(unless otherwise noted) “Radius” gives the average projected separation from the centroid position. “$\rho 1$” is M1 divided by the volume. “$\rho 2$” is M2 divided by the volume. In either case, this is probably closer to the central density than the average density because the mass is for the whole cluster while the radius is the average projected radius. “Age” is the assumed age for the cluster. “Luminosity” gives the total measured luminosity for observed stars. “Q” is the estimated Lyman continuum flux emitted by the cluster. | Log(M1) | Log(M2) | Radius | Log($\rho 1$) | Log($\rho 2$) | Age | Log(L) | Log(Q) ---|---|---|---|---|---|---|---|--- Cluster | M⊙ | M⊙ | pc | M⊙ pc-3 | M⊙ pc-3 | Myr | L⊙ | s-1 Quintuplet | 3.0 | 3.8 | 1.0 | 2.4 | 3.2 | 3$-$6 | 7.5 | 50.9 Arches111Mass estimates have been made based upon the number of stars having Minitial$>$20 M⊙ given in Figer et al. (1999b) and the mass function slope in Stolte et al. (2003). The age, luminosity and ionizing flux are from Figer et al. (2002). | 4.1 | 4.1 | 0.19 | 5.6 | 5.6 | 2$-$3 | 8.0 | 51.0 Center222Krabbe et al. (1995). The mass, “M2” has been estimated by assuming that a total 103.5 stars have been formed. The age spans a range covering an initial starburst, followed by an exponential decay in the star formation rate. | 3.0 | 4.0 | 0.23 | 4.6 | 5.6 | 3$-$7 | 7.3 | 50.5 The three clusters are similar in many respects, as they are all young and contain $\gtrsim$104 M⊙ in stars. They have very high central stellar mass densities, up to nearly 106 M⊙ pc-3, exceeding central densities in most globular clusters. They have luminosities of 107-8 L⊙, and are responsible for heating nearby molecular clouds. They also generate 1050-51 ionizing photons per second, enough to account for nearby giant HII regions. The primary difference between the clusters is likely to be age, where the Quintuplet and Central clusters are about twice the age of the Arches cluster. In addition, the Central cluster is unique for its population of evolved massive stars that have broad and strong helium emission lines (Krabbe et al., 1991, and referenes therein). While the Quintuplet cluster has a few similar stars (Geballe et al., 1994; Figer et al., 1999a), the Central cluster has far more as a fraction of its total young stellar population (Paumard et al., 2006). Table 2 summarizes the massive stellar content of the clusters. Table 2: Massive Stars in the Galactic Center Clusters | Age (Myr) | O | LBV | WN | WC | RSG | References ---|---|---|---|---|---|---|--- Quintuplet | 4 | 100 | 2 | 6 | 11 | 1 | Figer et al. (1999a); Geballe, Najarro, & Figer (2000); Homeier et al. (2003) Arches | 2 | 160 | 0 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$6 | 0 | 0 | Figer et al. (2002) Center | 4$-$7 | 100 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$1 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$18 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$12 | 3 | Paumard et al. (2006) Total | | 360 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$3 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$29 | $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$>$}}}$23 | 4 | ### 5.1 Central cluster The Central cluster contains many massive stars that have recently formed in the past 10 Myr (Becklin et al., 1978; Rieke, Telesco, & Harper, 1978; Lebofsky, Rieke, & Tokunaga, 1982; Forrest et al., 1987; Allen, Hyland, & Hillier, 1990; Krabbe et al., 1991; Najarro et al., 1994; Krabbe et al., 1995; Najarro, 1995; Libonate et al., 1995; Blum, Depoy, & Sellgren, 1995a; Blum, Sellgren, & Depoy, 1995b; Genzel et al., 1996; Tamblyn et al., 1996; Najarro et al., 1997). In all, there are now known to be at least 80 massive stars in the Central cluster (Eisenhauer et al., 2005; Paumard et al., 2006), including $\approx$50 OB stars on the main sequence and 30 more evolved massive stars (see Figure 5). These young stars appear to be confined to two disks (Genzel et al., 2003; Levin & Beloborodov, 2003; Paumard et al., 2006; Tanner et al., 2006; Beloborodov et al., 2006). There is also a tight collection of a dozen or so B stars (the “s” stars) in the central arcsecond, highlighted in the small box in the figure. The formation of so many massive stars in the central parsec remains as much a mystery now as it was at the time of the first infrared observations of the region. Most recently, this topic has largely been supplanted by the even more improbable notion that star formation can occur within a few thousand AU of the supermassive black hole, an idea that will be addressed in Section 7. See Alexander (2005) for a thorough review of the “s” stars and Paumard et al. (2006) for a review of the young population in the Central cluster. Figure 5: K-band image of the Central cluster obtained with NAOS/CONICA from Schödel et al. (2006). The 100 or so brightest stars in the image are evolved descendants from main sequence O-stars. The central box highlights the “s” stars that are presumably young and massive (Minitial$\approx$20 M⊙). ### 5.2 Arches cluster The Arches cluster is unique in the Galaxy for its combination of extraordinarily high mass, M$\approx$104 M⊙, and relatively young age, $\tau=2~{}Myr$ (Figer et al., 2002). Being so young and massive, it contains the richest collection of O-stars and WNL stars in any cluster in the Galaxy (Cotera et al., 1996; Serabyn, Shupe, & Figer, 1998; Figer et al., 1999b; Blum et al., 2001; Figer et al., 2002). It is ideally suited for testing theories that predict the shape of the IMF up to the highest stellar masses formed (see Section 6). The cluster is prominent in a broad range of observations. Figure 6 shows an HST/NICMOS image of the cluster – the majority of the bright stars in the image have masses greater than 20 M⊙. The most massive dozen or so members of the cluster have strong emission lines at infrared wavelengths (Harris et al., 1994; Nagata et al., 1995; Cotera, 1995; Figer, 1995; Cotera et al., 1996; Figer et al., 1999b; Blum et al., 2001; Figer et al., 2002). These lines are produced in strong stellar winds that are also detected at radio wavelengths (Lang, Goss, & Rodríguez, 2001; Yusef-Zadeh et al., 2003; Lang et al., 2005; Figer et al., 2002), and x-ray wavelengths (Yusef-Zadeh et al., 2002; Rockefeller et al., 2005; Wang, Dong, & Lang, 2006). Figure 6: F205W image of the Arches cluster obtained by Figer et al. (2002) using HST/NICMOS. The brightest dozen or so stars in the cluster have Minitial$\gtrsim$100 M⊙, and there are $\approx$160 O-stars in the cluster. The diameter is $\approx$1 lyr, making the cluster the densest in the Galaxy with $\rho>10^{5}~{}\hbox{\it M${}_{\odot}$}~{}pc^{-3}$. ### 5.3 Quintuplet cluster The Quintuplet cluster was originally noted for its five very bright stars, the Quintuplet Proper Members (QPMs) (Glass, Moneti, & Moorwood, 1990; Okuda et al., 1990; Nagata et al., 1990). Subsequently, a number of groups identified over 30 stars evolved from massive main sequence stars (Geballe et al., 1994; Figer, McLean, & Morris, 1995; Timmermann et al., 1996; Figer et al., 1999a). Given the spectral types of the massive stars identified in the cluster, it appears that the Quintuplet cluster is $\approx$4 Myr old and had an initial mass of $>$104 M⊙ (Figer et al., 1999a). An accounting of the ionizing flux produced by the massive stars in the cluster conclusively demonstrates that the cluster heats and ionizes the nearby “Sickle” HII region (see Figure 1). The Quintuplet is most similar to Westerlund 1 in mass, age, and spectral content (Clark et al., 2005; Negueruela & Clark, 2005; Skinner et al., 2006; Groh et al., 2006; Crowther et al., 2006). Of particular interest in the cluster, the QPMs are very bright at infrared wavelengths, mK $\approx$ 6 to 9, and have color temperatures between $\approx$ 600 to 1,000 K. They are luminous, L$\approx$105 L⊙, yet spectroscopically featureless, making their spectral classification ambiguous. Figer, Morris, & McLean (1996), Figer et al. (1999a), and Moneti et al. (2001) argue that these objects are not protostars, OH/IR stars, or protostellar OB stars. Instead, they claim that these stars are dust-enshrouded WCL stars (DWCLs), similar to other dusty Galactic WC stars (Williams, van der Hucht, & The, 1987), i.e. WR 104 (Tuthill, Monnier, & Danchi, 1999) and WR 98A (Monnier, Tuthill, & Danchi, 1999). Chiar et al. (2003) tentatively identify a weak spectroscopic feature at 6.2 $\mu$m that they attribute to carbon, further supporting the hypothesis that these stars are indeed DWCLs. The stars have also been detected at x-ray wavelengths (Law & Yusef-Zadeh, 2004), and at radio wavelengths (Lang et al., 1999, 2005). Recently, Tuthill et al. (2006) convincingly show that the QPMs are indeed dusty WC stars. Figure 8 shows data that reveal the pinwheel nature of their infrared emission, characteristic of binary systems containing WCL plus an OB star (Tuthill, Monnier, & Danchi, 1999; Monnier, Tuthill, & Danchi, 1999). This identification raises intruiging questions concerning massive star formation and evolution. With their identifications, it becomes clear that every WC star in the Quintuplet is dusty, and presumably binary. There are two possible explanations for this result. Either the binary fraction for massive stars is extremely high (Mason et al., 1998; Nelan et al., 2004), or only binary massive stars evolve through the WCL phase (van der Hucht, 2001). The Quintuplet cluster also contains two Luminous Blue Variables, the Pistol star (Harris et al., 1994; Figer et al., 1998, 1999c), and FMM362 (Figer et al., 1999a; Geballe, Najarro, & Figer, 2000). Both stars are extraordinarily luminous (L$>$106 L⊙), yet relatively cool (T$\approx$104 K), placing them in the “forbidden zone” of the Hertzsprung-Russell Diagram, above the Humphreys- Davidson limit (Humphreys & Davidson, 1994). The Pistol star is particularly intriguing, in that it is surrounded by one of the most massive (10 M⊙) circumstellar ejecta in the Galaxy (see Figure 7; Figer et al., 1999c; Smith, 2006). Both stars are spectroscopically (Figer et al., 1999a) and photometrically variable (Glass et al., 2001). They present difficulties for stellar evolution and formation models. Their inferred initial masses are $>$100 M⊙, yet such stars should have already gone supernova in a cluster that is so old, as evidenced by the existence of WC stars (Figer et al., 1999a) and the red supergiant, q7 (Moneti, Glass, & Moorwood, 1994; Ramírez et al., 2000). Figer & Kim (2002) and Freitag, Rasio, & Baumgardt (2006) argue that stellar mergers might explain the youthful appearance of these stars. Alternatively, these stars might be binary, although no evidence has been found to support this assertion. Note that in a similar case, LBV1806$-$20 is also surrounded by a relatively evolved cluster (Eikenberry et al., 2004; Figer et al., 2005), yet it does appear to be binary (Figer, Najarro, & Kudritzki, 2004). Figure 7: Paschen-$\alpha$ image of the region surrounding the Pistol star from Figer et al. (1999c). North is to the upper right, and east is to the upper left. The Pistol star ejected $\approx$10 M⊙ of material approximately 6,000 yr ago to form what now appears to be a circumstellar nebula that is ionized by two WC stars to the north of the nebula. Moneti et al. (2001) use ISO data to show that the nebula is filled with dust that is heated by the Pistol star. Figure 8: Tuthill et al. (2006) find that the Quintuplet Proper Members are dusty Wolf-Rayet stars in binary systems with OB companions. The insets in this illustration show high-resolution infrared imaging data for two Quintuplet stars, overlaid on the HST/NICMOS image from Figer et al. (1999b). All of the Quintuplet WC stars are dusty, suggesting that they are binary. ## 6 The initial mass function in the Galactic center The IMF in the Galactic center has primarily been estimated through observations of the Arches cluster (Figer et al., 1999b; Stolte et al., 2003), although there have been several attempts to extract such information through observations of the Central cluster (Genzel et al., 2003; Nayakshin & Sunyaev, 2005; Paumard et al., 2006) and the background population in the region (Figer et al., 2004). These studies suggest an IMF slope that is flatter than the Salpeter value. ### 6.1 The slope Figer et al. (1999b) and Stolte et al. (2003) estimate a relatively flat IMF slope in the Arches cluster (see Figure 9). Portegies Zwart et al. (2002) interpret the data to indicate an initial slope that is consistent with the Salpeter value, and a present-day slope that has been flattened due to dynamical evolution. Performing a similar analysis, Kim et al. (2000) arrive at the opposite conclusion – that the IMF truly was relatively flat. The primary difficulty in relating the present-day mass function to the initial mass function is the fact that n-body interactions operate on relatively short timescales to segregate the highest stellar masses toward the center of the cluster and to eject the lowest stellar masses out of the cluster. Most analysis is needed to resolve this issue. ### 6.2 Upper mass cutoff The Arches cluster is the only cluster in the Galaxy that can be used to directly probe an upper mass cutoff. It is massive enough to expect stars at least as massive as 400 M⊙, young enough for its most massive members to still be visible, old enough to have broken out of its natal molecular cloud, close enough, and at a well-established distance, for us to discern its individual stars (Figer, 2005). There appears to be an absence of stars with initial masses greater than 130 M⊙ in the cluster, where the typical mass function predicts 18 (see Figure 9). Figer (2005) therefore claim a firm upper mass limit of 150 M⊙. There is additional support for such a cutoff in other environments (Weidner & Kroupa, 2004; Oey & Clarke, 2005; Koen, 2006; Weidner & Kroupa, 2006). Figure 9: Figer (2005) find an apparent upper-mass cutoff to the IMF in the Arches cluster. Magnitudes are transformed into initial mass by assuming the Geneva models for $\tau_{\rm age}$=2 Myr, solar metallicity, and the canonical mass-loss rates. Error bars indicate uncertainty from Poisson statistics. Two power-law mass functions are drawn through the average of the upper four mass bins, one having a slope of $-$0.90, as measured from the data, and another having the Salpeter slope of $-$1.35. Both suggest a dramatic deficit of stars with Minitial$>$130 M⊙, i.e. 33 or 18 are missing, respectively. These slopes would further suggest a single star with very large initial mass (MMAX). The analysis suggests that the probability of there not being an upper-mass cutoff is $\approx$ 10-8. ### 6.3 Lower mass rollover Morris (1993) argue for an elevated lower mass rollover in the GC based on the environmental conditions therein, and only recently have observations been deep enough to address this claim. Stolte et al. (2005) claim observational evidence for an elevated cutoff around 6 M⊙ in the Arches cluster; however, in that case, confusion and incompleteness are serious problems. In addition, even if the apparent turn-down is a real indication of the initial cluster population, the lack of low mass stars might result from their ejection through n-body interactions (Kim et al., 2000; Portegies Zwart et al., 2002). Field observations should not suffer from such an effect, as the field should be the repository for low mass stars ejected from massive clusters in the GC. Figure 4 reveals a turn-down in the observed luminosity function of the field in the GC at a dereddened K-band magnitude greater than 16. This appears to not be a feature of incompleteness, as the data are greater than 50% complete at these magnitudes (Figer et al., 1999b). A more convincing argument, based on this type of data, will await even deeper observations (Kim et al., 2006). ## 7 The “s” stars Figure 5 shows a dense collection of about a dozen stars within 1 arcsecond (0.04 pc) of Sgr A* (Genzel et al., 1997; Ghez et al., 1998, 2000; Eckart et al., 2002; Schödel et al., 2002; Ghez et al., 2003; Schödel et al., 2003; Ghez et al., 2005). This cluster stands out for its high stellar density, even compared to the already dense field population in the GC. Schödel et al. (2003) and Ghez et al. (2005) (and refereces therein) have tracked the proper motions of the “s” stars, finding that they are consistent with closed orbits surrounding a massive, and dark, object having M$\approx 2-4(10^{6})$ M⊙, consistent with previous claims based on other methods (Lynden-Bell & Rees, 1971; Lacy et al., 1980; Serabyn & Lacy, 1985; Genzel & Townes, 1987; Sellgren et al., 1987; Rieke & Rieke, 1988; McGinn et al., 1989; Lacy, Achtermann, & Serabyn, 1991; Lindqvist, Habing, & Winnberg, 1992; Haller et al., 1996). The orbital parameters for these stars are well determined, as seen in Figure 10 (left), and they require the existence of a supermassive black hole in the Galactic center. While these stars are useful as gravitational test particles, they are also interesting in their own right, as they have inferred luminosities and temperatures that are similar to those of young and massive stars (Genzel et al., 1997; Eckart et al., 1999; Figer et al., 2000; Ghez et al., 2003; Eisenhauer et al., 2003, 2005; Paumard et al., 2006). Figure 10 (right) shows the absorption lines that suggest relatively high temperatures. Table 3: Chronologically sorted list of references that explore hypotheses on the origin of the “s” stars. Some of the references primarily concern the other young stars in the central parsec and are included in the table because they propose ideas that may relate to the origins of the “s” stars. Contributions to this table have been made by Tal Alexander (priv. communication). Reference | Description ---|--- Lacy, Townes, & Hollenbach (1982) | tidal disruption of red giants Morris (1993) | compact objects surrounded by material from red giant envelopes (“Thorne-Zytkow objects”) Davies et al. (1998) | red giant envelope stripping through n-body interactions Alexander (1999) | red giant envelope stripping through dwarf-giant interactions Bailey & Davies (1999) | colliding red giants Morris, Ghez, & Becklin (1999) | duty cycle of formation from infalling CND clouds and evaporation of gas reservoir by accretion and star-formation light Gerhard (2001) | decaying massive cluster Alexander & Morris (2003) | tidal heating of stellar envelopes to form “Squeezars” Genzel et al. (2003) | stellar rejuvination through red giant mergers Gould & Quillen (2003) | “exchange reaction” between massive-star binary and massive black hole Hansen & Milosavljević (2003) | stars captured by inspiraling intermediate mass black hole Levin & Beloborodov (2003) | formation in nearby gas disk McMillan & Portegies Zwart (2003); Kim, Figer, & Morris (2004) | inward migration of young cluster with IMBH Portegies Zwart, McMillan, & Gerhard (2003); Kim & Morris (2003) | inward migration of young cluster Alexander & Livio (2004) | orbital capture of young stars by MBH-SBH binaries Milosavljević & Loeb (2004) | formation in molecular disk Davies & King (2005) | tidal strippng of red giant (AGB) stars (but see critique in Goodman & Paczynski, 2005) Gürkan & Rasio (2005) | decaying cluster with formation of an IMBH Haislip & Youdin (2005) | formation in disk and orbital relaxation.(but see critique in Goodman & Paczynski, 2005) Levin, Wu, & Thommes (2005) | dynamical interactions with sinking IMBH Nayakshin & Sunyaev (2005) | _in-situ_ formation within central parsec Nayakshin & Cuadra (2005) | formation in a fragmenting star disk Subr & Karas (2005) | formation in disk and accelerated orbital relaxation Berukoff & Hansen (2006) | cluster inspiral and n-body interactions Hopman & Alexander (2006) | resonant relaxation of orbits Freitag, Amaro-Seoane, & Kalogera (2006) | mass segregation through interactions with compact remnants Levin (2006) | star formation in fragmenting disk Perets, Hopman, & Alexander (2006) | exchange reactions between massive star binaries induced by efficient relaxation by massive perturbers Dray, King, & Davies (2006) | tidal strippng of red giant (AGB) stars (but see critique in Figer et al., 2000) Oddly, the increased density of the young stars in the central arcsecond is not matched by the density distribution of old stars. Indeed, there is a curious absence of late-type stars in the central few arcseconds, as evidenced by a lack of stars with strong CO absorption in their K-band spectra (Lacy, Townes, & Hollenbach, 1982; Phinney, 1989; Sellgren et al., 1990; Haller et al., 1996; Genzel et al., 1996, 2003). This dearth of old stars represents a true “hole” in three dimensional space, and not just a projection effect. Even the late-type stars that are projected on to the central parsec generally have relatively low velocities, suggesting dynamical evidence that the region nearest to the black hole lacks old stars (Figer et al., 2003). The existence of such massive and young stars in the central arcseconds is puzzling, although it is perhaps only an extension of the original problem in understanding the origins of the young stars identified in the central parsec over 20 years ago. Table 3 gives a list of recent papers regarding the origin of the “s” stars. While there are over 30 papers listed in this table, they can be reduced to a few basic ideas. One class of ideas considers the “s” stars as truly young. In this case, the “origin” of the “s” stars is often reduced to the case of massive star formation in the Galactic center region and transportation of the products to the central arcsecond. The other class regards the “s” stars as old stars that only appear to be young, i.e. via atmospheric stripping, merging, or heating. Both classes require new mechanisms that would be unique to the GC, and they both have considerable weaknesses. For example, Figer et al. (2000) argue that stripped red giants would not be as bright as the “s” stars (see Dray, King, & Davies, 2006, for detailed confirmation). See Alexander (2005) for a more thorough discussion of the strengths and weaknesses of these ideas. If the “s” stars are truly young, then that would require massive clumps to form OB stars (Minitial$\gtrsim$20 M⊙). In addition, the clumps would have to form from very high density material in order for them to be stable against tidal disruption. Assuming that the stars formed as far away from the supermassive black hole as possible, while still permitting dynamical friction to transport them into the central arcsecond during their lifetimes, then the required densities must be $>$10${}^{11}~{}cm^{-3}$ (Figer et al., 2000). The average molecular cloud density in the GC is about five orders of magnitude less, so highly compressive events might be required to achieve the necessary densities. Alternatively, the required densities can be reduced if the stars are gravitationally bound to significant mass, i.e. a surrounding stellar cluster. Indeed, Gerhard (2001), Portegies Zwart, McMillan, & Gerhard (2003), and Kim & Morris (2003) showed that particularly massive clusters could form tens of parsecs outside of the center and be delivered into the central parsec in just a few million years. The efficiency of this method is improved with the presence of an intermediate black hole in the cluster (McMillan & Portegies Zwart, 2003; Kim, Figer, & Morris, 2004). It is key in any of these cluster transport models that the host system have extremely high densities of $>$106 M⊙ pc-3, comparable to the highest estimated central density of the Arches cluster after core collapse (Kim & Morris, 2003). Detailed n-body simulations suggest that while these ideas may be relevant for the origins of the young stars in the central parsec, it is unlikely that they could explain the existence of the “s” stars in the central arcsecond. Figure 10: (left) Figure 2 in Ghez et al. (2005) and (middle) Figure 6 in Schödel et al. (2003), stretched to the same scale. Both figures fit similar model orbits through separate proper motion data sets for the “s” stars. (right) Paumard et al. (2006) find that one of the “s” stars, S2, has a K-band spectrum that is similar to those of OB stars in the central parsec (see also Ghez et al., 2003). ## 8 Comparisons to other massive star populations in Galaxy There are relatively few clusters in the Galaxy with as many massive stars as in the GC clusters. NGC3603 has about a factor of two less mass than each of the GC clusters (Moffat et al., 2002); whereas, W1 has at least a factor of two greater mass (Clark et al., 2005; Negueruela & Clark, 2005; Skinner et al., 2006; Groh et al., 2006). The next nearest similarly massive cluster is R136 in the LMC (Massey & Hunter, 1998). All of these clusters, and the GC clusters, appear to have IMF slopes that are consistent with the Salpter value (or slightly flatter) and are young enough to still possess a significant massive star population. It is remarkable to note that these massive clusters appear quite similar in stellar content, whether in the Galactic disk, the GC, or even the lower metallicity environment of the LMC. Evidently, the star formation processes, and natal environments, that gave birth to these clusters must be similar enough to produce clusters that are virtually indistinguishable. There are probably more massive clusters yet to be found in the Galaxy. The limited sample of known massive clusters is a direct result of extinction, as most star formation sites in the Galaxy are obscurred by dust at optical wavelengths. While infrared observations have been available for over 30 years, they have not provided the necessary spatial resolution, nor survey coverage, needed to probe the Galactic disk for massive clusters. Recently, a number of groups have begun identifying candidate massive star clusters using near-infrared surveys with arcsecond resolution (Bica et al., 2003; Dutra et al., 2003; Mercer et al., 2005). Indeed, these surveys have already yielded a cluster with approximate initial mass of 20,000 to 40,000 M⊙ (Figer et al., 2006), and one would expect more to be discovered from them. The present-day sites of massive star formation in the Galaxy have been known for some time through radio and far-infrared observations, as their hottest members ionize and heat nearby gas in molecular clouds. As one of many examples, consider W49 which is the next most massive star formation site in the Galaxy compared to Sgr B2 (Homeier & Alves, 2005), and wherein the star formation appears to be progressing in stages over timescales that far exceed the individual collapse times for massive star progenitors. This suggests a stimulus that triggers the star formation, perhaps provided by a “daisy chain” effect in which newly formed stars trigger collapse in nearby parts of the cloud. Similar suggestions are proposed in the 30 Dor region surrounding R136 (Walborn, Maíz-Apellániz, & Barbá, 2002). While there is no evidence for an age-dispersed population in Sgr B2, Sato et al. (2000) suggest that the cloud was triggered to form stars through a cloud-cloud interaction. ## 9 Conclusions Massive star formation in the GC has produced an extraordinary sample of stars populating the initial mass function up to a cutoff of approximately 150 M⊙. The ranges of inferred masses and observed spectral types are as expected from stellar evolution models, and the extraordinary distribution of stars in the region is a direct consequence of the large amount of mass that has fed star formation in the GC. The origin of the massive stars in the central parsec, and especially the central arcsecond, remains unresolved. ###### Acknowledgements. I thank the following individuals for discussions related to this work: Mark Morris, Bob Blum, Reinhard Genzel, Paco Najarro, Sungsoo Kim, and Peter Tuthill. Tal Alexander made substantial contributions to Table 3. The material in this paper is based upon work supported by NASA under award No. NNG05-GC37G, through the Long Term Space Astrophysics program. Full resolution versions of the above images are available at http://www.cis.rit.edu/ dffpci/private/papers/stsci06/ ## References * Alexander (1999) Alexander, T. 1999 The Distribution of Stars near the Supermassive Black Hole in the Galactic Center. Astrophysical Journal 527, 835 * Alexander (2005) Alexander, T. 2005, Stellar processes near the massive black hole in the Galactic center. Physics Reports 419, 65 * Alexander & Livio (2004) Alexander, T., & Livio, M. 2004 Orbital Capture of Stars by a Massive Black Hole via Exchanges with Compact Remnants. Astrophysical Journal Letters 606, L21 * Alexander & Morris (2003) Alexander, T., & Morris, M. 2003 Squeezars: Tidally Powered Stars Orbiting a Massive Black Hole. Astrophysical Journal Letters 590, L25 * Allen, Hyland, & Hillier (1990) Allen, D. A., Hyland, A. R., & Hillier, D. J. 1990 The source of luminosity at the Galactic Centre. Monthly Notices of the Royal Astronomical Society 244, 706 * Bailey & Davies (1999) Bailey, V. C., & Davies, M. B. 1999 Red giant collisions in the Galactic Centre. Monthly Notices of the Royal Astronomical Society 308, 257 * Becklin et al. (1978) Becklin, E. E., Matthews, K., Neugebauer, G., & Willner, S. P. 1978 Infrared observations of the galactic center. I - Nature of the compact sources. Astrophysical Journal 219, 121 * Beloborodov et al. (2006) Beloborodov, A. M., Levin, Y., Eisenhauer, F., Genzel, R., Paumard, T., Gillessen, S., & Ott, T. 2006 Clockwise Stellar Disk and the Dark Mass in the Galactic Center. ArXiv Astrophysics e-prints , arXiv:astro-ph/0601273 * Berukoff & Hansen (2006) Berukoff, S. J., & Hansen, B. M. S. 2006 Cluster Core Dynamics at the Galactic Center. ArXiv Astrophysics e-prints , arXiv:astro-ph/0607080 * Bica et al. (2003) Bica, E., Dutra, C. M., Soares, J., & Barbuy, B. 2003 New infrared star clusters in the Northern and Equatorial Milky Way with 2MASS. Astronomy & Astrophysics 404, 223 * Blum (1995) Blum, R. D. 1995 The Stellar Population at the Galactic Center and the Mass Distribution in the Inner Galaxy. Ph.D. Thesis * Blum, Depoy, & Sellgren (1995a) Blum, R. D., Depoy, D. L., & Sellgren, K. 1995a A comparison of near-infrared spectra of the galactic center compact He I emission-line sources and early-type mass-losing stars. Astrophysical Journal 441, 603 * Blum et al. (2003) Blum, R. D., Ramírez, S. V., Sellgren, K., & Olsen, K. 2003 Really Cool Stars and the Star Formation History at the Galactic Center. Astrophysical Journal 597, 323 * Blum et al. (2001) Blum, R. D., Schaerer, D., Pasquali, A., Heydari-Malayeri, M., Conti, P. S., & Schmutz, W. 2001 2 Micron Narrowband Adaptive Optics Imaging in the Arches Cluster. Astronomical Journal 122, 1875 * Blum, Sellgren, & Depoy (1995b) Blum, R. D., Sellgren, K., & Depoy, D. L. 1995b Discovery of a possible Wolf-Rayet star at the galactic center. Astrophysical Journal Letters 440, L17 * Blum, Sellgren, & Depoy (1996a) Blum, R. D., Sellgren, K., & Depoy, D. L. 1996a Really Cool Stars at the Galactic Center. Astronomical Journal 112, 1988 * Blum, Sellgren, & Depoy (1996b) Blum, R. D., Sellgren, K., & Depoy, D. L. 1996b JHKL Photometry and the K-Band Luminosity Function at the Galactic Center. Astrophysical Journal 470, 864 * Carr, Sellgren, & Balachandran (2000) Carr, J. S., Sellgren, K., & Balachandran, S. C. 2000 The First Stellar Abundance Measurements in the Galactic Center: The M Supergiant IRS 7\. Astrophysical Journal 530, 307 * Chiar et al. (2003) Chiar, J. E., Adamson, A. J., Whittet, D. C. B., & Pendleton, Y. J. 2003 Spectroscopy of Hydrocarbon Grains toward the Galactic Center and Quintuplet Cluster. Astronomische Nachrichten Supplement 324, 109 * Chuss et al. (2003) Chuss, D. T., Davidson, J. A., Dotson, J. L., Dowell, C. D., Hildebrand, R. H., Novak, G., & Vaillancourt, J. E. 2003 Magnetic Fields in Cool Clouds within the Central 50 Parsecs of the Galaxy. Astrophysical Journal 599, 1116 * Clark et al. (2005) Clark, J. S., Negueruela, I., Crowther, P. A., & Goodwin, S. P. 2005 On the massive stellar population of the super star cluster Westerlund 1. Astronomy & Astrophysics 434, 949 * Cotera (1995) Cotera, A. S. 1995 Stellar Ionization of the Thermal Emission Regions in the Galactic Center. Ph.D. Thesis * Cotera et al. (1996) Cotera, A. S., Erickson, E. F., Colgan, S. W. J., Simpson, J. P., Allen, D. A., & Burton, M. G. 1996 The Discovery of Hot Stars near the Galactic Center Thermal Radio Filaments. Astrophysical Journal 461, 750 * Cotera et al. (1999) Cotera, A. S., Simpson, J. P., Erickson, E. F., Colgan, S. W. J., Burton, M. G., & Allen, D. A. 1999 Isolated Hot Stars in the Galactic Center Vicinity. Astrophysical Journal 510, 747 * Crowther et al. (2006) Crowther, P.A., Hadfield, L. J., Clark, J. S., Negueruela, I., & Vacca, W. D. 2006 A census of the Wolf-Rayet content in Westerlund 1 from near-infrared imaging and spectroscopy. ArXiv Astrophysics e-prints , arXiv:astro-ph/0608356 * Davies et al. (1998) Davies, M. B., Blackwell, R., Bailey, V. C., & Sigurdsson, S. 1998 The destructive effects of binary encounters on red giants in the Galactic Centre. Monthly Notices of the Royal Astronomical Society 301, 745 * Davies & King (2005) Davies, M. B., & King, A. 2005 The Stars of the Galactic Center. Astrophysical Journal Letters 624, L25 * de Pree et al. (1995) de Pree, C. G., Gaume, R. A., Goss, W. M., & Claussen, M. J. 1995 The Sagittarius B2 Star-forming Region. II. High-Resolution H66 alpha Observations of Sagittarius B2 North. Astrophysical Journal 451, 284 * de Pree et al. (1996) de Pree, C. G., Gaume, R. A., Goss, W. M., & Claussen, M. J. 1996 The Sagittarius B2 Star-forming Region. III. High-Resolution H52 alpha and H66 alpha Observations of Sagittarius B2 Main. Astrophysical Journal 464, 788 * de Vicente et al. (2000) de Vicente, P., Martín-Pintado, J., Neri, R., & Colom, P. 2000 A ridge of recent massive star formation between Sgr B2M and Sgr B2N. Astronomy & Astrophysics 361, 1058 * Dray, King, & Davies (2006) Dray, L. M., King, A. R., & Davies, M. B. 2006 Young stars in the Galactic Centre: a potential intermediate-mass star origin. ArXiv Astrophysics e-prints , arXiv:astro-ph/0607470 * Dutra et al. (2003) Dutra, C. M., Bica, E., Soares, J., & Barbuy, B. 2003 New infrared star clusters in the southern Milky Way with 2MASS. Astronomy & Astrophysics 400, 533 * Eckart et al. (2002) Eckart, A., Genzel, R., Ott, T., & Schödel, R. 2002 Stellar orbits near Sagittarius A*. Monthly Notices of the Royal Astronomical Society 331, 917 * Eckart et al. (1999) Eckart, A., Ott, T., & Genzel, R. 1999, The Sgr A* stellar cluster: New NIR imaging and spectroscopy. Astronomy & Astrophysics 352, L22 * Eckart et al. (2005) Eckart, A., Schödel, R., & Straubmeier, C. 2005 The black hole at the center of the Milky Way. Andreas Eckart, Rainer Schödel, Christian Straubmeier. London: Imperial College Press, ISBN 1-86094-567-8, 2005, XXII+284 * Eikenberry et al. (2004) Eikenberry, S. S., et al. 2004 Infrared Observations of the Candidate LBV 1806-20 and Nearby Cluster Stars1,. Astrophysical Journal 616, 506 * Eisenhauer et al. (2005) Eisenhauer, F., et al. 2005 SINFONI in the Galactic Center: Young Stars and Infrared Flares in the Central Light-Month. Astrophysical Journal 628, 246 * Eisenhauer et al. (2003) Eisenhauer, F., Schödel, R., Genzel, R., Ott, T., Tecza, M., Abuter, R., Eckart, A., & Alexander, T. 2003 A Geometric Determination of the Distance to the Galactic Center. Astrophysical Journal Letters 597, L121 * Figer (1995) Figer, D. F. 1995 A Search for Emission-Line Stars Near the Galactic Center. Ph.D. Thesis. * Figer (2003) Figer, D. F. 2003, Massive stars and the creation of our Galactic Center. IAU Symposium, 212, 487 * Figer (2004) Figer, D. F. 2004, Young Massive Clusters in the Galactic Center. ASP Conf. Ser. 322: The Formation and Evolution of Massive Young Star Clusters, 322, 49. * Figer (2005) Figer, D. F. 2005 An upper limit to the masses of stars. Nature 434, 192 * Figer et al. (2000) Figer, D. F., et al. 2000, 2 Micron Spectroscopy within 0.3” of Sagittarius A*. Astrophysical Journal Letters 533, L49 * Figer et al. (2002) Figer, D. F. et al. 2002 Massive Stars in the Arches Cluster. Astrophysical Journal 581, 258 * Figer et al. (2003) Figer, D. F., et al. 2003 High-Precision Stellar Radial Velocities in the Galactic Center. Astrophysical Journal 599, 1139 * Figer et al. (1994) Figer, D. F., Becklin, E. E., McLean, I. S., & Morris, M. 1994 Discovery of Luminous NIR Sources Associated With Ionized Gas Near the Galactic Center. ASSL Vol. 190: Astronomy with Arrays, The Next Generation , 545 * Figer & Kim (2002) Figer, D. F., & Kim, S. S. 2002 Stellar Collisions and Mergers in the Galactic Center. ASP Conf. Ser. 263: Stellar Collisions, Mergers and their Consequences 263, 287 * Figer et al. (1999b) Figer, D. F., Kim, S. S., Morris, M., Serabyn, E., Rich, R. M., & McLean, I. S. 1999b HST/NICMOS Observations of Massive Stellar Clusters Near the Galactic Center. Astrophysical Journal 525, 750. * Figer et al. (2006) Figer, D. F., MacKenty, J. W., Robberto, M., Smith, K., Najarro, F., Kudritzki, R. P., & Herrero, A. 2006 Discovery of an Extraordinarily Massive Cluster of Red Supergiants. Astrophysical Journal 643, 1166 * Figer, McLean, & Morris (1995) Figer, D. F., McLean, I. S., & Morris, M. 1995 Two New Wolf-Rayet Stars and a Luminous Blue Variable Star in the Quintuplet (AFGL 2004) near the Galactic Center. Astrophysical Journal Letters 447, L29 * Figer et al. (1999a) Figer, D. F., McLean, I. S., & Morris, M. 1999a, Massive Stars in the Quintuplet Cluster. Astrophysical Journal 514, 202 * Figer et al. (1999c) Figer, D. F., Morris, M., Geballe, T. R., Rich, R. M., Serabyn, E., McLean, I. S., Puetter, R. C., & Yahil, A. 1999c High-Resolution Infrared Imaging and Spectroscopy of the Pistol Nebula: Evidence for Ejection. Astrophysical Journal 525, 759 * Figer, Morris, & McLean (1996) Figer, D. F., Morris, M., & McLean, I. S. 1996 Hot Stars in the Quintuplet. ASP Conf. Ser. 102: The Galactic Center 102, 263 * Figer et al. (2005) Figer, D. F., Najarro, F., Geballe, T. R., Blum, R. D., & Kudritzki, R. P. 2005 Massive Stars in the SGR 1806-20 Cluster. Astrophysical Journal Letters 622, L49 * Figer, Najarro, & Kudritzki (2004) Figer, D. F., Najarro, F., & Kudritzki, R. P. 2004 The Double-lined Spectrum of LBV 1806-20. Astrophysical Journal Letters 610, L109 * Figer et al. (1998) Figer, D. F., Najarro, F., Morris, M., McLean, I. S., Geballe, T. R., Ghez, A. M., & Langer, N. 1998 The Pistol star. Astrophysical Journal 506, 384 * Figer et al. (2004) Figer, D. F., Rich, R. M., Kim, S. S., Morris, M., & Serabyn, E. 2004 An Extended Star Formation History for the Galactic Center from Hubble Space Telescope NICMOS Observations. Astrophysical Journal 601, 319 * Forrest et al. (1987) Forrest, W. J., Shure, M. A., Pipher, J. L., & Woodward, C. E. 1987 Brackett Alpha Images. AIP Conf. Proc. 155: The Galactic Center 155, 153 * Freitag, Amaro-Seoane, & Kalogera (2006) Freitag, M., Amaro-Seoane, P., & Kalogera, V. 2006 Stellar remnants in galactic nuclei: mass segregation. ArXiv Astrophysics e-prints , arXiv:astro-ph/0603280 * Freitag, Rasio, & Baumgardt (2006) Freitag, M., Rasio, F. A., & Baumgardt, H. 2006 Runaway collisions in young star clusters - I. Methods and tests. Monthly Notices of the Royal Astronomical Society 368, 121 * Frogel, Tiede, & Kuchinski (1999) Frogel, J. A., Tiede, G. P., & Kuchinski, L. E. 1999 The Metallicity and Reddening of Stars in the Inner Galactic Bulge. Astronomical Journal 117, 2296 * Garay & Lizano (1999) Garay, G., & Lizano, S. 1999 Massive Stars: Their Environment and Formation. Publications of the Astronomical Society of the Pacific 111, 1049 * Gaume et al. (1995) Gaume, R. A., Claussen, M. J., de Pree, C. G., Goss, W. M., & Mehringer, D. M. 1995 The Sagittarius B2 Star-forming Region. I. Sensitive 1.3 Centimeter Continuum Observations. Astrophysical Journal 449, 663 * Geballe et al. (1994) Geballe, T. R., Genzel, R., Krabbe, A., Krenz, T., & Lutz, D. 1994 Spectra of a Remarkable Class of Hot Stars in the Galactic Center. ASSL Vol. 190: Astronomy with Arrays, The Next Generation , 73 * Geballe, Najarro, & Figer (2000) Geballe, T. R., Najarro, F., & Figer, D. F. 2000 A Second Luminous Blue Variable in the Quintuplet Cluster. Astrophysical Journal Letters 530, L97 * Genzel et al. (2003) Genzel, R., et al. 2003 The Stellar Cusp around the Supermassive Black Hole in the Galactic Center. Astrophysical Journal 594, 812 * Genzel et al. (1997) Genzel, R., Eckart, A., Ott, T., & Eisenhauer, F. 1997, On the nature of the dark mass in the centre of the Milky Way. Monthly Notices of the Royal Astronomical Society 291, 219 * Genzel et al. (1994) Genzel, R., Hollenbach, D., & Townes, C. H. 1994, The nucleus of our Galaxy. Reports of Progress in Physics, 57, 417. * Genzel et al. (1996) Genzel, R., Thatte, N., Krabbe, A., Kroker, H., & Tacconi-Garman, L. E. 1996 The Dark Mass Concentration in the Central Parsec of the Milky Way. Astrophysical Journal 472, 153 * Genzel & Townes (1987) Genzel, R., & Townes, C. H. 1987 Physical conditions, dynamics, and mass distribution in the center of the Galaxy. Annual Review of Astronomy and Astrophysics 25, 377 * Gerhard (2001) Gerhard, O. 2001 The Galactic Center HE I Stars: Remains of a Dissolved Young Cluster?. Astrophysical Journal Letters 546, L39 * Ghez et al. (2003) Ghez, A. M., et al. 2003, The First Measurement of Spectral Lines in a Short-Period Star Bound to the Galaxy’s Central Black Hole: A Paradox of Youth. Astrophysical Journal Letters 586, L127 * Ghez et al. (1998) Ghez, A. M., Klein, B. L., Morris, M., & Becklin, E. E. 1998 High Proper-Motion Stars in the Vicinity of Sagittarius A*: Evidence for a Supermassive Black Hole at the Center of Our Galaxy. Astrophysical Journal 509, 678 * Ghez et al. (2000) Ghez, A. M., Morris, M., Becklin, E. E., Tanner, A., & Kremenek, T. 2000 The accelerations of stars orbiting the Milky Way’s central black hole. Nature 407, 349 * Ghez et al. (2005) Ghez, A. M., Salim, S., Hornstein, S. D., Tanner, A., Lu, J. R., Morris, M., Becklin, E. E., & Duchêne, G. 2005 Stellar Orbits around the Galactic Center Black Hole. Astrophysical Journal 620, 744 * Glass et al. (2001) Glass, I. S., Matsumoto, S., Carter, B. S., & Sekiguchi, K. 2001 Large-amplitude variables near the Galactic Centre. Monthly Notices of the Royal Astronomical Society 321, 77 * Glass, Moneti, & Moorwood (1990) Glass, I. S., Moneti, A., & Moorwood, A. F. M. 1990 Infrared images and photometry of the cluster near G 0.15 - 0.05. Monthly Notices of the Royal Astronomical Society 242, 55P * Goodman & Paczynski (2005) Goodman, J., & Paczynski, B. 2005 On the nature of the S stars in the Galactic Center. ArXiv Astrophysics e-prints, arXiv:astro-ph/0504079 * Goss et al. (1985) Goss, W. M., Schwarz, U. J., van Gorkom, J. H., & Ekers, R. D. 1985 The SGR A East H II complex at L = -0.02, B = -0.07 deg. Monthly Notices of the Royal Astronomical Society 215, 69P * Gould & Quillen (2003) Gould, A., & Quillen, A. C. 2003 Sagittarius A* Companion S0-2: A Probe of Very High Mass Star Formation. Astrophysical Journal 592, 935 * Groh et al. (2006) Groh, J. H., Damineli, A., Teodoro, M., & Barbosa, C. L. 2006 Detection of additional Wolf-Rayet stars in the starburst cluster Westerlund 1 with SOAR. ArXiv Astrophysics e-prints , arXiv:astro-ph/0606498 * Gürkan & Rasio (2005) Gürkan, M. A., & Rasio, F. A. 2005 The Disruption of Stellar Clusters Containing Massive Black Holes near the Galactic Center. Astrophysical Journal 628, 236 * Haislip & Youdin (2005) Haislip, G., & Youdin, A. N. 2005 A Disk Origin for S-Stars in the Galactic Center? American Astronomical Society Meeting Abstracts 207, #181.14 * Haller & Melia (1996) Haller, J. W., & Melia, F. 1996 Inferring Spherical Mass Distributions Using the Projected Mass Estimator. Astrophysical Journal 464, 774 * Haller et al. (1996) Haller, J. W., Rieke, M. J., Rieke, G. H., Tamblyn, P., Close, L., & Melia, F. 1996 Stellar Kinematics and the Black Hole in the Galactic Center. Astrophysical Journal 456, 194 * Hansen & Milosavljević (2003) Hansen, B. M. S., & Milosavljević, M. 2003 The Need for a Second Black Hole at the Galactic Center. Astrophysical Journal Letters 593, L77 * Harris et al. (1994) Harris, A. I., Krenz, T., Genzel, R., Krabbe, A., Lutz, D., Politsch, A., Townes, C. H., & Geballe, T. R. 1994 Spectroscopy of the Galactic Center Arches Region: Evidence for Massive Star Formation. NATO ASIC Proc. 445: The Nuclei of Normal Galaxies: Lessons from the Galactic Center , 223 * Hildebrand et al. (1993) Hildebrand, R. H., Davidson, J. A., Dotson, J., Figer, D. F., Novak, G., Platt, S. R., & Tao, L. 1993 Polarization of the Thermal Emission from the Dust Ring at the Center of the Galaxy. Astrophysical Journal 417, 565 * Homeier & Alves (2005) Homeier, N. L., & Alves, J. 2005 Massive star formation in the W49 giant molecular cloud: Implications for the formation of massive star clusters. Astronomy & Astrophysics 430, 481 * Homeier et al. (2003) Homeier, N. L., Blum, R. D., Pasquali, A., Conti, P. S., & Damineli, A. 2003 Results from a near infrared search for emission-line stars in the Inner Galaxy: Spectra of new Wolf-Rayet stars. Astronomy & Astrophysics 408, 153 * Hopman & Alexander (2006) Hopman, C., & Alexander, T. 2006 Resonant Relaxation near the Massive Black Hole in the Galactic Center. ArXiv Astrophysics e-prints , arXiv:astro-ph/0605457 * Humphreys & Davidson (1994) Humphreys, R. M., & Davidson, K. 1994 The luminous blue variables: Astrophysical geysers. Publications of the Astronomical Society of the Pacific 106, 1025 * Kennicutt (1998) Kennicutt, R. C., Jr. 1998 Star Formation in Galaxies Along the Hubble Sequence. Annual Review of Astronomy and Astrophysics 36, 189 * Kim et al. (2000) Kim, S. S., Figer, D. F., Lee, H. M., & Morris, M. 2000 N-Body Simulations of Compact Young Clusters near the Galactic Center. Astrophysical Journal 545, 301 * Kim, Figer, & Morris (2004) Kim, S. S., Figer, D. F., & Morris, M. 2004 Dynamical Friction on Galactic Center Star Clusters with an Intermediate-Mass Black Hole. Astrophysical Journal Letters 607, L123 * Kim et al. (2006) Kim, S. S., Figer, D. F., Kudritzki, R. P., & Najarro, F. N. 2006 Keck/LGS Observations of the Arches Cluster. Astrophysical Journal submitted * Kim & Morris (2003) Kim, S. S., & Morris, M. 2003 Dynamical Friction on Star Clusters near the Galactic Center. Astrophysical Journal 597, 312 * Koen (2006) Koen, C. 2006 On the upper limit on stellar masses in the Large Magellanic Cloud cluster R136. Monthly Notices of the Royal Astronomical Society 365, 590 * Krabbe et al. (1995) Krabbe, A., et al. 1995 The Nuclear Cluster of the Milky Way: Star Formation and Velocity Dispersion in the Central 0.5 Parsec. Astrophysical Journal Letters 447, L95 * Krabbe et al. (1991) Krabbe, A., Genzel, R., Drapatz, S., & Rotaciuc, V. 1991 A cluster of He I emission-line stars in the Galactic center. Astrophysical Journal Letters 382, L19 * Lacy, Achtermann, & Serabyn (1991) Lacy, J. H., Achtermann, J. M., & Serabyn, E. 1991 Galactic center gasdynamics - A one-armed spiral in a Keplerian disk. Astrophysical Journal Letters 380, L71 * Lacy et al. (1980) Lacy, J. H., Townes, C. H., Geballe, T. R., & Hollenbach, D. J. 1980 Observations of the motion and distribution of the ionized gas in the central parsec of the Galaxy. II. Astrophysical Journal 241, 132 * Lacy, Townes, & Hollenbach (1982) Lacy, J. H., Townes, C. H., & Hollenbach, D. J. 1982 The nature of the central parsec of the Galaxy. Astrophysical Journal 262, 120 * Lang et al. (1999) Lang, C. C., Figer, D. F., Goss, W. M., & Morris, M. 1999 Radio Detections of Stellar Winds from the Pistol star and Other Stars in the Galactic Center Quintuplet Cluster. Astronomical Journal 118, 2327 * Lang, Goss, & Rodríguez (2001) Lang, C. C., Goss, W. M., & Rodríguez, L. F. 2001 Very Large Array Detection of the Ionized Stellar Winds Arising from Massive Stars in the Galactic Center Arches Cluster. Astrophysical Journal Letters 551, L143 * Lang, Goss, & Wood (1997) Lang, C. C., Goss, W. M., & Wood, D. O. S. 1997 VLA H92 alpha and H115 beta Recombination Line Observations of the Galactic Center H II Regions: The Sickle (G0.18-0.04) and the Pistol (G0.15-0.05). Astrophysical Journal 474, 275 * Lang et al. (2005) Lang, C. C., Johnson, K. E., Goss, W. M., & Rodríguez, L. F. 2005 Stellar Winds and Embedded Star Formation in the Galactic Center Quintuplet and Arches Clusters: Multifrequency Radio Observations. Astronomical Journal 130, 2185 * Law & Yusef-Zadeh (2004) Law, C., & Yusef-Zadeh, F. 2004 X-Ray Observations of Stellar Clusters Near the Galactic Center. Astrophysical Journal 611, 858 * Lebofsky & Rieke (1987) Lebofsky, M. J., & Rieke, G. H. 1987 The Stellar Population. AIP Conf. Proc. 155: The Galactic Center 155, 79 * Lebofsky, Rieke, & Tokunaga (1982) Lebofsky, M. J., Rieke, G. H., & Tokunaga, A. T. 1982 M supergiants and star formation at the galactic center. Astrophysical Journal 263, 736 * Levin (2006) Levin, Y. 2006 Starbursts near supermassive black holes: young stars in the Galactic Center, and gravitational waves in LISA band. ArXiv Astrophysics e-prints , arXiv:astro-ph/0603583 * Levin & Beloborodov (2003) Levin, Y., & Beloborodov, A. M. 2003 Stellar Disk in the Galactic Center: A Remnant of a Dense Accretion Disk? Astrophysical Journal Letters 590, L33 * Levin, Wu, & Thommes (2005) Levin, Y., Wu, A., & Thommes, E. 2005 Intermediate-Mass Black Hole(s) and Stellar Orbits in the Galactic Center. Astrophysical Journal 635, 341 * Libonate et al. (1995) Libonate, S., Pipher, J. L., Forrest, W. J., & Ashby, M. L. N. 1995 Near-infrared spectra of compact stellar wind sources at the Galactic center. Astrophysical Journal 439, 202 * Lindqvist, Habing, & Winnberg (1992) Lindqvist, M., Habing, H. J., & Winnberg, A. 1992 OH/IR stars close to the Galactic Centre. II - Their spatial and kinematic properties and the mass distribution within 5-100 PC from the galactic centre. Astronomy & Astrophysics 259, 118 * Liu & Snyder (1999) Liu, S.-Y., & Snyder, L. E. 1999 Subarcsecond Resolution Observations of Sagittarius B2 at 85 GHZ. Astrophysical Journal 523, 683 * Lynden-Bell & Rees (1971) Lynden-Bell, D., & Rees, M. J. 1971 On quasars, dust and the galactic centre. Monthly Notices of the Royal Astronomical Society 152, 461 * Maeda et al. (2002) Maeda, Y., et al. 2002 A Chandra Study of Sagittarius A East: A Supernova Remnant Regulating the Activity of Our Galactic Center?. Astrophysical Journal 570, 671 * Mason et al. (1998) Mason, B. D., Gies, D. R., Hartkopf, W. I., Bagnuolo, W. G., Jr., ten Brummelaar, T., & McAlister, H. A. 1998 ICCD speckle observations of binary stars. XIX - an astrometric/spectroscopic survey of O stars. Astronomical Journal 115, 821 * Massey & Hunter (1998) Massey, P., & Hunter, D. A. 1998 Star Formation in R136: A Cluster of O3 Stars Revealed by Hubble Space Telescope Spectroscopy. Astrophysical Journal 493, 180 * McGinn et al. (1989) McGinn, M. T., Sellgren, K., Becklin, E. E., & Hall, D. N. B. 1989 Stellar kinematics in the Galactic center. Astrophysical Journal 338, 824 * McGrath, Goss, & De Pree (2004) McGrath, E. J., Goss, W. M., & De Pree, C. G. 2004 H2O Masers in W49 North and Sagittarius B2. Astrophysical Journal Supplements 155, 577 * McMillan & Portegies Zwart (2003) McMillan, S. L. W., & Portegies Zwart, S. F. 2003 The Fate of Star Clusters near the Galactic Center. I. Analytic Considerations. Astrophysical Journal 596, 314 * Mercer et al. (2005) Mercer, E. P., et al. 2005 New Star Clusters Discovered in the GLIMPSE Survey. Astrophysical Journal 635, 560 * Milosavljević & Loeb (2004) Milosavljević, M., & Loeb, A. 2004 The Link between Warm Molecular Disks in Maser Nuclei and Star Formation near the Black Hole at the Galactic Center. Astrophysical Journal Letters 604, L45 * Moffat et al. (2002) Moffat, A. F. J., et al. 2002 Galactic Starburst NGC 3603 from X-Rays to Radio. Astrophysical Journal 573, 191 * Moneti, Glass, & Moorwood (1994) Moneti, A., Glass, I. S., & Moorwood, A. F. M. 1994 Spectroscopy and Further Imaging of IRAS Sources Near the Galactic Centre. Monthly Notices of the Royal Astronomical Society 268, 194 * Moneti et al. (2001) Moneti, A., Stolovy, S., Blommaert, J. A. D. L., Figer, D. F., & Najarro, F. 2001 Mid-infrared imaging and spectroscopy of the enigmatic cocoon stars in the Quintuplet Cluster. Astronomy & Astrophysics 366, 106 * Monnier, Tuthill, & Danchi (1999) Monnier, J. D., Tuthill, P. G., & Danchi, W. C. 1999 Pinwheel Nebula around WR 98A. Astrophysical Journal Letters 525, L97 * Morris (1993) Morris, M. 1993 Massive star formation near the Galactic center and the fate of the stellar remnants. Astrophysical Journal 408, 496 * Morris, Ghez, & Becklin (1999) Morris, M., Ghez, A. M., & Becklin, E. E. 1999 The galactic center black hole: clues for the evolution of black holes in galactic nuclei. Advances in Space Research 23, 959 * Morris & Serabyn (1996) Morris, M., & Serabyn, E. 1996 The Galactic Center Environment. Annual Review of Astronomy and Astrophysics 34, 645 * Muno et al. (2006) Muno, M. P., Bower, G. C., Burgasser, A. J., Baganoff, F. K., Morris, M. R., & Brandt, W. N. 2006 Isolated, Massive Supergiants near the Galactic Center. Astrophysical Journal 638, 183 * Nagata et al. (1995) Nagata, T., Woodward, C. E., Shure, M., & Kobayashi, N. 1995 Object 17: Another cluster of emission-line stars near the galactic center. Astronomical Journal 109, 1676 * Nagata et al. (1990) Nagata, T., Woodward, C. E., Shure, M., Pipher, J. L., & Okuda, H. 1990 AFGL 2004 - an infrared quintuplet near the Galactic center. Astrophysical Journal 351, 83 * Najarro (1995) Najarro, F. 1995 Quantitative Optical and Infrared Spectroscopy of Extreme Luminous Blue Supergiants. Ph.D. Thesis , * Najarro et al. (2004) Najarro, F., Figer, D. F., Hillier, D. J., & Kudritzki, R. P. 2004 Metallicity in the Galactic Center: The Arches Cluster. Astrophysical Journal Letters 611, L105 * Najarro et al. (1994) Najarro, F., Hillier, D. J., Kudritzki, R. P., Krabbe, A., Genzel, R., Lutz, D., Drapatz, S., & Geballe, T. R. 1994 The nature of the brightest galactic center HeI emission line star. Astronomy & Astrophysics 285, 573 * Najarro et al. (1997) Najarro, F., Krabbe, A., Genzel, R., Lutz, D., Kudritzki, R. P., & Hillier, D. J. 1997 Quantitative spectroscopy of the HeI cluster in the Galactic center.. Astronomy & Astrophysics 325, 700 * Nayakshin & Cuadra (2005) Nayakshin, S., & Cuadra, J. 2005 A self-gravitating accretion disk in Sgr A* a few million years ago: Is Sgr A* a failed quasar?. Astronomy & Astrophysics 437, 437 * Narayanan, Gould, & Depoy (1996) Narayanan, V. K., Gould, A., & Depoy, D. L. 1996 Luminosity Function of the Perigalactocentric Region. Astrophysical Journal 472, 183 * Nayakshin & Sunyaev (2005) Nayakshin, S., & Sunyaev, R. 2005 The ‘missing’ young stellar objects in the central parsec of the Galaxy: evidence for star formation in a massive accretion disc and a top-heavy initial mass function. Monthly Notices of the Royal Astronomical Society 364, L23 * Negueruela & Clark (2005) Negueruela, I., & Clark, J. S. 2005 Further Wolf-Rayet stars in the starburst cluster Westerlund 1. Astronomy & Astrophysics 436, 541 * Nelan et al. (2004) Nelan, E. P., Walborn, N. R., Wallace, D. J., Moffat, A. F. J., Makidon, R. B., Gies, D. R., & Panagia, N. 2004 Resolving OB Systems in the Carina Nebula with the Hubble Space Telescope Fine Guidance Sensor. Astronomical Journal 128, 323 * Oey & Clarke (2005) Oey, M. S., & Clarke, C. J. 2005 Statistical Confirmation of a Stellar Upper Mass Limit. Astrophysical Journal Letters 620, L43 * Okuda et al. (1990) Okuda, H., et al. 1990 An infrared quintuplet near the Galactic center. Astrophysical Journal 351, 89 * Paumard et al. (2006) Paumard, T., et al. 2006, The Two Young Star Disks in the Central Parsec of the Galaxy: Properties, Dynamics, and Formation. Astrophysical Journal 643, 1011 * Perets, Hopman, & Alexander (2006) Perets, H. B., Hopman, C., & Alexander, T. 2006 Massive perturber-driven interactions of stars with a massive black hole. ArXiv Astrophysics e-prints , arXiv:astro-ph/0606443 * Phinney (1989) Phinney, E. S. 1989 Manifestations of a Massive Black Hole in the Galactic Center. IAU Symp. 136: The Center of the Galaxy 136, 543 * Plante, Lo, & Crutcher (1995) Plante, R. L., Lo, K. Y., & Crutcher, R. M. 1995 The magnetic fields in the galactic center: Detection of H1 Zeeman splitting. Astrophysical Journal Letters 445, L113 * Portegies Zwart et al. (2002) Portegies Zwart, S. F., Makino, J., McMillan, S. L. W., & Hut, P. 2002 The Lives and Deaths of Star Clusters near the Galactic Center. Astrophysical Journal 565, 265 * Portegies Zwart, McMillan, & Gerhard (2003) Portegies Zwart, S. F., McMillan, S. L. W., & Gerhard, O. 2003 The Origin of IRS 16: Dynamically Driven In-Spiral of a Dense Star Cluster to the Galactic Center? Astrophysical Journal 593, 352 * Ramírez et al. (2000) Ramírez, S. V., Sellgren, K., Carr, J. S., Balachandran, S. C., Blum, R., Terndrup, D. M., & Steed, A. 2000 Stellar Iron Abundances at the Galactic Center. Astrophysical Journal 537, 205 * Rieke & Rieke (1988) Rieke, G. H., & Rieke, M. J. 1988 Stellar velocities and the mass distribution in the Galactic center. Astrophysical Journal Letters 330, L33 * Rieke, Telesco, & Harper (1978) Rieke, G. H., Telesco, C. M., & Harper, D. A. 1978 The infrared emission of the Galactic center. Astrophysical Journal 220, 556 * Rockefeller et al. (2005) Rockefeller, G., Fryer, C. L., Melia, F., & Wang, Q. D. 2005 Diffuse X-Rays from the Arches and Quintuplet Clusters. Astrophysical Journal 623, 171 * Salpeter (1955) Salpeter, E. E. 1955 The Luminosity Function and Stellar Evolution. Astrophysical Journal 121, 161 * Sato et al. (2000) Sato, F., Hasegawa, T., Whiteoak, J. B., & Miyawaki, R. 2000 Cloud Collision-induced Star Formation in Sagittarius B2. I. Large-Scale Kinematics. Astrophysical Journal 535, 857 * Schödel et al. (2002) Schödel, R., et al. 2002 A star in a 15.2-year orbit around the supermassive black hole at the centre of the Milky Way. Nature 419, 694 * Schödel et al. (2003) Schödel, R., Ott, T., Genzel, R., Eckart, A., Mouawad, N., & Alexander, T. 2003 Stellar Dynamics in the Central Arcsecond of Our Galaxy. Astrophysical Journal 596, 1015 * Schödel et al. (2006) Schödel, R., et al. 2006 Astronomy & Astrophysics , in preparation * Sellgren et al. (1987) Sellgren, K., Hall, D. N. B., Kleinmann, S. G., & Scoville, N. Z. 1987 Radial velocities of late-type stars in the galactic center. Astrophysical Journal 317, 881 * Sellgren et al. (1990) Sellgren, K., McGinn, M. T., Becklin, E. E., & Hall, D. N. 1990 Velocity dispersion and the stellar population in the central 1.2 parsecs of the Galaxy. Astrophysical Journal 359, 112 * Serabyn & Lacy (1985) Serabyn, E., & Lacy, J. H. 1985 Forbidden NE II observations of the galactic center - Evidence for a massive block hole. Astrophysical Journal 293, 445 * Serabyn & Morris (1996) Serabyn, E., & Morris, M. 1996 Sustained star formation in the central stellar cluster of the Milky Way.. Nature 382, 602 * Serabyn, Shupe, & Figer (1998) Serabyn, E., Shupe, D., & Figer, D. F. 1998 An extraordinary cluster of massive stars near the centre of the Milky Way.. Nature 394, 448 * Shields & Ferland (1994) Shields, J. C., & Ferland, G. J. 1994 Nebular properties and the ionizing radiation field in the galactic center. Astrophysical Journal 430, 236 * Sjouwerman et al. (1999) Sjouwerman, L. O., Habing, H. J., Lindqvist, M., van Langevelde, H. J., & Winnberg, A. 1999 OH/IR Stars as Signposts for Ancient Starburst Activity in the Galactic Center. ASP Conf. Ser. 186: The Central Parsecs of the Galaxy 186, 379 * Skinner et al. (2006) Skinner, S. L., Simmons, A. E., Zhekov, S. A., Teodoro, M., Damineli, A., & Palla, F. 2006 A Rich Population of X-Ray-emitting Wolf-Rayet Stars in the Galactic Starburst Cluster Westerlund 1. Astrophysical Journal Letters 639, L35 * Smith (2006) Smith, N. 2006 Eruptive Mass Loss in Very Massive Stars and Population III Stars. ArXiv Astrophysics e-prints , arXiv:astro-ph/0607457 * Stark et al. (1989) Stark, A. A., Bally, J., Wilson, R. W., & Pound, M. W. 1989 Molecular Line Observations of the Galactic Center Region. IAU Symp. 136: The Center of the Galaxy 136, 129 * Stolte (2003) Stolte, A. 2003, PhD Thesis, University of Heidelberg * Stolte et al. (2003) Stolte, A., Brandner, W., Grebel, E. K., Figer, D. F., Eisenhauer, F., Lenzen, R., & Harayama, Y. 2003 NAOS-CONICA performance in a crowded field - the Arches cluster. The Messenger 111, 9 * Stolte et al. (2005) Stolte, A., Brandner, W., Grebel, E. K., Lenzen, R., & Lagrange, A.-M. 2005 The Arches Cluster: Evidence for a Truncated Mass Function? Astrophysical Journal Letters 628, L113 * Stolte et al. (2002) Stolte, A., Grebel, E. K., Brandner, W., & Figer, D. F. 2002 The mass function of the Arches cluster from Gemini adaptive optics data. Astronomy & Astrophysics 394, 459 * Subr & Karas (2005) Subr, L. & Karas 2005 On highly eccentric stellar trajectories interacting with a self-gravitating disc in Sgr A*. Astronomy & Astrophysics 433, 405 * Takagi, Murakami, & Koyama (2002) Takagi, S.-i., Murakami, H., & Koyama, K. 2002 X-Ray Sources and Star Formation Activity in the Sagittarius B2 Cloud Observed with Chandra. Astrophysical Journal 573, 275 * Tamblyn et al. (1996) Tamblyn, P., Rieke, G. H., Hanson, M. M., Close, L. M., McCarthy, D. W., Jr., & Rieke, M. J. 1996 The Peculiar Population of Hot Stars at the Galactic Center. Astrophysical Journal 456, 206 * Tanner et al. (2006) Tanner, A., et al. 2006 High Spectral Resolution Observations of the Massive Stars in the Galactic Center. Astrophysical Journal 641, 891 * Timmermann et al. (1996) Timmermann, R., Genzel, R., Poglitsch, A., Lutz, D., Madden, S. C., Nikola, T., Geis, N., & Townes, C. H. 1996 Far-Infrared Observations of the Radio Arc (Thermal Arches) in the Galactic Center. Astrophysical Journal 466, 242 * Tuthill, Monnier, & Danchi (1999) Tuthill, P. G., Monnier, J. D., & Danchi, W. C. 1999 A dusty pinwheel nebula around the massive star WR 104.. Nature 398, 487 * Tuthill et al. (2006) Tuthill, P., Monnier, J., Tanner, A., Figer, D., & Ghez, A. 2006 “Pinwheels” discovered in the Quintuplet cluster. _Science_ 313, 935. * Uchida & Guesten (1995) Uchida, K. I., & Guesten, R. 1995 The large-scale magnetic field in the Galactic Center.. Astronomy & Astrophysics 298, 473 * van der Hucht (2001) van der Hucht, K. A. 2001 The VIIth catalogue of galactic Wolf-Rayet stars. New Astronomy Review 45, 135 * Walborn, Maíz-Apellániz, & Barbá (2002) Walborn, N. R., Maíz-Apellániz, J., & Barbá, R. H. 2002 Further Insights into the Structure of 30 Doradus from the Hubble Space Telescope Instruments. Astronomical Journal 124, 1601 * Wang, Dong, & Lang (2006) Wang, Q. D., Dong, H., & Lang, C. 2006 The Interplay between Star Formation and the Nuclear Environment of our Galaxy: Deep X-ray Observations of the Galactic Center Arches and Quintuplet Clusters. ArXiv Astrophysics e-prints , arXiv:astro-ph/0606282 * Weidner & Kroupa (2004) Weidner, C., & Kroupa, P. 2004 Evidence for a fundamental stellar upper mass limit from clustered star formation. Monthly Notices of the Royal Astronomical Society 348, 187 * Weidner & Kroupa (2006) Weidner, C., & Kroupa, P. 2006 The maximum stellar mass, star-cluster formation and composite stellar populations. Monthly Notices of the Royal Astronomical Society 365, 1333 * Williams, van der Hucht, & The (1987) Williams, P. M., van der Hucht, K. A., & The, P. S. 1987 Variable dust emission from Wolf-Rayet stars. Quarterly Journal of the Royal Astronomical Society 28, 248 * Yusef-Zadeh et al. (2002) Yusef-Zadeh, F., Law, C., Wardle, M., Wang, Q. D., Fruscione, A., Lang, C. C., & Cotera, A. 2002 Detection of X-Ray Emission from the Arches Cluster near the Galactic Center. Astrophysical Journal 570, 665 * Yusef-Zadeh & Morris (1987) Yusef-Zadeh, F., & Morris, M. 1987 Structural details of the Sagittarius A complex - Evidence for a large-scale poloidal magnetic field in the Galactic center region. Astrophysical Journal 320, 545 * Yusef-Zadeh et al. (2003) Yusef-Zadeh, F., Nord, M., Wardle, M., Law, C., Lang, C., & Lazio, T. J. W. 2003 Nonthermal Emission from the Arches Cluster (G0.121+0.017) and the Origin of gamma-ray Emission from 3EG J1746-2851. Astrophysical Journal Letters 590, L103 * Zhao et al. (1993) Zhao, J.-H., Desai, K., Goss, W. M., & Yusef-Zadeh, F. 1993 VLA Radio Recombination Line Observations of Ionized Gas in the -30 Kilometers per Second Molecular Cloud (G0.04+0.03) near the Galactic Center. I. The Discrete Radio Sources. Astrophysical Journal 418, 235

arxiv-papers

2024-09-04T02:48:54.280437

0803.1687

# Exact Spherically Symmetric Solutions in Massive Gravity Z. Berezhiani,​a D. Comelli,​b F. Nesti,​a L. Piloa aDipartimento di Fisica, Università di L’Aquila, I-67010 L’Aquila, and INFN, Laboratori Nazionali del Gran Sasso, I-67010 Assergi, Italy bINFN, Sezione di Ferrara, I-35131 Ferrara, Italy ###### Abstract: A phase of massive gravity free from pathologies can be obtained by coupling the metric to an additional spin-two field. We study the gravitational field produced by a static spherically symmetric body, by finding the exact solution that generalizes the Schwarzschild metric to the case of massive gravity. Besides the usual $1/r$ term, the main effects of the new spin-two field are a shift of the total mass of the body and the presence of a new power-like term, with sizes determined by the mass and the shape (the radius) of the source. These modifications, being source dependent, give rise to a dynamical violation of the Strong Equivalence Principle. Depending on the details of the coupling of the new field, the power-like term may dominate at large distances or even in the ultraviolet. The effect persists also when the dynamics of the extra field is decoupled. ## 1 Introduction The search for large-distance modified theories of gravity, motivated by the evidence for the cosmological acceleration, has stimulated a number of studies in the recent years. The main goal has been to look for a massive deformation of standard general relativity, featuring a large distance (infrared) modification of the Newtonian gravitational potential, and massive gravitons. The idea of considering a Lorentz-invariant theory of a massive spin-two field dates back to 1939 [1]: the resulting theory is plagued by a number of diseases that make it unphysical, besides being phenomenologically excluded. In particular, the modification of the Newtonian potentials is not continuous when the mass $m^{2}$ vanishes, giving a large correction (25%) to the light deflection from the sun that is experimentally excluded [2]. A possible way to circumvent the physical consequences of the discontinuity was proposed in [3]; the idea is that in the Fierz Pauli theory (FP) the linearized approximation breaks down near the star and an improved perturbative expansion must be used, leading to a continuous result when $m\to 0$. Whether the solution associated with the improved perturbative expansion valid near the star can be extended up to infinity is an open problem [4]. In addition, FP is problematic as an effective theory at the quantum level. Regarding FP as a gauge theory where the gauge symmetry is broken by a explicit mass term $m$, one would expect a cutoff $\Lambda_{2}\sim mg^{-1}=(mM_{pl})^{1/2}$, however the real cutoff is $\Lambda_{5}=(m^{4}M_{pl})^{1/5}$ [5] much lower than $\Lambda_{2}$. A would- be Goldstone mode is responsible for the extreme ultraviolet sensitivity of FP theory, that becomes totally unreliable in the absence of proper completion. These issues cast a shadow on the the possibility of realizing a Lorentz- invariant theory of massive gravity [6]. It was recently noted that by allowing lorentz-breaking mass terms for the graviton the resulting theory can be physically viable [7], being free from pathologies such as ghosts or low strong coupling scales, while still leading to modified gravity. Since the mass terms break anyway the diffeomorphisms invariance, this possibility was analized mainly in a model-independent way, by reintroducing the goldstone fields of the broken gauge invariance, and by studying their dynamics [5, 8]; we refer to a recent review for the status and results in this direction [9]. This approach has the power of being model independent, but the advantage turns into a difficulty when investigating the concrete behavior of solutions. In [10] we considered a class of theories that generate Lorentz-breaking mass terms for the graviton, by coupling the metric to an additional spin-two field. This system was originally introduced and analyzed by Isham, Salam and Strathdee [11, 12] and reanalyzed more recently in [13, 14, 15]. While this approach may seem antieconomical, we stress that this is the simplest model that can explain dynamically the emergence of lorentz breaking and give mass to the graviton. What happens is that the two tensor fields lead in general to two coexisting and different backgrounds, inducing Lorentz-breaking mass terms at linearized order. For a general discussion on the consequences of Lorentz Breaking see [16]. The linearized analysis showed that only two gravitons propagate, one massive and the other massless, both with two polarization states, representing two kinds of gravitational waves (GW). These are the only states in the theory that feel the Lorentz breaking, showing a frame-dependence that may be measured at future GW detectors. In addition, the linearized gravitational potential differs in a crucial way from the Newtonian one: it contains a new term that is linearly growing with distance. Of course, this signals the breakdown of perturbation theory at large distances, and in this regime the theory should be treated fully nonlinearly. This fact is not surprising, since one has effectively introduced nonlinear interactions, and therefore antiscreening may be present also at classical level like in non-abelian gauge theories. We believe that this is a general feature of massive gravity theories due to the presence in the full theory of nonderivative interaction terms. In such situations the linearized analysis is of limited reach, and we are forced to find exact classical solutions to be compared with the standard Schwarzschild metric. Though in general this a very hard task, we have managed to find a whole class of interaction terms for which nontrivial and rather interesting exact solutions can be found. After describing the setup and the flat backgrounds in section 2 ad 2.1, we review the linearized analysis and its problems in section 2.2. We then describe the spherically symmetric solutions in section 3, that we match with an interior star solution to estimate the modifications of the gravitational potential as a function of the source parameters. We also comment on the properties of these solutions and of the whole theory in the interesting limit when the second metric decouples, leaving just one massive gravity theory with modified Schwarzschild solutions, as well as in the Lorentz-invariant limit. ## 2 The model Consider a gravity theory in which, besides our standard metric field, an additional rank-2 tensor is introduced in in the form of a bimetric theory. The action is taken as111We use the mostly plus convention for the metric. Indices of type 1(2) are raised/lowered with $g_{1}(g_{2})$. $S=\int\\!\\!d^{4}x\,\left[\sqrt{-g_{1}}\left(M_{pl1}^{2}R_{1}+{\cal L}_{1}\right)+\sqrt{-g_{2}}\left(M_{pl2}^{2}R_{2}+{\cal L}_{2}\right)-4(g_{1}g_{2})^{1/4}V(X)\right]\,,$ (1) and for symmetry each rank-2 field is coupled to its own matter with the respective Lagrangians ${\cal L}_{1,2}$. In the interaction term we only consider non-derivative couplings. The only invariant tensor, without derivatives, that can be written out of the two metrics is $X^{\mu}_{\nu}=g^{\mu\alpha}_{1}{g_{2}}_{\alpha\nu}$, and then $V$ is taken as a function of the four independent scalars $\\{\tau_{n}=\text{tr}(X^{n}),\ n=1,2,3,4\\}$ made out of $X$. The cosmological terms can be included in $V$, e.g. $V_{\Lambda_{1}}=\Lambda_{1}q^{-1/4}$, with $q=\det X=g_{2}/g_{1}$. Then the (modified) Einstein equations read $\displaystyle M_{pl1}^{2}\,{E_{1}}^{\mu}_{\nu}+Q_{1}{}^{\mu}_{\nu}=\frac{1}{2}\,{T_{1}}^{\mu}_{\nu}$ (2) $\displaystyle M_{pl2}^{2}\,{E_{2}}^{\mu}_{\nu}+Q_{2}{}^{\mu}_{\nu}=\frac{1}{2}\,{T_{2}}^{\mu}_{\nu}\,,$ (3) where we defined the effective energy-momentum tensors induced by the interaction: $\displaystyle Q_{1}{}_{\nu}^{\mu}$ $\displaystyle=$ $\displaystyle q^{1/4}\left[V\delta^{\mu}_{\nu}-4(V^{\prime}X)^{\mu}_{\nu}\right]$ (4) $\displaystyle Q_{2}{}_{\nu}^{\mu}$ $\displaystyle=$ $\displaystyle q^{-1/4}\left[V\delta^{\mu}_{\nu}+4(V^{\prime}X)^{\mu}_{\nu}\right],$ (5) with $(V^{\prime})^{\mu}_{\nu}=\partial V/\partial X_{\mu}^{\nu}$. Indeed, the field $g_{2}$ plays the role of matter in the equations of motion for $g_{1}$, and viceversa for $g_{1}$. The Einstein tensors satisfy the corresponding contracted Bianchi identities222$\nabla_{1/2}$ denotes the covariant derivative associated to the Levi-Civita connection of $g_{1/2}$. $g_{1}^{\alpha\nu}{\nabla_{1}}_{\alpha}{E_{1}}_{\mu\nu}=\nabla_{1}^{\nu}{E_{1}}_{\mu\nu}=0\qquad g_{2}^{\alpha\nu}{\nabla_{2}}_{\alpha}{E_{2}}_{\mu\nu}=\nabla_{2}^{\nu}{E_{2}}_{\mu\nu}=0\,.$ (6) that follows from the invariance of the respective Einstein-Hilbert terms under common diffeomorphisms $\delta{g_{1}}_{\mu\nu}=2{g_{1}}_{\alpha(\mu}{\nabla_{1}}_{\nu)}\xi^{\alpha}\qquad\delta{g_{2}}_{\mu\nu}=2{g_{2}}_{\alpha(\mu}{\nabla_{2}}_{\nu)}\xi^{\alpha}\,.$ (7) The interaction term is also separately invariant and we can derive conservation laws for $Q_{1}$ and $Q_{2}$ similar to the conservation of the energy-momentum tensor in GR: $\begin{split}&\nabla_{1}^{\nu}{Q_{1}}_{\mu\nu}=0\qquad\text{on shell for }g_{2}\\\ &\nabla_{2}^{\nu}{Q_{2}}_{\mu\nu}=0\qquad\text{on shell for }g_{1}\,.\end{split}$ (8) These identities are quite powerful; for instance they allow to solve completely the simplest of these models, when $V$ is a function of $q$ only. This peculiar case is discussed in appendix D. ### 2.1 Asymptotic solutions At infinity, far from all the sources, we expect that $g_{1}$ and $g_{2}$ are maximally symmetric, setting up the benchmark for the asymptotic behavior of all solutions of the EOM. Denoting with $-{\mathcal{K}}_{a}/4$ the constant scalar curvature of $g_{a\,\mu\nu}$, i.e. $M_{pl\,a}^{2}\,{E_{a}}_{\mu\nu}={\mathcal{K}}_{a}\,{g_{a}}_{\mu\nu}\,,\qquad(a=1,2)$ (9) the equations (2)-(3) read $\displaystyle 2V+\left(q^{-1/4}\mathcal{K}_{1}+q^{1/4}\mathcal{K}_{2}\right)=0$ (10) $\displaystyle 8\left(V^{\prime}X\right)^{\mu}_{\nu}+\delta^{\mu}_{\nu}\,\left(q^{1/4}\mathcal{K}_{2}-q^{-1/4}\mathcal{K}_{1}\right)=0\,.$ (11) and these equations can be solved for specific ansätze. In order to study the properties of this model for asymptotically flat spaces, we analyze first the biflat solutions, $\mathcal{K}_{1}=\mathcal{K}_{2}=0$. Eqs. (10)-(11) yield: $\displaystyle V^{\prime}{}_{\mu}^{\nu}=0\,,\qquad V=0\,.$ (12) Assuming that rotational symmetry is preserved and that the two metrics have the same signature, the biflat solution can written in the following form $\begin{split}&{\bar{g}}_{1\,\mu\nu}=\eta_{\mu\nu}\equiv\text{diag}(-1,1,1,1)\\\\[2.15277pt] &{\bar{g}}_{2\,\mu\nu}=\omega^{2}\,\text{diag}(-c^{2},1,1,1)\,,\end{split}$ (13) where $c$ parametrizes the speed of light in sector 2 and $\omega$ is the relative conformal factor. Eqs. (12) correspond to three independent equations $V=0$, $V^{\prime}{}_{0}^{0}=0$ and $V^{\prime}{}_{i}^{i}=0$, where $0$ and $i=1,2,3$ stand for temporal and spatial indices. Therefore, two of these equations determine the values of two parameters $c$ and $\omega$, while the third represents a fine tuning condition for the function $V$, necessary to ensure flatness. The same sort of fine tuning is necessary in the context of normal GR to set the cosmological term to zero. Therefore, for a generic function $V(X)$ we expect to have a Lorentz Breaking (LB) solution with $c\neq 1$, and hence a preferred reference frame (13) in which both metrics are diagonal. Asymptotic flat solutions should approach (13) in a suitable coordinate system. In addition, there exists also a Lorentz Invariant (LI) solution, with $c=1$: in this case two equations coincide $V^{\prime}{}_{0}^{0}=V^{\prime}{}_{i}^{i}$, and can be used to determine the value of $\omega$. Summarizing, asymptotically the solutions fall in two branches: LI with $c=1$, and LB with $c\neq 1$.333 In the special case when $V=V(\det X)$, Bianchi identities force $\det X$ to be constant, and there are additional gauge symmetries that allow to set $c=1$. As a result the branches are equivalent (appendix D). The LB branch is of particular interest since it naturally allows for consistent massive deformations of gravity [10]. ### 2.2 Review (and critics) of the linearized analysis In [10] we performed a linearized analysis around the biflat background, that we report here for the LB branch. In addition to the kinetic terms, the linearized action contains a Lorentz- breaking mass term for the fluctuations $h_{a\,\mu\nu}=g_{a\,\mu\nu}-\bar{g}_{a\,\mu\nu}$ ($a=1,2$). Since on the biflat background $V=V^{\prime}{}_{\mu}^{\nu}=0$, one can expand the potential at second order in the fluctuations $Y=\bar{X}{\bar{g}}_{2}^{-1}h_{2}-{\bar{g}}_{1}^{-1}h_{1}\,\bar{X}$, and define the mass lagrangian: $\displaystyle{\cal L}_{m}$ $\displaystyle=$ $\displaystyle-2\left({\bar{g}}_{1}{\bar{g}}_{2}\right)^{1/4}\,\text{tr}\left[Y\,V^{\prime\prime}(\bar{X})\,Y\right]$ (14) $\displaystyle\equiv$ $\displaystyle\frac{1}{4}\Big{(}h_{00}^{t}\,{\cal M}_{0}\,h_{00}+2h_{0i}^{t}\,{\cal M}_{1}\,h_{0i}-h_{ij}^{t}\,{\cal M}_{2}\,h_{ij}+h_{ii}^{t}\,{\cal M}_{3}\,h_{ii}-2h_{ii}^{t}\,{\cal M}_{4}\,h_{00}\Big{)}\,.$ Here $h_{\mu\nu}=\\{h_{1\mu\nu},h_{2\mu\nu}\\}$ is the column vector of fluctuations and ${\cal M}_{0,1,2,3,4}$ are 2$\times$2 mass matrices. It is then crucial to realize that, due to linearized gauge invariance (that we remark is never broken) these matrices are of rank-one; one can write $\displaystyle{\cal M}_{0}$ $\displaystyle=$ $\displaystyle\lambda_{0}\,{\cal C}^{-2}{\cal P}{\cal C}^{-2}$ $\displaystyle\qquad{\cal M}_{2,3}$ $\displaystyle=$ $\displaystyle\lambda_{2,3}\,\,{\cal P}\,,\qquad\qquad{\cal P}\equiv\begin{pmatrix}\ 1&-1\\\ -1\,&\ 1\end{pmatrix}\,,\quad{\cal C}\equiv\begin{pmatrix}\ 1\ &\\\ &\ c\ \end{pmatrix}$ (15) $\displaystyle{\cal M}_{4}$ $\displaystyle=$ $\displaystyle\lambda_{4}\,{\cal C}^{-2}{\cal P}$ where $\lambda_{0,2,3,4}$ depend on the potential. In addition, due to the LB, ${\cal M}_{1}$ vanishes regardless of the potential $V$. This fact leads to a well defined phase of linearized massive gravity.444The vanishing of the second eigenvalue of ${\cal M}_{1}$ can be understood by noting that $h_{1}{}_{0i}-h_{2}{}_{0i}$ is a goldstone direction, corresponding to the broken boosts in the LB background. In sec. 3.6 we comment on the fate of this condition on nontrivial backgrounds. In this phase, the only propagating states are the two spin-2 tensor components of the fluctuations (two polarizations each) corresponding to two gravitons, of which one is massless and the other has mass $\lambda_{2}$. Their dispersion relation is non-linear due to their mixing and their different propagating speeds [10]. The other components, scalars and vectors, do not propagate, and therefore discontinuity and strong coupling problems are absent in this phase. They however mediate instantaneous interactions, so the Newtonian potentials that one finds at linearized level are then drastically modified; for example in sector 1, the potential from a point-like source $M_{1}$ is: $\Phi_{1}=-\frac{GM_{1}}{r}+GM_{1}\mu^{2}r\,,$ (16) where $\mu^{2}\equiv\frac{\lambda_{2}}{2M_{1}^{2}}\frac{3\lambda_{4}^{2}-\lambda_{0}(3\lambda_{3}-\lambda_{2})}{\lambda_{4}^{2}-\lambda_{0}(\lambda_{3}-\lambda_{2})}$ (17) and for later reference note that $\mu^{2}$ may be negative. The linearly growing term in (16) signals the breakdown of perturbation theory at distances larger than $r_{IR}=(GM_{1}\mu^{2})^{-1}$ [8, 10], and one usually considers the solution to be valid as long as the potential stays in the weak field regime. However, one should note that the linear term in (16) is induced by an other scalar field having an instantaneous interaction and acting as a source for $\Phi$ (see e.g. [22, 20]). It is then easy to realize that non-linear corrections to this field can drastically modify the IR behavior, even in the weak-field regime. We can clarify this point by showing, as an example, two systems of differential equations that differ by nonlinear terms and have drastically different IR behavior $\left\\{\begin{array}[]{l}\Delta\Phi+\mu^{2}\sigma=M\delta^{3}(x)\\\ \Delta\sigma=M\delta^{3}(x)\end{array}\right.\qquad\quad\left\\{\begin{array}[]{l}\Delta\Phi+\mu^{2}\sigma=M\delta^{3}(x)\\\ \Delta\sigma+\lambda\,\sigma^{2}=M\delta^{3}(x)\,,\end{array}\right.$ (18) Here $\Phi$ is a scalar field (mimicking the gravitational potential) $\sigma$ is an additional scalar field coupled to it by a mass term (and $\Delta$ is the laplacian). While in the first system $\sigma\sim M/r$ and this induces a linear term in $\Phi$ like in (16), in the second system $\sigma$ drops to zero faster than $M/r$ so that the bad behavior of $\Phi$ is cured. What happens is that the IR behavior is dominated by a non-linear term, because effectively $\Delta\to 0$ at large distances. We incidentally point out that standard GR is safe in this respect, because nonlinear terms coming from the Einstein tensor are always accompanied by two derivatives and thus are equally suppressed at large distances. We are thus led to the conclusion that in massive gravity the situation is similar to non-abelian gauge theories, where the large distance behavior is generically non-trivial and inaccessible to the linearized approximation. We recall that in Yang-Mills theories, nonabelian configurations of charges can lead to non coulomb-like classical solutions, screening or even anti-screening the charge, leading also to infinite energy configurations [17]. In this situation what one may try is to really look at higher orders and maybe retain the first terms that are relevant at large distance. This approach would require the painful procedure of defining the gauge invariant fields at higher orders, and would also lead to nonlinear terms mixing scalars, vectors and tensors. Instead of following this approach, we find more instructive to study the exact spherically symmetric solutions. ## 3 Exact spherically symmetric static solutions The Schwarzschild solution describes the spherically symmetric gravitational field produced by a spherically symmetric source. It is crucial to understand what kind of modification is introduced in this theory by the presence of a new spin 2 field. Spherical symmetry allows us to choose a coordinate patch $(t,r,\theta,\varphi)$ where $g_{1}$ and $g_{2}$ have the form $\displaystyle ds^{2}_{1}=-J\,dt^{2}+K\,dr^{2}+r^{2}\,d\Omega^{2}$ (19) $\displaystyle ds^{2}_{2}=-C\,dt^{2}+A\,dr^{2}+2D\,dtdr+\,B\,d\Omega^{2}\,.$ (20) and all the functions $J,K,C,A,D,B$ entering $g_{1}$ and $g_{2}$ are function of $r$ only. Notice that the off-diagonal piece $D$ cannot be gauged away. ### 3.1 Black hole solutions In the absence of matter, a number of interesting properties follow from the form of the Einstein tensors $E_{1}{}_{\nu}^{\mu}$, $E_{2}{}_{\nu}^{\mu}$ derived from (19)-(20) and do not depend on the chosen $V$. Following [12, 18] the spherically symmetric solutions can be divided in two classes: type I with $D\neq 0$ and type II with $D=0$. We shall focus here mainly on type I solutions. Since $E_{1}{}_{\nu}^{\mu}$ is diagonal by the choice of the first metric, then also $(V^{\prime}X)^{\mu}_{\nu}$ must be diagonal because of the EOM (2). The only possible source of a off-diagonal term in the RHS of (3) would be $(V^{\prime}X)^{\mu}_{\nu}$, so as a result also $E_{2}{}_{\nu}^{\mu}$ must be diagonal, i.e. $E_{2}{}_{r}^{t}=0$. For type I solutions, this condition amounts to a single equation: $AC+D^{2}=d_{2}\frac{(B^{\prime})^{2}}{4B}\,,$ (21) where $d_{2}$ is a constant. Incidentally using this relation it turns out that ${E_{2}}_{t}^{t}={E_{2}}_{r}^{r}$, then using (3) also $(V^{\prime}X)^{t}_{t}=(V^{\prime}X)^{r}_{r}$, and by (2) we have also that ${E_{1}}_{t}^{t}={E_{1}}_{r}^{r}$. This relation determines $K$ in terms of $J$ $K=\frac{d_{1}}{J}\,,$ (22) with $d_{1}$ an other constant. The metric 2 can be brought in a diagonal form by a coordinate change $dt=dt^{\prime}+dr\,D/C$. Thanks to (21), in the new coordinates we have $ds_{2}^{2}=-C\,{dt^{\prime}}^{2}+\frac{(B^{\prime})^{2}}{4B}\frac{d_{2}}{C}\,dr^{2}+B\,d\Omega^{2}$ (23) (and of course the metric 1 in no longer diagonal). Then by a suitable change of $r$ the metric 2 can also be put in a Schwarzschild-like form; setting $r^{\prime}=\sqrt{B(r)}$, we find $ds_{2}^{2}=-C\,{dt^{\prime}}^{2}+\frac{d_{2}}{C}\,{dr^{\prime}}^{2}+r^{\prime 2}\,d\Omega^{2}\,,\qquad C(r)=C(r^{\prime})\,,$ (24) which shows that $C$ is the physically relevant potential in sector 2. To proceed further a choice of $V$ is needed. In the existing literature essentially all the results are based on a potential $V_{\text{IS}}$ introduced in [19] and [11] in the context of hadronic physics.555In the years preceding QCD, the proposal of Isham, Salam and Strathdee was that of a second metric mediating a strongly coupled interaction, responsible for confinement of quarks inside tiny black holes. The motivation for this choice is probably due to the fact that $V_{\text{IS}}$ is the simplest potential producing a FP mass term in the (Lorentz-invariant) linearized limit: $\displaystyle V_{IS}$ $\displaystyle=$ $\displaystyle(\tau_{-2}-\tau_{-1}^{2}+6\tau_{-1}-12)$ $\displaystyle=$ $\displaystyle(g^{2\,\mu\nu}-g^{1\,\mu\nu})(g^{2\,\rho\sigma}-g^{1\,\rho\sigma})(g_{1\,\mu\rho}g_{1\,\nu\sigma}-g_{1\,\mu\nu}g_{1\,\rho\sigma})$ $\displaystyle\simeq$ $\displaystyle\text{tr}(h_{-}^{2})-\text{tr}(h_{-})^{2}\qquad\qquad\text{for}\ \bar{g}_{1}=\bar{g}_{2}=\eta\,.$ where $h_{-}{}_{\mu\nu}=h_{2}{}_{\mu\nu}-h_{1}{}_{\mu\nu}$. For $V_{\text{IS}}$ it was shown in [12] that type-I solutions are always Schwarzschild-(A)dS, and it was recently realized that these solutions are present for any potential [20]. It turns out that $V_{\text{IS}}$ can be deformed and there exists a whole family of potentials for which the exact spherically symmetric solutions can be found. Let us consider the family of potentials $\displaystyle V$ $\displaystyle=$ $\displaystyle a_{0}+a_{1}V_{1}+a_{2}V_{2}+a_{3}V_{3}+a_{4}V_{4}+b_{1}V_{-1}+b_{2}V_{-2}+b_{3}V_{-3}+b_{4}V_{-4}$ (26) $\displaystyle{}+q^{-1/4}\Lambda_{1}+q^{1/4}\Lambda_{2}\,,$ where we introduced the following combinations involving the generalized determinants (again $\tau_{n}=\text{tr}(X^{n})$ and $\epsilon$ is the 4-index antisymmetric symbol) $\begin{split}&V_{0}=\frac{1}{24|g_{2}|}(\epsilon\epsilon g_{2}g_{2}g_{2}g_{2})=1\equiv\frac{1}{24q}(\tau_{1}^{4}-6\,\tau_{2}\tau_{1}^{2}+8\,\tau_{1}\tau_{3}+3\,\tau_{2}^{2}-6\,\tau_{4})\\\ &V_{1}=\frac{1}{6|g_{2}|}(\epsilon\epsilon g_{2}g_{2}g_{2}g_{1})=(\tau_{-1})\equiv\frac{1}{6q}(\tau_{1}^{3}-3\,\tau_{2}\tau_{1}+2\,\tau_{3})\\\ &V_{2}=\frac{1}{2|g_{2}|}(\epsilon\epsilon g_{2}g_{2}g_{1}g_{1})=(\tau_{-1}^{2}-\tau_{-2})\equiv q^{-1}(\tau_{1}^{2}-\tau_{2})\\\ &V_{3}=\frac{1}{|g_{2}|}(\epsilon\epsilon g_{2}g_{1}g_{1}g_{1})=(\tau_{-1}^{3}-3\tau_{-2}\tau_{-1}+2\,\tau_{-3})\equiv 6\,q^{-1}\,\tau_{1}\\\ &V_{4}=\frac{1}{|g_{2}|}(\epsilon\epsilon g_{1}g_{1}g_{1}g_{1})=(\tau_{-1}^{4}-6\,\tau_{-2}\tau_{-1}^{2}+8\,\tau_{-1}\tau_{-3}+3\,\tau_{-2}^{2}-6\,\tau_{-4})\equiv 24\,q^{-1}\\\ \end{split}$ (27) and where $V_{-n}=V_{n}(X\to X^{-1})$. The cosmological constants $\Lambda_{1}$ and $\Lambda_{2}$ have been added to simplify the asymptotic flatness conditions. The Isham-Storey potential is recovered by setting $a_{0}=-12$, $a_{1}=6$, $a_{2}=1$ and $b_{1}=b_{2}=b_{3}=b_{4}=a_{3}=a_{4}=0$. Remarkably, the general combination $V$ of (26) leads to solvable equations for type I spherically symmetric solutions, and these can be found in a closed form (these equations are the main result of the paper):666This solvability is linked to the fact that the combinations $V_{n}$ are actually the coefficients of the secular equation of $X$ and are (multi)linear combinations of its eigenvalues $\lambda_{i}$: $V_{n}=\sum_{i_{1}>i_{2}\cdots>i_{n}}\lambda_{i_{1}}\lambda_{i_{2}}\cdots\lambda_{i_{n}}$. $\displaystyle J$ $\displaystyle=$ $\displaystyle\Big{[}1-2\frac{Gm_{1}}{r}+{\mathcal{K}}_{1}r^{2}\Big{]}+2G\,S\,r^{\gamma}\,,\qquad\qquad\qquad\qquad\quad KJ=1\,,$ (28) $\displaystyle C$ $\displaystyle=$ $\displaystyle c^{2}\omega^{2}\Big{[}1-2\frac{Gm_{2}}{\kappa\,r}+{\mathcal{K}}_{2}r^{2}\Big{]}-\frac{2G}{c\,\omega^{2}\kappa}S\,r^{\gamma}\,,\qquad\quad D^{2}+AC=c^{2}\omega^{4}$ (29) $\displaystyle B$ $\displaystyle=$ $\displaystyle\omega^{2}r^{2}\,,\qquad\qquad\qquad A=\omega^{2}\frac{\tilde{J}-\tilde{C}-\tilde{J}\,\tilde{S}\,r^{\gamma-2}}{\tilde{J}^{2}}\,,$ (30) with $\\{\tilde{J},\tilde{C}\\}=\\{J,C\\}/\omega^{4}(c^{2}+1)$, $\tilde{S}=S/\lambda_{2}\,[(c^{2}-1)(\gamma+1)(\gamma-2)/16\omega^{2}c^{1/2}(c^{2}+1)]$. The solution depends on the integration constants $m_{1},m_{2}$ and $S$ and we have introduced $G=1/16\pi M_{pl1}^{2}$ and $\kappa=M_{pl2}^{2}/M_{pl1}^{2}$. The values of $c^{2}$, $\gamma$ and of the graviton mass $\lambda_{2}/M_{pl1}^{2}$ of the linearized analysis (15) are given in terms of the coupling constants: $c^{2}=-\frac{\tilde{a}_{1}+4\tilde{a}_{2}+6\tilde{a}_{3}}{\tilde{b}_{1}+4\tilde{b}_{2}+6\tilde{b}_{3}}\,,\quad\gamma=-\frac{4[(\tilde{a}_{2}+3\tilde{a}_{3})-c^{2}(\tilde{b}_{2}+3\tilde{b}_{3})]}{c^{2}(\tilde{b}_{1}+4\tilde{b}_{2}+6\tilde{b}_{3})}\,,\quad\lambda_{2}=\frac{2(\gamma-2)}{\gamma}(\alpha_{2}+3\alpha_{3})\,,$ (31) where $\tilde{a}_{n}=\omega^{-2n}a_{n}$, $\tilde{b}_{n}=\omega^{2n}b_{n}$ and $\alpha_{n}=(\tilde{a}_{n}-c^{2}\tilde{b}_{n})(c^{2}-1)/c^{2}$. Notice that one may trade $\tilde{a}_{1}$, $\tilde{b}_{1}$ and e.g. $\tilde{a}_{2}$ for, respectively, $c^{2}$, $\gamma$ and the graviton mass $\lambda_{2}$, showing that these may take any value for this class of potentials. When $\gamma<2$, the ${\mathcal{K}}_{i}$ are proportional to the constant asymptotic curvatures of $g_{i}$; the explicit expressions are given in appendix A. Finally, $\omega^{2}$ is also in general a free parameter that determines for example $\Lambda_{2}$ to have ${\mathcal{K}}_{2}=0$, after having fine tuned $\Lambda_{1}$ to set ${\mathcal{K}}_{1}=0$. The expression (28) resembles the Schwarzschild-dS(AdS) solution but with a crucial difference: a $r^{\gamma}$ term of magnitude $S$ is present and it may alter significantly the behavior of the gravitational field, depending on whether $\gamma<-1$, $-1<\gamma<0$ or $\gamma>0$. Before discussing these solutions, let us comment on the Isham-Storey potential $V_{\text{IS}}$ used traditionally. Since in this case all $b_{n}$ vanish, it leads to a singular situation where $c^{2}\to\infty$ unless an additional fine tuning $\omega^{2}=2/3$ is performed. Even choosing this case, the linearized analysis is ill defined due to an enhanced gauge invariance (see [10] for the case of $\lambda_{\eta}=0$), and moreover from (31) one has $\gamma\to\infty$. This is the reason why only standard Schwarzschild-(A)dS solutions were found. Now, in order to shed light on the physical meaning of the various constants in the solution let us also compute the total gravitational energy, as measured with respect to backgrounds 1 or 2. In the stationary case this is the Komar energy that can be calculated as a surface integral on a sphere of large radius $r_{outer}\to\infty$ (see appendix C). We find, for the two fields: $\begin{split}&{\mathcal{E}}_{1}=m_{1}+S\,\gamma\,r_{\text{outer}}^{\gamma+1}\\\ &{\mathcal{E}}_{2}=m_{2}-\frac{c}{\omega^{2}}\,S\,\gamma\,r_{\text{outer}}^{\gamma+1}\,.\end{split}$ (32) From these expressions we see that only IR modifications with $\gamma<-1$ will lead to finite total energy. #### Case $\gamma<-1$: At very large distance the solution reduce to a maximally symmetric solution parametrized by ${\mathcal{K}}_{i}$. In particular one can set ${\mathcal{K}}_{1}={\mathcal{K}}_{2}=0$ with a single fine tuning, determining the asymptotic conformal factor $\omega^{2}$ as discussed above, so that the solution describes asymptotically flat metrics. Clearly because $\gamma<-1$, at large distances gravity is Newtonian, while at short/intermediate distances, depending on $S$, the presence of the additional spin 2 field has changed the nature of the gravitational force. Since the large distance behavior is Newtonian, the total energy is finite; taking $r_{outer}\to\infty$, we find ${\mathcal{E}}_{1}=m_{1}$, ${\mathcal{E}}_{2}=m_{2}$. In black hole solutions like these, $m_{1}$ and $m_{2}$ are just parameters, that can be related to the mass of a material object only when the solution is considered as the outer part of, for instance, a star. In the case of standard GR, for a star of radius $R$ and mass density $\rho$, the total gravitational energy $E$ is the total mass $M=4\pi R^{3}\rho/3$. Here, the interaction with $g_{2}$ is turned on and we expect a contribution to this energy given by the interaction term $Q_{1}$. Its size should be controlled by $V$ and by the matter itself, because this interaction energy also is turned on by the source. Moreover, by dimensional analysis the coefficient $S$ of the $r^{\gamma}$ term should also be a function of the _size_ of the object, and not only on its mass. This can be understood intuitively as the failure of the Gauss theorem due to the presence of $Q$ in the EOMs and of the $r^{\gamma}$ term in the solution. Accordingly, the separate contribution of $Q_{1}$ to the energy is not expressible as a flux on a 2-surface at infinity, as it happens for the total Komar energy. The explicit computation of this interaction energy for a star will be performed in section 3.3. #### Case $\gamma>-1$: For simplicity, also in this case we set ${\mathcal{K}}_{1}={\mathcal{K}}_{2}=0$ as discussed above, but note that because the new term induces a curvature $R\sim r^{\gamma-2}$, only when $\gamma<2$ we have that $g_{1}$ and $g_{2}$ are asymptotically flat and $\omega^{2}$ can be interpreted as an asymptotical conformal factor. For these choices of $\gamma$, we have a solution such that $Q_{1}$ does not vanish rapidly as $r\to\infty$, and compensates a slow fall-off (or rise!) of the gravitational field. However on dimensional grounds any fall-off slower than $1/r$ makes the Komar total energy infinite, and indeed when $r_{\text{outer}}\to\infty$ both ${\mathcal{E}}_{1}$ and ${\mathcal{E}}_{2}$ diverge making this configuration physically unfeasible. If spherically symmetric solutions of infinite energy are surely not physical, this may only suggest that solutions will not be spherically symmetric, as it happens in non-abelian gauge theories. For example one may speculate that finite energy configurations will arrange in flux tubes of gravitation at large distance, between sources of type 1 and 2, as suggested by the different signs of $S$ in $J$ and $C$. In a similar ’confinement-like’ scenario, the term $r^{\gamma}$ may be screened dynamically by the self-arrangement of configurations of matter 1 and 2, so that effectively $S\to 0$ at large distances, as suggested by the full star solutions that we will describe later. To summarize, we found that exact black-hole solutions are modified in the IR or in the UV depending on the choice of the potential, and that this behavior is not captured by the linearized approximation. There are even cases where the behavior of the potential is not modified at all with respect to GR (e.g. $\gamma=2$ or $\gamma=-1$) while the linearized approximation still shows a linear term. It is also interesting to observe that in the limit in which the second metric decouples, $M_{pl2}\to\infty$ (i.e. $\kappa\to\infty$), and assuming that $m_{2}$ and $S$ remains finite, from the solution (28) we find that the term $S\,r^{\gamma}$ remains in $g_{1}$ while $g_{2}$ becomes exactly flat. We will discuss the decoupling limit as well as the limit $c^{2}\to 1$ in section 3.4 and 3.5. ### 3.2 Comparison with the linearized solution It is interesting to comment on the perturbative origin of the exact solutions. This can be addressed by looking at the asymptotical weak-field limit of the solution (for $\gamma<-1$): $J,\ K,\,C,\,A\sim\text{const}+O(1/r)\qquad D\sim O(\sqrt{1/r})\,.$ (33) The crucial observation is that $D$ vanishes more slowly (and non- analytically) than the other components of the perturbations. As a result, this solution is not captured by the standard linearization, where all the perturbations have the same large distance fall-off. Technically, the origin of this behavior can be traced back to the equation $(Q_{1,2})^{t}_{r}=0$, that is algebraic: $D(r)\left[\frac{A(r)C(r)+D(r)^{2}}{J(r)K(r)}+\frac{a_{1}+4a_{2}r^{2}B(r)^{-1}+6a_{3}r^{4}B(r)^{-2}}{b_{1}+4b_{2}r^{-2}B(r)+6b_{3}r^{-4}B(r)^{2}}\right]=0$ (34) This equation can be solved either with $D=0$ (type-II solutions) or with $D\neq 0$ (type-I). In this last case, for the exterior solution, since it turns out that $B=\omega^{2}r^{2}$, asymptotically the equation turns into the definition of the speed of light of $g_{2}$ (as in (31)), while the deviations give the mentioned behavior of $D\sim 1/\sqrt{r}$. From equation (34) we can also understand that standard linearization (around the LB background) can not distinguish between type-I and type-II at leading order: considering the standard perturbative expansion (with parameter $\epsilon$) where $D\sim\epsilon$, this equation starts from order $\epsilon^{2}$. Also, at first order $D$ can be gauged away: it does not appear neither in the linearized Einstein tensors (due to separate gauge invariances) nor in the mass terms (due to ${\cal M}_{1}=0$). At higher orders however one must choose $D=0$, otherwise there is a constraint on the fields $A$, $C$, $J$, $K$, $B$ that are already determined at previous-orders. We reach the conclusion that the standard perturbation theory around the LB background may only approximate the solutions in the type-II branch (if any exist: we recall that nontrivial type-II solutions are not known). On the other hand, it is interesting that the $r^{\gamma}$ term can be recovered in a semi-linearized approach, where one solves exactly equation (34) and treats the remaining ones perturbatively. This will be done for the interior star solution and the result containing the $r^{\gamma}$ terms can be found in appendix B, e.g. equation 54. Alternatively, if one insists in solving perturbatively all the equations, the correct result can also be recovered by assuming $D\sim\sqrt{\epsilon}$ while all the other fluctuations are still of order $\epsilon$, and retaining the first nonvanishing order.777A similar approach was envisaged in [6] to find an asymptotically flat modified Schwarzschild solution for LI massive gravity, valid in the $m^{2}\to 0$ limit. Also in that case a field that is not determined at linearized level for $m=0$, is found to vanish non-analitically as $\sqrt{1/r}$. However that solution is not valid beyond some distance scale, and there is probably no global extension [6]. In the present work, it is remarkable that the semi- linearized solution is also extendable to the exact one. Exactly as in the comparison between the Newtonian and Schwarzschild solutions, the final difference between this semi-linearized and the exact solution is just that $K=1+2\Phi$ instead of $K=J^{-1}=1/(1-2\Phi)$ (and similarly for $g_{2}$). ### 3.3 Interior solution In order to determine the integration constants $m_{1},m_{2}$ and $S$ one can imagine that (28) is the exterior portion of the solution describing a spherically symmetric star. We aim at finding the interior solution and then determine $m_{1}$, $m_{2}$, $S$ by matching with the exterior one. It is instructive to consider first in full generality a spherical star made of fluids of type 1 and 2, extending from the origin to radii $R_{1}$, $R_{2}$, stationary with respect to the respective metrics: $T_{1}{}_{\mu}^{\nu}=\begin{pmatrix}-\rho_{1}\\\ &p_{1}\\\ &&p_{1}\\\ &&&p_{1}\end{pmatrix}\,,\qquad T_{2}{}_{\mu}^{\nu}=\begin{pmatrix}-\rho_{2}&\frac{D}{C}(p_{2}+\rho_{2})\\\ 0&p_{2}\\\ &&p_{2}\ \ \\\ &&&\ \ p_{2}\ \end{pmatrix}\,.$ (35) Like in the vacuum, since $g_{1}$, $T_{1}$ and $E_{1}$ are diagonal, so should be $Q_{1/2}$, i.e. $Q_{1}{}^{t}_{r}=0$. This equation, being the same as in the vacuum case, is exactly solvable for the class of potentials (26). The remaining equations are more involved in the presence of matter, but in linearized approximation the solution can be found analytically. According to the discussion of section 3.2 this partial linearization corresponds to choosing the type-I class ($D\neq 0$) also for the interior solution. For simplicity, we consider a star made of an incompressible fluid of constant density and small pressure, $p\ll\rho$. The interior solution is then matched by requiring continuity of $C$, $J$, $B$, $K$ and of the derivatives $C^{\prime}$, $B^{\prime}$. This procedure, in the physical case when $\gamma<-1$, determines exactly the exterior constants $m_{1}$, $m_{2}$, $S$.888In the unphysical case $\gamma>-1$, although the potentials $J$ and $C$ are regular, the field $B$ develops a singularity $r^{\gamma+1}$ in the origin, calling probably for a fully nonlinear interior solution. While the detailed solution is given in appendix B, we present here the instructive case $R_{1}=R_{2}=R$, $M_{1}$,$M_{2}\neq 0$, and then discuss the phenomenologically interesting case of only matter 1, $M_{2}=0$. #### Case with both kinds of matter: For $R_{1}=R_{2}=R$, and setting $M_{1,2}=4\pi\rho_{1,2}R_{1,2}^{3}/3$, the matching condition gives: $\displaystyle\qquad m_{1}$ $\displaystyle=$ $\displaystyle M_{1}+\Delta M\,,\qquad\qquad\qquad\Delta M=\alpha\,\mu^{2}R^{2}\left(M_{1}-\frac{\omega^{2}}{\kappa}M_{2}\right)$ $\displaystyle m_{2}$ $\displaystyle=$ $\displaystyle M_{2}-\Delta M/{c\,\kappa\,\omega^{2}}$ (36) $\displaystyle S$ $\displaystyle=$ $\displaystyle\quad\Delta M\,R^{-(\gamma+1)}\;\;15/(2\gamma-1)(\gamma-4)\,,$ where $\alpha=8c^{1/2}\omega^{2}/5(\gamma+1)(\gamma-2)$ and $\mu^{2}$ is the same constant that appears in the linearly growing potential (16) of the linearized analysis. In the exterior solution (28) $S$ and $\Delta M$ modify the form of standard Schwarzschild solution. The first modification is in the Newtonian terms, and amounts to a _mass shift_ with respect to the standard values $m_{1,2}=M_{1,2}$. The second is the _new term_ $r^{\gamma}$. Both are proportional to the same combination $\Delta M$.999But see appendix B for the full case $R_{1}\neq R_{2}$. The mass shift $\Delta M$ can be understood as the contribution of the interaction terms $Q$ to the total energy, i.e. to the total mass as measured by the Newton law at large distance. To clarify this, it is useful to recall that in standard GR the Komar energy, written as a spatial volume integral $(8\pi)^{-1}\int R\xi{\rm d}v$ (see appendix C) can be rewritten as a volume integral of the matter energy-momentum tensor by means of the Einstein equations: ${\mathcal{E}}={\mathcal{E}}_{T}=(8\pi)^{-1}\int(2T_{\mu}^{\nu}-T\delta_{\mu}^{\nu})\xi^{\mu}{\rm d}v_{\nu}$. This result is modified in massive gravity because the Einstein equations contain the additional energy-momentum tensor of interactions $Q$, and the additional contribution can be evaluated with its volume integral ${\mathcal{E}}_{Q}=(8\pi)^{-1}\int(2Q_{\mu}^{\nu}-Q\delta_{\mu}^{\nu})\xi^{\mu}{\rm d}v_{\nu}$. We remark that while the sum of these two integrals, being the total energy, can be expressed as a surface integral at infinity (see appendix C), they separately can not, and they can only be evaluated using the smooth interior and exterior solution. The result of the volume integral is (again finite only for $\gamma<-1$): $\displaystyle{\mathcal{E}}_{T_{1,2}}=$ $\displaystyle=$ $\displaystyle M_{1,2}\,,\qquad{\mathcal{E}}_{Q_{1}}=-\kappa\,{\mathcal{E}}_{Q_{2}}=\Delta M\,.$ (37) This confirms that the mass shift $\Delta M$ is a screening effect, due to the energy of the interacting fields in $Q$, and corresponding to the nonzero Ricci curvature even outside the source (see e.g. [21]). As a side remark, looking at the matching (36), we observe that we did _not_ linearize in $V$, but neglecting terms higher order in the matter density one has effectively neglected higher orders in $V\sim\mu^{2}$. Indeed, the result depends on the two dimensionless parameters $R^{2}\mu^{2}$ and $GM/R$, but at first order in $GM/R$ only the first order in $R^{2}\mu^{2}$ appears, i.e. $GMR\mu^{2}$, and the final result is smooth when the interaction vanishes, $V\sim\mu^{2}\to 0$. This is opposed to the singular massless limit of Lorentz-Invariant (Fierz-Pauli) massive gravity. #### Case of normal matter: Turning off $M_{2}=0$, we can focus on sector 1 and discuss more phenomenologically how normal gravity is modified by the presence of the additional spin two field. From the matching condition we have, with $M_{1}=M$: $\begin{split}m_{1}=&\,M(1+\alpha\,\mu^{2}R^{2})\,,\qquad\qquad m_{2}=-\alpha\,\mu^{2}R^{2}M/{c\,\kappa\,\omega^{2}}\\\\[4.30554pt] S=&\,\mu^{2}MR^{1-\gamma}\,\;\;15\alpha/(2\gamma-1)(\gamma-4)\end{split}$ (38) and from (28) we find for the modified potential (ignoring the numerical factors): $\Phi\sim G\,M\,\left[\frac{1}{r}(1+\mu^{2}R^{2})+\mu^{2}R\left(\frac{r}{R}\right)^{\gamma}\right]\,.$ (39) The mass shift is now equivalent to a rescaling of the Newton constant $G(1+\alpha\mu^{2}R^{2})$, that depends on the source radius!! We observe that for the sun101010$R_{\odot}\simeq 5\cdot 10^{5}\,$Km$\,=5\cdot 10^{15}\,$eV-1 and $M_{\odot}=10^{66}\,$eV. we have $\mu^{2}R^{2}\sim 10^{-10}$, assuming all coupling constants to be of the same order so that $\mu\sim m_{g}\lesssim(10^{-20}\,$eV$)\sim(100$AU$)^{-1}$ (this limit corresponding to the rough experimental bound on the graviton mass from pulsar GW emission [8, 22, 10]). We thus see that for the sun the size dependence is negligible and unobservable, and even more so for the planets. The effect becomes important for objects of size $\mu^{-1}$. For instance, for large objects with $R\gtrsim 10^{5}R_{sun}$ (red giants, large gas clouds, galaxies…) the effect may be of order one, and induces a macroscopic modification of the Newton constant. For low density objects that we consider here, this modification does not depend on the mass but just on the object size; therefore, given the mass, a large sphere of gas has a larger effective newton constant. In the limit $\mu^{2}R^{2}>1$, the surface potential would even scale as $R^{4}$, instead of $R^{2}$ as in standard gravity. Moreover, remembering that $\mu^{2}$ may be negative, the negative interaction energy could cause large fluids to antigravitate, hinting toward the acceleration of the cosmological solutions. Then, the new term in the potential is of the form $\delta\Phi\sim G\,M\,\mu^{2}R\left(\frac{r}{R}\right)^{\gamma}\,,$ (40) replacing the linear term $G\,M\,\mu^{2}r$ of the linearized analysis. The Newtonian and the new term will be competing at a critical distance $r_{c}$ that also depends on $\gamma$: $r_{c}=R\left|\frac{\mu^{2}R^{2}}{1+\mu^{2}R^{2}}\right|^{-\frac{1}{\gamma+1}}\,.$ (41) Of course since $\gamma<-1$ the relevant modification is ultraviolet, and is evident for $r<r_{c}$ (while for $\gamma>-1$ it would be infrared, for $r>r_{c}$). To estimate $r_{c}$, we observe that since the exponent $-1/(\gamma+1)$ is positive, one always has $r_{c}<R$ for $\mu^{2}>0$, and the critical distance is inside the star. This does not mean that there will be no observable effects, since even subleading modifications to the newton potential may be measured (for example modifications of the gravitational potential of relative magnitude $10^{-3/-5}$ are at the level the current solar-system tests). On the other hand for negative $\mu^{2}$, and in particular for $\mu^{2}R^{2}<-1/2$, one has $r_{c}>R$ so that in a UV region near the source the gravitational potential has stronger fall-off. For $\mu^{2}R^{2}\simeq-1$ we even find that $r_{c}$ becomes infinite, so that the region of UV modification expands to larger and larger distances! We can summarize the results in the physical phase $\gamma<-1$: * • For sources of dimension $R<\mu^{-1}$, the effects are: a mass shift equivalent to a small Newton-constant renormalization $(1+\mu^{2}R^{2})$, and a subleading correction to the Newtonian potential, $\delta\Phi\lesssim(r/R)^{\gamma}$. * • For large sources, of dimension $R>\mu^{-1}$, the mass shift is more pronounced, and for negative $\mu^{2}$ even the new $r^{\gamma}$ term can become dominant in a region near the source. As we see, even discarding the nonphysical and possibly confining branch $\gamma>-1$, the phenomenology of these modified static solutions appears to be quite rich, and deserves a separate analysis to confront their features with real physical systems, e.g. modified galactic gravitational field, gravitation of large sources, post-Newtonian analysis. ### 3.4 Decoupling the second metric The idea of introducing a second metric and considering its decoupling limit, to have a second background at hand while disposing of its fluctuations, is not new and was indeed considered to tackle the problem of the nonlinear continuation of the FP massive gravity [4]. However, due to the singular Isham-Storey potential, or due to the ill-defined nature of the Lorentz- Invariant theory, this did not lead to significant advance. In this work we found some nonperturbative solutions of the full system, so we are in a position to control the decoupling limit $M_{pl2}\to\infty$ ($\kappa\to\infty$) in which the second gravity is effectively switched off. First, as far as the propagating states are concerned, we recall from the linearized analysis [10] that in the flat background out of two gravitons only the first graviton survives the decoupling limit: it is massive (with two polarization states and mass $G\lambda_{2}$) and has a normal dispersion relation. For the nontrivial solutions, as anticipated, one may take this limit in the exterior solutions, once one checks that $m_{1}$, $m_{2}$ and $S$ stay finite. This is indeed shown by the interior solution (36), therefore we directly find the result: $\displaystyle\Phi_{1}$ $\displaystyle=$ $\displaystyle\frac{GM_{1}}{r}(1+R^{2}\mu^{2}\alpha)+GM_{1}\mu^{2}R\left(\frac{r}{R}\right)^{\gamma}\left[15\alpha/(2\gamma-1)(\gamma-4)\right]\,,$ (42) $\displaystyle\Phi_{2}$ $\displaystyle=$ $\displaystyle 0\,.$ (43) Both the mass shift and the new term remain, but the second gravity disappeared: here the other metric is flat! A look at the exact solution (28) in the decoupling limit shows that the limiting metric 2 is still nondiagonal ($D\neq 0$). This means that $g_{2}$ is only _gauge-equivalent_ to $\eta_{2}=\omega^{2}\text{diag}\\{-c^{2},1,1,1\\}$, and that to make contact with this traditional minkowski diagonal vacuum one has to choose the gauge (23), where $\bar{g}_{1}$ is not diagonal. Explicitly we find: $ds_{1}^{2}=-\bar{J}\,dt^{2}+2\bar{D}\,dt\,dr+\bar{K}\,dr^{2}+r^{2}d\Omega^{2}\,,\qquad ds_{2}^{2}=\omega^{2}(-c^{2}dt^{2}+dr^{2}+r^{2}d\Omega^{2})\,,$ (44) with $\bar{J}=J$, $\bar{K}=J^{-1}(1-\bar{D}^{2})$, $\bar{D}=-(c\omega)^{-1}J\sqrt{\omega^{2}-A}$. Notice that $\bar{D}$ is defined by the deviation of $A$ from $\omega^{2}$, and that still $\bar{J}\bar{K}+\bar{D}^{2}=1$. We therefore note that, to recover the present solutions in effective massive gravity theories, where only $g_{1}$ is dynamical and the Lorentz breaking is an external diagonal metric, one should look for nondiagonal configurations. We also remark that while taking the decoupling limit has left us with a flat auxiliary metric, still there is curvature for metric 1 in the vacuum outside the sources, due to $Q_{1}$ and $Q_{2}$ being nonzero there, because of the $r^{\gamma}$ term. Therefore the order of the limit matters, and one would not get the correct result if one were to assume a flat second metric _before_ taking the decoupling limit. In other words, setting $g_{2}$ to be flat in advance: $g_{2}=\bar{g}_{2}$, we have in vacuum $\displaystyle E_{2\nu}^{\mu}=0=V\delta_{\nu}^{\mu}+4(V^{\prime}X)_{\nu}^{\mu}\quad\text{so that}$ (45) $\displaystyle M_{pl1}^{2}E_{1\nu}^{\mu}=2\;V\delta_{\nu}^{\mu}\qquad\longrightarrow V=\text{const by Bianchi}\longrightarrow\text{(Anti)deSitter}\,.$ Instead, in the limit $M_{pl2}\to\infty$ we have still $g_{2\mu\nu}\to\bar{g}_{2\mu\nu}$, but different solutions for $g_{1}$: $\displaystyle E_{2\nu}^{\mu}\to 0\qquad\text{but}\quad V\delta_{\nu}^{\mu}+4(V^{\prime}X)^{\mu}_{\nu}\neq 0\,,\qquad\text{and then}$ (46) $\displaystyle M_{pl1}^{2}E_{2\nu}^{\mu}=(V\delta_{\nu}^{\mu}-4V_{\nu}^{{}^{\prime}\mu})\neq\text{const}\qquad\longrightarrow\text{non- trivial solutions.}$ Summarizing, the decoupling limit shows that the theory remains well behaved, consisting of a modified gravity with massive gravitons, while the auxiliary metric is flat and decoupled. ### 3.5 Lorentz-Invariant limit The Lorentz Invariant limit $c^{2}\to 1$ is also interesting to address the Vainshtein’s claim that nonlinear corrections actually cure the discontinuity problem in Pauli-Fierz theory [6]. Indeed, we find that the limit $c^{2}\to 1$ is well behaved, and the solutions retain their validity. In this limiting phase therefore, gravity is modified, but lorentz breaking disappears. The linearized mass term is accordingly of the form $a\,h_{\mu\nu}^{2}+b\,h^{2}$, however the limiting theory reached in this way is _not_ the Fierz-Pauli one, where $a+b=0$ (and for this reason FP is free from coupled ghosts). Here, we get to a theory where $a=0$; in fact, since we approach the LI phase from the $\lambda_{1}=0$ branch, and because in the LI limit one has $\lambda_{1}=\lambda_{2}=a$, we see that also the graviton mass vanishes in this limit, as can be checked with the expression (31). It is nevertheless worth to point out that also $a=0$ is a ghost-free theory like $a+b=0$. This case is not usually considered because at the linearized level there is no massive graviton as a consequence of an additional gauge symmetry (three transverse diffeomorphisms), see [23] and also the PF0 phase in [10]. Accordingly, no strong coupling problems are expected and no Vainshtein issues. This matches nicely with our model having good properties along all the LB branch, that survive also in the LI limit. ### 3.6 Local Lorentz-breaking in nontrivial background While the asymptotic biflat metrics are Lorentz breaking, one may ask about the situation at finite distance. This will have definite interest when addressing the nonpropagation of ghosts in the described nontrivial background. In fact, we recall (section 2.2) that on flat background this is a consequence of ${\cal M}_{1}=0$ , and this follows from gauge invariance together with the fact that locally boosts are spontaneously broken. Now, even in nontrivial background a spontaneous breaking of Lorentz will lead to flat directions of the potential. Whether this fact will be enough to lead to absence of ghosts and to stable configurations is under scrutiny, and goes beyond the scope of the present work. To describe the local breaking of Lorentz at any given point in the nontrivial background, one chooses a local Lorentz frame ($g_{1}=\eta$) and simultaneously diagonalizes $g_{2}$. The Lorentz breaking is given by the entries of $g_{2}$, that are actually the eigenvalues of $X$. These are easily calculated (in polar coordinates $t$, $r$, $\theta$, $\phi$): $\displaystyle\hat{X}=\omega^{2}\big{\\{}c^{2}\xi^{-1},\xi,1,1\big{\\}}\,,$ (47) $\displaystyle\text{with}\ \xi=\frac{1}{2}\left[c^{2}f_{-}+f_{+}+\sqrt{(c^{2}-1)(c^{2}f_{-}^{2}-f_{+}^{2})}\right],\qquad f_{\pm}=1\pm\tilde{S}r^{\gamma-2}\,.$ Quite remarkably, that they do not depend on the masses $m_{1,2}$ of the newtonian terms, and this is due to the nondiagonal structure given by $D$. One can easily check that for $r\to\infty$ we have $\hat{X}=\omega^{2}\\{c^{2},1,1,1\\}$, reproducing the asymptotical lorentz breaking ($\gamma<-1$). Then we see that in the case $S=0$ the eigenvalues are constant, so that at any distance the Lorentz breaking is the same: $\hat{X}=\omega^{2}\\{c^{2},1,1,1\\}$. We have two pure Schwarzschild solutions in a configuration that at any point breaks local boosts but preserves rotations (these are the solutions found in [23]). On the other hand, since in general $S\neq 0$, a star solution will break not only boosts but also local rotations, because the $rr$ term is different from the $\theta\theta$ and $\phi\phi$ ones. For example at large but finite distance, where $Sr^{\gamma-2}$ is small, we have: $\hat{X}\simeq\omega^{2}\\{c^{2}(1-\tilde{S}r^{\gamma-2}),1+\tilde{S}r^{\gamma-2},1,1\\}\,.$ (48) To compare the situation with standard GR, we recall that in GR Lorentz- invariance at any given point is always valid, in the Lorentz frame, and is broken in a finite neighbourhood only by the curvature (tidal) effects. Here on the contrary in the gravitational sector a Lorentz breaking is felt also locally, and for $\tilde{S}\neq 0$ also rotations are broken. The physical effect is that gravitons will propagate differently in direction of the source. We strongly believe that this breaking of (local) boosts _and_ rotations at finite distance from a source is a general feature of nontrivial solutions in massive gravity, due to the presence of additional fields that can not be ‘gauged away’. ## 4 Conclusions In this paper we approached the problem of finding a consistent massive deformation of gravity by introducing an additional spin 2 field $g_{2}$ coupling non-derivatively to the standard metric field. This allows us to explore both the Lorentz invariant (LI) and Lorentz breaking (LB) phases working with consistent and dynamically determined backgrounds. Preserving diffeomorphisms and breaking Lorentz is also important; at the linearized level it forces ${\cal M}_{1}$ to vanish and no dangerous scalar mode is propagating. Still at the linearized level, it was shown [8] that in the case ${\cal M}_{1}=0$ the vDVZ discontinuity is absent, but a new linearly growing term is present in the static gravitational potential [10, 24], that seems to invalidate perturbation theory beyond some distance scale. To address this and the vDVZ discontinuity problem, we thus studied the exact spherically symmetric configurations. The exact solution that we found, valid for a large class of interaction potentials, shows that the linear term is replaced in the full solution by a power-like term $r^{\gamma}$, with $\gamma$ depending on the nonlinear couplings in the interaction potential. Phenomenologically, when $\gamma<-1$ the total energy of the solution is finite and the space is asymptotically flat; Lorentz is broken in the gravitational sector by the asymptotic value of $g_{2}$, but normal matter only feels the modification of the gravitational potential. Using the full solution one can check that the absence of the vDVZ discontinuity is an exact result. The effect of the interaction manifests in the $r^{\gamma}$ term whose size $S$ was determined for a star by matching the exterior solution with an interior one. In addition to this, by the presence of the additional spin 2 field, the total mass of the star appearing in the Newton term gets a finite renormalization that depends on the object size, and may screen or even antiscreen the star mass. We believe that this is a general feature of massive gravity. When $g_{1}$ describes a black hole the solution depends not only on the collapsed mass but also on an other constant, probably remnant of the original shape; notice that there is no contradiction with the no-hair theorem because the Einstein equations are modified by the presence of $Q_{1/2}$. In the case $\gamma>-1$ the total energy is infinite and this may indicate only that solutions will not be spherically symmetric. For example the solution may be unstable under axially symmetric perturbations and drop to a flux tube in a sort of mass confinement scenario. Indeed, regarding stability, even for the physical case $\gamma<-1$ the final word would be given by studying the small fluctuations also around the exact solution, to check that the non-propagation of the (ghost) scalars and vectors is preserved on a nontrivial background. To this aim, we have discussed how the spontaneous Lorentz-breaking is present also in the nontrivial background, where we note that in general also rotations are locally broken. One expects this also to be a generic feature of massive gravity. We showed that we can reach the LI phase by tuning $c^{2}\to 1$ in the exact solutions, and this results into a well behaved phase, though not the Fierz- Pauli one (gravitons are massless). The fate of the discontinuity and Vainshtein claim for the PF case is thus still an open problem and exact solutions of type II with $c=1$ are presently under investigation. Finally, it would be interesting to speculate on the role of the mass screening in cosmology, that may change the form of the Hubble expansion. ###### Acknowledgments. Work supported in part by the MIUR grant for the Projects of National Interest PRIN 2006 “Astroparticle Physics”, and in part by the European FP6 Network “UniverseNet” MRTN-CT-2006-035863. ## Appendix A Background for solvable potentials For the solvable potential (26), we report here the biflat solution as it results from solving the EOM (12), as well as the $\mu^{2}$ and $\lambda_{2}$ constants of the linearized analysis. The fine tuning conditions to ensure flatness ${\mathcal{K}}_{1}={\mathcal{K}}_{2}=0$ turn in two relations for the two cosmological constants. Defining $\tilde{a}_{n}=\omega^{-2n}a_{n}$, $\tilde{b}_{n}=\omega^{2n}b_{n}$ and $\alpha_{n}=(\tilde{a}_{n}-c^{2}\tilde{b}_{n})(c^{2}-1)/c^{2}$, and $\beta_{n}=(\tilde{a}_{n}+c^{2}\tilde{b}_{n})$, we have: $\displaystyle 3{\mathcal{K}}_{1}$ $\displaystyle=$ $\displaystyle{}-2\omega^{-2}\Lambda_{1}+\frac{c^{-3/2}}{(c^{2}-1)\gamma}\bigg{[}-8c^{2}(3c^{2}+\gamma+1)\alpha_{2}-12c^{2}((\gamma+6)c^{2}+3\gamma+2)\alpha_{3}\bigg{]}$ $\displaystyle{}\qquad\qquad+c^{-3/2}\bigg{[}(6c^{2}-2)\beta_{2}+24(c^{2}-1)\beta_{3}-\tilde{a}_{0}c^{2}+3\tilde{b}_{4}c^{4}-5\tilde{a}_{4}\bigg{]}$ $\displaystyle 3{\mathcal{K}}_{2}$ $\displaystyle=$ $\displaystyle{}-2\omega^{2}\Lambda_{2}\kappa^{-1}+\frac{c^{-5/2}}{(c^{2}-1)\gamma\kappa}\bigg{[}8((\gamma+1)c^{2}+3)\alpha_{2}c^{2}+12((3\gamma+2)c^{2}+\gamma+6)\alpha_{3}c^{2}\bigg{]}$ (49) $\displaystyle{}\qquad\qquad-c^{-5/2}\bigg{[}2(c^{2}-3)\beta_{2}+24(c^{2}-1)\beta_{3}+\tilde{a}_{0}c^{2}+5\tilde{b}_{4}c^{4}-3\tilde{a}_{4})\bigg{]}$ and we remind that one of these is a genuine fine tuning to achieve flatness, as is usual in General Relativity, while the other is a complicated equation that may be used to find $\omega$. We prefer thinking in reverse and consider $\omega$ a free parameter determining the right $\Lambda_{2}$. Finally, the lorentz breaking speed of light turns out to be $c^{2}=-\frac{\tilde{a}_{1}+4\tilde{a}_{2}+6\tilde{a}_{3}}{\tilde{b}_{1}+4\tilde{b}_{2}+6\tilde{b}_{3}}\,.$ (50) The relevant quantities entering in the linearized analysis are the graviton mass $\lambda_{2}$ and the $\mu^{2}$ parameter, that have the following expressions: $\displaystyle M_{pl1}^{2}\mu^{2}$ $\displaystyle=$ $\displaystyle\frac{(\gamma-2)^{2}}{32}\bigg{\\{}\tilde{a}_{0}-\frac{3}{c^{2}(c^{2}-1)(\gamma-2)\gamma}\left[5(c^{2}-1)(\gamma-2)\gamma(\tilde{a}_{4}+c^{4}\tilde{b}_{4})\right.$ (51) $\displaystyle\qquad\qquad\quad\left.\left.{}+6c^{2}(1+c^{2})(\gamma-2)\gamma\beta_{2}+32c^{2}(1+c^{2})(\gamma-2)\gamma\beta_{3}\right.\right.$ $\displaystyle\qquad\qquad\quad\left.{}-4c^{2}(c^{2}-1)\left[(\gamma+2)^{2}\alpha_{2}+(12+4\gamma+7\gamma^{2})\alpha_{3}\right]\right]\bigg{\\}}$ $\displaystyle M_{pl1}^{2}m_{g}^{2}=\lambda_{2}$ $\displaystyle=$ $\displaystyle G\frac{2(\gamma-2)}{\gamma}(\alpha_{2}+3\alpha_{3})\,.$ (52) ## Appendix B Interior solution For generality we report here the case of a star composed of two spherical regions filled with incompressible fluids of kind 1 and 2 extending from the origin to different radii $R_{1}$ and $R_{2}$; of constant densities $\rho_{1,2}=\frac{M_{1,2}}{4/3\,\pi R_{1,2}^{3}}$ and negligible pressures. In general we have two scenarios, for the three different regions: $a$) for $\;R_{2}<R_{1}$ we have $0<r<R_{2}$, $R_{2}<r<R_{1}$ or $r>R_{1}$; $b$) for $\;R_{1}<R_{2}$ we have $0<r<R_{1}$, $R_{1}<r<R_{2}$ or $r>R_{2}$. We give only the analitic results for $J[r]$, that is the gravitational potential in $g_{1}$: * • Starting from the exterior solutions $r>R_{1,2}$, we find a common value for the exterior potential $J$ in both scenarios $a$) and $b$): $\displaystyle J(r)=$ $\displaystyle 1$ $\displaystyle{}-\frac{2G}{r}\left[M_{1}+\frac{16\mu^{2}\left(M_{1}R_{1}^{2}-\omega^{2}\kappa^{-1}M_{2}R_{2}^{2}\right)\sqrt{c}\,\omega^{2}}{5(\gamma-2)(\gamma+1)}\right]+$ (53) $\displaystyle{}+r^{\gamma}\left[\frac{96G\mu^{2}\left(\omega^{2}M_{2}R_{2}^{1-\gamma}-\kappa M_{1}R_{1}^{1-\gamma}\right)\sqrt{c}\,\omega^{2}}{(\gamma-4)(\gamma-2)(\gamma+1)(2\gamma-1)\kappa}\right]\,.$ * • The intermediate solutions are different in the two scenarios: for $a$) i.e. $R_{2}<r<R_{1}$ $\displaystyle 1$ $\displaystyle-$ $\displaystyle\frac{3GM_{1}}{R_{1}}+\frac{32G\sqrt{c}\mu^{2}\omega^{4}M_{2}R_{2}^{2}}{5r(2-\gamma)(\gamma+1)\kappa}+\frac{Gr^{2}M_{1}\left(1-\frac{16\sqrt{c}\mu^{2}\omega^{2}R_{1}^{2}}{(2-\gamma)(\gamma+1)}\right)}{R_{1}^{3}}-\frac{48Gr^{4}\sqrt{c}\mu^{2}\omega^{2}M_{1}}{5(\gamma-4)(\gamma+3)R_{1}^{3}}+$ (54) $\displaystyle\frac{96G\sqrt{c}\mu^{2}\omega^{2}M_{1}R_{1}^{\gamma}r^{1-\gamma}}{(\gamma-2)(\gamma+1)(\gamma+3)(2\gamma-1)}+\frac{96G\sqrt{c}\mu^{2}\omega^{4}M_{2}R_{2}^{1-\gamma}r^{\gamma}}{(\gamma-4)(\gamma-2)(\gamma+1)(2\gamma-1)\kappa}$ for $b$) i.e. $R_{1}<r<R_{2}$ $\displaystyle 1$ $\displaystyle-$ $\displaystyle\frac{2GM_{1}}{r}-\frac{32G\sqrt{c}\mu^{2}\omega^{2}M_{1}R_{1}^{2}}{5r(2-\gamma)(\gamma+1)}-\frac{16Gr^{2}\sqrt{c}\mu^{2}\omega^{4}M_{2}}{(\gamma-2)(\gamma+1)\kappa R_{2}}+\frac{48Gr^{4}\sqrt{c}\mu^{2}\omega^{4}M_{2}}{5(\gamma-4)(\gamma+3)\kappa R_{2}^{3}}-$ (55) $\displaystyle\frac{96G\sqrt{c}\mu^{2}\omega^{4}M_{2}R_{2}^{\gamma}r^{1-\gamma}}{(\gamma-2)(\gamma+1)(\gamma+3)(2\gamma-1)\kappa}-\frac{96G\sqrt{c}\mu^{2}\omega^{2}M_{1}R_{1}^{1-\gamma}r^{\gamma}}{(\gamma-4)(\gamma-2)(\gamma+1)(2\gamma-1)}.$ * • The inner solutions $r<R_{1,2}$ have again a common form in both scenarios $a$) and $b$): $\displaystyle 1-\frac{3GM_{1}}{R_{1}}+Gr^{2}\left[M_{1}\left(\frac{16\sqrt{c}\mu^{2}\omega^{2}}{(\gamma-2)(\gamma+1)R_{1}}+\frac{1}{R_{1}^{3}}\right)-\frac{16\sqrt{c}\mu^{2}\omega^{4}M_{2}}{(\gamma-2)(\gamma+1)\kappa R_{2}}\right]+$ $\displaystyle\frac{48G\sqrt{c}\mu^{2}\omega^{2}\left(\omega^{2}M_{2}R_{1}^{3}-\kappa M_{1}R_{2}^{3}\right)r^{4}}{5(\gamma-4)(\gamma+3)\kappa R_{1}^{3}R_{2}^{3}}+\frac{96G\sqrt{c}\mu^{2}\omega^{2}\left(\kappa M_{1}R_{1}^{\gamma}-\omega^{2}M_{2}R_{2}^{\gamma}\right)r^{1-\gamma}}{(\gamma-2)(\gamma+1)(\gamma+3)(2\gamma-1)\kappa}.\ \ \ $ (56) As one checks, the solution is regular at the origin. ## Appendix C Energy integrals In the presence of a time-like Killing vector in GR on can define the notion of total gravitational energy as a flux from an asymptotic 2-surface that involve only the gravitational field at large distance, far from the sources. Consider the following metric $ds^{2}=-C(r)\,dt^{2}+2D(r)\,drdt+A(r)\,dr^{2}+B(r)\,d\Omega^{2}\,,$ (57) with the time-like Killing vector $K=\frac{\partial}{\partial t}$. From the Killing equation we have $\nabla^{\mu}J_{\mu}=0\,,\qquad J_{\mu}=\Box K_{\mu}\,.$ (58) The Komar energy ${\mathcal{E}}$ is defined by ${\mathcal{E}}=w\int_{t=t_{1}}\sqrt{h}\,n^{\mu}J_{\mu}\,,$ (59) where $h_{\mu\nu}$ is the induced metric in the hyper-surface $t=const.$ with unit normal $n^{\mu}$ and $w$ is a normalization constant. According to Stokes theorem, given a 3-surface $V$, for any antisymmetric tensor $F_{\mu\nu}$ we have $\int_{V}d^{3}x\,\sqrt{h}\,n^{\mu}\nabla^{\nu}F_{\mu\nu}=\int_{\partial V}d^{2}x\,\sqrt{\gamma}\,\left(n^{\alpha}v^{\beta}-n^{\beta}v^{\alpha}\right)\,F_{\alpha\beta}\,,$ (60) where $v^{\alpha}$ is the unit normal to $\partial V$ and $\gamma_{\alpha\beta}$ is the induced metric in $\partial V$. Then Stokes theorem gives ${\mathcal{E}}=-\frac{w}{2}\,\int_{t=t_{1},r=r_{1}.}\sqrt{\gamma}\,\left(n^{\alpha}v^{\beta}-v^{\alpha}n^{\beta}\right)\nabla_{\alpha}K_{\beta}\,.$ (61) In general the Komar energy will depend on 2-surface that bounds the $t=const.$ slice. Indeed, from Einstein equations it easy to show that the difference $\Delta{\mathcal{E}}$ between the Komar energy computed with two different bounding 2-surfaces $\Sigma_{1}$ and $\Sigma_{2}$ is proportional to the integral of the Ricci tensor over the 3-volume bounded by $\Sigma_{1}$ and $\Sigma_{2}$. As a result the Komar energy does not depend on $\Sigma$ in a region where the Ricci tensor is vanishing. This is indeed the case in a region far from any source. Then ${\mathcal{E}}=-\frac{w}{2}\,\int_{t=const.,r\to\infty.}\sqrt{\gamma}\,\left(n^{\alpha}v^{\beta}-v^{\alpha}n^{\beta}\right)\nabla_{\alpha}K_{\beta}\,;$ (62) for the induced metric on the 3-surface $t=\text{const.}$ and its normal $n$ we get $\begin{split}&dl^{2}=A(r)\,dr^{2}+B(r)\,d\Omega^{2}\;;\\\ &n=(AC+D^{2})^{-1/2}\left(-A^{1/2}\frac{\partial}{\partial t}+DA^{-1/2}\frac{\partial}{\partial r}\right)\end{split}$ (63) and for the induced metric on $t,r=const.$ and its normal $v$ ( v is normalized with h ) $ds^{2}=B(r)\,d\Omega^{2}\,,\qquad\qquad v=A^{1/2}\frac{\partial}{\partial r}\;.$ (64) We have then ${\mathcal{E}}=-\lim_{r\to\infty}\,{\frac{4\pi w\,C^{\prime}\,B}{\sqrt{D^{2}+AC}}}\;.$ (65) One can recover the same result using the language of differential forms. Introducing the 1-form $J=-(\delta d+d\delta)\tilde{K}$ in terms of the 1-form $\tilde{K}$ associated with the Killing vector $K$. From the Killing equation $\delta\tilde{K}=0$, then $J=-(\delta d+d\delta)\tilde{K}\equiv\delta d\tilde{K}=\ast d\ast d\tilde{K}=\Box K_{\mu}dx^{\mu}\;.$ (66) Now $\int d\ast J=0\,\Rightarrow\int_{\text{t=\hbox to0.0pt{const.\hss}}}\ast\tilde{J}\ \ \ \text{ is time-independent}$ (67) and finally $\displaystyle{\mathcal{E}}$ $\displaystyle=$ $\displaystyle-w\int_{t=t_{1}}\\!\\!\\!\\!\ast J=-w\int_{t=t_{1}}\\!\\!\\!\\!\ast\delta d\tilde{K}=w\int_{t=t_{1}}\\!\\!\\!\\!\ast\ast d\ast d\tilde{K}=w\int_{t=t_{1}}\\!\\!\\!\\!d\ast d\tilde{K}$ (68) $\displaystyle=$ $\displaystyle w\int_{t=t_{1},r=r_{1}}\\!\\!\\!\\!\ast d\tilde{K}=\int_{t=t_{1},r=r_{1}}\\!\\!\\!\\!\sqrt{g}\,\frac{C^{\prime}}{2}\,g^{\mu t}\,g^{\nu r}\,\epsilon_{\mu\nu\alpha\beta}\,\,dx^{\alpha}\wedge dx^{\beta}$ $\displaystyle=$ $\displaystyle-\frac{4\pi w\,BC^{\prime}}{\sqrt{D^{2}+AC}}\,.$ ## Appendix D Simplest bigravity Here we analyze the system in the simpler particular case when $V$ is a function of $q$ only: $V=f(q)$. The Bianchi identities (8) for $Q_{1,2}$, can be written as111111And recall that for any vector field $v_{\mu}$ one has $(\nabla_{2\mu}-\nabla_{1\mu})v_{\nu}=C_{\mu\nu}^{\sigma}v_{\sigma}$, with the tensor $C_{\mu\nu}^{\sigma}=g_{2}^{\sigma\beta}\left(\nabla_{1\mu}g_{2\nu\beta}+\nabla_{1\nu}g_{2\mu\beta}-\nabla_{1\beta}g_{2\mu\nu}\right)/2$. $\displaystyle\partial_{\mu}V-\left[\partial_{\nu}\log q+2\,\left(\nabla_{1\nu}-\nabla_{2\nu}\right)\right]\left(V^{\prime}X\right)^{\nu}_{\mu}=0$ (69) $\displaystyle 8\left(\nabla_{1\nu}+\nabla_{2\nu}\right)\left(V^{\prime}X\right)^{\nu}_{\mu}-V\,\partial_{\mu}\log q=0$ (70) and because in the case at hand $\left(V^{\prime}X\right)^{\nu}_{\mu}=f^{\prime}q\delta^{\nu}_{\mu}$, the only non-trivial equation is $\left[16\,\frac{d^{2}}{d(\log q)^{2}}f-f\right]\partial_{\mu}q=0\;.$ (71) Thus, either $q=$const. or $V=V_{0}=c_{1}q^{1/4}+c_{2}q^{-1/4}$. However, $(g_{1}g_{2})^{1/4}V_{0}=c_{1}\sqrt{g1}+c_{2}\sqrt{g2}$ would imply that the two sectors do not see each other and we are left with two independent copies of GR + cosmological term. The theory with $q=$const is the simplest of all possible bigravity theories, the EOM reduce to $\displaystyle M_{pl1}^{2}{E_{1}}^{\mu}_{\nu}+{\mathcal{K}}_{1}\,\delta^{\mu}_{\nu}=\frac{1}{2}{T_{1}}^{\mu}_{\nu}$ (72) $\displaystyle M_{pl2}^{2}{E_{2}}^{\mu}_{\nu}+{\mathcal{K}}_{2}\,\delta^{\mu}_{\nu}=\frac{1}{2}{T_{2}}^{\mu}_{\nu}\,,$ (73) with $\displaystyle{\mathcal{K}}_{1}=q^{1/4}\left[f(q)-4q\,f^{\prime}(q)\right]$ (74) $\displaystyle{\mathcal{K}}_{2}=q^{-1/4}\left[f(q)+4q\,f^{\prime}(q)\right].$ (75) The effective cosmological constants are thus related; moreover, this simplest bigravity, due to the constraint $q=$const, is equivalent to a single GR + unimodular GR, and the two sectors share the conformal mode. Finally, besides the diagonal diff also two independent volume-preserving diffs are present. As an example of exact solution, we present the solution for the potential $V=\text{tr}\ln X=\ln\det X$ (76) (and we may add also the two cosmological constant terms $\Lambda_{1}q^{-1/4}$ and $\Lambda_{2}q^{1/4}$ to achieve flatness). With this potential we have $V^{\prime}X={\mathbf{1}}$ by construction. The solution in general is Schwarzschild-deSitter for both metrics, but $g_{2}$ is in a different gauge: $\displaystyle J$ $\displaystyle=$ $\displaystyle\Delta_{1}\left(1-2\frac{m_{1}}{r}+c_{1}r^{2}\right)\,,\qquad\qquad K=\Delta_{1}/J\,,$ (77) $\displaystyle C$ $\displaystyle=$ $\displaystyle\Delta_{2}\left(1-2\frac{m_{2}}{\rho}+c_{2}\rho^{2}\right)\,,\qquad\qquad\rho=(r^{3}+\lambda^{3})^{1/3}$ (78) $\displaystyle B$ $\displaystyle=$ $\displaystyle\omega^{2}\rho^{2}\,,\qquad D^{2}+AC=\Delta_{2}\frac{(B^{\prime})^{2}}{B}=c^{2}\omega^{4}\Delta_{1}\,(\rho^{\prime})^{2}\,,$ (79) $\displaystyle A$ $\displaystyle=$ $\displaystyle\text{free}\,,\qquad\qquad c^{2}=\frac{4\Delta_{2}}{\omega^{2}\Delta_{1}}\,.$ (80) This is a family of solutions because $A(r)$ is a free function (!), remnant of the spatial diffs. The determinant $AC+D^{2}$ is fixed by $B(r)$, and for $\lambda\neq 0$ it is not constant, at finite distance. Then one can also use $A$ to set $D=0$ and get a bidiagonal solution like (23). Notice that $\omega^{2}$ and $\Delta_{2}/\Delta_{1}$ are free constants, and so also the relative speed of light $c^{2}$ is free. ## References * [1] M. Fierz and W. Pauli, _Proc. Roy. Soc. Lond._ A 173, 211 (1939). * [2] H. van Dam and M. J. G. Veltman, Nucl. Phys. B 22 (1970) 397; Y. Iwasaki, Phys. Rev. D 2 (1970) 2255; V.I.Zakharov, Sov. Phys. JETP Lett. 12 (1971) 198. * [3] A. I. Vainshtein, Phys. Lett. B 39 (393) 1972. * [4] T. Damour, I. I. Kogan and A. Papazoglou, Phys. Rev. D 67 (2003) 064009. * [5] N. Arkani-Hamed, H. Georgi and M. D. Schwartz, Ann. Phys. (NY) 305 (2003) 96. * [6] D. G. Boulware and S. Deser, Phys. Rev. D 6 (1972) 3368; G. Dvali, _New J. Phys._ 8 (2006) 326; A. Vainshtein, _Surveys High Energ. Phys._ 20, 5 (2006); P. Creminelli, A. Nicolis, M. Papucci, E. Trincherini, J. High Energy Phys. 09 (2005) 003. * [7] V. A. Rubakov, hep-th/0407104. * [8] S. L. Dubovsky, J. High Energy Phys. 10 (2004) 076. * [9] V.A. Rubakov, P.G. Tinyakov, arXiv:0802437 [hep-th]. * [10] Z. Berezhiani, D. Comelli, F. Nesti and L. Pilo, Phys. Rev. Lett. 99 (2007) 131101. * [11] C. J. Isham, A. Salam and J. A. Strathdee, Phys. Rev. D 3 (1971) 867; A. Salam and J. A. Strathdee, Phys. Rev. D 16 (1977) 2668; C. Aragone and J. Chela-Flores, Nuovo Cim. A10 (1972) 818. * [12] C. J. Isham and D. Storey, Phys. Rev. D 18 (1978) 1047; M. Gurses, Phys. Rev. D 20 (1979) 1019. * [13] T. Damour and I. I. Kogan, Phys. Rev. D 66 (2002) 104024. * [14] D. Blas, C. Deffayet and J. Garriga, Class. and Quant. Grav. 23 (2006) 1697; D. Blas, Int. J. Theor. Phys. 46 (2007) 2258. * [15] S. Groot Nibbelink and M. Peloso, Class. and Quant. Grav. 22 (2005) 1313; S. G. Nibbelink, M. Peloso and M. Sexton, Eur. Phys. J. C 51 (2007) 741. * [16] V.A. Kostelecky, Phys. Rev. D 69 (2004) 105009. * [17] Point-source solutions are usually discarded because they have infinite energy, R. Jackiw, L. Jacobs, C. Rebbi, Phys. Rev. D 20 (1979) 474, but they can be retained if one allows the flux lines to end on other charges, see for example S.L. Adler and T. Piran, Rev. Mod. Phys. 56 (1984) 1. * [18] J. Chela-Flores, Int. J. Theor. Phys. 10 (1974) 103. * [19] B. Zumino, in Lectures on elementary particles and quantum field theory, edited by S. Deser, M. Grisaru and H. Pendleton. MIT Press 1970, Vol. 2 , p. 437. * [20] D. Blas, C. Deffayet and J. Garriga, Phys. Rev. D 76 (2007) 104036; * [21] G. Gabadadze and A. Iglesias, arXiv:0712.4086 [hep-th]. * [22] S. L. Dubovsky, P. G. Tinyakov and I. I. Tkachev, Phys. Rev. D 72 (2005) 084011. * [23] E. Alvarez, D. Blas, J. Garriga and E. Verdaguer, Nucl. Phys. B 756 (2006) 148. * [24] S. L. Dubovsky, P. G. Tinyakov and I. I. Tkachev, Phys. Rev. Lett. 94 (2005) 181102.

arxiv-papers

2024-09-04T02:48:54.288841

0803.1690

# VLBI Observations of SiO Masers around AH Scorpii Xi Chen11affiliation: Shanghai Astronomical Observatory, 80 Nandan Road, Shanghai 200030, P.R.China & Zhi-Qiang Shen11affiliation: Shanghai Astronomical Observatory, 80 Nandan Road, Shanghai 200030, P.R.China ###### Abstract We report the first Very Long Baseline Array (VLBA) observations of 43 GHz $v$=1, $J$=1–0 SiO masers in the circumstellar envelope of the M-type semi- regular supergiant variable star AH Sco at 2 epochs separated by 12 days in March 2004. These high-resolution VLBA images reveal that the distribution of SiO masers is roughly on a persistent elliptical ring with the lengths of the major and minor axes of about 18.5 and 15.8 mas, respectively, along a position angle of $150^{\circ}$. And the red-shifted masers are found to be slightly closer to the central star than the blue-shifted masers. The line-of- sight velocity structure of the SiO masers shows that with respect to the systemic velocity of $-6.8$ km s-1 the higher velocity features are closer to the star, which can be well explained by the simple outflow or infall without rotation kinematics of SiO masers around AH Sco. Study of proper motions of 59 matched features between two epochs clearly indicates that the SiO maser shell around AH Sco was undergoing an overall contraction to the star at a velocity of $\approx$13 km s-1 at a distance of 2.26 kpc to AH Sco. Our 3-dimensional maser kinematics model further suggests that such an inward motion is very likely due to the gravitation of the central star. The distance to AH Sco of 2.26$\pm$0.19 kpc obtained from the 3-dimensional kinematics model fitting is consistent with its kinematic distance of 2.0 kpc. circumstellar matter — masers — stars: individual (AH Sco) (catalog ) — stars: kinematics (catalog ) ## 1 Introduction AH Scorpii (AH Sco) is a semi-regular variable with an optical period of 714 days (Kukarkin et al. 1969) and a spectral type of M5Ia-Iab (Humphreys 1974). The systemic velocity of AH Sco is estimated to be about $-7$ and $-3$ km s-1 based on the observations of OH maser (Baudry, Le Squeren & Lépine 1977) and H2O maser (Lépine, Pase de Barros & Gammon 1976), respectively. The distance to AH Sco remains uncertain. A photometric distance of 4.6 kpc has been derived by Humphreys & Ney (1974) based on their infrared data. However, Baudry, Le Squeren & Lépine (1977) led to a photometric distance of 2.6 kpc under an assumption of an absolute visual magnitude of $-5.8$ for Iab stars, whereas its kinematic distance was about $1.5\sim 2.0$ kpc for its systemic velocity $-5.5\sim-7.5$ km s-1. Late type stars often exhibit circumstellar maser emission in molecules OH, H2O, and SiO. The supergiant variable AH Sco is such a star that has been detected strong maser emission with single-dish in all three species (e.g. Lépine, Pase de Barros & Gammon 1976; Baudry, Le Squeren & Lépine 1977; Balister et al. 1977, Gómez Balboa & Lépine 1986). The interferometric observations of these masers would be useful in determining the structure and kinematics of the circumstellar envelop (CSE) and understanding the physical circumstance and mass loss procedure for this supergiant variable. Unfortunately, there has been so far no any published interferometric map of OH, H2O and SiO masers toward this source. Especially, SiO masers provide a good probe of the morphology of CSE and kinematics of gas in the extended atmosphere which is a complex region located between the photosphere and the inner dust formation shell. Previous VLBI experiments have demonstrated ringlike configurations (e.g. Diamond et al. 1994; Greenhill et al. 1995; Boboltz, Diamond & Kemball 1997; Yi et al. 2005, Chen et al. 2006), or elliptical distributions (e.g. Boboltz & Marvel 2000; Sánchez et al. 2002; Boboltz & Diamond 2005) of SiO masers, and also revealed complex kinematics in SiO maser regions, e.g. contraction and expansion at the different phase of stellar pulsation (Boboltz, Diamond & Kemball 1997; Diamond & Kemball 2003; Chen et al. 2006) and even rotation (Boboltz & Marvel 2000; Hollis et al. 2001; Sánchez et al. 2002; Cotton et al. 2004; Boboltz & Diamond 2005). The ringlike or elliptical distribution that is assumed to be centered at the stellar position with a radius of 2–4 R∗ suggests that SiO masers are amplified in tangential rather than radial path. In this paper, we present the first VLBI maps of SiO maser emission toward AH Sco observed at two epochs separated by 12 days in March 2004. The observations and data reduction are described in $\S$ 2; results and discussions are presented in $\S$ 3, followed by conclusions in $\S$ 4. ## 2 Observations and data reduction The observations of the $v=$1, $J$=1–0 SiO maser emission toward AH Sco ($\alpha=$17h11m16.98s, $\delta=-32\arcdeg 19\arcmin 31.2\arcsec$, J2000) were performed at two epochs on March 8, 2004 (hereafter epoch A) and March 20, 2004 (hereafter epoch B) using the 10 stations of the Very Long Baseline Array (VLBA) of the NRAO111The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.. A reference frequency of 43.122027 GHz was adopted for the $v$=1, $J$=1–0 SiO transition. The data were recorded in left circular polarization in an 8 MHz band and correlated with the FX correlator in Socorro, New Mexico. The correlator output data had 256 spectral channels, corresponding to a velocity resolution of 0.22 km s-1. The system temperatures and sensitivities were on the order of 150 K and 11 Jy K-1, respectively, for both epochs. For the data reduction, we followed the standard procedure for VLBA spectral line observations using the Astronomical Image Processing System (AIPS) package. The bandpass response was determined from scans on the continuum calibrator (NRAO530). The amplitude calibration was achieved using the total- power spectra of AH Sco based on the “template spectrum” method. The template spectra at each epoch were obtained from the Mauna Kea (MK) station at a high elevation. A zenith opacity of about 0.05 at MK station estimated from the variation of system temperature with zenith angle was applied to correct the atmosphere absorption for each epoch. Residual group delays determined from a fringe fitting to the continuum calibrator were applied to the spectral line data. Residual fringe-rates were obtained by performing a fringe-fitting on a reference channel (at V${}_{\rm LSR}=-$9.4 km s-1), which has a relatively simple structure in the maser emission. An iterative self-calibration on the reference channel was performed to remove any structural phase. The solutions of fringe-fitting and self-calibration were then applied to the whole spectral line data. Image cubes were produced for all the velocity channels between 14 and $-$18 km s-1 with a synthesized beam of 0.69 mas $\times$ 0.20 mas at a position angle $-6^{\circ}$. Off-source rms noise ($\sigma_{\rm rms}$) in channel maps ranges from 20 mJy beam-1 in the maps with weak or no maser emission to 50 mJy beam-1 in the maps containing strong maser emission. The flux densities, and positions in right ascension (R.A.) and declination (Dec.) for all emission components with intensity above 8 $\sigma_{\rm rms}$ in each channel maps were determined by fitting a two-dimension Gaussian brightness distribution using the AIPS task SAD. Errors in R.A. and Dec. obtained from above fitting procedure range from 1 $\mu$as for components with high SNR, to 168 $\mu$as for components with low SNR, and the typical uncertainty of the fitted position of maser components was smaller than 10 $\mu$as. The remaining analysis of the maser component identifications was performed outside of the AIPS package. As described in our previous work (Chen et al. 2006), a maser spot is a single velocity component of the maser emission in each velocity channel map; a maser feature is a group of the maser spots within a small region in both space and Doppler velocity, typically 1 AU and 1 km s-1, and is expected to be a physical feature consisting of a single gas clump. In order to study the characteristics of SiO masers, it is necessary to identify maser features for each epoch. The maser spots in different channels were deemed as the same feature according to the criterion that these spots appear in at least three adjacent channels and lie within an angular separation of 0.5 mas. Finally, 82 and 87 maser features were identified for epochs A and B, respectively. ## 3 Results and Discussions ### 3.1 The spatial structure of the SiO masers The full lists of parameters for each identified feature are given in Tables 1 and 2, for epochs A and B, respectively. We fit a Gaussian curve to the velocity profile of a feature containing at least four spots to determine VLSR at the peak of velocity profile. For some features which can not be well represented by a Gaussian profile (labelled by a “$\ast$” in Tables 1 and 2), intensity weighted mean VLSR was adopted. The velocity range across the feature $\Delta u$, is defined to be the difference between the maximal and minimal velocities of the spots in the feature. Feature positions ($x$, $y$) in R.A. and Dec. were determined from an intensity weighted average over maser spots in the feature. The uncertainty ($\sigma_{x},\sigma_{y}$) of a feature position was defined as squared root of the square sum of (1) the mean spot distance from the defined feature position and (2) the mean measurement error of the spot positions. The weights proportional to the intensity of the spot were applied in the uncertainty estimation. The typical position uncertainties of features are 0.01 mas and 0.02 mas for R.A. and Dec., respectively. The positions are measured with respect to the reference feature at (0, 0) for aligning the maps in the two epochs (labelled by “R” in Tables 1 and 2; see Sect. 3.2.2). The distance of a maser feature, $r$, is measured with respect to the fitted center obtained from the ellipse model fitting to the distribution of maser features (see below). The flux density of the brightest spot in each feature was deemed as its peak flux density, P. In two top panels of Figure 1, we compare the total power imaged by the VLBA (open circle) to the total power (solid line) obtained from the MK antenna for each epoch. The total power imaged by VLBA is obtained by summing all fitted flux of spots belonging to features. The fractional power representing the ratio of the total flux imaged by VLBA to the total flux of maser emission is shown in two bottom panels. The fraction is mostly between 0.4 and 0.8. That is, on average about $\sim$60% of the total luminosity of masers was detected in our observations. Actually, such a fractional power reflects the degree of extension of the maser emission. If the apparent sizes of maser components are larger than that of the interferometric beam, the maser would be partly resolved (i.e. the fractional power is less than 1) and the fractional power should decrease with the increase of apparent size of masers. The typical size of maser spots estimated from geometric mean of sizes of the major and minor axes of the spots, which were obtained by fitting to an elliptical Gaussian brightness distribution in the CLEAN map, is 0.5 mas. This typical scale size is slightly larger than the geometric mean of the VLBA beam of 0.4 mas. Thus $\sim$40% missing flux in our map is mainly due to the high spatial resolution of the interferometric array. Figure 2 shows the distributions of maser features toward AH Sco for the two epochs. These high resolution VLBI images reveal a persistent elliptical structure of SiO masers around AH Sco druing an interval of 12 days. We characterized this morphology by performing a least-squares fit of an ellipse to the distribution of masers weighted by the flux density of each feature for each of two epochs. The best-fitting ellipses and the ellipse centers are also shown in Figure 2. The lengths of the major and minor axes were found to be 18.6 and 15.7 mas for epoch A, and 18.4 and 15.9 mas for epoch B, respectively, with the major axis of the ellipse oriented similarly at $\sim 150^{\circ}$ at both epochs. And the fitted centers of the elliptical distributions are almost the same at both epochs of $-7.9$ and 5.8 mas in R.A. and Dec., respectively. At a distance to AH Sco of 2.26 kpc (see Sect. 3.2.3), the distribution of SiO masers corresponds to about $42\times 35$ AU for both two epochs. However, an axial ratio of 1.18 suggests that the distribution of maser features around AH Sco can be viewed approximately as the ringlike structure with an average diameter of about 17.2 mas (obtained from the geometric average of the major and minor axes for both epochs). The simultaneous near-infrared interferometry and radio interferometry imaging of circumstellar SiO maser for late type stars have been done recently (e.g. Boboltz & Wittkowski 2005; Wittkowski et al. 2007; Cotton et al. 2004, 2006). These observations reveal that the ratio of the maser ring radius to the photospheric radius of the central star is about $1.5-4.0$. Unfortunately, there has been no any published photospheric radius measurement for AH Sco. Thus we can not directly compare SiO maser ring radius with stellar radius for AH Sco. We also notice that the red-shifted SiO masers lie slightly closer to the center than the blue-shifted masers (see Figure 2). This can be seen more clearly in Figure 3, showing the maser feature distance from the fitted center versus its line-of-sight (LOS) velocity (see Sect. 3.2.1). This phenomenon seems to be explained under the assumption that maser gas was undergoing infall to the central star. This is because along the same LOS path the red- and blue-shifted masers would appear in front of and behind the star, respectively, as long as the coherence path lengths satisfy the requirement of maser excitation, under the condition that the maser gas was undergoing infall to the central star during our observations. However the blue-shifted masers generated behind star would be obscured by the stellar disc projected on the LOS, while the red-shifted masers would not. Thus only red-shifted maser emission will be seen closer to the center. Actually, we have confirmed that the SiO maser shell contracts to the star during our observations in Sect. 3.2.2. Moreover, from Figure 3, some systemic masers with velocity $-7$ km s-1 (see Sect. 3.2.1) locate at the distance of 7 mas which can be viewed as an upper limit to the photospheric radius, then the extreme blue-shifted masers with the distance of less than 7 mas would be obscured. This is consistent with our data (see Figure 3) showing that all blue-shifted masers locate at the distance of larger than 7 mas. ### 3.2 The kinematics of the SiO masers #### 3.2.1 The kinematics obtained from LOS velocities From Figure 2, we notice that there appears a velocity gradient at both epochs, with the bluest- and reddest-shifted maser features lying closer to the center of the distribution than those with intermediate velocities. To verify this, we plotted in Figure 3 feature distance from the fitted center (marked by the red star in Figure 2) versus its LOS velocity for both epochs. Both epochs appear to have the same distribution with a peak near the velocity of $-7$ km s-1 and decreasing maser distance with increasing deviation of velocity from this peak velocity. This has also been seen in some OH maser sources (e.g. Reid et al. 1977, Chapman & Cohen 1986), H2O maser sources (e.g. Yates & Cohen 1994) and SiO maser sources (e.g. Boboltz & Marvel 2000; Wittkowski et al. 2007). A widely used simple model to explain this phenomenon is that of a uniformly expanding thin shell (e.g. Reid et al. 1977; Yates & Cohen 1994; Wittkowski et al. 2007). In this model the projected distance $r$ (in the fourth column of Tables 1 and 2) of a maser on the shell from the center is related to its LOS velocity $V_{\rm LSR}$ by the expression $(\frac{r}{r_{s}})^{2}+(\frac{V_{\rm LSR}-V_{*}}{V_{exp}})^{2}=1,$ (1) where $V_{*}$ is the systemic velocity of maser source, $r_{s}$ is the shell radius and Vexp is the expanding velocity. Apparently, this model traces an ellipse on the $r-V_{\rm LSR}$ plot. The uniformly expanding thin shell model was used to characterize the expansion or contraction kinematics of a circular maser distribution. For the case of AH Sco, even though we characterize the maser distribution as an ellipse, an axial ratio of $\sim$1.2 suggests that the distribution of maser features is approximately a circular structure as discussed in Section 3.1. Thus, we can also apply the uniformly expanding thin shell model to AH Sco. Moreover, most maser features locate in the northwest and southeast (i.e. the direction of major axis), and only few maser features locate in the northeast and southwest (i.e. the direction of minor axis). These make the assumption of uniformly expansion/contraction in all directions of the uniformly expanding thin shell model to be suitable for the case of AH Sco. We performed a least- squares fit of the uniformly expanding thin shell model to the distribution of epochs A and B. The LSR stellar velocity of AH Sco, which is estimated to be about $-7$ and $-3$ km s-1 based on the observations of OH maser (Baudry, Le Squeren & Lépine 1977) and H2O maser (Lépine, Pase de Barros & Gammon 1976), has not been measured particularly well yet. Thus the LSR stellar velocity $V_{*}$, together with the shell radius $r$ and expansion velocity Vexp, is treated as a free parameter in the fitting procedure. The values of $V_{*}$, Vexp and $r_{s}$ were found to be $-6.8\pm$0.5 km s-1, 18.8$\pm$2.0 km s-1 and 9.3$\pm$0.1 mas, respectively, for both epochs, where the uncertainties are their standard errors. The best fitted $r-V_{\rm LSR}$ ellipse is also plotted in Figure 3\. The systemic velocity of AH Sco of $-6.8$ km s-1 is consistent with the value of $-7$ km s-1 from the OH maser observations. The shell radius of 9.3 mas determined from above model fitting is larger than the radius of maser distribution of 8.6 mas (see Sect. 3.1). A note is that the definition of the shell radius determined from the uniformly expanding thin shell model is different from that of the radius of the maser distribution. The shell radius reflects the scale of a 3-dimensional maser spherical shell, whereas the radius of maser distribution reflects the scale of maser distribution in the sky plane, and is the projected size on the sky plane of the 3-dimensional spherical shell. Thus it is not surprising that the shell radius is a bit larger than the radius of maser distribution. For comparison, the escape velocity calculated at the shell radius of 9.3 mas, assuming a typical mass of 10 $M_{\odot}$ for supergiant and a distance to AH Sco of 2.26 kpc, is about 29 km s-1. Thus, the expansion/contraction velocity at the location of SiO maser shell is less than the corresponding escape velocity, suggesting that the maser gas is still gravitational bound to the star. However, the current fitting can not tell the sign of Vexp term (as can be seen in Eq. (1)), and thus can not differentiate between expansion and contraction of the maser shell. The dominant expansion or contraction of maser shell can be clarified by the SiO maser proper motion analysis to be discussed below. #### 3.2.2 Maser Proper Motions We can study proper motion and the kinematics of the CSE of AH Sco by tracing the matched features that appeared in both epochs. Because the absolute position of the phase center in each image is not kept during the data reduction, we must align two-epoch maps for studying the proper motion. The feature used for registration is the one with a velocity V${}_{\rm LSR}\approx-9.3$ km s-1 (labelled by “R” in Tables 1 and 2) at both epochs. And then we shift the coordinate frames for both epochs to align the origin (0, 0) with this feature. At an assumed distance of 2.6 kpc (Baudry et al. 1977) and a maximum expansion/contraction velocity of 20 km s-1 (see Sect. 3.2.1), the maser proper motion should be less than 0.05 mas in an interval of 12 days. Thus we can match these features from one epoch to another epoch using the criterion that the angular separation of the matched features between two epochs should not exceed 0.15 mas after allowing the maximum position uncertainty of (0.05, 0.10) mas (see Tables 1 and 2) and they have similar velocity profile and flux density. As a result, we identified 59 commonly matched maser features between two epochs (see Tables 1 and 2). Actually, using a reference feature located on the maser shell to align maps for two epochs could introduce a constant offset vector representing the motion of the reference feature in the individual maser proper motions. We assumed that the vector-average of the proper motions for all the matched features represents the motion of the aligned feature. In order to present a better representation of the real motions of individual features, the mean proper motion was subtracted from each of the determined proper motion vectors. The proper motions of matched maser features are shown in Figure 4. Here we adopt the distance to AH Sco of 2.26 kpc (see Sect. 3.2.3) for denoting the velocity values of the proper motions. From Figure 4, we can clearly see that the maser shell shows an overall contraction toward the central star. In order to better characterize the net contraction of the masers, we computed the separations between pairwise combinations of features. This technique has previously been applied to analyse proper motions of OH masers (Chapman, Cohen & Saika 1991; Bloemhof, Reid & Moran 1992), H2O masers (Boboltz & Marvel 2007), and SiO masers (Boboltz, Diamond & Kemball 1997; Chen et al. 2006), and has no dependence on the alignment of maps. The procedure involves computing the angular separation between two features at the first epoch and the separation between the corresponding two features at the second epoch. The difference between the two values of separation is referred to as the pairwise separation. The procedure is repeated for all the possible pair combinations. However, the inclusion of all possible pair combinations often results in decreasing toward zero due to the bias caused by calculating pairs of closely spaced maser features. For the sake of clarity, and to determine representative values for the angular shifts due to the contraction, we have included only those pairs separated by more than 9 mas (corresponding to the radius of maser distribution). We obtained the mean value of these pairwise separations of $-0.039\pm 0.002$ mas, in an interval of 12 days, corresponding to a proper motion of $-1.186\pm 0.061$ mas yr-1 or a velocity of $-12.7\pm 0.7$ km s-1 at a distance of 2.26 kpc, where the uncertainties are the standard errors. The negative value of proper motion implies an overall contraction of the maser shell. The contraction value of $1.186\pm 0.061$ mas yr-1 of maser shell derived from the pairwise separation is significantly less than the scalar-averaged value of SiO proper motions of $1.96\pm 0.15$ mas yr-1 (where its uncertainty is the standard error). This is because some of maser proper motions do not completely point to the center or even a few proper motions show outflow motion as can be seen in Figure 4. The contraction of SiO maser shell has been reported in two Mira variables R Aqr (Boboltz, Diamond & Kemball 1997) and TX Cam (Diamond & Kemball 2003) and one red supergiant VX Sgr (Chen et al. 2006; 2007). Our observations provide an inward motion of SiO maser shell around another red supergiant AH Sco. In Table 3 we list these four sources. We also estimated the stellar optical phase of AH Sco at our observation sessions to be $\phi\simeq$ 0.55 (i.e. at the optical minimum phase) based on the American Association of Variable Star Observers (AAVSO) data. Interestingly, the optical phase at which the SiO maser shell around the red supergiant AH Sco contracts is nearly the same as that seen in other three sources: VX Sgr ($\phi=0.75-0.80$; Chen et al. 2006), R Aqr ($\phi=0.78-0.04$; Boboltz, Diamond & Kemball 1997), TX Cam ($\phi=0.50-0.65$; Diamond & Kemball 2003). This infers that the contraction of the SiO maser shell would occur during an optical stellar phase of $0.5-1$, which agrees with the previous conclusion reported by Chen et al. (2006) and the theoretical kinematical model results of Humphreys et al. (2002). Moreover, from Table 3 we can find that the contraction velocity of about 13 km s-1 of maser shell around AH Sco is the largest among the four sources. #### 3.2.3 3-dimensional kinematics model for SiO masers In order to estimate further the kinematical parameters of SiO masers and the distance to AH Sco, we made model-fitting to analyze spatial distribution and proper motion of SiO maser features as done by Gwinn, Moran & Reid (1992) and Imai et al. (2000; 2003). The model fitting requires to minimize the squared sum of the differences between the observed and model velocities, $\chi^{2}=\sum_{i}\\{\frac{[\mu_{ix}-V_{ix}/(a_{0}d)]^{2}}{\sigma_{ix}^{2}}+\frac{[\mu_{iy}-V_{iy}/(a_{0}d)]^{2}}{\sigma_{iy}^{2}}+\frac{[u_{iz}-V_{iz}]^{2}}{\sigma_{iz}^{2}}\\},$ (2) where, $\mu_{ix}$ and $\mu_{iy}$ are the observed proper motions in R.A. and Dec., respectively, and $u_{iz}$ the observed velocity along LOS, $d$ the distance to the maser source, $a_{0}\equiv 4.74$ km s-1 mas -1 yr kpc-1, ($\sigma_{ix},\sigma_{iy},\sigma_{iz}$) the standard deviations of the observed velocity vectors which are determined in the similar manner of Imai et al. (2002). In this work, we assume a spherically expanding flow in SiO maser region, thus the model velocity vector $\textbf{V}_{i}$ (including $V_{ix}$, $V_{iy}$ in R.A. and Dec., and $V_{iz}$ in LOS) for the $i$th maser feature can be expressed as $\mathbf{V}_{i}=\mathbf{V_{0}}+V_{exp}(i)\frac{\mathbf{r}_{i}}{r_{i}}\mathrm{,}$ (3) $\mathbf{r}_{i}=\mathbf{x}_{i}-\mathbf{x}_{0}\ (\mathrm{or}\ r_{ix}=x_{i}-x_{0},r_{iy}=y_{i}-y_{0},r_{iz}=z_{i}),$ (4) where $\textbf{V}_{0}$ ($v_{0x}$, $v_{0y}$, $v_{0z}$) is a systemic velocity vector of the stellar system, reflecting the motion of the central star; $V_{exp}(i)$ is an expanding velocity as a function of the distance from the origin of the flow, ri; x0 ($x_{0}$, $y_{0}$, 0) is the position vector of the flow origin with respect to the position of reference maser feature (here, we assumed that the origin of the flow was at the star, whose positions of $x_{0}=-7.94$ mas and $y_{0}=-5.83$ mas have been derived from the least- squares fit of an ellipse to the distribution of masers in Sect. 3.1.); and ($x_{i}$, $y_{i}$) is an observed position of maser feature on the sky plane; the position of a maser feature along the LOS, $z_{i}$ is estimated as one of the free parameters too (Imai et al. 2000). In the model fitting procedure, we adopted the expanding velocity as $V_{exp}(i)=V_{1}(r_{i}/r_{0})^{\alpha}$, where $V_{1}$ is the expansion velocity at a unit distance of $r_{0}\equiv 10$ mas, and $\alpha$ a power-law index indicating the apparent acceleration or deceleration of the flow. Moveover, we excluded those maser features with large positive expansion velocities in the fitting. Finally, we used 48 proper motion data and obtained the best solutions with their standard errors, which are given in Table 4. From Table 4, we can see that a negative expansion velocity of $-14.1\pm 1.4$ km s-1 at a shell radius of 10 mas was estimated from the best-fit model, supporting the presence of a real contracting flow in the SiO maser region around AH Sco. This is consistent with the conclusion derived from the pairwise separation calculation in Sect. 3.2.2. For comparison, we also estimate the corresponding velocity of $-15.0\pm 1.5$ km s-1 at the mean SiO maser radius of 9 mas according to the power-law index of the acceleration of the flow $\alpha$ obtained in our model. However, the velocity of $-15.0\pm 1.5$ km s-1 seems larger than that of $-12.7\pm 0.7$ km s-1 obtained from the pairwise separation analysis, which is because that we excluded maser features showing outflow motion in the 3-dimensional kinematics model fitting procedure. And the velocity of $15.0\pm 1.5$ km s-1 is slightly smaller than that of 18.8$\pm$2.0 km s-1 obtained from a least-squares fit of the uniformly expanding thin shell model to SiO maser LOS velocity structure (see Sect. 3.2.1), which is due to that some maser proper motions used in model fitting still deviate from the originating point of the inflow (i.e. the position of central star; see Figure 4), whereas the model fitting method involves one critical assumption that velocity vectors are in the radial direction from the commonly originating point of the flow. The systemic velocity along LOS of $v_{0z}=-5.2$ km s-1 from the 3-dimensional kinematics model is also roughly consistent with that of $-6.8$ km s-1 derived from the uniformly expanding thin shell model in Sect. 3.2.1, suggesting that the assumption of the origin of the flow located at the star is reasonable. More interestingly, we obtain a negative power-law index of the acceleration of the flow, $\alpha=-0.54\pm 0.16$, indicating that the contracting flow was accelerating in the SiO maser region with a similar form of $v\sim r^{-0.5}$ of the gravitational contraction. Thus the 3-dimensional maser kinematics model suggests that the infall motions of SiO masers can be achieved under the gravitational effect of the central star. The velocity gradient across the SiO maser region usually used in the numerical simulation of SiO masers (e.g. Doel et al. 1995; Humphreys et al. 2002) can be expressed by $\rm\varepsilon=\frac{dlnV}{dlnr}=\frac{rdV}{Vdr}.$ (5) A value of $\varepsilon=0$ corresponds to a constant velocity expansion, while $\varepsilon\geq 1$ corresponds to a velocity field with large radial accelerations. A value of $\varepsilon=1$ was usually adopted in the SiO maser numerical simulations (e.g. Doel et al. 1995). Chapman & Cohen (1986) estimated the velocity gradient $\varepsilon$ for the OH, H2O and SiO maser emission around VX Sgr and found that $\varepsilon\approx 0.2$ in the 1612 MHz OH maser region, $\varepsilon\approx 0.5$ in the region of the H2O and mainline OH masers, and $\varepsilon\approx 1$ in SiO maser region. However, the power-law index of the acceleration of the flow ($\alpha=-0.54\pm 0.16$) derived from our best-fit kinematic model suggests a negative velocity gradient value of $\varepsilon=-0.54$ in SiO maser region around AH Sco. This is different from those flows that apparently exhibit the accelerations in their SiO maser kinematics (e.g. VX Sgr, Chapman & Cohen 1986; S Ori, Wittkowski et al. 2007) and the positive velocity gradient value used in the maser simulations. The content of maser simulations is beyond this work. However, we think that such a negative velocity gradient $\varepsilon$ adpoted in maser simulation may be necessary for understanding the SiO maser emission, especially during the infall stage of the SiO maser shell. The 3-dimensional kinematics model fitting shows a best solution for the distance to AH Sco of $2.26\pm 0.19$ kpc. This distance value seems reasonable. Firstly, this distance of AH Sco is in a good agreement with its estimated ‘near’ kinematic distance of about 2.0 kpc at the systemic velocity of $-6.8$ km s-1 with the adopted galactic constants, R${}_{\odot}=$ 8.5 kpc and $\Theta_{\odot}=$ 220 km s-1. Secondly, at the distance of 2.26 kpc, the scalar-averaged value of $1.96\pm 0.15$ mas yr-1 of proper motions of matched features (shown in Figure 4) would correspond to a velocity of $21.0\pm 1.6$ km s-1. This velocity is consistent with the expansion/contraction velocity of $18.8\pm 2.0$ km s-1 in SiO maser region obtained from a least-squares fit of the uniformly expanding thin shell model to LOS velocity structure. Thus we adopt the distance to AH Sco of 2.26 kpc throughout this work. ## 4 Conclusions We summarize the main results obtained from 2-epoch (at an interval of 12 days) monitoring observations of the 43 GHz $v=1,J=1-0$ SiO maser emission toward AH Sco performed in March 2004, corresponding to a stellar optical phase of $\sim$ 0.55. 1. (1). Our observations revealed a persistent elliptical structure of SiO masers with the sizes of the major and minor axes of about 18.5 and 15.8 mas, respectively, along a position angle of $150^{\circ}$. We notice that the red- shifted SiO maser emission lies slightly closer to the center than the blue- shifted one. 2. (2). The LOS velocity structure of the SiO masers shows a velocity gradient at both epochs, with masers towrad the blue- and red-shifted ends of the spectrum lying closer to the center of the maser distribution than masers at intermediate velocities, which can be explained by the outflow or infall kinematics of SiO maser shell. By analyzing the uniformly expanding thin shell model to the LOS velocity of SiO masers, we estimated the expansion/contraction velocity of about 19 km s-1 in SiO maser region around AH Sco. 3. (3). The proper motions of 59 matched features between two epochs show that the SiO maser shell around AH Sco was undergoing inward motion to the central star. Computing pairwise separation of these matched features, we obtained that the maser shell contracts toward AH Sco with a velocity of about $13$ km s-1 at a distance to AH Sco of 2.26 kpc. The stellar optical phase of red supergiant AH Sco is very close to that of Mira variables R Aqr and TX Cam and red supergiant VX Sgr when the SiO maser shell contracts. And the contraction velocity of about 13 km s-1 of maser shell around AH Sco is the largest one among the known four sources showing contraction of SiO maser shell. 4. (4). We made a 3-dimensional kinematics model to analyze spatial distribution and proper motion of SiO maser features. The 3-dimensional maser kinematics model further suggested that the contraction of SiO maser shell around AH Sco is mainly due to the gravitation of the central star. And the distance to AH Sco of $2.26\pm 0.19$ kpc estimated from this kinematics model fitting is consistent with the kinematic distance of 2.0 kpc at the systemic velocity of AH Sco of $-7$ km s-1. We thank an anonymous referee for helpful comments that improved the manuscript. We also acknowledge with thanks data from the AAVSO International Database based on observations submitted to the AAVSO by variable star observers worldwide. This work was supported in part by the National Natural Science Foundation of China (grants 10573029, 10625314, and 10633010) and the Knowledge Innovation Program of the Chinese Academy of Sciences (Grant No. KJCX2-YW-T03), and sponsored by the Program of Shanghai Subject Chief Scientist (06XD14024) and the National Key Basic Research Development Program of China (No. 2007CB815405). X. Chen thanks the support by the Knowledge Innovation Program of the Chinese Academy of Sciences. ZQS acknowledges the support by the One-Hundred-Talent Program of Chinese Academy of Sciences. ## References * (1) Balister, M., Batchelor, R. A., Haynes, R. F., Knowles, S. H., Mc Culloch, M. G., Robinson, B. J., Wellington, K. J., & Yabsley, D. E. 1977, MNRAS, 180, 415 * (2) Baudry, A., Le Squeren, A. M., & Lépine, J. R. D. 1977, A&A. 54,593 * (3) Bloemhof, E. E., Reid, M. J., & Moran, J. M. 1992, ApJ, 397, 500 * (4) Boboltz, D. A., Diamond, P. J., & Kemball, A. J. 1997, ApJ, 487, L147 * (5) Boboltz, D. A., & Diamond, P. J. 2005, ApJ, 625, 978 * (6) Boboltz, D. A., & Marvel, K. B., 2007, ApJ, 665, 680 * (7) Boboltz, D. A., & Marvel, K. B. 2000, ApJ, 545, L149 * (8) Boboltz, D. A., & Wittkowski, M. 2005, ApJ, 618, 953 * (9) Chapman, J. M., & Cohen, R. J. 1986, MNRAS, 220, 513 * (10) Chapman, J. M., Cohen, R. J., & Saika, D. J. 1991, MNRAS, 249, 227 * (11) Chen, X., Shen, Z.-Q., Imai, H., & Kamohara, R. 2006, ApJ, 640, 982 * (12) Chen, X., Shen, Z.-Q., & Xu, Y. 2007, ChJA&A, 7, 531 * (13) Cotton, W. D., et al. 2006, A&A 456, 339 * (14) Cotton, W. D., et al. 2004, A&A, 414, 275 * (15) Diamond, P. J., Kemball, A. J., Junor, W., Zensus, A. Benson, J., & Dhawan, V. 1994, ApJ, 430, L61 * (16) Diamond, P. J., & Kemball, A. J. 2003, ApJ, 599, 1372 * (17) Doel, R. C., Gray, M. D., Humphreys, E. M. L., Braithwaite, M. F., & Field, D. 1995, A&A, 302, 797 * (18) Gómez Balboa, A. M., & Lepine, J. R. D. 1986, A&A, 159, 166 * (19) Greenhill, L. J., Colomer, F., Moran, J. M., Danchi, W. C., & Bester, M. 1995, ApJ, 449, 365 * (20) Gwinn, C. R., Moran, J. M., & Reid, M. J. 1992, ApJ, 393, 149 * (21) Hollis, J. M., Boboltz, D. A., Pedelty, J. A., White, S. M., & Forster, J. R. 2001, ApJ, 559, L37 * (22) Humphreys, R. M. 1974, ApJ, 188, 75 * (23) Humphreys, E. M. L., & Gray, M. D., Yates, J. A., Field, D., Bowen, G. H., & Diamond, P. J. 2002, A&A, 386, 256 * (24) Humphreys, R. M., Ney, E. P. 1974, ApJ, 194, 623 * (25) Imai, H., Kameya, O., Sasao, T., Miyoshi, M., Deguchi, S., Horiuchi, S., & Asaki, Y. 2000, ApJ, 538, 751 * (26) Imai, H., et al. 2003, ApJ, 590, 460 * (27) Kukarkin, B. V., et al. 1969. General Catalogue of Variable Stars, 3rd edn., Astronomical Council of the Academy of Sciences in the USSR, Moscow. * (28) Lépine, J. R. D., Pase de Barros, M. H., & Gammon, R. H. 1976, A&A, 48, 269 * (29) Reid, M. J., Muhleman, D. O., Moran, J. M., Johnston, K. J., & Schwartz, P. R. 1977, ApJ, 214, 60 * (30) Sánchez Contreras, C., Desmurs, J. F., Bujarrabal, V., Alcolea, J., & Colomer, F. 2002, A&A, 385, L1 * (31) Wittkowski, M., Boboltz, D. A., Ohnaka, K., Driebe, T., & Scholz, M. 2007, A&A, 470, 191 * (32) Yates, J. A., & Cohen, R. J. 1994, MNRAS, 270, 958 * (33) Yi, J., Booth, R. S., Conway, J. E., & Diamond, P. J. 2005, A&A, 432, 531 Table 1: 43 GHz SiO maser features around AH Sco observed by VLBA on March 8, 2004. ID | VLSR | $\Delta u$ | r | $x$ | $\sigma_{x}$ | $y$ | $\sigma_{y}$ | P | S | Match ID ---|---|---|---|---|---|---|---|---|---|--- | (km s-1) | (mas) | (mas) | (mas) | (Jy) | (Jy km s-1) | Epoch 2 (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) 1 | -17.28 | 1.52 | 8.10 | -2.862 | 0.002 | -0.452 | 0.005 | 11.7 | 36.8 | 1 2∗ | -16.32 | 0.65 | 6.96 | -1.356 | 0.009 | 7.974 | 0.032 | 0.7 | 2.1 | … 3 | -15.16 | 3.26 | 8.31 | -1.637 | 0.004 | 0.448 | 0.008 | 19.7 | 147.5 | 2 4∗ | -14.94 | 0.87 | 8.76 | -1.679 | 0.007 | -0.250 | 0.022 | 1.5 | 4.9 | … 5 | -14.54 | 2.82 | 8.42 | -1.304 | 0.003 | 0.686 | 0.008 | 16.6 | 84.3 | 6 6∗ | -14.16 | 0.87 | 8.56 | -1.724 | 0.004 | -0.018 | 0.015 | 8.2 | 21.0 | … 7 | -13.23 | 0.87 | 8.34 | -3.051 | 0.004 | -0.909 | 0.009 | 3.1 | 8.1 | 8 8 | -12.56 | 1.09 | 8.45 | -1.228 | 0.007 | 0.738 | 0.007 | 10.3 | 28.7 | 11 9 | -12.22 | 3.04 | 8.87 | -0.792 | 0.003 | 0.621 | 0.005 | 25.5 | 151.3 | 9 10 | -11.98 | 2.17 | 8.38 | -1.790 | 0.002 | 0.176 | 0.007 | 9.4 | 52.9 | … 11 | -10.94 | 2.17 | 9.47 | -0.330 | 0.011 | 0.243 | 0.015 | 9.8 | 59.9 | 13 12 | -9.97 | 1.30 | 9.04 | -1.329 | 0.008 | -0.305 | 0.019 | 4.5 | 18.9 | 14 13 | -9.77 | 2.82 | 9.30 | -0.957 | 0.003 | -0.280 | 0.008 | 15.9 | 103.8 | 16 14 | -9.57 | 1.09 | 8.39 | -16.011 | 0.007 | 8.277 | 0.013 | 5.2 | 21.6 | … 15R | -9.31 | 2.39 | 9.88 | 0.000 | 0.005 | 0.000 | 0.007 | 35.9 | 254.8 | 18 16 | -9.20 | 1.09 | 9.58 | 0.030 | 0.019 | 0.567 | 0.012 | 1.9 | 6.6 | 19 17∗ | -9.05 | 1.09 | 8.95 | -15.261 | 0.009 | 11.052 | 0.036 | 3.5 | 10.1 | 20 18∗ | -9.04 | 0.87 | 8.93 | -15.099 | 0.034 | 11.227 | 0.021 | 5.1 | 12.7 | … 19 | -8.96 | 1.95 | 9.39 | -15.667 | 0.006 | 11.237 | 0.018 | 6.2 | 29.5 | 21 20 | -8.68 | 1.74 | 10.21 | 0.412 | 0.008 | 0.014 | 0.026 | 6.7 | 23.3 | 22 21 | -8.58 | 1.09 | 11.93 | -16.001 | 0.009 | 14.662 | 0.029 | 1.4 | 6.0 | 23 22 | -8.33 | 2.82 | 8.37 | -15.791 | 0.009 | 8.848 | 0.016 | 5.3 | 40.4 | 27 23∗ | -8.32 | 0.65 | 11.59 | -1.897 | 0.012 | -4.038 | 0.028 | 1.2 | 3.1 | … 24 | -8.23 | 1.52 | 9.79 | -12.858 | 0.006 | 14.322 | 0.019 | 4.4 | 19.2 | 25 25 | -8.12 | 1.09 | 8.58 | -14.846 | 0.005 | 10.985 | 0.014 | 7.5 | 26.2 | 24 26 | -8.02 | 2.82 | 9.74 | 0.438 | 0.009 | 0.925 | 0.018 | 7.2 | 45.9 | 26 27 | -7.80 | 1.74 | 7.91 | -9.327 | 0.007 | 13.627 | 0.013 | 5.3 | 29.1 | 29 28 | -7.56 | 1.09 | 11.10 | -15.678 | 0.024 | 13.829 | 0.041 | 1.5 | 6.3 | … 29 | -7.55 | 1.09 | 7.13 | -7.390 | 0.009 | 12.937 | 0.029 | 1.9 | 6.9 | … 30 | -7.07 | 1.74 | 11.44 | -15.386 | 0.006 | 14.560 | 0.015 | 4.3 | 22.1 | 30 31 | -6.72 | 1.09 | 8.24 | -9.195 | 0.010 | 13.984 | 0.025 | 2.3 | 8.6 | 31 32 | -6.70 | 2.39 | 6.90 | -5.873 | 0.003 | 12.406 | 0.014 | 8.8 | 53.1 | … 33 | -6.45 | 1.74 | 9.34 | -12.505 | 0.012 | 14.010 | 0.018 | 2.8 | 14.8 | … 34∗ | -6.14 | 0.65 | 7.81 | -7.069 | 0.011 | 13.593 | 0.032 | 1.2 | 2.8 | … 35∗ | -6.10 | 0.65 | 9.38 | -16.537 | 0.015 | 9.687 | 0.032 | 1.8 | 4.9 | 32 36 | -5.53 | 1.52 | 8.82 | -8.159 | 0.005 | 14.657 | 0.003 | 22.6 | 92.2 | 34 37 | -4.58 | 1.74 | 9.07 | -16.715 | 0.003 | 8.303 | 0.006 | 10.3 | 48.5 | 36 38 | -4.20 | 1.09 | 8.88 | -7.921 | 0.006 | 14.718 | 0.003 | 8.6 | 41.4 | 37 39∗ | -3.70 | 0.65 | 9.51 | -17.329 | 0.024 | 4.082 | 0.039 | 2.4 | 6.4 | 39 40 | -3.10 | 2.39 | 8.75 | -16.673 | 0.007 | 4.814 | 0.016 | 11.2 | 67.2 | 43 41 | -3.03 | 1.96 | 9.13 | -7.780 | 0.006 | 14.967 | 0.008 | 12.4 | 75.7 | 41 42 | -2.88 | 2.17 | 9.25 | -16.938 | 0.008 | 8.141 | 0.009 | 6.7 | 44.6 | 42 43 | -2.82 | 1.74 | 8.39 | -16.237 | 0.013 | 4.292 | 0.024 | 3.5 | 13.5 | 47 44 | -2.78 | 1.52 | 11.08 | -14.008 | 0.016 | 15.137 | 0.026 | 2.2 | 10.0 | 46 45 | -2.32 | 2.39 | 9.17 | -16.406 | 0.006 | 9.461 | 0.009 | 12.9 | 79.8 | 45 46 | -1.72 | 2.17 | 8.03 | -15.938 | 0.009 | 4.691 | 0.018 | 7.2 | 46.0 | 48 47∗ | -1.50 | 0.87 | 8.27 | -16.227 | 0.022 | 5.114 | 0.047 | 4.6 | 6.4 | 50 48∗ | -0.92 | 0.65 | 7.73 | -15.366 | 0.015 | 3.546 | 0.043 | 2.0 | 5.5 | … 49 | -0.73 | 1.52 | 8.66 | -15.946 | 0.017 | 9.235 | 0.038 | 3.8 | 15.5 | … 50 | -0.64 | 1.52 | 9.15 | -11.534 | 0.012 | -2.610 | 0.022 | 4.2 | 18.8 | 52 51 | -0.57 | 1.09 | 9.96 | 1.832 | 0.021 | 7.554 | 0.029 | 1.6 | 7.2 | 53 52 | 0.25 | 1.09 | 7.10 | -14.770 | 0.015 | 3.747 | 0.032 | 2.1 | 7.6 | 54 53 | 1.08 | 2.17 | 8.67 | -11.042 | 0.014 | -2.283 | 0.025 | 5.4 | 30.3 | 57 54∗ | 1.30 | 0.65 | 8.07 | -11.478 | 0.047 | 13.113 | 0.047 | 4.8 | 8.2 | … 55 | 1.43 | 1.30 | 10.73 | 2.032 | 0.010 | 1.950 | 0.014 | 3.5 | 14.1 | 56 56 | 1.44 | 1.09 | 8.50 | -10.621 | 0.005 | -2.252 | 0.016 | 3.5 | 10.1 | 55 57∗ | 1.48 | 0.65 | 8.23 | -11.841 | 0.012 | 13.107 | 0.050 | 1.8 | 4.3 | 58 58 | 2.34 | 1.52 | 7.57 | -11.300 | 0.005 | 12.637 | 0.021 | 8.4 | 36.4 | 61 59 | 2.65 | 1.95 | 8.11 | -12.171 | 0.009 | 12.774 | 0.030 | 3.1 | 20.3 | … 60 | 2.87 | 1.74 | 7.72 | -2.150 | 0.014 | 0.760 | 0.021 | 1.8 | 11.9 | 65 61 | 3.19 | 1.09 | 7.08 | -11.118 | 0.018 | 12.181 | 0.023 | 7.4 | 24.4 | 64 62 | 3.77 | 1.52 | 6.84 | -10.486 | 0.006 | 12.202 | 0.016 | 4.0 | 16.7 | 66 63 | 3.86 | 1.52 | 7.73 | -12.410 | 0.021 | 12.175 | 0.024 | 3.1 | 17.1 | … 64∗ | 3.86 | 0.65 | 7.38 | -10.185 | 0.010 | -1.221 | 0.045 | 1.6 | 4.0 | 68 65 | 4.04 | 1.30 | 7.43 | -1.074 | 0.007 | 3.081 | 0.023 | 1.2 | 6.0 | … 66 | 4.10 | 3.04 | 8.66 | 0.508 | 0.008 | 4.077 | 0.013 | 2.8 | 29.4 | 70 67 | 4.70 | 1.52 | 7.40 | -12.565 | 0.009 | 11.650 | 0.020 | 5.8 | 30.0 | 71 68 | 4.83 | 1.30 | 6.67 | -9.377 | 0.012 | -0.696 | 0.033 | 2.3 | 10.2 | … 69 | 5.20 | 1.95 | 7.22 | -3.566 | 0.014 | 0.108 | 0.027 | 2.4 | 14.6 | … 70 | 5.43 | 1.30 | 6.05 | -10.355 | 0.003 | 11.399 | 0.012 | 5.5 | 19.7 | 75 71∗ | 5.59 | 0.65 | 6.82 | -4.651 | 0.008 | -0.124 | 0.023 | 1.2 | 3.2 | 76 72 | 5.97 | 1.52 | 5.75 | -13.283 | 0.009 | 3.593 | 0.016 | 2.5 | 15.5 | 74 73 | 7.00 | 1.30 | 6.21 | -7.430 | 0.011 | 12.023 | 0.025 | 1.9 | 8.1 | 79 74 | 7.33 | 3.04 | 6.39 | -9.017 | 0.006 | -0.477 | 0.013 | 4.0 | 41.0 | 78 75 | 7.42 | 1.74 | 6.51 | -8.278 | 0.008 | 12.338 | 0.022 | 2.9 | 16.6 | … 76 | 7.73 | 2.17 | 5.08 | -12.707 | 0.005 | 3.946 | 0.008 | 4.0 | 27.5 | 82 77∗ | 7.99 | 0.65 | 7.79 | -0.359 | 0.017 | 4.208 | 0.034 | 1.3 | 3.5 | … 78∗ | 8.01 | 1.09 | 6.16 | -8.120 | 0.008 | 11.995 | 0.023 | 2.3 | 7.8 | 83 79∗ | 8.35 | 1.09 | 3.98 | -4.070 | 0.014 | 6.607 | 0.026 | 1.5 | 6.5 | … 80 | 8.92 | 1.95 | 5.85 | -8.965 | 0.006 | 0.059 | 0.023 | 4.9 | 31.9 | 84 81 | 9.17 | 1.30 | 5.85 | -8.289 | 0.007 | 11.678 | 0.013 | 3.9 | 17.9 | 85 82 | 12.62 | 0.87 | 2.50 | -5.479 | 0.009 | 5.706 | 0.017 | 2.2 | 7.4 | 86 **footnotetext: Feature which can not be well represented by a Gaussian curve. RRfootnotetext: Reference feature. Note. — column (1): ID number; columns (2): VLSR at the peak of velocity profile of feature; column (3): the velocity range across the feature $\Delta u$; column (4): distance of maser feature, $r$, from the fitted position of central star; columns (5) and (7): the intensity weighted centroid of each feature ($x,y$); columns (6) and (8): the corresponding uncertainties ($\sigma_{x},\sigma_{y}$); column (9): the peak flux density of each feature $P$; column (10): the integrated flux density of all spots in the feature S; and column (11): the ID numbers of matched features at another epoch. Table 2: 43 GHz SiO maser features around AH Sco observed by VLBA on March 20, 2004. ID | VLSR | $\Delta u$ | r | $x$ | $\sigma_{x}$ | $y$ | $\sigma_{y}$ | P | S | Match ID ---|---|---|---|---|---|---|---|---|---|--- | (km s-1) | (mas) | (mas) | (mas) | (Jy) | (Jy km s-1) | Epoch 2 (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) 1 | -17.23 | 1.09 | 8.14 | -2.861 | 0.003 | -0.509 | 0.014 | 5.8 | 16.0 | 1 2 | -15.14 | 2.39 | 8.33 | -1.650 | 0.004 | 0.385 | 0.009 | 14.1 | 90.0 | 3 3∗ | -14.73 | 0.87 | 7.97 | -1.642 | 0.012 | 0.953 | 0.021 | 2.5 | 3.8 | … 4∗ | -14.60 | 0.87 | 8.03 | -1.351 | 0.018 | 1.243 | 0.011 | 1.3 | 3.4 | … 5∗ | -14.59 | 0.65 | 8.90 | -1.271 | 0.010 | -0.055 | 0.011 | 1.2 | 3.1 | … 6 | -14.50 | 2.39 | 8.39 | -1.351 | 0.003 | 0.642 | 0.009 | 21.7 | 121.4 | 5 7∗ | -13.74 | 0.65 | 9.04 | -0.905 | 0.052 | 0.159 | 0.091 | 2.1 | 5.5 | … 8∗ | -13.31 | 0.65 | 8.37 | -3.046 | 0.005 | -0.946 | 0.017 | 2.0 | 4.4 | 7 9 | -12.41 | 4.13 | 8.85 | -0.813 | 0.006 | 0.591 | 0.015 | 18.9 | 144.1 | 9 10 | -12.17 | 2.82 | 8.44 | -1.777 | 0.004 | 0.080 | 0.016 | 6.6 | 42.7 | … 11 | -12.14 | 1.52 | 8.56 | -1.139 | 0.016 | 0.640 | 0.012 | 15.4 | 61.6 | 8 12 | -11.07 | 0.87 | 9.09 | -0.930 | 0.010 | 0.054 | 0.018 | 7.9 | 24.6 | … 13 | -10.89 | 1.74 | 9.33 | -0.423 | 0.021 | 0.318 | 0.014 | 11.1 | 53.6 | 11 14 | -10.08 | 1.52 | 8.99 | -1.366 | 0.006 | -0.296 | 0.022 | 5.8 | 20.6 | 12 15∗ | -9.88 | 0.87 | 9.80 | -0.906 | 0.017 | -0.988 | 0.021 | 1.4 | 3.6 | … 16 | -9.72 | 3.48 | 9.27 | -1.002 | 0.005 | -0.301 | 0.010 | 14.7 | 109.9 | 13 17 | -9.50 | 1.30 | 8.36 | -15.994 | 0.010 | 8.099 | 0.029 | 3.6 | 18.3 | … 18R | -9.36 | 3.04 | 9.86 | 0.000 | 0.005 | 0.000 | 0.008 | 32.0 | 218.4 | 15 19 | -9.16 | 1.52 | 9.54 | -0.015 | 0.007 | 0.519 | 0.017 | 1.9 | 7.4 | 16 20 | -8.98 | 1.30 | 8.97 | -15.241 | 0.007 | 11.070 | 0.021 | 5.5 | 21.6 | 17 21 | -8.83 | 2.61 | 9.35 | -15.673 | 0.007 | 11.118 | 0.029 | 7.5 | 38.1 | 19 22∗ | -8.55 | 0.65 | 10.21 | 0.354 | 0.006 | -0.109 | 0.026 | 5.9 | 11.4 | 20 23 | -8.11 | 1.09 | 11.85 | -16.001 | 0.006 | 14.542 | 0.021 | 2.6 | 11.5 | 21 24 | -8.09 | 1.52 | 8.60 | -14.867 | 0.006 | 10.958 | 0.016 | 8.8 | 36.0 | 25 25 | -8.08 | 1.52 | 9.75 | -12.873 | 0.009 | 14.265 | 0.026 | 3.1 | 15.2 | 24 26 | -8.07 | 1.96 | 9.72 | 0.428 | 0.012 | 0.887 | 0.027 | 6.7 | 37.0 | 26 27 | -8.00 | 2.17 | 8.39 | -15.780 | 0.009 | 8.829 | 0.017 | 9.6 | 49.4 | 22 28 | -7.75 | 1.96 | 6.78 | -5.908 | 0.005 | 12.333 | 0.017 | 4.2 | 22.0 | … 29 | -7.41 | 2.17 | 7.81 | -9.290 | 0.009 | 13.558 | 0.024 | 3.8 | 21.6 | 27 30 | -7.09 | 1.30 | 11.35 | -15.398 | 0.011 | 14.414 | 0.031 | 3.2 | 17.1 | 30 31∗ | -6.99 | 1.30 | 8.21 | -9.175 | 0.012 | 13.978 | 0.033 | 3.2 | 7.1 | 31 32 | -6.34 | 1.52 | 9.39 | -16.524 | 0.017 | 9.659 | 0.045 | 2.2 | 11.9 | 35 33 | -6.31 | 1.52 | 9.17 | -12.445 | 0.012 | 13.846 | 0.027 | 2.4 | 12.8 | … 34 | -5.74 | 1.30 | 8.74 | -8.193 | 0.002 | 14.598 | 0.007 | 18.7 | 55.1 | 36 35∗ | -5.51 | 0.65 | 8.24 | -8.555 | 0.022 | 14.083 | 0.071 | 1.8 | 4.8 | … 36 | -4.62 | 1.74 | 9.12 | -16.734 | 0.004 | 8.241 | 0.011 | 6.2 | 33.0 | 37 37 | -4.30 | 1.74 | 8.76 | -7.981 | 0.010 | 14.622 | 0.009 | 12.4 | 69.8 | 38 38 | -4.08 | 1.52 | 9.47 | -7.822 | 0.014 | -3.618 | 0.048 | 2.1 | 11.2 | … 39∗ | -3.72 | 0.65 | 9.58 | -17.347 | 0.037 | 4.073 | 0.048 | 2.8 | 7.5 | 39 40∗ | -3.30 | 0.65 | 10.93 | -13.892 | 0.019 | 15.022 | 0.059 | 1.2 | 3.0 | … 41 | -3.16 | 2.61 | 9.03 | -7.811 | 0.009 | 14.892 | 0.019 | 9.4 | 62.0 | 41 42 | -3.09 | 2.39 | 9.26 | -16.914 | 0.011 | 8.125 | 0.023 | 3.9 | 26.3 | 42 43 | -3.05 | 2.17 | 8.75 | -16.624 | 0.011 | 4.791 | 0.021 | 9.0 | 60.1 | 40 44∗ | -2.67 | 0.65 | 8.33 | -7.722 | 0.033 | 14.190 | 0.027 | 1.2 | 2.9 | … 45 | -2.40 | 2.61 | 9.14 | -16.370 | 0.008 | 9.381 | 0.023 | 13.2 | 95.3 | 45 46∗ | -2.40 | 0.87 | 11.11 | -14.067 | 0.013 | 15.123 | 0.049 | 1.4 | 3.6 | 44 47∗ | -2.36 | 1.09 | 8.42 | -16.199 | 0.033 | 4.248 | 0.097 | 4.6 | 14.1 | 43 48 | -1.83 | 1.96 | 8.08 | -15.931 | 0.013 | 4.670 | 0.034 | 5.2 | 38.6 | 46 49∗ | -1.61 | 0.65 | 8.63 | -16.528 | 0.012 | 5.067 | 0.036 | 2.2 | 6.0 | … 50∗ | -1.16 | 0.65 | 8.27 | -16.157 | 0.011 | 5.014 | 0.022 | 2.0 | 5.1 | 47 51 | -0.54 | 1.30 | 8.48 | -15.780 | 0.019 | 9.080 | 0.040 | 1.7 | 12.6 | … 52 | -0.47 | 1.74 | 9.27 | -11.556 | 0.008 | -2.679 | 0.026 | 3.0 | 18.7 | 50 53 | -0.33 | 0.87 | 9.83 | 1.767 | 0.033 | 7.487 | 0.068 | 2.2 | 6.2 | 51 54∗ | 0.39 | 0.65 | 7.12 | -14.744 | 0.034 | 3.762 | 0.065 | 1.8 | 4.1 | 52 55 | 1.49 | 1.30 | 8.52 | -10.627 | 0.009 | -2.226 | 0.030 | 3.6 | 14.6 | 56 56 | 1.49 | 1.30 | 10.65 | 1.969 | 0.011 | 1.927 | 0.019 | 2.7 | 12.1 | 55 57 | 1.55 | 1.96 | 8.71 | -11.052 | 0.013 | -2.276 | 0.033 | 3.4 | 24.7 | 53 58 | 1.58 | 1.30 | 8.10 | -11.766 | 0.013 | 12.997 | 0.035 | 3.7 | 13.7 | 57 59∗ | 1.70 | 0.65 | 9.30 | -1.642 | 0.023 | 12.725 | 0.052 | 1.3 | 3.1 | … 60 | 1.85 | 0.87 | 8.33 | -12.225 | 0.033 | 13.001 | 0.036 | 5.3 | 9.5 | … 61 | 2.30 | 1.95 | 7.55 | -11.316 | 0.010 | 12.614 | 0.023 | 8.0 | 52.1 | 58 62 | 2.85 | 1.74 | 8.01 | -12.161 | 0.008 | 12.666 | 0.029 | 2.5 | 13.7 | … 63 | 3.07 | 1.09 | 8.78 | 0.646 | 0.006 | 3.962 | 0.020 | 2.3 | 8.7 | … 64 | 3.19 | 1.74 | 7.03 | -11.131 | 0.017 | 12.125 | 0.029 | 10.8 | 34.3 | 61 65 | 3.35 | 1.30 | 7.68 | -2.163 | 0.020 | 0.787 | 0.047 | 1.3 | 6.8 | 60 66 | 3.50 | 1.52 | 6.85 | -10.396 | 0.011 | 12.258 | 0.025 | 5.1 | 23.6 | 62 67 | 4.13 | 1.95 | 7.66 | -12.283 | 0.017 | 12.167 | 0.028 | 2.9 | 12.7 | … 68 | 4.17 | 0.87 | 7.46 | -10.183 | 0.011 | -1.255 | 0.053 | 1.7 | 5.5 | 64 69 | 4.27 | 1.09 | 7.42 | -3.501 | 0.016 | -0.097 | 0.053 | 1.4 | 5.1 | … 70 | 4.29 | 1.74 | 8.55 | 0.434 | 0.010 | 4.042 | 0.026 | 2.5 | 14.7 | 66 71 | 4.78 | 1.52 | 7.37 | -12.552 | 0.013 | 11.604 | 0.020 | 8.0 | 35.3 | 67 72 | 5.04 | 1.74 | 6.40 | -1.921 | 0.007 | 8.073 | 0.026 | 2.5 | 13.7 | … 73∗ | 5.44 | 0.87 | 6.93 | -3.062 | 0.011 | 0.916 | 0.038 | 1.2 | 3.6 | … 74 | 5.56 | 1.30 | 5.79 | -13.234 | 0.012 | 3.534 | 0.023 | 2.9 | 12.2 | 72 75 | 5.67 | 1.52 | 5.94 | -10.333 | 0.004 | 11.302 | 0.014 | 5.5 | 22.0 | 70 76∗ | 5.72 | 0.87 | 6.84 | -4.629 | 0.011 | -0.132 | 0.038 | 1.5 | 5.7 | 71 77 | 6.06 | 0.87 | 7.13 | -3.665 | 0.008 | 0.137 | 0.026 | 1.7 | 5.4 | … 78 | 7.05 | 1.74 | 6.46 | -8.987 | 0.013 | -0.517 | 0.027 | 5.3 | 25.7 | 74 79 | 7.23 | 1.30 | 6.08 | -7.467 | 0.014 | 11.923 | 0.042 | 1.4 | 6.3 | 73 80 | 7.51 | 1.09 | 7.87 | -0.253 | 0.016 | 4.125 | 0.048 | 1.9 | 6.6 | … 81 | 7.91 | 1.74 | 6.48 | -9.070 | 0.010 | -0.522 | 0.033 | 4.7 | 26.1 | … 82 | 7.92 | 2.39 | 5.13 | -12.679 | 0.008 | 3.906 | 0.022 | 2.8 | 19.6 | 76 83 | 8.09 | 2.39 | 6.21 | -8.201 | 0.013 | 12.065 | 0.036 | 3.7 | 25.2 | 78 84 | 9.15 | 1.96 | 5.89 | -8.943 | 0.006 | 0.052 | 0.020 | 5.1 | 31.3 | 80 85 | 9.31 | 1.30 | 5.73 | -8.309 | 0.008 | 11.587 | 0.017 | 3.5 | 16.1 | 81 86 | 12.98 | 1.09 | 2.41 | -5.522 | 0.008 | 5.632 | 0.026 | 2.9 | 10.8 | 82 87 | 12.99 | 1.09 | 4.44 | -6.258 | 0.007 | 9.977 | 0.029 | 1.8 | 6.4 | … **footnotetext: Feature which can not be well represented by a Gaussian curve. RRfootnotetext: Reference feature. Note. — The representations of columns (1)-(11) are the same as in Table 1. Table 3: Sources with the detected contraction of SiO maser shell. Source | stellar phase | Contraction velocity | Reference ---|---|---|--- | $\phi$ | km s-1 | R Aqr | $0.78-1.04$ | 4.2$\pm$0.9 | Boboltz et al. (1997) TX Cam | $0.50-0.65$ | $5-10$ | Diamond & Kemball (2003) VX Sgr | $0.75-0.80$ | 4.1$\pm$0.6 | Chen et al. (2006) AH Sco | $\sim$ 0.55 | 12.7$\pm$0.7 | this work Table 4: Best-fit model for the SiO maser kinematics in AH Sco. $v_{0x}$ | $v_{0y}$ | $v_{0z}$ | $V_{1}$ | $\alpha$ | d | Reduced $\chi^{2}$ ---|---|---|---|---|---|--- (km s-1) | (km s-1) | (km s-1) | (km s-1) | | (kpc) | $3.8\pm 0.3$ | $-0.9\pm 0.4$ | $-5.2\pm 0.4$ | $-14.1\pm 1.4$ | $-0.54\pm 0.16$ | $2.26\pm 0.19$ | 3.58 Figure 1: Top: Comparison of total power (solid line) to cross power (open circle) of 43 GHz v=1, J=1–0 SiO maser emission toward AH Sco obtained on (a) March 8, 2004 and (b) March 20, 2004. Bottom: The corresponding fraction power (cross/total) detected by the high-resolution VLBA observations. Figure 2: VLBI images of 43 GHz v=1, J=1–0 SiO maser emission toward AH Sco obtained on (a) March 8, 2004 and (b) March 20, 2004. Each maser feature is represented by a filled circle whose area is proportional to the logarithm of the flux density, and the color indicates its Doppler velocity with respect to the local standard of rest. Its stellar velocity is about $-7$ km s-1. Errors in the positions of the features are smaller than the data points. The ellipse indicates the least-squares fit to the maser distribution for each epoch. The fitted center of ellipse model is marked by the red star. Figure 3: Distance of maser features from the fitted position of central star (in Fig. 2) versus their LOS velocity for epochs A and B. Maser features of the different epochs are denoted by different color symbols whose area is proportional to the logarithm of the flux density. The plot suggests that the higher-velocity maser features lie closer to the central star, which can be well explained by the uniformly expanding thin shell model. Indicated by the downward arrow is the systemic velocity of AH Sco of $-6.8$ km s-1, obtained from the best- fitting thin-shell model (shown by the curve). Figure 4: Distribution of proper motion velocity vectors of the matched maser features at an assumed distance of 2.26 kpc. The length of the vector is proportional to the velocity. The mean proper motion vector has been subtracted from each of the determined proper motion vectors. The color and size of symbols are the same as that shown in Fig. 2. Red star represents the fitted center of ellipse model to maser distribution (see Sect. 3.1).

arxiv-papers

2024-09-04T02:48:54.297085

0803.1740

# Primes in the form $\lfloor{\alpha p+\beta}\rfloor$ Hongze Li lihz@sjtu.edu.cn and Hao Pan haopan79@yahoo.com.cn Department of Mathematics, Shanghai Jiaotong University, Shanghai 200240, People’s Republic of China ###### Abstract. Let $\beta$ be a real number. Then for almost all irrational $\alpha>0$ (in the sense of Lebesgue measure) $\limsup_{x\to\infty}\pi_{\alpha,\beta}^{*}(x)(\log x)^{2}/x\geq 1,$ where $\pi_{\alpha,\beta}^{*}(x)=\\{p\leq x:\,\text{both }p\text{ and }\lfloor{\alpha p+\beta}\rfloor\text{ are primes}\\}.$ ###### 2000 Mathematics Subject Classification: Primary 11N05; Secondary 11N36, 11P32 This work was supported by the National Natural Science Foundation of China (Grant No. 10771135). Recently Jia [4] solved a conjecture of Long and showed that for any irrational number $\alpha>0$, there exist infinitely many primes not in the form $2n+2\lfloor{\alpha n}\rfloor+1$, where $\lfloor{x}\rfloor$ denotes the largest integer not exceeding $x$. Subsequently, in [2] Banks and Shparlinski investigated the distribution of primes in the Beatty sequence $\\{\lfloor{\alpha n+\beta}\rfloor:\,n\geq 1\\}$. Motivated by the binary Goldbach conjecture and the twin primes conjecture, we have the following conjecture: ###### Conjecture 1. Let $\alpha>0$ be an irrational number and $\beta$ be a real number. Then there exist infinitely many primes $p$ such that $\lfloor{\alpha p+\beta}\rfloor$ is also prime. On the other hand, Deshouillers [3] proved that for almost all (in the sense of Lebesgue measure) $\gamma>1$ there exist infinitely many primes $p$ in the form $[n^{\gamma}]$. Furthermore, Balog [1] showed that for almost all $\gamma>1$ $\limsup_{x\to\infty}\frac{|\\{p\leq x:\,\text{both }p\text{ and }\lfloor{p^{\gamma}}\rfloor\text{ are primes}\\}|}{x/(\log x)^{2}}\geq\gamma.$ In this note we shall show that Conjecture 1 holds for almost all $\alpha$. Define $\pi_{\alpha,\beta}^{*}(x)=\\{p\leq x:\,\text{both }p\text{ and }\lfloor{\alpha p+\beta}\rfloor\text{ are primes}\\}.$ ###### Theorem 1. Let $\beta$ be a real number. Then $\limsup_{x\to\infty}\pi_{\alpha,\beta}^{*}(x)(\log x)^{2}/x\geq 1$ (1) for almost all irrational $\alpha>0$. For a set $X\subseteq\mathbb{R}$, let ${\rm mes}(X)$ denote its Lebesgue measure. Without the additional mentions, the constants implied by $\ll$, $\gg$ and $O(\cdot)$ will be always absolute. ###### Lemma 1. Let $I\subseteq[0,1)$ be an interval. Suppose that $b,l>0$. Then ${\rm mes}(\\{\alpha\in(0,b):\,\\{\alpha/l\\}\in I\\}\\})=O(b+l){\rm mes}(I),$ where $\\{x\\}=x-\lfloor{x}\rfloor$. ###### Proof. Without loss of generality, we may assume that $I=(c_{1},c_{2})$ with $0\leq c_{1}<c_{2}\leq 1$. Let $J=\\{\alpha\in(0,b):\,\\{\alpha/l\\}\in I\\}$. Clearly $\displaystyle J\subseteq\bigcup_{0\leq j\leq b/l}((j+c_{1})l,\min\\{b,(j+c_{2})l\\}).$ If $l\leq b$, then ${\rm mes}(J)\leq(\lfloor{b/l}\rfloor+1)(c_{2}-c_{1})l=O(b){\rm mes}(I).$ And if $l>b$, then ${\rm mes}(J)={\rm mes}((c_{1}l,\min\\{b,c_{2}l\\}))\leq(c_{2}-c_{1})l.$ ∎ ###### Lemma 2. Suppose that $b_{2}>b_{1}>0$ and $\beta$ are arbitrarily real numbers. Let $\epsilon>0$ be a small number and $x$ be a sufficiently large (depending on $b_{1}$, $b_{2}$, $\beta$ and $\epsilon$) integer. Then there exists an exceptional set $J_{E}\subseteq(b_{1},b_{2})$ with ${\rm mes}(J_{E})=O(x^{-\epsilon})$ such that for any square-free $d\leq x^{1/3-2\epsilon}$ and irrational $\alpha\in(b_{1},b_{2})\setminus J_{E}$, $|\\{1\leq n\leq x:\,n\lfloor{\alpha n+\beta}\rfloor\equiv 0\ ({\rm mod}\ d)\\}|=\frac{x}{d}\prod_{p\mid d}\bigg{(}2-\frac{1}{p}\bigg{)}+O(x^{1-\epsilon}/d).$ (2) ###### Proof. For an irrational $\alpha\in(b_{1},b_{2})$, let $\mathscr{A}(x;\alpha)=\\{n\lfloor{\alpha n+\beta}\rfloor:\,1\leq n\leq x\\}$ and $\mathscr{A}_{d}(x;\alpha)=\\{a\in\mathscr{A}:\,a\equiv 0\ ({\rm mod}\ d)\\}.$ For a square-free $d$, we have $\displaystyle|\mathscr{A}_{d}(x;\alpha)|=$ $\displaystyle\sum_{s\mid d}|\\{1\leq n\leq x/s:\,\lfloor{\alpha sn+\beta}\rfloor\equiv 0\ ({\rm mod}\ d/s),(n,d/s)=1\\}|$ $\displaystyle=$ $\displaystyle\sum_{s\mid d}\sum_{t\mid d/s}\mu(t)|\\{1\leq n\leq x/s:\,\lfloor{\alpha sn+\beta}\rfloor\equiv 0\ ({\rm mod}\ d/s),t\mid n\\}|$ $\displaystyle=$ $\displaystyle\sum_{s\mid d,t\mid s}\mu(t)|\\{1\leq n\leq x/s:\,\lfloor{\alpha sn+\beta}\rfloor\equiv 0\ ({\rm mod}\ dt/s)\\}|.$ Clearly $\displaystyle\lfloor{\alpha sn+\beta}\rfloor\equiv 0\ ({\rm mod}\ td/s)\Longleftrightarrow\\{\alpha ns^{2}/td+\beta s/td\\}\in[0,s/td).$ Let $\alpha^{\prime}=\alpha s^{2}/td$, $\beta^{\prime}=\beta s/td$, $d^{\prime}=td/s$ and $y=x/s$. Clearly $y\geq x^{2/3+2\epsilon}$ and $d^{\prime}\leq d$. Let $I_{a,q}=\\{\theta\in[0,1):\,|\theta q-a|\leq x^{2\epsilon}/y\\}.$ Suppose that $d^{\prime}x^{2\epsilon}\leq q\leq y/x^{2\epsilon}$ and $1\leq a\leq q$ with $(a,q)=1$. If $\\{\alpha^{\prime}\\}\in I_{a,q}$, $\displaystyle|\\{1\leq n\leq y:\,\\{an/q+\beta^{\prime}\\}\in[1/q,1/d^{\prime}-1/q)\\}|$ $\displaystyle\leq$ $\displaystyle|\\{1\leq n\leq y:\,\\{\alpha^{\prime}n+\beta^{\prime}\\}\in[0,1/d^{\prime})\\}|$ $\displaystyle\leq$ $\displaystyle|\\{1\leq n\leq y:\,\\{an/q+\beta^{\prime}\\}\in[0,1/d^{\prime}+1/q)\cup[1-1/q,1)\\}|.$ Hence $|\\{1\leq n\leq y:\,\\{\alpha^{\prime}n+\beta^{\prime}\\}\in[0,1/d^{\prime})\\}|=y/d^{\prime}+O(q/d^{\prime})+O(y/q).$ Let $\mathcal{I}_{d^{\prime}}=\bigcup_{\begin{subarray}{c}1\leq a\leq q\leq d^{\prime}x^{2\epsilon}\\\ (a,q)=1\end{subarray}}I_{a,q}$ Clearly ${\rm mes}(\mathcal{I}_{d^{\prime}})\leq\sum_{\begin{subarray}{c}1\leq a\leq q\leq d^{\prime}x^{2\epsilon}\\\ (a,q)=1\end{subarray}}{\rm mes}(I_{a,q})\ll\frac{d^{\prime}x^{4\epsilon}}{y}=tdx^{4\epsilon-1}.$ If $\alpha s^{2}/td\not\in\mathcal{I}_{td/s}$ for each $s,t$ with $s\mid d,t\mid s$, then $\displaystyle|\mathscr{A}_{d}(x;\alpha)|=\sum_{s\mid d,t\mid s}\mu(t)x/td(1+O(x^{-2\epsilon}))=\frac{x}{d}\prod_{p\mid d}(2-1/p)+O(x^{1-\epsilon}/d).$ Let $\mathcal{J}_{d}=\\{\alpha\in(0,b):\,\\{\alpha s^{2}/td\\}\in\mathcal{I}_{td/s}\text{ for some }s,t\text{ with }s\mid d,t\mid s\\}.$ Applying Lemma 1, ${\rm mes}(\mathcal{J}_{d})\ll b_{2}\sum_{\begin{subarray}{c}s\mid d,t\mid s\\\ b_{2}\geq td/s^{2}\end{subarray}}{\rm mes}(\mathcal{I}_{td/s})+\frac{td}{s^{2}}\sum_{\begin{subarray}{c}s\mid d,t\mid s\\\ b_{2}<td/s^{2}\end{subarray}}{\rm mes}(\mathcal{I}_{td/s})=O(x^{-1/3+\epsilon}).$ Finally, Let $J_{E}=\bigcup_{d\leq x^{1/3-2\epsilon}}\mathcal{J}_{d}.$ Clearly we have ${\rm mes}(J_{E})=O(x^{-\epsilon})$. ∎ ###### Lemma 3. Suppose that $b_{2}>b_{1}>0$, $\epsilon>0$ and $\beta$ are arbitrarily real numbers. Then there exists an exceptional set $J_{E}\subseteq(b_{1},b_{2})$ with ${\rm mes}(J_{E})=O(x^{-\epsilon})$ such that for any irrational $\alpha\in(b_{1},b_{2})\setminus J_{E}$, $|\\{1\leq p\leq x:\,\text{both }p\text{ and }\lfloor{\alpha p+\beta}\rfloor\text{ are primes}\\}|\ll\frac{x}{(\log x)^{2}}$ (3) for sufficiently large (depending on $b_{1}$, $b_{2}$, $\beta$ and $\epsilon$) $x$. ###### Proof. Let $z=x^{1/8}$. Define $P(z)=\prod_{\begin{subarray}{c}p<z\\\ p\text{ prime}\end{subarray}}p$ and $\mathcal{S}(A,z)=\\{a\in A;\,(a,P(z))=1\\}.$ Let $\mathscr{A}(\alpha)=\\{n\lfloor{\alpha n+\beta}\rfloor:\,1\leq n\leq x\\}$. Clearly $\\{p\lfloor{\alpha p+\beta}\rfloor:\,z+\alpha^{-1}(z+1-\beta)\leq p\leq x,\ \text{both }p\text{ and }\lfloor{\alpha p+\beta}\rfloor\text{ are primes}\\}$ is a subset of $\mathcal{S}(\mathscr{A}(\alpha),z)$. Furthermore, by Lemma 2, we know that there exists a set $J_{E}\subseteq(b_{1},b_{2})$ with ${\rm mes}(J_{E})=O(x^{-\epsilon})$ such that for any square-free $1\leq d\leq x^{1/3-2\epsilon}$ and irrational $\alpha\in(b_{1},b_{2})\setminus J_{E}$, $|\mathscr{A}_{d}(\alpha)|=\frac{x}{d}\prod_{p\mid d}\bigg{(}2-\frac{1}{p}\bigg{)}+O(x^{1-\epsilon}/d),$ where $\mathscr{A}_{d}(\alpha)=\\{y\in\mathscr{A}(\alpha):\,d\mid y\\}$. Let $g(m)$ be the completely multiplicative function such that $g(p)=2/p-1/p^{2}$ for each prime $p$. Define $G(z)=\sum_{\begin{subarray}{c}m<z\\\ m|P(z)\end{subarray}}g(m)$. By Selberg’s sieve method, $|\mathcal{S}(\mathscr{A}(\alpha),z)|\leq\frac{|\mathscr{A}(\alpha)|}{G(z)}+O(\sum_{\begin{subarray}{c}d<z^{2}\\\ d\text{ square-free}\end{subarray}}3^{\omega(d)}x^{1-\epsilon}/d),$ where $\omega(d)$ denotes the number of distinct prime divisors of $d$. Since $3^{\omega(d)}\ll d^{\epsilon}$, $\sum_{\begin{subarray}{c}d<z^{2}\end{subarray}}\frac{3^{\omega(d)}}{d}\ll z^{2\epsilon}.$ So it suffices to show $G(z)\gg(\log z)^{2}$. By Theorem 7.14 in [5], we know $G(z)=\sum_{\begin{subarray}{c}m<z\\\ m\mid P(z)\end{subarray}}g(m)\gg\prod_{p<z}(1-g(p))^{-1}=\prod_{p<z}(1-2/p+1/p^{2})^{-1}\gg(\log z)^{2}.$ ∎ ###### Proof of Theorem 1. Suppose that $b_{2}>b_{1}>0$. Let $\mathscr{F}=\\{\alpha\in(b_{1},b_{2}):\,\limsup_{x\to\infty}\pi_{\alpha,\beta}^{*}(x)(\log x)^{2}/x<1\\}$ and $\mathscr{F}_{n}=\\{\alpha\in(b_{1},b_{2}):\,\limsup_{x\to\infty}\pi_{\alpha,\beta}^{*}(x)(\log x)^{2}/x\leq 1-1/n\\}.$ Clearly $\mathscr{F}=\bigcup_{n>1}\mathscr{F}_{n}$. So it suffices to show that ${\rm mes}(\mathscr{F}_{n})=0$ for every $n>1$. (The measurability of $\mathscr{F}_{n}$ will be proven later.) Assume on the contrary that there exists $n>1$ such that ${\rm mes}(\mathscr{F}_{n})>0$. Let $I=(c_{1},c_{2})$ be an arbitrary sub-interval of $(b_{1},b_{2})$. Clearly $\displaystyle\int_{c_{1}}^{c_{2}}\pi_{\alpha,\beta}^{*}(x)d\alpha=$ $\displaystyle\int_{c_{1}}^{c_{2}}\bigg{(}\sum_{\begin{subarray}{c}p\leq x\\\ p\text{ prime}\end{subarray}}\sum_{\begin{subarray}{c}\alpha p+\beta-1<q\leq\alpha p+\beta\\\ q\text{ prime}\end{subarray}}1\bigg{)}d\alpha$ $\displaystyle=$ $\displaystyle\sum_{\begin{subarray}{c}p\leq x\\\ p\text{ prime}\end{subarray}}\sum_{\begin{subarray}{c}c_{1}p+\beta-1<q\leq c_{2}p+\beta\\\ q\text{ prime}\end{subarray}}{\rm mes}([(q-\beta)/p,(q+1-\beta)/p)\cap[c_{1},c_{2}])$ $\displaystyle\geq$ $\displaystyle\sum_{\begin{subarray}{c}p\leq x\\\ p\text{ prime}\end{subarray}}\sum_{\begin{subarray}{c}c_{1}p+\beta<q\leq c_{2}p+\beta-1\\\ q\text{ prime}\end{subarray}}\frac{1}{p}$ $\displaystyle=$ $\displaystyle(c_{2}-c_{1})\sum_{\begin{subarray}{c}p\leq x\\\ p\text{ prime}\end{subarray}}\frac{1}{\log p}\bigg{(}1+O\bigg{(}\frac{1}{\log(c_{1}p)}\bigg{)}\bigg{)}$ $\displaystyle\geq$ $\displaystyle(c_{2}-c_{1})\frac{x}{(\log x)^{2}}\bigg{(}1+O\bigg{(}\frac{1}{\log x}\bigg{)}\bigg{)},$ (4) provided that $x$ is sufficiently large (depending on $b_{1}$ and $b_{2}$). Suppose that $C>1$ is the implied constant in Lemma 3. Let $\mathscr{L}_{I}=\mathscr{F}_{n}\cap I$ and $\mathscr{L}_{I,\delta}(x)=\\{\alpha\in I:\,\pi_{\alpha,\beta}^{*}(x)\leq(1-\delta)x/(\log x)^{2}\\}.$ For any two primes $p$ and $q$, clearly $J_{p,q}:=\\{\alpha\in I:\,\lfloor{\alpha p+\beta}\rfloor=q\\}$ is an interval or empty set. Hence $\mathscr{L}_{I,\delta}(x)=I\setminus\bigg{(}\bigcup_{\begin{subarray}{c}k>(1-\delta)x/(\log x)^{2}\\\ p_{1},\ldots,p_{k}\leq x\text{ are distinct primes}\\\ q_{1},\ldots,q_{k}\text{ are primes}\end{subarray}}\bigcap_{j=1}^{k}J_{p_{j},q_{j}}\bigg{)}$ is measurable in the sense of Lebesgue measure. Let $\epsilon>0$ be a very small number. By Lemma 3, $\int_{c_{1}}^{c_{2}}\pi_{\alpha,\beta}^{*}(x)d\alpha\leq O(x^{1-\epsilon})+$ $+{\rm mes}(\mathscr{L}_{I,\delta}(x))\frac{(1-\delta)x}{(\log x)^{2}}+(c_{2}-c_{1}-{\rm mes}(\mathscr{L}_{I,\delta}(x)))\frac{Cx}{(\log x)^{2}}$ (5) provided that $x$ is sufficiently large. Combining (Proof of Theorem 1.) and (5), we have ${\rm mes}(\mathscr{L}_{I,\delta}(x))\leq\frac{C-1}{C-1+\delta/2}{\rm mes}(I).$ (6) We claim that $\mathscr{L}_{I}=\bigcap_{m>n}\bigcup_{y\geq 1}\bigcap_{x\geq y}\mathscr{L}_{I,1/n-1/m}(x).$ (7) In fact, for any $m>n$, if $\limsup_{x\to\infty}\frac{\pi_{\alpha,\beta}^{*}(x)}{x/(\log x)^{2}}<1-\frac{1}{n}+\frac{1}{m},$ then there exists $y_{0}$ such that for any $x\geq y_{0}$ $\pi_{\alpha,\beta}^{*}(x)\leq\bigg{(}1-\frac{1}{n}+\frac{1}{m}\bigg{)}\frac{x}{(\log x)^{2}}.$ On the other hand, if $\alpha\in\bigcup_{y}\bigcap_{x\geq y}\mathscr{L}_{I,1/n-1/m}(x)$, clearly we have $\limsup_{x\to\infty}\frac{\pi_{\alpha,\beta}^{*}(x)}{x/(\log x)^{2}}\leq 1-\frac{1}{n}+\frac{1}{m}.$ By (6) and (7), we get ${\rm mes}(\mathscr{L}_{I})\leq\limsup_{x\to\infty}\mathscr{L}_{I,2/3n}(x)\leq\frac{C-1}{C-1+1/3n}{\rm mes}(I).$ Since ${\rm mes}(\mathscr{F}_{n})>0$, there exist open intervals $I_{1},I_{2},\ldots\subseteq(b_{1},b_{2})$ such that $\mathscr{F}_{n}\subseteq\bigcup_{k=1}^{\infty}I_{k}$ and $\sum_{k=1}^{\infty}{\rm mes}(I_{k})\leq\frac{C-1+1/4n}{C-1}{\rm mes}(\mathscr{F}_{n}).$ But by (6), ${\rm mes}(\mathscr{F}_{n})=\sum_{k=1}^{\infty}{\rm mes}(\mathscr{L}_{I_{k}})\leq\frac{C-1}{C-1+1/n}\sum_{k=1}^{\infty}{\rm mes}(I_{k})\leq\frac{C-1+1/4n}{C-1+1/3n}{\rm mes}(\mathscr{F}_{n}).$ This evidently leads to a contradiction. ∎ ###### Remark. In [6] and [8], Harman proved that for almost all real $\alpha>0$ there are infinitely many pairs of $(p,q)$ satisfying $|\alpha p-q|<\psi(p),\qquad p,q\text{ are primes},$ provided that $\psi$ is a non-increasing positive function and $\sum_{\begin{subarray}{c}2\leqslant p\leqslant\infty\\\ p\text{ primes}\end{subarray}}\frac{\psi(p)}{\log p}$ (8) diverges. (In fact, in [8] Harman established a quantitative version of the above result, on condition that $\psi(n)\in(0,1/2)$ for each $n$.) As an immediate consequence, for almost all $\alpha>0$, there exists infinitely many pair of primes $(p,q)$ such that $[\\![\alpha p]\\!]=q$, where $[\\![x]\\!]$ is the nearest integer to $x$. For more related results, the readers may refer to [7, Chapter 6]. ###### Acknowledgment. We are grateful to Professor Glyn Harman for his very helpful discussions and kindly sending us the copies of the references [6] and [8]. ## References * [1] A. Balog, On a variant of the Piatetski-Shapiro prime number problem. Publ. Math. Orsay, 1989, 3-11. * [2] W. D. Banks and I. E. Shparlinski, Prime numbers with Beatty sequences, preprint, arXiv:0708.1015. * [3] J. Deshouillers, Nombres premiers de la forme $[n^{c}]$. C. R. Acad. Sci., Paris, Ser. A, 282 (1976), 131-133. * [4] C.-H. Jia, On a conjecture of Yiming Long. Acta Arith., 122 (2006), 57-61. * [5] C.-D. Pan and C.-B. Pan, Goldbach Conjecture. Science Press, Beijing, 1992. * [6] G. Harman, Metric diophantine approximation with two restricted variables. III: Two prime numbers. J. Number Theory, 29 (1988), 364-375. * [7] G. Harman, Metric number theory. London Mathematical Society Monographs. New Series 18. Oxford, Clarendon Press. 1998. * [8] G. Harman, Variants of the second Borel-Cantelli lemma and their applications in metric number theory. Bambah, R. P. (ed.) et al., Number theory. Basel, Birkhäuser. Trends in Mathematics. 121-140, 2000.

arxiv-papers

2024-09-04T02:48:54.305658

0803.1749

# An isomorphism between the completion of an algebra and its Caratheodory Extension Jun Tanaka University of California, Riverside, USA juntanaka@math.ucr.edu, yonigeninnin@gmail.com, junextension@hotmail.com (Date: January, 10, 2008) ###### Abstract. Let $\Omega$ denote an algebra of sets and $\mu$ a $\sigma$-finite measure. We then prove that the completion of $\Omega$ under the pseudometric $d(A,B)$ = $\mu^{\ast}(A\triangle B)$ is $\sigma$-algebra isomorphic and isometric to the Caratheodory Extension of $\Omega$ under the equivalence relation $\sim$. ###### Key words and phrases: Measure Theory, Caratheodory Extension Theorem, Metric ###### 2000 Mathematics Subject Classification: Primary: 28A12, 28B20 ## 1\. Introduction This paper shows a new result by combining two papers authored by P.F. Mcloughlin and myself ([4], [5]). Let $\mu$ be a $\sigma$-finite measure and let $\Omega$ denote an algebra of sets; i.e., $\Omega$ is closed under unions and complements. Let ($X$, $\Omega$, $\mu$) denote a measure space where $\mu(X)$ is finite from the $\sigma$-finite property. Let $\mu^{\ast}$ denote the outer measure defined by $\mu^{\ast}(A)$ = $\inf\\{\sum\mu(A_{i})\mid E\subseteq\cup A_{i}$ and $A_{i}\in\Omega$ for all i $\geq$ 1$\\}$, for any $A\in\mathbf{P}(X)$ where $\mathbf{P}(X)$ is the power set of $X$. Clearly, $d(A,B)$ = $\mu^{\ast}(A\triangle B)$ is a pseudometric, where $\triangle$ is the symmetric difference of sets. In addition, $d$ is a metric on $\mathbf{P}(X)_{\diagup_{\sim}}$, where A $\sim$ B iff $\mu^{\ast}(A\triangle B)$ =0. From [1], pg 292, $\mu^{\ast}|_{\Omega}=\mu$. In [5], we defined a $\mu$-Cauchy sequence $\\{B_{n}\\}$, $B_{n}\in\Omega$, if $\lim\mu(B_{n}\triangle B_{m})$ $\rightarrow$ 0 as $n,m\rightarrow\infty$. Let $\widetilde{\mathbf{S}}$ = $\\{S\in\mathbf{P}(X)\ |$ $\exists\ \mu$-Cauchy sequence $\\{B_{n}\\}$ s.t. $\lim\mu^{\ast}(B_{n}\triangle S)=0\\}$. In the first joint paper [5], we proved that $\widetilde{\mathbf{S}}$ is a $\sigma$-algebra where, for any $\mu$-Cauchy sequence $\\{B_{n}\\}$ such that $\lim\mu^{\ast}(B_{n}\triangle S)=0$, the measure $\widetilde{\mu}(S)$ on $\widetilde{\mathbf{S}}$ is defined as $\widetilde{\mu}(S)$ = $\lim\mu(B_{n})$. In addition, we proved that $\widetilde{\mu}$ is a countably additive measure on $\widetilde{\mathbf{S}}$. Thus, ($\widetilde{\mu}$, $\widetilde{\mathbf{S}}$) is a measure space. We showed that the Caratheodory Extension of $\Omega$ can be expressed as the set of limit points of $\mu$-Cauchy sequences under the pseudometric $d(A,B)$ = $\mu^{\ast}(A\triangle B)$. Moreover, when the measure is a sigma finite measure, we obtained an equivalent expression of the Caratheodory Extension, $\\{S\in\mathbf{P}(X)$ $|$ $\exists\ \mu$-Cauchy sequence $\\{B_{n}\\}$ s.t. $\lim\mu^{\ast}(B_{n}\triangle S)=0\\}$. Theorem 2 in [5] shows that $E$ is a measurable set iff $E$ is in $\widetilde{\mathbf{S}}$. Thus, the measure space ($\widetilde{\mu}$, $\widetilde{\mathbf{S}}$) agrees with the Caratheodory Extension when $\mu$ is a finite measure. Moreover, it shows that measurable sets are exactly limit points of $\mu$-Cauchy sequences. The $\sigma$-finite case follows from the finite case. From the second joint paper [4], we denoted by ($\overline{d}$, $\overline{\Omega}$) the completion of ($d$, $\Omega_{\diagup_{\sim}}$). Let $\mathcal{S}$ be the set of all $\mu$-Cauchy sequences in ($d$, $\Omega_{\diagup_{\sim}}$). By the completion procedures, we know $\\{B^{\alpha}_{n}\\}$ $\sim$ $\\{B^{\gamma}_{n}\\}$ iff $\lim d(B^{\alpha}_{n},B^{\gamma}_{n})$ = 0 defines an equivalence relation on $\mathcal{S}$. Moreover, $\overline{\Omega}$ = $\mathcal{S}_{\diagup_{\sim}}$ and $\overline{d}(\overline{\\{B^{\alpha}_{n}\\}},\overline{\\{B^{\gamma}_{n}\\}})$ = $\lim d(\\{B^{\alpha}_{n}\\},\\{B^{\gamma}_{n}\\})$, where $\overline{\\{B^{\alpha}_{n}\\}}$ is the class of $\\{B^{\alpha}_{n}\\}$. Let $E_{\alpha}$ = $\overline{\\{B^{\alpha}_{n}\\}}$ and $E_{A}$ = $\overline{\\{A\\}}$ when $A\in\Omega$. Let $\overline{\mu}(E_{\alpha})$ = $\overline{d}(E_{\alpha},E_{\emptyset})$ = $\lim d(B^{\alpha}_{n},\emptyset)$ = $\lim\mu(B^{\alpha}_{n})$. Note that in [4] $d(A,B)$ := $\mu(A\triangle B)$, whereas in [5] $d(A,B)$ := $\mu^{\ast}(A\triangle B)$; the completion of $\Omega$ will remain the same due to the property $\mu^{\ast}|_{\Omega}=\mu$. Note that $\overline{\mu}(E_{A})$ = $\mu(A)$ when $A\in\Omega$. In [4], we defined set-theoretic notations for unions, intersections, and complements on $\overline{\Omega}$ as follows: $\bigcup$ : $\overline{\Omega}\times\overline{\Omega}\rightarrow\overline{\Omega}$ where $\bigcup(E_{\alpha}\times E_{\gamma})$ = $E_{\alpha}\bigcup E_{\gamma}$ = $\overline{\\{B^{\alpha}_{n}\cup B^{\gamma}_{n}\\}}$; similarly for intersections on $\overline{\Omega}$. $\cdot^{\textbf{C}}$: $\overline{\Omega}\rightarrow\overline{\Omega}$ where $(\overline{\\{B^{\alpha}_{n}\\}})^{\textbf{C}}$ = $\overline{\\{(B^{\alpha}_{n})^{\textbf{C}}\\}}$ and we showed the set theoretic notations are well defined on $\overline{\Omega}$ in [4]. We showed the set theoretic notations are well defined on $\overline{\Omega}$ in [4]. Note that $E_{\alpha}\bigcap E_{\gamma}$ = $E_{\emptyset}$ iff $\overline{\mu}(E_{\alpha}\bigcap E_{\gamma})$ = 0 iff $\lim\mu(B^{\alpha}_{n}\cap B^{\gamma}_{n})$ = 0. We say $E_{\alpha}$ and $E_{\gamma}$ are disjoint iff $E_{\alpha}\bigcap E_{\gamma}$ = $E_{\emptyset}$. Thus, if $E_{\alpha_{1}}$ and $E_{\alpha_{2}}$ are disjoint, then $\overline{\mu}(E_{\alpha_{1}}\bigcup E_{\alpha_{2}})$ = $\overline{\mu}(E_{\alpha_{1}})$ \+ $\overline{\mu}(E_{\alpha_{2}})$ . As for the infinite union on $\overline{\Omega}$; if $E_{\alpha_{i}}$ $\in\overline{\Omega}$ for i $\geq$ 1, there exists a unique E := $\bigcup_{i=1}^{\infty}E_{\alpha_{i}}$ in $\overline{\Omega}$ such that $\bigcup_{i=1}^{n}E_{\alpha_{i}}\subset$ E for all n, and $\lim\overline{\mu}(E\bigcap(\bigcup_{i=1}^{n}E_{\alpha_{i}})^{\textbf{C}})$ = 0. In addition, we showed that for any $\mu$-Cauchy sequence $\\{B_{n}\\}$, there exists a $f(n)$ $>$ n such that $\lim\mu^{\ast}(B_{n}\triangle\overline{\lim}B_{f(n)})=0$. In this paper, I define a $\sigma$-algebra isomorphism between two $\sigma$-algebras, and define a map $F:$ $\overline{\Omega}\rightarrow\mathbf{P}(X)$ given by $F(\overline{\\{B_{n}\\}})=\overline{\lim}B_{f(n)}$ where $f(n)$ is defined as above. We will show that $F$ is an isometry and a $\sigma$-algebra isomorphism between the completion $\overline{\Omega}$ and the Caratheodory Extension of $\Omega$ under the equivalence relation $\sim$ defined as A $\sim$ B iff $\mu^{\ast}(A\triangle B)$ =0. ## 2\. Main Result ###### Definition 1. For A, B in $\mathbf{P}(X)$, A = B a.e. iff $\mu^{\ast}(A\triangle B)=0.$ ###### Definition 2. Define a map $F:$ $\overline{\Omega}\rightarrow\mathbf{P}(X)$ given by $F(\overline{\\{B_{n}\\}})=\overline{\lim}B_{f(n)}$ where $\lim\mu^{\ast}(B_{n}\triangle\overline{\lim}B_{f(n)})=0$. Note that such $f(n)$ always exists by Lemma 20 in [4]. ###### Remark 1. $F$ is a map into $\widetilde{\mathbf{S}}$ by the definition of $\widetilde{\mathbf{S}}$. ###### Lemma 1. $F$ is well defined. ###### Proof. Suppose that $\overline{\\{A_{n}\\}}=\overline{\\{B_{n}\\}}$. There exist f(n) and g(n) such that $\lim\mu^{\ast}(A_{n}\triangle\overline{\lim}A_{f(n)})=0$ and $\lim\mu^{\ast}(B_{n}\triangle\overline{\lim}B_{g(n)})=0$ by Lemma 20 in [4]. $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})\leq$ $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle A_{f(n)})+\mu^{\ast}(A_{f(n)}\triangle B_{g(n)})+\mu^{\ast}(B_{g(n)}\triangle\overline{\lim}B_{g(n)})$ by the triangle inequality. By taking the limit on both sides, $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})=0$. Thus, $\overline{\lim}A_{f(n)}=\overline{\lim}B_{g(n)}$ a.e.. Therefore, F is well-defined. ∎ ###### Theorem 1. F is an isometry between $\overline{\Omega}$ and $\widetilde{\mathbf{S}}_{\diagup_{\sim}}$. ###### Proof. First, we show F is onto $\widetilde{\mathbf{S}}$. Let $X\in\widetilde{\mathbf{S}}$. Then there exists a $\mu$-Cauchy sequence $\\{B_{n}\\}$ such that $\lim\mu^{\ast}(B_{n}\triangle X)=0$. Then there exist f(n) such that $\lim\mu^{\ast}(B_{n}\triangle\overline{\lim}B_{f(n)})=0$. Thus $F(\overline{\\{B_{n}\\}})=\overline{\lim}B_{f(n)}=X$ a.e.. Therefore, F is onto. Second, we will show F preserves the metric. Let $\overline{\\{A_{n}\\}}$, $\overline{\\{B_{n}\\}}$ $\in$ $\overline{\Omega}$. Then we have f(n) and g(n) as before. $\displaystyle\mu(A_{f(n)}\triangle B_{g(n)})=$ $\displaystyle\mu^{\ast}(A_{f(n)}\triangle B_{g(n)})$ $\displaystyle\leq$ $\displaystyle\mu^{\ast}(A_{f(n)}\triangle\overline{\lim}A_{f(n)})+\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})+\mu^{\ast}(\overline{\lim}B_{g(n)}\triangle B_{g(n)})$ By taking the limit on both sides, $\lim\mu(A_{n}\triangle B_{n})=\lim\mu(A_{f(n)}\triangle B_{g(n)})\leq\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)}).$ In addition, $\displaystyle\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})$ $\displaystyle\leq\mu^{\ast}(A_{f(n)}\triangle\overline{\lim}A_{f(n)})+\mu(A_{f(n)}\triangle B_{g(n)})+\mu^{\ast}(\overline{\lim}B_{g(n)}\triangle B_{g(n)}).$ By taking the limit on both sides, $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})\leq\lim\mu(A_{f(n)}\triangle B_{g(n)}).$ Therefore, $\overline{d}(\overline{\\{A_{n}\\}},\overline{\\{B_{n}\\}})$ = $\lim\mu(A_{n}\triangle B_{n})$ = $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})$ = $\mu^{\ast}(F(\overline{\\{A_{n}\\}})\triangle F(\overline{\\{B_{n}\\}}))$ = $d(F(\overline{\\{A_{n}\\}}),F(\overline{\\{B_{n}\\}}))$. Lastly, we will show that F is one to one. Let $F(\overline{\\{A_{n}\\}}),F(\overline{\\{B_{n}\\}})\in\widetilde{\mathbf{S}}$ such that $F(\overline{\\{A_{n}\\}})=F(\overline{\\{B_{n}\\}})$ a.e.. Then $\overline{\lim}A_{f(n)}=\overline{\lim}B_{g(n)}$ a.e. implies $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})=0$. Then, as in the proof of F being onto, $\lim\mu(A_{n}\triangle B_{n})$ = $\mu^{\ast}(\overline{\lim}A_{f(n)}\triangle\overline{\lim}B_{g(n)})$. Thus $\overline{\\{A_{n}\\}}$ = $\overline{\\{B_{n}\\}}$. Thus, F is one to one. Therefore, F is an isometry between $\overline{\Omega}$ and $\widetilde{\mathbf{S}}_{\diagup_{\sim}}$. ∎ ###### Definition 3. Suppose X and Y are $\sigma$-algebras, and F: X $\rightarrow$ Y is a one to one, onto well defined map. Then F is called a $\sigma$-algebra isomorphism if $\displaystyle F(\cdot\bigcup\cdot)=$ $\displaystyle F(\cdot)\cup F(\cdot),\ \ \ \ \ \ \ F(\bigcup^{\infty}_{i=1}\cdot)$ $\displaystyle=$ $\displaystyle\cup^{\infty}_{i=1}F(\cdot),$ $\displaystyle F(\cdot\bigcap\cdot)=$ $\displaystyle F(\cdot)\cap F(\cdot),\ \ \ \ \ \ \ F(\cdot^{\textbf{C}})$ $\displaystyle=$ $\displaystyle F(\cdot)^{\textbf{C}}.$ ###### Lemma 2. Let $E_{i}$ = $\overline{\\{B^{i}_{n}\\}}$ $\in\overline{\Omega}$ for i $\geq$ 1 and by following Lemma 8 in [5], construct $Y_{L}$ = $\cup_{i=1}^{N_{L}}B^{i}_{K_{L}}$ for each $L$ such that $\mu^{\ast}(\cup_{i=1}^{\infty}S_{i}\triangle\cup_{i=1}^{N_{L}}B^{i}_{K_{L}})\leq\mu^{\ast}(\cup_{i=N_{L}+1}^{\infty}S_{i})+\mu^{\ast}(\cup_{i=1}^{N_{L}}S_{i}\triangle\cup_{i=1}^{N_{L}}B^{i}_{K_{L}})<\frac{1}{L}.$ . Then $\overline{\\{Y_{L}\\}}$ = $\bigcup_{i=1}^{\infty}E_{i}$. ###### Proof. Note that $E_{i}$ = $\overline{\\{B^{i}_{n}\\}}$ = $\overline{\\{B^{i}_{K_{L}}\\}}$. $(\bigcup_{i=1}^{n}E_{i})\bigcap\overline{\\{Y_{L}\\}}$ = $\overline{\\{\cup_{i=1}^{n}B^{i}_{K_{L}}\cap Y_{L}\\}}$ = $\overline{\\{\cup_{i=1}^{n}B^{i}_{K_{L}}\\}}$ = $\bigcup_{i=1}^{n}E_{i}$ for any n. Let $N_{L}>n$. $\displaystyle\mu(\cup_{i=1}^{N_{L}}B^{i}_{K_{L}}\cap(\cup_{i=1}^{n}B^{i}_{K_{L}})^{\textbf{C}})=$ $\displaystyle\mu(\cup_{i=1}^{N_{L}}B^{i}_{K_{L}}\triangle\cup_{i=1}^{n}B^{i}_{K_{L}})=\mu^{\ast}(\cup_{i=1}^{N_{L}}B^{i}_{K_{L}}\triangle\cup_{i=1}^{n}B^{i}_{K_{L}})$ $\displaystyle\leq$ $\displaystyle\mu^{\ast}(\cup_{i=1}^{\infty}S_{i}\triangle\cup_{i=1}^{N_{L}}B^{i}_{K_{L}})+\mu^{\ast}(\cup_{i=1}^{\infty}S_{i}\triangle\cup_{i=1}^{n}B^{i}_{K_{L}}).$ This implies that $\lim\overline{\mu}(\overline{\\{Y_{L}\\}}\bigcap(\bigcup_{i=1}^{n}E_{\alpha_{i}})^{\textbf{C}})$ = 0. Therefore by the uniqueness of $\bigcup_{i=1}^{\infty}E_{i}$, $\overline{\\{Y_{L}\\}}$ = $\bigcup_{i=1}^{\infty}E_{i}$. ∎ ###### Theorem 2. F is a $\sigma$-algebra isomorphism between $\overline{\Omega}$ and $\widetilde{\mathbf{S}}_{\diagup_{\sim}}$. ###### Proof. We already showed that F is a one to one, onto map in Theorem 1. Since, in general, $\overline{\lim}A_{n}\cup B_{n}$ = $\overline{\lim}A_{n}\cup\overline{\lim}B_{n}$, $F(\cdot\bigcup\cdot)$ = $F(\cdot)\cup F(\cdot)$ follows immediately. Let $\overline{\\{B_{n}\\}}\in\overline{\Omega}$. Then, $F(\overline{\\{B_{n}\\}}^{\textbf{C}})=F(\overline{\\{(B_{n})^{\textbf{C}}\\}})=\overline{\lim}(B_{f(n)})^{\textbf{C}}=(\underline{\lim}B_{f(n)})^{\textbf{C}}=(\overline{\lim}B_{f(n)})^{\textbf{C}}a.e..$ Note: by the construction of f(n), $\underline{\lim}B_{f(n)}=\overline{\lim}B_{f(n)}$ a.e. Thus, $F(\cdot^{\textbf{C}})$ = $F(\cdot)^{\textbf{C}}$ in $\widetilde{\mathbf{S}}_{\diagup_{\sim}}$. Similarly, $F(\cdot\bigcap\cdot)$ = $F(\cdot^{\textbf{C}}\bigcup\cdot^{\textbf{C}})^{\textbf{C}}$ = $[F(\cdot^{\textbf{C}})\cup F(\cdot^{\textbf{C}})]^{\textbf{C}}$= $F(\cdot)\cap F(\cdot)$. Let $E_{\alpha_{i}}$ $\in\overline{\Omega}$ for i $\geq$ 1 and $E_{\alpha_{i}}$ = $\overline{\\{B^{\alpha_{i}}_{n}\\}}$. Then for each i, there exists a $S_{i}=\overline{\lim}B^{\alpha_{i}}_{f(n)}\in\widetilde{\mathbf{S}}$ such that $\lim\mu^{\ast}(B^{\alpha_{i}}_{n}\triangle S_{i})=0$. Now suppose we have $\\{Y_{L}\\}$ in the same manner as Lemma 2. By design, $\\{Y_{L}\\}$ converges to $\cup_{i=1}^{\infty}S_{i}$. Then $\overline{\\{Y_{L}\\}}$ = $\bigcup_{i=1}^{\infty}E_{\alpha_{i}}$ by Lemma 2. Now we have $F(\bigcup_{i=1}^{\infty}E_{\alpha_{i}})=F(\bigcup_{i=1}^{\infty}\overline{\\{B^{\alpha_{i}}_{n}\\}})=F(\overline{\\{Y_{L}\\}})=\overline{\lim}Y_{f(L)}.$ Since $\lim\mu^{\ast}(Y_{L}\triangle\overline{\lim}Y_{f(L)})=0$ and $\lim\mu^{\ast}(Y_{L}\triangle\cup^{\infty}_{i=1}S_{i})=0$, we have $\overline{\lim}Y_{f(L)}$ = $\cup^{\infty}_{i=1}S_{i}$ a.e.. In addition, $\cup^{\infty}_{i=1}S_{i}$ = $\cup^{\infty}_{i=1}F(\overline{\\{B^{\alpha_{i}}_{n}\\}})$. Thus, $F(\bigcup_{i=1}^{\infty}E_{\alpha_{i}})=\cup^{\infty}_{i=1}F(E_{\alpha_{i}}).$ Therefore, the claim follows. ∎ ## 3\. Conclusion Theorem 1 and 2 show that the completion of $\Omega$ is isometric and $\sigma$-algebra isomorphic to $\widetilde{\mathbf{S}}_{\diagup_{\sim}}$. Thus the completion of $\Omega$ is isometric and $\sigma$-algebra isomorphic to the Catheordory Extension under the equivalence relation $\sim$ by the conclusion in [5]. ## 4\. Acknowledgement I would like to thank my grandfather Waichi Tanaka for his inspiration and financial assistance and Andrew Aames for encouraging him to progress through the graduate program. With the kind support of both, I have progressed further than I ever thought possible. In addition, I would like to thank my friends Richard Han and Eli Depalma for their editing assistance and Vincent Davis, Aaron Hudson, Mark Tseselsky for representing me and for their professional advice. ## References * [1] 1\. H.L.Royden, Real Analysis Third Edition, Prentice-Hall Inc. 1988. * [2] 2\. N. Dunford and J. T. Schwartz, Liner Operators Part 1 General Theory, Willy Interscience Publication, 1988. * [3] 3\. Walter Rudin, Real and Complex Analysis McGraw-Hill Publishing Co, 1987. * [4] 4\. P.F. Mclaughlin and J. Tanaka, A Relationship Between the Completion of a Metric Space and the Caratheodory Extension, will be submitted soon * [5] 5\. J. Tanaka and P.F. Mclaughlin, A realization of measurable sets as limit points, submitted * [6]

arxiv-papers

2024-09-04T02:48:54.308733

0803.1804

# Bardeen-Stephen flux flow law disobeyed in the high-$T_{c}$ superconductor Bi2Sr2CaCu2O8+δ Á. Pallinger1 B. Sas1 I. Pethes1 K. Vad2 F. I. B.Williams1,3 G. Kriza1,4 1Research Institute for Solid State Physics and Optics, PO Box 49, H-1525 Budapest, Hungary 2Institute of Nuclear Research, PO Box 51, H-4001 Debrecen, Hungary 3CEA-Saclay, Service de Physique de l’Etat Condensé, Comissariat à l’Energie Atomique, Saclay, F-91191 Gif-sur-Yvette, France 4Institute of Physics, Budapest University of Technology and Economics, Budafoki út 8, H-1111 Budapest, Hungary ###### Abstract Pulsed high current experiments in single crystals of the high-$T_{c}$ superconductor Bi2Sr2CaCu2O8+δ in $c$-axis directed magnetic field $H$ reveal that the $ab$-face resistance in the free flux flow regime is a solely logarithmic function of H, devoid of any power law component. Re-analysis of published data confirms this result and leads to empirical analytic forms for the $ab$-plane and $c$-axis resistivities: $\rho_{ab}\propto$ $H^{3/4}$, which does not obey the expected Bardeen-Stephen result for free flux flow, and $\rho_{c}\propto H^{-3/4}\log^{2}H.$ ###### pacs: 74.72.Hs, 74.25.Fy, 74.25.Qt, 74.25.Sv Free flux flow (FFF) resistivity describes how fast the vortices in a type II superconductor move in the direction of an applied force kop . It is a measure of how the momentum of the superfluid is transferred to the host lattice via quasiparticle excitations. The velocity-force relation expressed by the FFF resistivity has to be taken into account in interpreting any vortex transport, be it global or local. Since the primary source of dissipation in a type II superconductor in magnetic field is vortex motion, FFF resistivity is also of great importance for technical applications. A transport current exerts a Lorentz-Magnus force on a vortex and if other forces like vortex-defect interaction (pinning) are negligible (i.e., the vortex motion is “free”), the velocity-force relation can be inferred from the resistivity $\rho_{\mathrm{FFF}}$. The Bardeen-Stephen (BS) law bar states that $\rho_{\mathrm{FFF}}$ is proportional to the density of vortices and therefore to the magnetic field $H$: $\rho_{\mathrm{FFF}}=\gamma\rho_{n}(H/H_{c2})^{\beta},\ \ \beta=1,$ (1) where $\rho_{n}$ is the normal state resistivity, $H_{c2}$ the upper critical field, and $\gamma$ a constant $\approx 1$. This law has been experimentally established par for a number of conventional superconductors. In high-$T_{c}$ materials the quasi-two-dimensional (2d) electronic structure, the nodes of the $d$-wave order parameter, and the structure of the vortex system may potentially influence FFF. The motion of 2d “pancake” vortices along their well conducting $ab$ plane leads to dissipation by flux flow, whereas in the poorly conducting $c$ direction dissipation is governed not by flux flow but by tunneling between weakly coupled $ab$ planes. A quasiclassical calculation Kopnin and Volovik (1997) suggests that the BS law is valid also in the $d$-wave case. Experiments to test the validity of Eq. (1) in high-$T_{c}$ superconductors—and especially in Bi2Sr2CaCu2O8+δ (BSCCO), the model system of this study—are contradictory. Data on low frequency transport in single crystals bs- and thin films thi ; xia as well as microwave and millimeter wave impedance mic are inconsistent with one another, agreeing only that the BS law is not obeyed. The resistivity often resembles a sublinear power law in field. The situation is similar in other high-$T_{c}$ materials with the notable exception of the results of Kunchur et al. kun who find agreement with BS law in thin film resistivity measurements in YBa2Cu3O7. However, the resistance they measure does not saturate, i.e., it increases with current, up to the highest current they use, leading to an uncertainty in the value of $\rho_{\mathrm{FFF}}$. A re-analysis of the differential resistance indicates again a sublinear field dependence of $\rho_{\mathrm{FFF}}$. The main difficulty in measuring the velocity-force relation is to take account of the pinning force about which one has little detail. One way is to model pinning to interpret the surface impedance arising from local vortex motion. Another approach is to create experimental conditions where pinning is irrelevant as occurs in a true (unpinned) vortex liquid. We extend our experiments to the non-ohmic regime by applying sufficiently high current that the pinning force is negligible compared with the Lorentz-Magnus force from the transport current. To this end we have made pulsed high-current transport measurements on single crystal BSCCO with electrode contacts on the face parallel to the well conducting $ab$ planes in a $c$-axis directed magnetic field. The global single crystal resistance measured on the $ab$-face in the ohmic regime and the asymptotic high-current differential resistance in the non-ohmic regime show the same logarithmic magnetic field dependence devoid of any power law. We combine this result with published data from other experiments Busch et al. (1992); Morozov et al. (2000) and set up empirical functional forms for the local resistivities $\rho_{ab}$ and $\rho_{c}$, valid over a broad range of temperature and field in the vortex liquid phase. The most striking and important of our conclusions is that for BSCCO $\beta=3/4$ in Eq. (1). We selected for experiment three single crystals from 3 different batches of Bi2Sr2CaCu2O8+δ with typical dimensions $1\times 0.5\times 0.003$ mm3, the shortest corresponding to the poorly conducting $c$ axis. All were close to optimal doping with a resistance-determined critical temperature $T_{c}\approx 89$ K and transition width about $2$ K in zero field; the diamagnetism in a 1 mT field set in progressively below $T_{c}$ to near $100\%$ at low temperature. Voltage-current ($V$-$I$) response was measured in the usual four-point configuration on an $ab$ face in perpendicular magnetic field with two current contacts across the width near the ends and two point voltage contacts near each edge of the same face. The contacts were made by bonding 25 $\mu$m gold wires with silver epoxy fired at $900$ K in an oxygen atmosphere resulting in current contact resistances of less than 3 $\Omega$. To avoid significant Joule heating, we employed short ($\leq 50$ $\mu$s) current pulses of isosceles triangular shape at $0.2$ to $1$ s intervals. Technical details and the issue of Joule heating are treated in Ref. sas with the conclusion that the temperature change in the area between the voltage contacts is negligible for the duration of the pulse. Figure 1: Typical voltage-current characteristics at selected temperatures in $B=3$ T. Inset: current dependence of differential resistance $dV/dI$ at the same temperatures. At high currents $dV/dI$ saturates at $R_{ab}$. Typical $V$-$I$ characteristics at different temperatures in a field of $3$ T are shown in Fig. 1. Above a temperature $T_{\mathrm{lin}}<T_{c}$ the $V$-$I$ curves are linear (see the lower inset of Fig. 2 for the field dependence of the characteristic temperatures). Below $T_{\mathrm{lin}}$ nonlinearity develops and the $I\rightarrow 0$ resistance decreases faster than exponentially until it becomes unmeasurably small even with the most sensitive technique. At low temperature dissipation sets in abruptly at a threshold current $I_{\mathrm{th}}$; for higher temperatures a marked upturn in the $V$-$I$ curve (a “knee”) is seen at $I_{\mathrm{k}}\lesssim I_{\mathrm{th}}$. Throughout the nonlinear range the differential resistance increases with increasing current; for currents several times $I_{\mathrm{th}}$ or $I_{\mathrm{k}}$ it saturates (becomes current independent) at a value $dV/dI=R_{ab}$ as shown in the inset of Fig. 1. Since $I_{\mathrm{th}}$ and $I_{\mathrm{k}}$ are hallmarks of depinning, the current-independent $R_{ab}$ observed at many times this current suggests that at these high currents pinning is irrelevant and $R_{ab}$ reflects FFF non . The focus of this article is the behavior of $R_{ab}$. Figure 2: High-current differential resistance $R_{ab}$ at several temperatures as a function of the logarithm of magnetic field normalized to the upper critical field $H_{c2}(T)$. Upper inset: Same data on a linear field scale. Lower inset: Phase diagram measured on same samples except for the first order transition $T_{\mathrm{FOT}}$ taken from Ref. zel for comparison. $T_{\mathrm{irr}}$ refers to magnetic irreversibility, $T_{\mathrm{2nd}}$ to second magnetization peak, $T_{\mathrm{lin}}$ to the beginning of linear $V$-$I$. $H_{c2}=120[1-(T/T_{c})^{2}]$ tesla. The vertical black (grey) lines show the range of full (open) symbols in the main panel. The open circle is crossover in the field dependence of $R_{ab}$. The upper inset of Fig. 2 shows the field dependence of $R_{ab}$ for the temperatures indicated in the phase diagram of the lower inset. The resistance is field and temperature independent at low temperature. With increasing temperature there is a crossover to a field-dependent behavior at $T_{\mathrm{co}}$ (open circles in the phase diagram) situated between the magnetic irreversibility line and $T_{\mathrm{lin}}$. Having in mind that in the BS law the characteristic field is $H_{c2}$, we interpolate the upper critical field using the form $H_{c2}=(120\ \mathrm{tesla})[1-(T/T_{c})^{2}]$ constructed from $dH_{c2}/dT_{T_{c}}=-2.7$ T/K li (1) and use it to plot $R_{ab}$ against $H/H_{c2}$ on a logarithmic field scale in the main panel of Fig. 2. The high-temperature curves all collapse into a master curve representing a logarithmic field dependence: $R_{ab}(H,T)=R_{ab}^{n}[1+\alpha\log(H/H_{c2}(T))]\ ,$ (2) where $R_{ab}^{n}$ is the zero field normal resistance at $T_{c}$ and $\alpha$ is a constant. This scaling only contains the temperature through $H_{c2}(T)$. Equation (2) provides an excellent description of all three samples; for the parameter $\alpha$ we find 0.16, 0.19 and 0.21, essentially the same values, insensitive to the presumably different disorder between samples. We emphasize that $R_{ab}$ cannot be compared directly to the BS law because the strong anisotropy of the electronic properties makes the current distribution very inhomogeneous and $R_{ab}$ reflects both $ab$-plane and $c$-axis properties. However, if the sample is thick in the $c$ direction, a simple scaling argument for a current independent local resistivity tensor yields $R_{ab}=A\sqrt{\rho_{ab}\rho_{c}}$ where $A$ is a geometrical factor. This relation is valid in the linear region $T>T_{\mathrm{lin}}(B)$ but also below $T_{\mathrm{lin}}$ if the current is sufficiently high that the current density is well in the upper differentially linear portion of the response near the top surface of the crystal non . The analysis and experimental checks of Ref. Busch et al. (1992) indicate that with the sample size, shape, and contact geometry used in their and our single crystal $ab$ plane studies in BSCCO, the thick sample limit provides a good description. Independent confirmation of our results emerges from analysis of other experiments. Although high-current data are absent, the fact that the $V$-$I$ curves are linear for $T>T_{\mathrm{lin}}(H)$ allows comparison with low- current data in this temperature and field range. In Fig. 3(a) we show $R_{ab}$ vs. $H$ curves extracted from the $R_{ab}$ vs. $T$ data taken by Busch et al. Busch et al. (1992) in different magnetic fields. Excellent agreement with Eq. (2) is seen for $T>T_{\mathrm{lin}}$ (full symbols in the figure) with $\alpha=0.23$. Figure 3: Single crystal resistance $R_{ab}$ (a); $ab$ plane resistivity $\rho_{ab}$ (b); and $c$-axis resistivity $\rho_{c}$ (c) as a fuction of magnetic field normalized to the upper critical field $H_{c2}$. Data from Ref. Busch et al. (1992). For full symbols $T>T_{\mathrm{lin}}$. Solid lines in panels (a) to (c) are fits to Eqs. (2) to (4), respectively. In panel (b) 77-K thin film data from Ref. xia is also shown by triangles. Inset of panel (c): $c$-axis conductivity data from Ref. Morozov et al. (2000). $\sigma_{c}=1/\rho_{c}$ and $\sigma_{0}$ a constant. The solid line is a fit to Eq. (3). Busch et al. Busch et al. (1992) were able to disentangle $\rho_{ab}$ and $\rho_{c}$ by using data from two additional contacts on the bottom of the crystal. $\rho_{ab}$ and $\rho_{c}$ results extracted from their work for a series of temperatures are shown as a function of $H/H_{c2}$ in Fig. 3(b) and (c). In the temperature and field range $T>T_{\mathrm{lin}}(H)$ both quantities individually exhibit $H/H_{c2}$ scaling. The in-plane resistivity does not agree with the $\beta=1$ BS law of Eq. (1), but is well described by a $\beta=3/4$ exponent (best fit $\beta=0.75\pm 0.02$). Having definite analytic forms for both $R_{ab}$ and $\rho_{ab}$, we can use the relation $(R_{ab}/A)^{2}=\rho_{ab}\rho_{c}$ to write an expression for the $c$-axis resistivity. In summary: $\displaystyle\rho_{ab}$ $\displaystyle\equiv$ $\displaystyle\rho_{\mathrm{FFF}}=\rho_{ab}^{n}(H/H_{c2})^{\beta},$ (3) $\displaystyle\rho_{c}$ $\displaystyle=$ $\displaystyle\rho_{c}^{n}(H/H_{c2})^{-\beta}[1+\alpha\log(H/H_{c2})]^{2},$ (4) $\displaystyle\beta$ $\displaystyle=$ $\displaystyle 3/4,~{}\alpha=0.2~{}.$ The prefactors $\rho_{ab}^{n}$ and $\rho_{c}^{n}$ are in good agreement with the respective normal resistivities at $T_{c}$. Are these forms corroborated by other types of measurement? In principle $\rho_{ab}$ can be measured in thin films where the current density is expected to be homogeneous. In Fig. 3(b) we show the $77$ K thin film resistivity obtained by digitizing the $V$-$I$ curves in Ref. xia . Data above about 1 T are reasonably well described by a $H^{3/4}$ dependence, but a closer look reveals that the $\log\rho_{ab}$ vs. $\log H$ curves are concave from below at every $H$, i.e., there is a systematic deviation from power law. This “logarithm like” (but not logarithmic) dependence is shared with other thin film results thi but there are significant quantitative differences between data measured by different groups. A possible reason is that macroscopic defects like steps on the surface or mosaic boundaries force $c$-axis currents and the measured resistance is a sample-dependent combination of $\rho_{ab}$ and $\rho_{c}$. The expression for $\rho_{c}$ reproduces well the maximum (at $H_{\mathrm{max}}=H_{c2}\exp(8/3-1/\alpha)\sim 0.1H_{c2}$) observed in the high field $c$-axis magnetoresistance Morozov et al. (2000) and the overall field dependence of these independently measured data is very well described by Eq. (4). We demostrate this in Fig. 3(c) where we plot $70$ K data for $\sigma_{c}(B)-\sigma_{0}$ from Fig. 5 of Ref. Morozov et al. (2000) where $\sigma_{c}=1/\rho_{c}$ and the constant $\sigma_{0}$ is interpreted as the zero-field quasiparticle conductivity. Using our estimate of $H_{c2}(T)\approx 46$ T for $T=70$ K, Eq. (4) fits the measured data with parameters $\sigma_{c}^{n}=1/\rho_{c}^{n}=6.6$ (k$\Omega$cm)-1, $\sigma_{0}=3.6$ (k$\Omega$cm)-1 and $\alpha=0.20$ to obtain the curve indicated by the continuous line in the figure. The value of $\alpha$ is in excellent agreement with that inferred from $R_{ab}$ measurements. It should be pointed out, however, that $\sigma_{0}$ is significantly smaller than the value $\sigma_{0}\approx 8$ (k$\Omega$cm)-1 inferred in Ref. Morozov et al. (2000). In terms of resistivities, this means that $\rho_{c}$ decreases more slowly in high fields than described by Eq. (4) with $\beta=3/4$ and is in fact best described with an exponent $\beta=0.51$. In the high-current limit the same form for $R_{ab}(H)$ also holds below $T_{\mathrm{lin}}(H)$ where the $V$-$I$ curves are nonlinear. Since no change in the behavior of $R_{ab}$ is observed when the $T_{\mathrm{lin}}(H)$ line is crossed, it is reasonable to assume the same for $\rho_{ab}$ and $\rho_{c}$. In the low field direction a lower limit of the validity of Eq. (2) is the zero of the equation at $H_{0}/H_{c2}=e^{-1/\alpha}\sim 10^{-3}-10^{-2}$, higher but in the order of the first order transition in the static vortex system. In the high-field direction Eq. (2) is valid up to the highest field $\approx 0.3H_{c2}$ we investigated. The temperature $T_{\mathrm{co}}$ of the crossover from $R_{ab}=\mathrm{const}$ to $R_{ab}\propto\log H$ is distinctly higher than the onset of magnetic irreversibility at $T_{\mathrm{irr}}$, and also above the vortex glass transition $T_{g}\approx T_{\mathrm{irr}}$ inferred from scaling analysis gla of the $V$-$I$ curves. On the other hand, no change in the behavior of $R_{ab}$ is observed when the $T_{\mathrm{irr}}(H)$ and $T_{g}(H)$ lines are crossed. This suggests that because the pinning potential is smoothed at high velocities, the phase diagram pha of the far-from- equilibrium dynamic vortex system dyn is different from that of the unperturbed thermodynamic phases. Since $R_{ab}$ behaves the same in the pinned ($T<T_{\mathrm{lin}}$) and unpinned ($T>T_{\mathrm{lin}}$) liquid phases, we propose that the unpinned phase, otherwise observed only above $T_{\mathrm{lin}}$, may be restored in the range $T_{\mathrm{co}}<T<T_{\mathrm{lin}}$. Then $T_{\mathrm{co}}$ may approximate the melting transition in a hypothetical defect-free crystal. Our most robust finding, invariably observed not only in our 3 batches but also in the data of Ref. Busch et al. (1992), is the logarithmic field dependence of the high-current single crystal resistance $R_{ab}$. Although a power of $H$ factor is expected both in $\rho_{ab}$ kop and $\rho_{c}$ vek , no such factor is present in $R_{ab}\propto\sqrt{\rho_{ab}\rho_{c}}$. The most likely reason is that the power-law factors in $\rho_{ab}$ and $\rho_{c}$ cancel (exponents 3/4 and -3/4 in our analysis). The cancellation is very accurate; we estimate that a power law factor with exponent as low as 0.1 could be observed in our $R_{ab}$ data. Moreover, because we find no logarithmic correction to $\rho_{ab}$, the logarithmic dependence of $R_{ab}$ is carried by $\rho_{c}$. Arguing that both $\rho_{ab}$ and $\sigma_{c}$ are proportional to the quasiparticle density of states at the Fermi level, $N(0)$, it cancels in the product $\rho_{ab}\rho_{c}$. In conventional superconductors $N(0)$ is proportional to the number of vortices therefore to $H$, leading to the $H$-linear resistivity of the BS law. In nodal gap superconductors near-nodal quasiparticles lead to a sublinear dependence; for line nodes $N(0)\propto H^{1/2}$ vol , as evidenced in recent low-temperature thermodynamic measurements the . Although delocalized near-nodal quasiparticles are not expected to contribute significantly to $\rho_{\mathrm{FFF}}$ because of the weak spectral flow force Kopnin and Volovik (1997) they experience, the result may be different in the diffusive limit in the liquid phase. A possible reason for the $\beta=3/4$ exponent is the different structure factors of the solid and liquid phases. Nonlocal effects Levin (1997); Koshelev (1996) may influence the evaluation of the 6-contact measurements Busch et al. (1992) and therefore the validity of Eqs. (3) and (4) (but not of Eq. (2)). This seems, however, unlikely in the light of the good agreement of $\rho_{c}$ inferred from independent $ab$-plane Busch et al. (1992) and $c$-axis Morozov et al. (2000) measurements and of the broad temperature range of validity of Eq. (2). In conclusion, we have set up empirical rules for the analytic form of single crystal resistance as well as for the $ab$ plane and $c$ axis resistivities in the high-current free flux flow limit in the vortex liquid state of BSCCO, valid over a broad range of temperature and field. Both the logarithmic field dependence of the single crystal resistance and the 3/4-power law in the $ab$-plane free flux flow resistance are in disagreement with the current theoretical understanding of high-$T_{c}$ superconductors. We acknowledge with pleasure fruitful discussions with F. Portier, I. Tüttő, L. Forró and T. Fehér and the help and technical expertise of F. Tóth. L. Forró and the EPFL laboratory in Lausanne have contributed in a very essential way to sample preparation and characterization. Finally we acknowledge with gratitude the Hungarian funding agency OTKA (grant no. K 62866). ## References * (1) For a review, see N. B. Kopnin, Theory of Nonequilibrium Superconductivity (Oxford University Press, 2001). * (2) J. Bardeen and M. J. Stephen, Phys. Rev. 140, A1197 (1965); P. Nozières and W. F. Vinen, Philos. Mag. 14, 667 (1966). * (3) For a review, see D. Parks, Ed., Superconductivity (Dekker, New York, 1969). * Kopnin and Volovik (1997) N. B. Kopnin and G. E. Volovik, Phys. Rev. Lett. 79, 1377 (1997). * (5) I. Pethes et al., Synth. Met. 120, 1013 (2000). * (6) H. Raffy et al., Phys. Rev. Lett. 66, 2515 (1991); P. Wagner et al., Phys. Rev. B49, 13184 (1994); M. Giura et al., Phys. Rev. B50, 12920 (1994). * (7) Z. L. Xiao, P. Voss-de Haan, G. Jakob, and H. Adrian Phys. Rev. B57, R736 (1998). * (8) R. Mallozzi et al., Phys. Rev. Lett. 81, 1485 (1998); Tetsuo Hanaguri et al., Phys. Rev. Lett. 82, 1273 (1999). * (9) M. N. Kunchur, D. K. Christen, and J. M. Phillips, Phys. Rev. Lett. 70, 998 (1993); M. N. Kunchur, Phys. Rev. Lett. 89, 137005 (2002). * Busch et al. (1992) R. Busch, G. Ries, H. Werthner, G. Kreiselmeyer, and G. Saemann-Ischenko, Phys. Rev. Lett. 69, 522 (1992). * Morozov et al. (2000) N. Morozov, L. Krusin-Elbaum, T. Shibauchi, L. N. Bulaevskii, M. P. Maley, Y. I. Latyshev, and T. Yamashita, Phys. Rev. Lett. 84, 1784 (2000). * (12) B. Sas et al., Phys. Rev. B61, 9118 (2000). * (13) In a numerical simulation with nonlinear local conductivity we find that at $I>3I_{\mathrm{k}}$, $R_{ab}=dV/dI$ approximates the high-current linear resistance within a few percent. $dV/dI$ is always closer to the linear resistance than $V/I$. * (14) H. Beidenkopf et al., Phys. Rev. Lett. 95, 257004 (2005). * li (1) Qiang Li et al., Phys. Rev. B48, 9877 (1993). * (16) H. Safar et al., Phys. Rev. Lett. 68, 2672 (1992); H. Yamasaki et al., Phys. Rev. B50, 12959 (1994). * (17) Á. Pallinger et al. (unpublished). * (18) For an introduction and further references see P. Le Doussal and T. Giamarchi, Phys. Rev. B57, 11356 (1998). * (19) I. Vekhter et al., Phys. Rev. Lett. 84, 1296 (2000). * (20) G. E. Volovik, Pis’ma Zh. Eksp. Teor. Fiz. 58, 457 (1993) [JETP Lett. 58, 469 (1993)]. * (21) K. A. Moler et al., Phys. Rev. Lett. 73, 2744 (1994); B. Revaz et al., Phys. Rev. Lett. 80, 3364 (1998); D. A. Wright et al., Phys. Rev. Lett. 82, 1550 (1999). * Levin (1997) G. A. Levin, Phys. Rev. Lett. 79, 5299 (1997). * Koshelev (1996) A. E. Koshelev, Phys. Rev. Lett. 76, 1340 (1996).

arxiv-papers

2024-09-04T02:48:54.312929

0803.1834

# Precise Measurement of the Spin Parameter of the Stellar-Mass Black Hole M33 X-7 Jifeng Liu11affiliation: Harvard-Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Jeffrey E. McClintock11affiliation: Harvard- Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Ramesh Narayan11affiliation: Harvard-Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Shane W. Davis22affiliation: Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540 , and Jerome A. Orosz33affiliation: Department of Astronomy, San Diego State University, 5500 Campanile Drive, San Diego, CA 92182 # Erratum: “Precise Measurement of the Spin Parameter of the Stellar-mass Black Hole M33 X-7” (ApJL, 679, 37L [2008]) Jifeng Liu11affiliation: Harvard-Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Jeffrey E. McClintock11affiliation: Harvard- Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Ramesh Narayan11affiliation: Harvard-Smithsonian Center for Astrophysics,60 Garden Street, Cambridge, MA 02138 , Shane W. Davis22affiliation: Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540 , and Jerome A. Orosz33affiliation: Department of Astronomy, San Diego State University, 5500 Campanile Drive, San Diego, CA 92182 ###### Abstract In prior work, Chandra and Gemini-North observations of the eclipsing X-ray binary M33 X-7 have yielded measurements of the mass of its black hole primary and the system’s orbital inclination angle of unprecedented accuracy. Likewise, the distance to the binary is known to a few percent. In an analysis based on these precise results, fifteen Chandra and XMM-Newton X-ray spectra, and our fully relativistic accretion disk model, we find that the dimensionless spin parameter of the black hole primary is $a_{*}=0.77\pm 0.05$. The quoted 1-$\sigma$ error includes all sources of observational uncertainty. Four Chandra spectra of the highest quality, which were obtained over a span of several years, all lead to the same estimate of spin to within statistical errors (2%), and this estimate is confirmed by 11 spectra of lower quality. There are two remaining uncertainties: (1) the validity of the relativistic model used to analyze the observations, which is being addressed in ongoing theoretical work; and (2) our assumption that the black hole spin is approximately aligned with the angular momentum vector of the binary, which can be addressed by a future X-ray polarimetry mission. Galaxies: individual(M33) — X-rays: binaries — black hole physics — binaries: individual (M33 X-7) ††slugcomment: Published in ApJ Letters: 679, 37-40 (2008) ## 1 INTRODUCTION M33 X-7 is the first stellar-mass black hole to be discovered that is eclipsed by its companion [Pietsch et al., 2006]. The X-ray eclipse and the precisely known distance of this system, $D=840\pm 20$ kpc, underpin the most accurate dynamical model that has been achieved for any of the 21 known black hole binaries (Orosz et al. 2007, hereafter O07). The two dynamical parameters of interest in this Letter are the black hole mass $M=15.65\pm 1.45M_{\odot}$ and the orbital inclination angle $i=74.6^{\circ}\pm 1.0^{\circ}$ (O07). Our group has published spin estimates for three stellar-mass black holes using the X-ray continuum fitting method: GRO J1655-40, $a_{*}=0.65-0.75$; 4U 1543-47, $a_{*}=0.75-0.85$; and GRS 1915+105, $a_{*}=0.98-1.0$ (Shafee et al. 2006, hereafter S06; McClintock et al. 2006, hereafter M06). For LMC X-3, Davis et al. [2006] find $a_{*}<0.26$. Meanwhile, the Fe line method has been used to obtain two additional estimates of black hole spin (Brenneman and Reynolds 2006; Miller et al. 2008). The dimensionless spin parameter $a_{*}\equiv a/M=cJ/GM^{2}$, where $M$ and $J$ are the mass and angular momentum of the black hole; $-1\leq a_{*}\leq 1$ [Shapiro and Teukolsky, 1986]. The continuum-fitting method, which was pioneered by [Zhang et al., 1997] (also see Gierliński et al. 2001), is based on the existence of an innermost stable circular orbit (ISCO) for a particle orbiting a black hole, inside which the particle suddenly plunges into the hole. In the continuum-fitting method, one identifies the inner edge of the black hole’s accretion disk with the ISCO and estimates the radius $R_{\rm ISCO}$ of this orbit by fitting the X-ray continuum spectrum. Since the dimensionless radius $r_{\rm isco}\equiv R_{\rm ISCO}/(GM/c^{2})$ is solely a monotonic function of the black hole spin parameter [Shapiro and Teukolsky, 1986], knowing its value allows one to immediately infer the black hole spin parameter $a_{*}$. Our estimates of spin are based on our fully relativistic accretion disk model [Li et al., 2005] and an advanced treatment of spectral hardening [Davis et al., 2005]. We consider only rigorously-selected thermal-state data [Remillard and McClintock, 2006], which are largely free of the effects of Comptonization. Furthermore, we only accept data for which the bolometric disk luminosity is moderate, $L/L_{\rm Edd}<0.3$, in order to ensure that the standard geometrically-thin thermal disk model is applicable (Shafee et al. 2008; M06). For the continuum-fitting method to succeed, it is essential to have accurate values of the black hole mass, orbital inclination and distance (M06), quantities that are known precisely in the case of M33 X-7. Other virtues of M33 X-7 for the determination of spin, are the abundance of Chandra and XMM data, the remarkably thermal and featureless spectrum of the X-ray source, and its moderate luminosity (§3). ## 2 DATA SELECTION AND REDUCTION There have been 17 Chandra ACIS observations and 12 XMM-Newton EPIC observations of M33 X-7. We analyzed Chandra observations (downloaded from the Chandra Data Archive) with CIAO 3.4 and extracted the spectra from source ellipses enclosing 95% of the source photons as reported by wavdetect. The XMM observations were downloaded from the HEASARC archive and analyzed with SAS 7.0.0 [Gabriel et al., 2004]. Using the standard procedures and excluding intervals of high background, we extracted separate spectra from the PN and two MOS chips using radii of 400 pixels (i.e., $20^{\prime\prime}$) and fitted them independently. The resultant count rate data were folded on the M33 X-7 X-ray eclipse ephemeris [Pietsch et al., 2006]: HJD $(2453639.119\pm 0.005)\pm N(3.453014\pm 0.000020)$. The folded light curve for all the Chandra data are shown in Figure $2a$ in O07. For the purpose of measuring the spin of M33 X-7, we excluded (1) spectra obtained in the phase range –0.3 to 0.2, i.e., during the eclipse or the pre-eclipse period of erratic X-ray variability (which is presumably caused by the accretion stream; O07); (2) four ACIS spectra (ObsIDs 6384, 6385, 7170, 7171) that are severely affected by pile-up; and (3) all spectra that contain less than 1000 counts. The 15 spectra so selected are listed in Table 1 and comprise eight Chandra ACIS spectra and seven EPIC PN/MOS spectra from five XMM observations. We refer throughout to the four ACIS spectra with $\buildrel{\scriptstyle>}\over{\scriptstyle\sim}$ 5,000 counts as the “gold” spectra and to the rest as the “silver” spectra. ## 3 ANALYSIS AND RESULTS The procedures used here are precisely the same as those that are described fully in M06. Briefly, the relativistic accretion disk model kerrbb2 has just two fit parameters, namely the black hole spin $a_{*}$ and the mass accretion rate $\dot{M}$ (or equivalently, $a_{*}$ and the Eddington-scaled bolometric luminosity, $l\equiv L_{\rm bol}(a_{*},\dot{M})/L_{\rm Edd}$; M06). In the case of M33 X-7, we also fit for a third parameter, $N_{\rm H}$, the hydrogen column density (phabs in XSPEC). The spectral hardening factor $f\equiv T_{\rm col}/T_{\rm eff}$ was computed as a function of $l$ for the appropriate metallicity of M33 X-7 ($Z=0.1Z_{\odot}$; O07) using the model of Davis and Hubeny 2006 (bhspec in XSPEC). These values of $f$ are contained in a pair of lookup tables, which correspond to two representative values of the viscosity parameter ($\alpha=0.01,0.1$; M06) for a wide range of the spin parameter (e.g., $0<a_{*}<0.99$). We find that our results are quite insensitive to the choice of $\alpha$ or an increase in metallicity. We have also experimented with varying the input parameters $M$, $D$, $i$ and $N_{\rm H}$, and we find that the values of $f$ are scarcely affected. All of the spectra were well-fitted using a simple absorbed kerrbb2 model [i.e., phabs(kerrbb2) in XSPEC]. Notably, neither Fe line/edge components nor an additional nonthermal component was required, as they were in our earlier work (R06, M06). We fitted each spectrum for $a_{*}$, the mass accretion rate $\dot{M}$, and the neutral hydrogen column density $N_{\rm H}$ with the input parameters fixed at their baseline values (see §1; O07). The normalization was fixed at unity (as appropriate when $M$, $i$ and $D$ are held fixed). We included the effects of limb darkening (lflag = 1) and returning radiation effects (rflag = 1), and we set the torque at the inner boundary of the accretion disk to zero ($\eta=0$). The fits obtained for all 15 spectra are quite acceptable with $\chi^{2}_{\nu}<1.2$; results for fits over the energy range 0.3–8 keV are summarized in Table 1. An inspection of the fitting residuals for all the spectra show them to be free of any systematic effects, as illustrated in Figure 1. Given the modest luminosities, $0.07<l<0.11$ (Table 1), which are well below our selection limit of $l=0.3$, the accretion disk in M33 X-7 is quite thin, $H/R\leq 0.04$ (see Fig. 17 in Shafee et al. 2008), and our assumption of zero torque at the inner boundary is likely to be valid. Figure 2 shows plots of $a_{*}$ for all 15 observations, which are ordered by the number of counts detected. Each of the four panels corresponds to a different choice for the energy interval used in fitting the data (e.g., 0.3–8 keV, 0.5–8 keV, etc.); a comparison of the results in the four panels shows that this choice is quite unimportant. The four gold spectra with $\lower 2.0pt\hbox{$\buildrel{\scriptstyle>}\over{\scriptstyle\sim}$}\ 5,000$ counts each (solid symbols) yield spin estimates that agree with their mean value (indicated by the dotted lines) typically to within their $\approx 2$% statistical uncertainties. The stability of these four gold spectra is especially remarkable given that three of the observations were separated by 3-month intervals in 2005–2006, and one of them was obtained five years earlier in 2000 (Table 1). The dispersion for the 11 silver spectra that have $\lower 2.0pt\hbox{$\buildrel{\scriptstyle<}\over{\scriptstyle\sim}$}\ 3,000$ counts (open symbols) is much larger. However, in each panel, the mean of these 11 spin values agrees with the mean determined using the gold spectra to within $\approx 1$%. As concluded in the caption of Figure 2, our adopted average spin for the four gold spectra is $\bar{a}_{*}=0.77$ with a standard deviation of $\Delta a_{*}=0.02$. In order to determine the error in $a_{*}$ due to the combined uncertainties in $M$, $i$ and $D$ (§1), we performed Monte Carlo simulations assuming that the uncertainties in these parameters are normally and independently distributed. The results for 3,000 simulation runs are plotted in Figure 3. The histogram of $\bar{a}_{*}-\bar{a}_{*0}$ shows that the $1\sigma$ error in the spin due to the combined uncertainties of the three input parameters is about $\Delta a_{*}=0.05$. The error is dominated by the uncertainty in $M$; the uncertainties in $i$ and $D$ are relatively unimportant. This error is based on a readily available table that was computed for solar metallicity. Despite this limitation, we believe that our error estimate is accurate because the effects of going from $Z=0.1Z_{\odot}$ to $Z=Z_{\odot}$ are very small at the luminosities in question, $l\approx 0.1$. ## 4 DISCUSSION The largest error in our spin estimate arises from the uncertainties in the validity of the disk model we employ. For example, the spin depends on accurate model determinations of the hardening factor $f$; this problem is quite tractable and vigorous theoretical efforts are underway (Davis et al. 2005, 2006, Blaes et al. 2006). Possibly more problematic is our assumption that the viscous torque vanishes at the ISCO and that there is no significant emission from the gas inside the ISCO. Hydrodynamic models of the accretion disk indicate that the viscous torque at the ISCO as well as emission from inside the ISCO should both be negligible for the geometrically thin disks and low luminosities ($l\leq 0.3$) that we restrict ourselves to (Afshordi and Paczyński 2003; S08). The emission from inside the ISCO causes rather modest errors in spin estimates; in the case of M33 X-7, the estimated error is $\Delta a_{*}\leq 0.01$ since $l\leq 0.1$ and hence $R/H\leq 0.04$ (M06; S08). On the other hand, MHD simulations of accretion flows around black holes [Hawley and Krolik, 2002, Beckwith et al., 2008] find a large torque at the ISCO and substantial dissipation inside the ISCO. We note, however, that these simulations carried out so far are for geometrically thick systems, with $H/R\sim 0.2$; these flows are nearly an order of magnitude thicker than the disk in M33 X-7. In the hydrodynamic models of Shafee et al. [2008] the stress at the ISCO increases rapidly with increasing disk thickness, so it is conceivable that there is no serious disagreement between the hydrodynamic and MHD results. Numerical MHD simulations of truly thin disks are necessary to resolve this issue. We note that a recent MHD simulation of a geometrically thin accretion disk for a pseudo-Newtonian potential does show a dramatic drop in the mid-plane density and vertical column density over a narrow range of radii close to the ISCO [Reynolds and Fabian, 2008]. Although there is theoretical uncertainty about conditions near the ISCO, there is a long history of evidence suggesting that fitting the X-ray continuum is a promising approach to measuring black hole spin. This history begins in the mid-1980s with the simple non-relativistic multicolor disk model [Mitsuda et al., 1984], which returns the color temperature $T_{\rm in}$ at the inner-disk radius $R_{\rm in}$. Tanaka and Lewin [1995] summarize examples of the steady decay (by factors of 10–100) of the thermal flux of transient sources during which $R_{\rm in}$ remains quite constant (see their Fig. 3.14). More recently, this evidence for a constant inner radius in the thermal state has been presented for a number of sources in several papers via plots showing that the bolometric luminosity of the thermal component is approximately proportional to $T^{4}$ (McClintock et al. 2007, and references therein). Obviously, this non-relativistic analysis cannot provide a secure value for the radius of the ISCO nor even establish that this stable radius is the ISCO. Nevertheless, the presence of a fixed radius indicates that the continuum-fitting method is a well-founded approach to measuring black hole spin. It is reasonable to assume that the inner X-ray-emitting portion of the disk is aligned with the spin axis of the black hole by the Bardeen-Petterson effect [lod05]. Throughout, in making use of the orbital inclination angle, we have assumed that the black hole spin is aligned with the angular momentum vector of the binary system. As Figure 3 indicates, if any misalignment is $\lower 2.0pt\hbox{$\buildrel{\scriptstyle<}\over{\scriptstyle\sim}$}\ 3^{\circ}$, then it will contribute an error in $a_{*}$ that is no larger than our total observational error of $\Delta a_{*}=0.05$. There is no evidence for significant misalignments despite the often-cited examples of GRO J1655-40 and SAX J1819.3-2525 (see §2.2 in Narayan and McClintock 2005; but see Maccarone 2002). The clear-cut way to assess the degree of alignment is via X-ray polarimetric observations of black hole systems in the thermal state (Li et al. 2008, in preparation). What is the origin of the spin of M33 X-7? Was the black hole born with its present spin, or was it torqued up gradually via the accretion flow supplied by its companion? In order to achieve a spin of $a_{*}=0.77$ via disk accretion, an initially non-spinning black hole must accrete $4.9M_{\odot}$ from its donor [King and Kolb, 1999] in becoming the $M=15.65M_{\odot}$ that we observe today (O07). However, to transfer this much mass even in the case of Eddington-limited accretion ($\dot{M}_{\rm Edd}\equiv L_{\rm Edd}/c^{2}\approx 4\times 10^{-8}M_{\odot}/{\rm yr}$) requires $\sim 120$ million years, whereas the age of the system is only 2–3 million years (O07). Thus, it appears that the spin of M33 X-7 must be natal, which is the same conclusion that has been reached for two other stellar black holes (S06, M06; but see Bethe et al. 2003 on the possibility of hypercritical accretion) M33 X-7’s secure dynamical data and distance, the X-ray source’s clean thermal-state spectrum and moderate luminosity, and an abundance of Chandra and XMM data have provided arguably the most secure estimate of black hole spin that has been achieved to date: $a_{*}=0.77\pm 0.05$, where the error estimate includes all sources of observational error. Since an astrophysical black hole can be described by just the two parameters that specify its mass and spin [Shapiro and Teukolsky, 1986], we now have a complete description of an asteroid-size object that is situated at a distance of about one Mpc. JFL and SWD acknowledge support from NASA through the Chandra Fellowship Program, grants PF6-70043 and PF6-70045. JEM acknowledges support from NASA grant AR8-9006X. We thank Rebecca Shafee for technical advice and Jack Steiner for critical comments on the manuscript. In the paper “Precise Measurement of the Spin Parameter of the Stellar-Mass Black Hole M33 X-7” by Jifeng Liu, Jeffrey E. McClintock, Ramesh Narayan, Shane W. Davis, and Jerome A. Orosz (ApJ, 679, L37 [2008]), the reported value of the black-hole spin parameter $a_{*}=0.77\pm 0.05$ is in error. The correct value is larger by 0.068 and is $a_{*}=0.84\pm 0.05$. The error is the result of a bug in the XSPEC accretion- disk model kerrbb.111http://heasarc.gsfc.nasa.gov/docs/xanadu/xspec/issues/archive/issues.12.5.0an.html (patch 12.5.0a). Prior to 1 December 2008, the model’s two parameter flags that switch limb darkening and self-irradiation of the disk on/off were reversed (e.g., “par8” incorrectly controlled limb darkening rather than self- irradiation). In computing tables of the spectral hardening factor $f$, we use both kerrbb and the disk atmosphere model bhspec [McClintock et al., 2006]. Because the latter model does not include the effect of self-irradiation, we switch this feature off in kerrbb when computing the $f$-tables. In this instance, because of the bug we switched off limb darkening instead of self- irradiation, which corrupted our results. Meanwhile, our earlier spin results for GRS 1915+105 (McClintock et al. 2006) and for 4U 1543–47 and GRO J1655–40 [Shafee et al., 2006] are unaffected by the bug. The figures and tabular data in the paper are essentially unaffected, apart from the increase in $a_{*}$ and corresponding decreases in $f$ and the Eddington-scaled luminosity $l$ (8.7% and 4.5%, respectively, for the four gold spectra). The higher spin increases somewhat our estimate of how much mass ($4.9~{}M_{\odot}$) and time ($\sim 120$ million years) would be required to spin up an initially nonspinning black hole to the present spin of M33 X-7. In order to achieve $a_{*}=0.84$, the black hole must accrete $5.7~{}M_{\odot}$, which would require $\sim 140$ million years. Because the age of the binary system is only 2–3 million years this change does not at all affect our conclusion that the spin of M33 X-7 is natal. ## References * Afshordi and Paczyński [2003] N. Afshordi and B. Paczyński. Geometrically Thin Disk Accreting into a Black Hole. _ApJ_ , 592:354–367, July 2003. 10.1086/375559. * Beckwith et al. [2008] K. Beckwith, J. F. Hawley, and J. H. Krolik. Where is the radiation edge in magnetized black hole accretion discs? _MNRAS_ , 390:21–38, October 2008. 10.1111/j.1365-2966.2008.13710.x. * Bethe et al. [2003] H. A. Bethe, G. E. Brown, and C.-H. Lee. _Formation and evolution of black holes in the Galaxy : selected papers with commentary_. 2003\. * Blaes et al. [2006] O. M. Blaes, S. W. Davis, S. Hirose, J. H. Krolik, and J. M. Stone. Magnetic Pressure Support and Accretion Disk Spectra. _ApJ_ , 645:1402–1407, July 2006. 10.1086/503741. * Brenneman and Reynolds [2006] L. W. Brenneman and C. S. Reynolds. Constraining Black Hole Spin via X-Ray Spectroscopy. _ApJ_ , 652:1028–1043, December 2006. 10.1086/508146. * Davis and Hubeny [2006] S. W. Davis and I. Hubeny. A Grid of Relativistic, Non-LTE Accretion Disk Models for Spectral Fitting of Black Hole Binaries. _ApJS_ , 164:530–535, June 2006. 10.1086/503549. * Davis et al. [2005] S. W. Davis, O. M. Blaes, I. Hubeny, and N. J. Turner. Relativistic Accretion Disk Models of High-State Black Hole X-Ray Binary Spectra. _ApJ_ , 621:372–387, March 2005. 10.1086/427278. * Davis et al. [2006] S. W. Davis, C. Done, and O. M. Blaes. Testing Accretion Disk Theory in Black Hole X-Ray Binaries. _ApJ_ , 647:525–538, August 2006. 10.1086/505386. * Gabriel et al. [2004] C. Gabriel, M. Denby, D. J. Fyfe, J. Hoar, A. Ibarra, E. Ojero, J. Osborne, R. D. Saxton, U. Lammers, and G. Vacanti. The XMM-Newton SAS - Distributed Development and Maintenance of a Large Science Analysis System: A Critical Analysis. In F. Ochsenbein, M. G. Allen, & D. Egret, editor, _Astronomical Data Analysis Software and Systems (ADASS) XIII_ , volume 314 of _Astronomical Society of the Pacific Conference Series_ , pages 759–+, July 2004. * Gierliński et al. [2001] M. Gierliński, A. Maciołek-Niedźwiecki, and K. Ebisawa. Application of a relativistic accretion disc model to X-ray spectra of LMC X-1 and GRO J1655-40. _MNRAS_ , 325:1253–1265, August 2001. 10.1046/j.1365-8711.2001.04540.x. * Hawley and Krolik [2002] J. F. Hawley and J. H. Krolik. High-Resolution Simulations of the Plunging Region in a Pseudo-Newtonian Potential: Dependence on Numerical Resolution and Field Topology. _ApJ_ , 566:164–180, February 2002. 10.1086/338059. * King and Kolb [1999] A. R. King and U. Kolb. The evolution of black hole mass and angular momentum. _MNRAS_ , 305:654–660, May 1999. 10.1046/j.1365-8711.1999.02482.x. * Li et al. [2005] L.-X. Li, E. R. Zimmerman, R. Narayan, and J. E. McClintock. Multitemperature Blackbody Spectrum of a Thin Accretion Disk around a Kerr Black Hole: Model Computations and Comparison with Observations. _ApJS_ , 157:335–370, April 2005. 10.1086/428089. * Maccarone [2002] T. J. Maccarone. On the misalignment of jets in microquasars. _MNRAS_ , 336:1371–1376, November 2002. 10.1046/j.1365-8711.2002.05876.x. * McClintock et al. [2006] J. E. McClintock, R. Shafee, R. Narayan, R. A. Remillard, S. W. Davis, and L.-X. Li. The Spin of the Near-Extreme Kerr Black Hole GRS 1915+105. _ApJ_ , 652:518–539, November 2006. 10.1086/508457. * McClintock et al. [2007] J. E. McClintock, R. Narayan, and R. Shafee. Estimating the Spins of Stellar-Mass Black Holes. _ArXiv e-prints_ , July 2007. * Miller et al. [2008] J. M. Miller, C. S. Reynolds, A. C. Fabian, E. M. Cackett, G. Miniutti, J. Raymond, D. Steeghs, R. Reis, and J. Homan. Initial Measurements of Black Hole Spin in GX 339-4 from Suzaku Spectroscopy. _ApJ_ , 679:L113–L116, June 2008. 10.1086/589446. * Mitsuda et al. [1984] K. Mitsuda, H. Inoue, K. Koyama, K. Makishima, M. Matsuoka, Y. Ogawara, K. Suzuki, Y. Tanaka, N. Shibazaki, and T. Hirano. Energy spectra of low-mass binary X-ray sources observed from TENMA. _PASJ_ , 36:741–759, 1984. * Narayan and McClintock [2005] R. Narayan and J. E. McClintock. Inclination Effects and Beaming in Black Hole X-Ray Binaries. _ApJ_ , 623:1017–1025, April 2005. 10.1086/428709. * Orosz et al. [2007] J. A. Orosz, J. E. McClintock, R. Narayan, C. D. Bailyn, J. D. Hartman, L. Macri, J. Liu, W. Pietsch, R. A. Remillard, A. Shporer, and T. Mazeh. A 15.65-solar-mass black hole in an eclipsing binary in the nearby spiral galaxy M 33. _Nature_ , 449:872–875, October 2007. 10.1038/nature06218. * Pietsch et al. [2006] W. Pietsch, F. Haberl, M. Sasaki, T. J. Gaetz, P. P. Plucinsky, P. Ghavamian, K. S. Long, and T. G. Pannuti. M33 X-7: ChASeM33 Reveals the First Eclipsing Black Hole X-Ray Binary. _ApJ_ , 646:420–428, July 2006. 10.1086/504704. * Remillard and McClintock [2006] R. A. Remillard and J. E. McClintock. X-Ray Properties of Black-Hole Binaries. _ARA &A_, 44:49–92, September 2006. 10.1146/annurev.astro.44.051905.092532. * Reynolds and Fabian [2008] C. S. Reynolds and A. C. Fabian. Broad Iron-K$\alpha$ Emission Lines as a Diagnostic of Black Hole Spin. _ApJ_ , 675:1048–1056, March 2008. 10.1086/527344. * Shafee et al. [2006] R. Shafee, J. E. McClintock, R. Narayan, S. W. Davis, L.-X. Li, and R. A. Remillard. Estimating the Spin of Stellar-Mass Black Holes by Spectral Fitting of the X-Ray Continuum. _ApJ_ , 636:L113–L116, January 2006. 10.1086/498938. * Shafee et al. [2008] R. Shafee, R. Narayan, and J. E. McClintock. Viscous Torque and Dissipation in the Inner Regions of a Thin Accretion Disk: Implications for Measuring Black Hole Spin. _ApJ_ , 676:549–561, March 2008. 10.1086/527346. * Shapiro and Teukolsky [1986] S. L. Shapiro and S. A. Teukolsky. _Black Holes, White Dwarfs and Neutron Stars: The Physics of Compact Objects_. June 1986. * Tanaka and Lewin [1995] Y. Tanaka and W. H. G. Lewin. Black hole binaries. In W. H. G. Lewin, J. van Paradijs, & E. P. J. van den Heuvel, editor, _X-ray binaries, p. 126 - 174_ , pages 126–174, 1995. * Zhang et al. [1997] S. N. Zhang, W. Cui, and W. Chen. Black Hole Spin in X-Ray Binaries: Observational Consequences. _ApJ_ , 482:L155+, June 1997. 10.1086/310705. Table 1: Kerrbb2 fit results for M33 X-7 in 0.3-8 keVaaThe columns are (1) ID number; (2) date of observation; (3) exposure time in ksec; (4) no. of counts; (5) spin parameter; (6) mass accretion rate in $10^{18}$ g s-1; (7) hydrogen column density in $10^{20}$ cm-2; (8) spectral hardening factor; (9) Eddington-scaled bolometric luminosity; and (10) reduced $\chi^{2}$ per dof. spectrum | obs-date | Texp | counts | $a_{*}$ | $\dot{M}$ | $n_{\rm H}$ | $f_{\rm col}$ | $\lg l$ | $\chi^{2}_{\nu}$/dof ---|---|---|---|---|---|---|---|---|--- acis6376 | 2006-03-03 | 93.1 | 9748 | 0.751 $\pm$ 0.026 | 1.88 $\pm$ 0.12 | 11.1 $\pm$ 1.1 | 1.78 | -1.01 | 1.07/180 acis6387 | 2006-06-26 | 77.3 | 7271 | 0.782 $\pm$ 0.019 | 1.64 $\pm$ 0.10 | 11.4 $\pm$ 1.2 | 1.76 | -1.05 | 0.93/157 acis6382 | 2005-11-23 | 72.3 | 6515 | 0.772 $\pm$ 0.030 | 1.72 $\pm$ 0.14 | 9.9 $\pm$ 1.4 | 1.78 | -1.04 | 1.17/152 acis1730 | 2000-07-12 | 49.5 | 4855 | 0.800 $\pm$ 0.026 | 1.37 $\pm$ 0.11 | 6.1 $\pm$ 1.3 | 1.77 | -1.12 | 1.15/126 acis7344 | 2006-07-01 | 21.5 | 1711 | 0.873 $\pm$ 0.031 | 1.05 $\pm$ 0.14 | 8.6 $\pm$ 2.8 | 1.75 | -1.16 | 0.69/55 acis6386 | 2005-10-31 | 14.9 | 1491 | 0.786 $\pm$ 0.041 | 1.55 $\pm$ 0.21 | 12.8 $\pm$ 3.2 | 1.77 | -1.07 | 0.95/49 acis7197 | 2005-11-03 | 12.7 | 1117 | 0.892 $\pm$ 0.043 | 0.97 $\pm$ 0.20 | 9.1 $\pm$ 4.1 | 1.75 | -1.17 | 0.99/37 acis7208 | 2005-11-21 | 11.5 | 1014 | 0.678 $\pm$ 0.110 | 1.73 $\pm$ 0.39 | 13.9 $\pm$ 4.6 | 1.78 | -1.10 | 0.81/33 PN0102642301 | 2002-01-27 | 10.0 | 2724 | 0.832 $\pm$ 0.031 | 1.34 $\pm$ 0.14 | 7.3 $\pm$ 1.0 | 1.75 | -1.09 | 0.84/103 PN0102641201 | 2000-08-02 | 10.3 | 1836 | 0.618 $\pm$ 0.056 | 2.51 $\pm$ 0.29 | 12.0 $\pm$ 1.5 | 1.78 | -0.97 | 0.87/69 PN0141980801 | 2003-02-12 | 8.4 | 1596 | 0.636 $\pm$ 0.074 | 2.50 $\pm$ 0.36 | 10.7 $\pm$ 1.5 | 1.78 | -0.96 | 0.90/60 PN0141980601 | 2003-01-23 | 11.6 | 1545 | 0.656 $\pm$ 0.077 | 2.21 $\pm$ 0.32 | 12.5 $\pm$ 1.6 | 1.77 | -1.00 | 1.00/59 M10102642301 | 2002-01-27 | 12.3 | 1199 | 0.841 $\pm$ 0.042 | 1.30 $\pm$ 0.20 | 6.7 $\pm$ 2.1 | 1.75 | -1.10 | 0.89/40 PN0102640401 | 2000-08-02 | 9.2 | 1136 | 0.838 $\pm$ 0.039 | 1.42 $\pm$ 0.20 | 6.8 $\pm$ 1.8 | 1.75 | -1.06 | 1.15/44 M20102642301 | 2002-01-27 | 12.3 | 1135 | 0.839 $\pm$ 0.043 | 1.25 $\pm$ 0.20 | 7.2 $\pm$ 2.2 | 1.75 | -1.12 | 0.83/39 Figure 1: X-ray spectrum of M33 X-7. (upper panel) This spectrum (ObsID 6376) is representative of the four gold spectra (see text). The histogram shows a model that has been fitted to the spectrum (0.3–8.0 keV), which is comprised of only the thermal disk component (kerrbb2) and a low-energy absorption model (phabs). The fit parameters are summarized in Table 1. (lower panel) The fit is good ($\chi_{\nu}^{2}/dof=1.074/180$) and the fit residuals show no systematic structure; in particular, there is no evidence for an Fe-line, absorption edges, or a nonthermal power-law/Comptonization component of emission at higher energies. Figure 2: Spin results for all 15 spectra ordered by total 0.3–8 keV counts. (a) Results based on spectral fits over the energy interval 0.3–8 keV. Filled circles are for the gold Chandra spectra, and the other plotting symbols are for the 11 silver spectra, which include both Chandra ACIS spectra (open circles) and XMM-Newton EPIC spectra (crosses). The indicated uncertainties are at the 90% level of confidence. The dotted line indicates the average spin for the 4 gold spectra, $\bar{a}_{*}=0.776\pm 0.018$. The average for the 11 silver spectra is almost identical, although the dispersion is much greater, $\bar{a}_{*}=0.772\pm 0.098$. (b–d) Same as panel a except the fit interval is as indicated rather than 0.3–8 keV. Figure 3: Effect on the spin parameter $a_{*}$ of varying the input parameters $M$, $i$ and $D$. (a) Spin versus mass $M$ for 3,000 sets of parameters drawn at random. The black filled circle indicate our final adopted estimate of the spin ($\bar{a}_{*0}=0.77\pm 0.02$) for $M$, $i$ and $D$ at their baseline values. (b) Spin versus distance $D$. (c) Spin versus inclination angle $i$. (d) Histogram of spin displacements for 3,000 parameters sets. The vertical solid line indicates the average spin ($\bar{a}_{*}=0.77$). The two dotted lines enclose 68.3% of the spin values centered on the solid line; the half- separation, $\Delta(\bar{a}_{*}-\bar{a}_{*0})=0.053$, represents the $1\sigma$ error in the average spin of $\bar{a}_{*0}=0.77$.

arxiv-papers

2024-09-04T02:48:54.317032

0803.1838

# Python - All a Scientist Needs Julius B. Lucks ###### Abstract Any cutting-edge scientific research project requires a myriad of computational tools for data generation, management, analysis and visualization. Python is a flexible and extensible scientific programming platform that offered the perfect solution in our recent comparative genomics investigation [1]. In this paper, we discuss the challenges of this project, and how the combined power of Biopython [2], Matplotlib [3] and SWIG [4] were utilized for the required computational tasks. We finish by discussing how python goes beyond being a convenient programming language, and promotes good scientific practice by enabling clean code, integration with professional programming techniques such as unit testing, and strong data provenance. ## 1 The Scientists Dilemma A typical scientific research project requires a variety of computational tasks to be performed. At the very heart of every investigation is the generation of data to test hypotheses. An experimental physicist builds instruments to collect light scattering data; a crystallographer collects X-ray diffraction data; a biologist collects fluorescence intensity data for reporter genes, or DNA sequence data for these genes; and a computational researcher writes programs to generate simulation data. All of these scientists use computer programs to control instruments or perform simulations to collect and manage data in an electronic format. Once data is collected, the next task is to analyze it in the context of hypothesis-driven models that help them understand the phenomenon they are studying. In the case of light, or X-ray scattering data, there is a well- proven physical theory that is used to process the data and calculate the observed structure function of the material being studied [5]. This structure function is then compared to predictions made by the hypotheses begin tested. In the case of biological reporter gene data, light intensity is matched up with phenotypic traits or DNA sequences, and statistically analyzed for trends that might explain the observed patterns. As these examples illustrate, across science, the original raw data of each investigation is extensively processed by computational programs in an effort to understand the underlying phenomena. Visualization tools to create a variety of scientific plots are often a preferred tool for both troubleshooting ongoing experiments, and creating publication-quality scientific plots and charts. These plots and charts are often the final product of a scientific investigation in the form of data-rich graphics that demonstrate the truth of a hypothesis compared to its alternatives [6]. Unfortunately, all too often scientists resort to a grab-bag of tools to perform these varied computational tasks. For physicists and theoretical chemists, it is common to use C or FORTRAN to generate simulation data, and C code is used to control experimental apparatus; for biologists, perl is the language of choice to manipulate DNA sequence data [7]. Data analysis is performed in separate, external software packages such as Matlab or Mathematica for equation solving [8, 9], or Stata, SPSS or R for statistical calculations [10, 11, 12]. Furthermore, separate data visualization packages can be used, making the scientific programming toolset extremely varied. Such a mixed bag of tools is an inadequate solution for a variety of reasons. From a computational perspective, most of these tools cannot be pipelined easily which necessitates many manual steps or excessive glue code that most scientists are not trained to write. Far more important than just an inconvenience associated with gluing these tools together is the extreme burden placed on the scientist in terms of data management. In complicated systems, there are often a plethora of different data files in several different formats residing at many different locations. Most tools do not produce adequate metadata for these files, and scientists typically fall back on cryptic file naming schemes to indicate what type of data the files contain and how it was generated. Such complications can easily lead to mistakes. This in turn provides poor at best data provenance when it is in fact of utmost importance in scientific studies where data integrity is the foundation of every conclusion reached and every fact established. Furthermore, when data files are manually moved around from tool to tool, it is not clear if an error is due to program error, or human error in using the wrong file. Analyses can only be repeated by following work flows that have to be manually recorded in a paper or electronic lab notebook. This practice makes steps easily forgotten, and hard to pass on to future generations of scientists, or current peers trying to reproduce scientific results. The Python programming language and associated community tools [13] can help scientists overcome some of these problems by providing a general scientific programming platform that allows scientists to generate, analyze, visualize and manage their data within the same computational framework. Python can be used to generate simulation data, or control instrumentation to capture data. Data analysis can be accomplished in the same way, and there are graphics libraries that can produce scientific charts and graphs. Furthermore python code can be used to glue all of these python solutions together so that visualization code resides alongside the code that generates the data it is applied to. This allows streamlined generation of data and its analysis, which makes data management feasible. Most importantly, such a uniform tool set allows the scientist to record the steps used in data work flows to be written down in python code itself, allowing automatic provenance tracking. In this paper, we outline a recent comparative genomics case study where python and associated community libraries were used as a complete scientific programming platform. We introduce several specific python libraries and tools, and how they were used to facilitate input of standardized biological data, create scientific plots, and provide solutions to speed bottle-necks in the code. Throughout, we provide detailed tutorial-style examples of how these tools were used, and point to resources for further reading on these topics. We conclude with ideas about how python promotes good scientific programing practices, and tips for scientists interested in learning more about python. ## 2 A Comparative Genomics Case Study Figure 1: Lambda phage GenBank file snippet. The full file can be found online - see [15]. LOCUS NC_001416 48502 bp DNA linear PHG 28-NOV-2007 DEFINITION Enterobacteria phage lambda, complete genome. ACCESSION NC_001416 VERSION NC_001416.1 GI:9626243 PROJECT GenomeProject:14204 KEYWORDS . SOURCE Enterobacteria phage lambda ORGANISM Enterobacteria phage lambda Viruses; dsDNA viruses, no RNA stage; Caudovirales; Siphoviridae; Lambda-like viruses. REFERENCE 1 (sites) AUTHORS Chen,C.Y. and Richardson,J.P. TITLE Sequence elements essential for rho-dependent transcription termination at lambda tR1 JOURNAL J. Biol. Chem. 262 (23), 11292-11299 (1987) PUBMED 3038914 ... FEATURES Location/Qualifiers source 1..48502 /organism="Enterobacteria phage lambda" /mol_type="genomic DNA" /specific_host="Escherichia coli" /db_xref="taxon:10710" gene 191..736 /gene="nu1" /locus_tag="lambdap01" /db_xref="GeneID:2703523" CDS 191..736 /gene="nu1" /locus_tag="lambdap01" /codon_start=1 /transl_table=11 /product="DNA packaging protein" /protein_id="NP_040580.1" /db_xref="GI:9626244" /db_xref="GeneID:2703523" /translation="MEVNKKQLADIFGASIRTIQNWQEQGMPVLRGGGKGNEVLYDSA AVIKWYAERDAEIENEKLRREVEELRQASEADLQPGTIEYERHRLTRAQADAQELKNA RDSAEVVETAFCTFVLSRIAGEIASILDGLPLSVQRRFPELENRHVDFLKRDIIKAMN KAAALDELIPGLLSEYIEQSG" ... ORIGIN 1 gggcggcgac ctcgcgggtt ttcgctattt atgaaaattt tccggtttaa ggcgtttccg 61 ttcttcttcg tcataactta atgtttttat ttaaaatacc ctctgaaaag aaaggaaacg 121 acaggtgctg aaagcgaggc tttttggcct ctgtcgtttc ctttctctgt ttttgtccgt 181 ggaatgaaca atggaagtca acaaaaagca gctggctgac attttcggtg cgagtatccg 241 taccattcag aactggcagg aacagggaat gcccgttctg cgaggcggtg gcaagggtaa 301 tgaggtgctt tatgactctg ccgccgtcat aaaatggtat gccgaaaggg atgctgaaat 361 tgagaacgaa aagctgcgcc gggaggttga agaactgcgg caggccagcg aggcagatct 421 ccagccagga actattgagt acgaacgcca tcgacttacg cgtgcgcagg ccgacgcaca ... Recently we performed a comparative genomics study of the genomic DNA sequences of the 74 sequenced bacteriophages that infect _E. coli_ , _P. aeruginosa_ , or _L. lactis_ [1]. Bacteriophages are viruses that infect bacteria. The DNA sequences of these bacteriophages contain important clues as to how the relationship with their host has shaped their evolution. Each virus that we examined has a DNA genome that is a long strand of four nucleotides called Adenine (A), Threonine (T), Cytosine (C), and Guanine (G). The specific sequences of A’s, T’s, C’s and G’s encode for proteins that the virus uses to take over the host bacteria and create more copies of itself. Each protein is encoded in a specific region of the genomic DNA called a gene. Proteins are made up of linear strings of 20 amino acids. There are 4 bases encoding for 20 amino acids, and the translation table that governs the encoding, called the genetic code, is comprised of 3 base triplets called codons. Each codon encodes a specific amino acid. Since there are 64 possible codons, and only 20 amino acids, there is a large degeneracy in the genetic code. For more information on the genetic code, and the biological process of converting DNA sequences into proteins, see [14]. Because of this degeneracy, each protein can be ‘spelled’ as a sequence of codons in many possible ways. The particular sequence of codons used to spell a given protein in a gene is called the gene’s ‘codon usage’. As we found in [1], bacteriophages genomes favor certain codon spellings of genes over the other possibilites. The primary question of our investigation was - does the observed spellings of the bacteriophage genome shed light onto the relationship between the bacteriophage and its host [1]? To address this question, we examined the codon usage of the protein coding genes in these bacteriophages for any non-random patterns compared to all the possible spellings, and performed statistical tests to associate these patterns with certain aspects about the proteins. The computational requirements of this study included: * • Downloading and parsing the genome files for viruses from GenBank in order to get the genomic DNA sequence, the gene regions and annotations: GenBank [16] is maintained by the National Center of Biotechnology Information (NCBI), and is a data wharehouse of freely available DNA sequences. For each virus, we needed to obtain the genomic DNA sequence, the parts of the genome that code for genes, and the annotated function of these genes. Figure 1 displays this information for lambda phage, a well-studied bacteropphage that infects _E. coli_ [14], in GenBank format, obtained from NCBI. Once these files were downloaded and stored, they were parsed for the required information. * • Storing the genomic information: The parsed information was stored in a custom genome python class which also included methods for retrieving the DNA sequences of specific genes. * • Drawing random genomes to compare to the sequenced genome: For each genome, we drew random genomes according to the degeneracy rules of the genetic code so that each random genome would theoretically encode the same proteins as the sequenced genome. These genomes were then visually compared to the sequenced genome through zero-mean cumulative sum plots discussed below. * • Visualize the comparisons through ‘genome landscape’ plots: Genome landscapes are zero-mean cumulative sums, and are useful visual aids when comparing nucleotide frequency properties of the genomes they are constructed from (see [1] for more information). Genome landscapes were computed for both the sequenced genome, and each drawn genome. The genome landscape of the sequenced genome was compared to the distribution of genome landscapes generated from the random genomes to detect regions of the genomes that have extremely non- random patterns in codon usage. * • Statistically analyzing the non-random regions with annotation and host information: To understand the observed trends, we performed analysis of variance (ANOVA) [17] analysis to detect correlations between protein function annotation or host lifestyle information with these regions. Python was used in every aspect of this computational work flow. Below we discuss in more detail how python was used in several of these areas specifically, and provide illustrative tutorial-style examples. For more information on the details of the computational work flow, and the biological hypotheses we tested, see [1]. For specific details on the versions of software used in this paper, and links to free downloads, see Materials and Methods. ## 3 Biopython Biopython is an open-source suite of bioinfomatics tools for the python language [2]. The suite is comprehensive in scope, and offers python modules and routines to parse bio-database files, facilitate the computation of alignments between biological sequences (DNA and protein), interact with biological web-services such as those provided by NCBI, and examine protein crystallographic data to name a few. In this project, Biopython was used both to download and parse genomic viral DNA sequence files from the NCBI Genbank database [16] as outlined in Listing 1. Listing 1: Downloading and parsing the GenBank genome file for lambda phage (refseq number NC_001416). ⬇ #genbank.py - utilities for downloading and parsing GenBank files from Bio import GenBank #(1) from Bio import SeqIO def download(accession_list): '''Download and save all GenBank records in accession_list.''' try: handle = GenBank.download_many(accession_list) #(2) except: print ”Are you connected to the internet?” raise genbank_strings = handle.read().split('//\n') #(3) for i in range(len(accession_list)): #Save raw file as .gb gb_file_name = accession_list[i]+'.gb' f = open(gb_file_name,'w') f.write(genbank_strings[i]) #(4) f.write('//\n') f.close() def parse(accession_list): '''Parse all records in accession_list.''' parsed = [] for accession_number in accession_list: gb_file_name = accession_number+'.gb' print 'Parsing … ',accession_number try: gb_file = file(gb_file_name,'r') except IOError: print 'Is the file %s downloaded?' % gb_file_name raise gb_parsed_record = SeqIO.parse(gb_file,”genbank”).next() #(5) gb_file.close() print gb_parsed_record.id #(6) print gb_parsed_record.seq parsed.append(gb_parsed_record) #(7) return parsed import genbank #(8) genbank.download(['NC_001416']) genbank.parse(['NC_001416']) #(1) The biopython module is called Bio. The Bio.Genbank module is used to download records from GenBank, and the Bio.SeqIO module provides a general interface for parsing a variety of biological formats, including GenBank. #(2) The Bio.GenBank.download_many method is used in the genbank.download method to download Genbank records over the internet. It takes a list of GenBank accession numbers identifying the records to be downloaded. #(3) GenBank records are separated by the character string ``//=. Here we manually separate GenBank files that are part of the same character string. #(4) When we save the GenBank records as individual files to disk, we include the ``//= separator again. #(5) The Bio.SeqIO.parse method can parse a variety of formats. Here we use it to parse the GenBank files on our local disk using the ”genbank” format parameter. The method returns a generator, who’s next() method is used to retrieve an object representing the parsed file. #(6) The object representing the parsed GenBank file has a variety of methods to extract the record id and sequence. See Listing 2 for more details. #(7) The genbank.parse method returns a listed of parsed objects, one for each input sequence file. #(8) To run the code in genbank.py, Biopython 1.44 must first be installed (see Materials and Methods). Executing the following code should create a file called ‘NC_001416.gb’ on the local disk (see Figure 1), as well as produce the following output: ''' Parsing … NC_001416 NC_001416.1 Seq('GGGCGGCGACCTCGCGGGTTTTCGCTATTTATGAAAATTTTCCGGTTTAAGGCGTTTCCG …', IUPACAmbiguousDNA()) ''' The benefits of using Biopython in this project were several including: 1. 1. Not having to write or maintain this code ourselves. This is an important point as the number of web-available databases and services grows. These often change rapidly, and require rigorous maintenance to keep up with tweaks to API’s and formats - a monumental task that is completed by an international group of volunteers for the Biopython project. 2. 2. The Biopython parsing code can be wrapped in custom classes that make sense for a particular project. Listing 2 illustrates the latter by outlining a custom genome class used in this project to store the location of coding sequences for genes (CDS_seq). Listing 2: A custom Genome class which wraps the biopython parsing code outlined in Listing 1. ⬇ #genome.py - a custom genome class which wraps biopython parsing code import genbank #(1) from Bio import Seq from Bio.Alphabet import IUPAC class Genome(object): ”””Genome - representing a genomic DNA sequence with genes Genome.genes[i] returns the CDS sequences for each gene i.””” def __init__(self, accession_number): genbank.download([accession_number]) #(2) self.parsed_genbank = genbank.parse([accession_number])[0] self.genes = [] self._parse_genes() def _parse_genes(self): ”””Parse out the CDS sequence for each gene.””” for feature in self.parsed\_genbank.features: #(3) if feature.type == 'CDS': #Build up a list of (start,end) tuples that will #be used to slice the sequence in self.parsed_genbank.seq # #Biopython locations are zero-based so can be directly #used in sequence splicing locations = [] if len(feature.sub_features): #(4) #If there are sub_features, then this gene is made up #of multiple parts. Store the start and end positins #for each part. for sf in feature.sub_features: locations.append((sf.location.start.position, sf.location.end.position)) else: #This gene is made up of one part. Store its start and #end position. locations.append((feature.location.start.position, feature.location.end.position)) #Store the joined sequence and nucleotide indices forming #the CDS. seq = '' #(5) for begin,end in locations: seq += self.parsed_genbank.seq[begin:end].tostring() #Reverse complement the sequence if the CDS is on #the minus strand if feature.strand == -1: #(6) seq_obj = Seq.Seq(seq,IUPAC.ambiguous_dna) seq = seq_obj.reverse_complement().tostring() #append the gene sequence self.genes.append(seq) #(7) #(1) Here we import the genbank module outlined in Listing 1, along with two more biopython modules. The Bio.Seq module has methods for creating DNA sequence objects used later in the code, and the Bio.Alphabet module contains definitions for the types of sequences to be used. In particular we use the Bio.Alphabet.IUPAC definitions. #(2) We use the genbank methods to download and parse the GenBank record for the input accession number. #(3) The parsed object stores the different parts of the GenBank file as a list of features. Each feature has a type, and in this case, we are looking for features with type ’CDS’, which stores the coding sequence of a gene. #(4) For many organisms, genes are not contiguous stretches of DNA, but rather are composed of several parts. For GenBank files, this is indicated by a feature having sub_features. Here we gather the start and end positions of all sub features, and store them in a list of 2-tuples. In the case that the gene is a contiguous piece of DNA, there is only one element in this list. #(5) Once the start and end positions of each piece of the gene are obtained, we use them to slice the seq of the parsed_genbank object, and collect the concatenated sequence into a string. #(6) Since DNA has polarity, there is a difference between a gene that is encoded on the top, plus strand, and the bottom, minus strand. The strand that the gene is encoded in is stored in feature.strand. If the strand is the minus strand, we need to reverse compliment the sequence to get the actual coding sequence of the gene. To do this we use the Bio.Seq module to first build a sequence, then use the reverse_complement() method to return the reverse compliment. # (7) We store each gene as an element of the Genome.genes list. The CDS of the ith gene is then retrievable through Genome.genes[i]. For a more detailed introduction to the plethora of biopython features, as well as introductory information into python see [18] . ## 4 Matplotlib Matplotlib [3] is a suite of open-source python modules that provide a framework for creating scientific plots similar to the Matlab [8] graphical tools. In this project, matplotlib was used to create genome landscape plots both to have a quick look at data as it was generated, and to produce publication quality figures. Genome landscapes are cumulative sums of a zero- mean sequence of numbers, and are useful visualization tools for understanding the distribution of nucleotides across a genome (see [1] for more information). Listing 3 outlines how matplotlib was used to quickly generate graphics to test raw simulation data as it was being generated. Listing 3: Sample matplotlib script that calculates and plots the zero-mean cumulative sum of the numbers listed in a single column of an input file. ⬇ #landscape.py - plotting a zero-mean cumulative sum of numbers import fileinput #(1) import numpy from matplotlib import pylab def plot(filename): ”””Read single-column numbers in filename and plot zero-mean cumulative sum””” numbers = [] for line in fileinput.input(filename): #(2) numbers.append(float(line.split('\n')[0])) mean = numpy.mean(numbers) #(3) cumulative_sum = numpy.cumsum([number - mean for number in numbers]) pylab.plot(cumulative_sum[0::10],'k-') #(4) pylab.xlabel('i') pylab.title('Zero Mean Cumulative Sum') pylab.savefig(filename+'.png') #(5) pylab.show() #(1) We use several python community modules to plot the zero-mean cumulative sum. As part of the python standard library, fileinput can be used as a quick an easy solution to reading in a file containing a column of entries. numpy is a comprehensive python project aimed at providing numerical routines for scientific applications [19]. Finally we import the matplotlib.pylab module which provides a Matlab-like plotting environment. #(2) Here we use fileinput to read successive lines of the input file, which takes care of opening and closing the input file automatically. Notice that we split each line by the newline character ``=, and take everything to the left of it, assuming that each line contains a single number. #(3) The numpy module provides many convenient methods such as mean to compute the ``mean= of a list of numbers, and ``cumsum= which computes the cumulative sum. To shift the input numbers by the mean, we use a python list comprehension to subtract the mean from each number, and then input the shifted list to numpy.cumsum. #(4) The pylab module presents a Matlab-like plotting environment. Here we use several methods to create a basic line plot with an xlabel and title. #(5) To view the plot, we use pylab.show(), after we have saved the figure as a PNG file using pylab.savefig. The following script uses the genome class outlined in Listing 2, along with the landscape class to plot the GC-landscape for the lambda phage genome. The genome class is used to download and parse the GenBank file for lambda phage. Each gene sequence is then scanned for ’G’ or ’C’ nucleotides. For every ’G’ or ’C’ nucleotide encountered, a 1 is appended to the list GC; for every ’A’ or ’T’ encountered, a 0 is appended. This sequence of 1’s and 0’s representing the GC-content of the lambda phage genome is saved in a file, and input into the landscape.plot method. A plot corresponding to executing this script is shown in Figure 2. import genome,landscape lambda_phage = genome.Genome('NC_001416') GC = [] for gene_sequence in lambda_phage.genes: for nucleotide in gene_sequence: if nucleotide == 'G' or nucleotide == 'C': GC.append(1) else: GC.append(0) f = file('NC_001416.GC','w') for num in GC: f.write('%i\n' % num) f.close() landscape.plot('NC_001416.GC') Figure 2: The lambda phage GC-landscape generated by the sample code in Listing 3. Matplotlib was also used to make custom graphics classes for creating publication-quality plots. To do this, we used the object oriented interface to matplotlib plotting routines to inherit funcionality in our classes. The benefits of using matplotlib in this project were several: 1. 1. The code that produced the scientific plots resided alongside the code that produced the underlying data for the plots. The importance of this cannot be stressed enough as having the code structured in this way removed many opportunities for human error involved in manually shuffling raw data files into separate graphical programs. Moreover, the instructions for producing the plots from the underlying raw data was _python code_ , which not only described these instructions, but could be executed to produce the plots. Imagine instead the often practiced use of spreadsheets to create plots from raw data - in these spreadsheets, formulas are hidden by the results of the calculations, and it is often very confusing to construct a picture of the computational flow used to produce a specific plot. 2. 2. Having the graphics instructions in code allowed for quick trouble shooting when creating the plots, or evaluating raw data as it was generated. 3. 3. Complicated plots were easily regenerated by tweaking the code for particular graphical plots. ## 5 SWIG The Simple Wrapper and Interface Generator (SWIG) [4], is an easy-to-use system for extending python. In particular, it allows the speed up of selected parts of an application by writing these routines in another more low-level language such as C or C++. Furthermore, SWIG implements the use of this low- level code using the standard python module importing structure. This allows developers to first prototype code in python, then re-implement the code in C and SWIG causing _no change_ in the python code that uses the re-implemented module. This project relied heavily on drawing random numbers from an input discrete distribution. For example, we often needed to draw a sequence of A’s, T’s, C’s or G’s corresponding to the nucleotide sequence of the genome, but preserving the genomic distribution of these four nucleotide bases. For some viruses, the distribution might look like: $P_{A}=0.2$, $P_{T}=0.2$, $P_{C}=0.3$, $P_{G}=0.3$, with $P_{A}+P_{T}+P_{C}+P_{G}=1.0$. Listing 4 illustrates the outline of a python module that has methods to draw numbers according to a discrete distribution with 4 possible outcomes. It also illustrates how this module could be implemented in C, and included in a python module with SWIG. Listing 4: Drawing random numbers from a specified discrete distribution with four possibilities implemented in python. ⬇ #module discrete_distribution.py - drawing numbers from a discrete probability distribution import random #(1) def seed(): #(2) random.seed() def draw(distribution): #(3) '''Drawing an index according to distribution. distribution is a list of floating point numbers, one for each index number, representing the probability of drawing that index number. Example: [0.5, 0.5] would represent equal probabilities of returning a 0 or 1. ''' sum = 0 #(4) r = random.random() for i in range(0,len(distribution)): sum += distribution[i] if r ¡ sum: return i import discrete_distribution #(5) discrete_distribution.seed() print sum([discrete_distribution.draw([0.2,0.2,0.3,0.3]) for x in range(10000)])/10000. #(1) Import the random number generator. #(2) We use the discrete_distribution.seed() method to seed the random number generator. If no arguments are supplied to random.seed(), the system time is used to seed the number generator [20]. #(3) The draw function takes an argument distribution, which is a list of floating point numbers. #(4) The algorithm for drawing a number according to a discrete distribution is to draw a number, r, from a uniform distribution on [0,1]; compute a cumulative sum of the probabilities in the discrete distribution for successive indices of the distribution; when r is less than this cumulative sum, return the index that the cumulative sum is at. # (5) To test this code, plug in a distribution [0.2,0.2,0.3,0.3], draw 10000 numbers from this distribution, and compute the mean, which theoretically should be $0*0.2 + 1*0.2 + 2*0.3 + 3*0.3 = 1.7$. In this case, when this code was executed, the result $1.7013$ was returned. In Listing 5, we implement this routine using C, and use SWIG to create a python module of the C implementation. Listing 5: Drawing random numbers from a specified discrete distribution with four possibilities implemented in C with SWIG. ⬇ //c_discrete_distribution.c - A C implementation of the discrete_distribution.py module #include ”stdlib.h” //(1) #include ”stdio.h” #include ”time.h” void seed() { srand((unsigned) time(NULL) * getpid()); } int draw(float distribution[4]) { //(2) float r= ((float) rand() / (float) RAND_MAX); float sum = 0.; int i = 0; for(i = 0; i ¡ 4; i++) { sum += distribution[i]; if (r lt sum) { return i; } } } //(1) Here we define two functions, seed and draw, which correspond to the python methods in discrete_distribution.py. Note that the python implementation of discrete_distribution.draw() worked with distributions of arbitrary numbers of elements. For simplicity, we are restricting the C implementation to work with distributions of length 4. //(2) The draw routine is implemented using the same algorithm as in the python implementation. For simplicity, we use the C standard library rand() routine, although there are more advanced random number generators that would be more appropriate for scientific applications [21]. (Note that the ‘lt’ symbol should be replaced by ‘¡’ when executing the code.) //c_discrete_distribution.i - A Swig interface file for the c_discrete_distribution module // (3) %module c_discrete_distribution //(4) //Grab a 4 element array as a Python 4-list // (5) %typemap(in) float[4](float temp[4]) { //temp[4] becomes a local variable int i; if (PyList_Check($input)) { PyObject* input_to_tuple = PyList_AsTuple($input); if (!PyArg_ParseTuple(input_to_tuple,”ffff”,temp,temp+1,temp+2,temp+3)) { PyErr_SetString(PyExc_TypeError,”tuple must have 4 elements”); return NULL; } $1 = &temp[0]; } else { PyErr_SetString(PyExc_TypeError,”expected a tuple.”); return NULL; } } void seed(); //(6) int draw(float distribution[4]); //(3) To use SWIG, we create a swig interface file that describes how to translate python inputs to the C code, and C outputs to the python code. //(4) SWIG directives are preceded by the % sign. Here we declare that the module we are going to make is called c_discrete_distribution. In general, the module name, the C source name, and the interface file name should all be the same outside of the file extension. //(5) SWIG will automatically handle the conversion of many data-types from python to C and C to python. For illustration purposes, we create an explicit typemap which converts a 4-element python list into a 4 element C list of floats. Since we are using the typemap(in) directive, SWIG knows that we are converting python to C. The rest of the code checks that a list was passed from python to C, and the list has 4 elements. If these conditions are not met, python errors are thrown. If they are met, an array of floats called temp is called, and passed to C. This conversion is adapted from the SWIG reference manual [4]. // (6) The last thing to do in the SWIG interface file is to declare the function signatures of the C implementation. To use this module outlined in Listing 5, we have to call swig to generate wrapper code, then compile and link our code with the wrapper code. With SWIG installed, the procedure would look something like swig -python -o c_discrete_distribution_wrap.c c_discrete_distribution.i We first use SWIG to generate the wrapper code. Using the c_discrete_distribution.i interface file, SWIG will generate c_discrete_distribution_wrap.c using the Python C API, since we specified the -python flag. In addition, SWIG will also generate c_discrete_distribution.py, which we will use to import the module into our code. gcc -c c_discrete_distribution.c c_discrete_distribution_wrap.c -I/usr/include/python2.5 -I/usr/lib/python2.5 Next we use a C compiler to compile each of the C files (our C source, and the SWIG generated wrapper). We have to include the python header files and libraries for the python version we are using. In our case, we used python 2.5. After this procedure completes, we should have two additional files: c_discrete_distribution.o and c_discrete_distribution_wrap.o . gcc -bundle -flat_namespace -undefined suppress -o _c_discrete_distribution.so c_discrete_distribution.o c_discrete_distribution_wrap.o The final step is to link them all together. The linking options are platform dependent, and the official SWIG documentation should be consulted [4]. For Mac OS X, we use the “-bundle -flat_namespace -undefined suppress” options for gcc. When this step is done, the file _c_discrete_distribution.so is created. The python module file c_discrete_distribution.py can be used in the same way as in Listing 4 above, import c_discrete_distribution as discrete_distribution discrete_distribution.seed() print sum([discrete_distribution.draw([0.2,0.2,0.3,0.3]) for x in range(10000)])/10000. which produces the number 1.6942. The benefits of using SWIG in this project were several: 1. 1. We used all the benefits of python with the increased speed for critical bottlenecks of our simulation code. 2. 2. The parts that were sped up were used in the exact same context through the python module import structure, removing the need for glue code to tie in external C-programs. More generally, SWIG allows scientists using python to leverage experience in other languages that they typically have, while staying within the python framework with all its benefits outlined above. This promotes a scientific work flow which consists of prototyping simulation code using the more simple python, then profiling the python code to identify the speed bottlenecks. These can then be re-implemented in C or C++ and wrapped into the existing python code using SWIG. This is a much preferred methodology than writing unnecessarily complicated and error-prone C programs, and using glue code to integrate them within the larger simulation methodology. ## 6 Conclusions There are several practical conclusions to draw for scientists. The first is that python, and its associated modules supported by the python community, offer a general platform for computing that is useful accross a broad range of scientific disciplines. We have only outlined several such tools in this article, but there exist many more relevant to scientists [22]. The second is that python and its community modules can _easily_ be used by scientists. The clean nature of the code is quick to learn, and its high-level features make complicated tasks quick to accomplish. We have not discussed the interactive programming environments offered by python[23, 24], which when combined with the power of the language makes prototyping ideas and algorithms extremely easy. The bigger picture conclusion is that python promotes good scientific practice. The code readability and package structure enables code to be easily understood by different researchers working on the same project. In fact, python code is often self-documenting which allows researchers to go back to code they wrote in the past and easily understand it. Python and its community modules provide a consistent framework to generate data, and shuttle it to the various analysis tasks. This in turn promotes data provenance through a written record _in code_ of every step used to analyze specific data, which removes many manual steps, and thus many errors. Finally, by using python, scientists can start to use other community tools and practices originally designed for professional programmers, but also useful to scientists. The most important of these, but not discussed in this article, is unit testing, whereby test code is written alongside scientific code that tests to see if that code is working properly. This allows scientists to re-write aspects of the code, perhaps using a different algorithm, and to re-run the tests to see if it still works as they think it should. For large projects this is critical, and removes the need for often- used ad-hoc practices of looking at some sample data by eye, which is not only tedious, but not guaranteed to uncover subtle numerical bugs that could cause crucial mis-interpretation of scientific data. Since python is a well-established language and has a large and active community, the resources available for beginners can be overwhelming. For the scientist interested in learning more about scientific programming in python, we recommend visiting the web page and mailing lists of the SciPy project for an introduction to scientific modules [22], and [25, 26] for excellent introductory python tutorials. ## 7 Materials and Methods All code examples in this paper were written by the author. The particular versions of the relevant software used were: Python 2.5, Biopython 1.44, MatPlotLib 0.91.2, and SWIG 1.3.33. Documentation and free downloads of this software are available at the following URLs: * • Python - http://python.org * • Biopython - http://biopython.org * • MatPlotLib - http://matplotlib.sourceforge.net * • SWIG - http://www.swig.org/ The source code for all the listings above, as well as the original and maintained version of this article can be found at `http://openwetware.org/wiki/Julius_B._Lucks/Projects/Python_All_A_Scientist_Needs`. ## 8 Acknowledgements The author would like to thank Adrian Del Maestro, Joao Xavier, David Thompson and Stanley Qi for helpful comments during the preparation of this manuscript. The author also thanks the Miller Institute for Basic Research in Science at the University of California, Berkeley for support. ## 9 References and Resources ## References * [1] J. B. Lucks, D. R. Nelson, G. Kudla, J. B. Plotkin. _Genome landscapes and bacteriophage codon usage_ , PLoS Computational Biology, 4, .1000001, 2008. (doi:10.1371/journal.pcbi.1000001) * [2] The Biopython project homepage is at http://biopython.org, and the documentation can be found at http://biopython.org/wiki/Documentation. * [3] The Matplotlib project homepage is at http://matplotlib.sourceforge.net, where the documentation can also be found. * [4] The Simple Wrapper and Interface Generator (SWIG) project homepage is at http://www.swig.org, and the documentation can be found at http://www.swig.org/doc.html. * [5] N. Ashcroft and N. Mermin _Solid State Physics_ (Holt, Reinhart and Winston, New York 1976). * [6] E. Tufte _The Visual Display of Quantitative Information 2nd Ed._ (Graphics Press, Cheshire CT). * [7] L. Stein _How Perl Saved the Human Genome Project_ (http://www.bioperl.org/wiki/How_Perl_saved_human_genome) * [8] The Matlab programming environment is developed by Mathworks - http://www.mathworks.com/. * [9] The Mathematica software is developed by Wolfram Research - http://www.wolfram.com/. * [10] The Stata statistical software is developed by StataCorp- http://www.stata.com/. * [11] The SPSS statistical software is developed by SPSS - http://www.spss.com/. * [12] The R statistical programming project homepage is at http://www.r-project.org/. * [13] The Python project homepage is at http://python.org. * [14] Alberts, Johnson, Lewis, Raff, Roberts, Walter _Molecular Biology of the Cell 4th Ed_ (Garland Science). * [15] The GenBank file for lambda phage can be downloaded by querying the Genbank nucleotide database with the keyword NC_001416 at http://www.ncbi.nlm.nih.gov/entrez/ * [16] The GenBank data repository can be found at http://www.ncbi.nlm.nih.gov/Genbank/. * [17] For information on the Analysis of Variances Statistical Method, see Julian Faraway _Practical Regression and Anova using R_ , which can be found at http://cran.r-project.org/other-docs.html. * [18] S. Bassi. _A Primer on Python for Life Science Researchers_ , PLoS Comput Biol, 3, e199, 2007. (doi:10.1371/journal.pcbi.0030199) * [19] The numpy project provides a numerical backend for scientific applications. The project homepage is at http://numpy.scipy.org/. * [20] The python random module documentation can be found at http://docs.python.org/lib/module-random.html. * [21] For an extensive discussion of random numbers, see _Numerical Recipes in C_ , Chapter 7, which can be found at http://www.nrbook.com/a/bookcpdf.php. * [22] The SciPy project is aimed at collecting and developing scientific tools for python. The project homepage is at http://www.scipy.org/. * [23] The ipython project can be found at http://ipython.scipy.org/. * [24] F. Perez and B. Granger. _IPython: A System for Interactive Scientific Computing_ ”, Computing in Science and Engineering, 2007. http://ieeexplore.ieee.org/iel5/5992/4160244/04160251.pdf?arnumber=4160251. * [25] C. H. Swaroop _A Byte of Python_. http://www.ibiblio.org/swaroopch/byteofpython/read/. * [26] M. Pilgrim _Dive Into Python_. http://www.diveintopython.org/.

arxiv-papers

2024-09-04T02:48:54.321722

0803.1986

# Companion Stars of Type Ia Supernovae Z. Han National Astronomical Observatories / Yunnan Observatory, the Chinese Academy of Sciences, Kunming, 650011, China zhanwenhan@hotmail.com ###### Abstract The WD+MS channel of the single-degenerate scenario is currently favourable for progenitors of type Ia supernovae (SNe Ia). Incorporating the results of detailed binary evolution calculations for this channel into the latest version of a binary population synthesis code, I obtained the distributions of many properties of the companion stars at the moment of SN explosion. The properties can be verified by future observations. binaries: close — stars: evolution — supernovae: general — white dwarfs ††slugcomment: submitted to ApJ Letters ## 1 Introduction Type Ia supernovae (SNe Ia) have been used as a calibrated candle to probe the dynamics of the universe, leading to a significant progress in cosmology, i.e. the determination of $\Lambda$ and $\Omega$ (Riess et al., 1998; Perlmutter et al., 1999). They are believed to be thermonuclear explosions of carbon-oxygen (CO) white dwarfs (WDs). However the nature of their progenitors has remained unclear, and this still raises doubts as to the calibration which is purely empirical and based on the nearby SN Ia sample. Based on the characteristics of observed SNe Ia (e.g. light curves, chemical stratification), it seems most likely that they occur when the accreting CO WDs reach the Chandrasekhar limit, and sub-Chandrasekhar models appear not to be consistent with the observations of SNe Ia. Chandrasehar-mass WDs can be created through the single-degenerate scenario, where the CO WD accretes mass from a non- degenerate companion (Nomoto, Thielemann & Yokoi, 1984; Hachisu, Kato & Nomoto, 1999), or the double degenerate scenario, where two CO WDs with a total mass larger than the Chandrasekhar mass coalesce (Iben & Tutukov, 1984; Webbink & Iben, 1987)111Note, it is quite likely that the merger product experiences core collapse rather than a thermonuclear explosion.. Recent observations indicate that there is a wide spread of delay time between star formation and SN Ia explosion, and this imply that there exist two populations of progenitors, a “prompt” one with a delay time less than $\sim 0.1\,{\rm Gyr}$ and a “tardy” one with a delay time of $\sim 3\,{\rm Gyr}$ (Mannucci, Della Valle & Panagia, 2006). Using the archival data prior to explosion from Hubble Space Telescope (HST) and Chandra, Voss & Nelemans (2008) tried to search for a possible progenitor of SN2007on (type Ia) in elliptical galaxy NGC 1404. They discovered an object at the position of SN2007on. The X-ray luminosity of the object and the non- detection in the optical images are fully consistent with the single- degenerate scenario, and the the discovery therefore favours the single- degenerate scenario. However, new Chandra X-ray observations (Roelofs et al., 2008) and detailed astrometry of the site of SN2007on showed that there appears to be an offset between the SN and the X-ray source, and the offset means a small probability ($\sim 1\%$) of the X-ray source being related to the SN. However, the X-ray source is not unconnected with the SN, as the X-ray source has dimmed after the explosion and the source before the explosion showed an excess of soft X-ray photons relative to the other sources in the field. Kepler’s 1604 supernova is suggested to be of type Ia. Reynolds et al. (2007) made a deep Chandra observation of Kepler’s supernova remnant and concluded that Kepler’s SN is a SN Ia with circumstellar medium (CSM) interaction. This might indicate that the progenitor is a massive (young) star. Maoz & Mannucci (2008) searched HST pre-explosion images for NGC 1316, a radio galaxy in the Fornax cluster and a prolific producer of SNe Ia, for evolved intermediate-mass progenitor stars. The pre-explosion images (3 years before explosion) of the sites of SN2006dd (a normal SN Ia) and SN2006mr (likely a subluminous SN Ia) show no potential luminous stellar progenitors. More effort is still needed in identifying the progenitors of SNe Ia. Among various possible progenitor models, the WD+MS channel of the single- degenerate scenario is the most widely accepted one, in which a CO WD accretes mass from its main-sequence (MS) star or a slightly evolved subgiant in a close binary system until it reaches a mass of $\sim 1.378M_{\odot}$ and explodes as a SN Ia (Nomoto, Thielemann & Yokoi, 1984; Li & van den Heuvel, 1997; Hachisu, Kato & Nomoto, 1999; Langer et al., 2000; Han & Podsiadlowski, 2004). The companion star should survive the explosion and show distinguishing properties, and it is therefore a promising method to test progenitor models by identifying the surviving companion stars of SNe Ia. Tycho Brahe’s 1572 supernova is a Galactic SN Ia. Ruiz-Lapuente et al. (2004) found in the remnant region that Tycho G, a star similar to the Sun but with a lower gravity, moves at more than three times the mean velocity of the stars there. They argued that Tycho G could be the surviving companion of the supernova222However, Fuhrmann (2005) argued that there exists a possibility of Tycho G being a thick disk star coincidentally passing in the vicinity of the remnant of SN 1572.. Indeed, a surviving companion would have a high space velocity and evolve to a WD finally, and the single-degenerate scenario could potentially explain the properties of halo WDs observed by Oppenheimer et al. (2001), e.g. their space density and ages (Hansen, 2003; Bergeron, 2003). Furthermore, Justham et al. (2008) argued that the ultra-cool WDs observed by Wolf (2005) might have been formed via the single-degenerate scenario, i.e. they might be the remnants of the non-degenerate donor stars. Note, however, there have been no conclusive proof yet that any individual object is the surviving companion of a SN Ia. Langer et al. (2000) did detailed binary evolution calculations for part of the WD+MS channel of the single-degenerate scenario, and presented the properties, e.g. orbital velocities, luminosities, effective temperatures, of the companion stars at the moment of SN Ia explosion for the sample WD binaries they studied, in which the Roche lobe overflow (RLOF) starts when the companion star is in MS. However, the distributions of the properties have not been obtained. Canal, Méndez & Ruiz-Lapuente (2001) run a Monte Carlo scenario code to calculate the distributions of masses, luminosities and velocities of the companion stars of SNe Ia. However, they have not done detailed binary evolution calculations for the production of SNe Ia, and this results in big uncertainties (see Han & Podsiadlowski, 2004), e.g. the orbital velocities obtained for the companion stars from the WD+MS channel are over 450 km/s, which is far too high. Han & Podsiadlowski (2004, hereafter HP04) carried out detailed binary evolution calculations for the WD+MS channel for about 2300 close WD binaries, in which RLOF starts when the companion star is in MS or in Hertzsprung gap (i.e. slightly evolved). The study is comprehensive, and various properties of the companion stars were obtained but not sorted for publishing. In this Letter, I extract the properties from the data files of the calculations and incorporate them into the latest version of the binary population synthesis (BPS) code developed for the study of various binary-related objects (Han, Podsiadlowski & Eggleton, 1995; Han et al., 1995; Han, 1998; Han et al., 2002, 2003), including the progenitors of SNe Ia (Han & Podsiadlowski, 2004, 2006), and obtain the distributions of the properties. ## 2 The distributions of properties of the companion stars In the single-degenerate scenario, the progenitor of a SN Ia is a close WD binary system, which has most likely emerged from common envelope (CE) evolution (Paczyński, 1976) of a giant binary system. During the CE evolution, the envelope engulfs the core (here a CO WD) of the giant and the secondary, and the orbital energy released in the spiral-in process (i.e. orbital decay) is used to overcome the binding energy of the CE in order to eject it. For the CE evolution, I have two parameters: $\alpha_{\rm CE}$ the CE ejection efficiency, i.e. the fraction of the released orbital energy used to overcome the binding energy, and $\alpha_{\rm th}$, which defines the fraction of the thermal energy contributing to the binding energy of the CE. As in previous studies, I adopted $\alpha_{\rm CE}=\alpha_{\rm th}=1.0$, which gives good matches between theory and observations for many binary-related objects 333 The prescription adopted here for the CE evolution is different from the $\lambda$ prescription, but appears to be more physical. See Han, Podsiadlowski & Eggleton (1995), Dewi & Tauris (2000) and Podsiadlowski, Rappaport & Han (2003) for details. . To obtain the distributions of properties of companion stars at the moment of SN explosion, I have performed a detailed Monte Carlo simulation with the latest version of the BPS code. The code follows the evolution of binaries with their properties being recorded at every step. If a binary system evolves to a WD+MS system, and if the system, at the beginning of the RLOF phase, is located in the SN Ia production regions in the plane of ($\log P^{\rm i}$, $M_{2}^{\rm i}$) for its $M_{\rm WD}^{\rm i}$, where $P^{\rm i}$, $M_{2}^{\rm i}$ and $M_{\rm WD}^{\rm i}$ are, respectively, the orbital period, the secondary’s mass and the WD’s mass of the WD+MS system at the beginning of the RLOF (see Fig. 3 of HP04), I assume that a SN Ia is resulted, and the properties of the WD binary at the moment of SN explosion are obtained by interpolation in the 3-dimensional grid ($M_{\rm WD}^{\rm i}$, $M_{2}^{\rm i}$, $\log P^{\rm i}$) of the $\sim 2300$ close WD binaries calculated in HP04. In the simulation, I follow the evolution of 100 million sample binaries according to grids of stellar models of metallicity $Z=0.02$ and the evolution channels leading to SNe Ia as described in HP04. I adopted the following input for the simulation (see Han, Podsiadlowski & Eggleton, 1995). (1) The star- formation rate (SFR) is taken to be constant over the last 15 Gyr. (2) The initial mass function (IMF) of Miller & Scalo (1979) is adopted. (3) The mass- ratio distribution is taken to be constant. (4) The distribution of separations is taken to be constant in $\log a$ for wide binaries, where $a$ is the orbital separation. (5) The orbits are assumed to be circular. The simulation gives current-epoch-snapshot distributions of many properties of companion stars at the moment of SN explosion, e.g. the masses, the orbital periods, the orbital separations, the orbital velocities, the effective temperatures, the luminosities, the surface gravities, the surface abundances, the mass transfer rates, the mass loss rates of the optically thick stellar winds. The simulation also shows the initial parameters of the primordial binaries and the WD binaries that lead to SNe Ia. Figs 1\- 7 are selected distributions that may be helpful to identify the progenitors of SNe Ia. ## 3 Discussion Figs 1 and 2 are the distributions of the masses, the orbital velocities 444The Chandrasekhar-mass WD has an orbital velocity of $\sim 50$ to $\sim 200\,{\rm km/s}$ for a corresponding companion star’s mass of $\sim 0.6$ to $\sim 2.0M_{\odot}$ at the moment of SN explosion., the effective temperatures and the surface gravities of companion stars at the moment of SN explosion. Tycho G was taken as the surviving companion star of Tycho Brahe’s 1572 supernova by Ruiz-Lapuente et al. (2004). It has a space velocity of $136\,{\rm km/s}$, more than 3 times the mean velocity there. Its surface gravity is $\log\,(g/{\rm cm}\,{\rm s}^{-2})=3.5\pm 0.5$, while the effective temperature is $T_{\rm eff}=5750\pm 250{\rm K}$. The parameters are compatible with Figs 1 and 2. As from our simulation, the recorded properties at each step show that a primordial binary system with a primary mass $M_{\rm 1i}\sim 4-5.5M_{\odot}$, a secondary mass $M_{\rm 2i}\sim 2-3M_{\odot}$ and an orbital period $P_{\rm i}\sim 100-250\,{\rm d}$ would evolve to a close WD binary system with a WD mass $M^{\rm i}_{\rm WD}\sim 0.8-1.2M_{\odot}$, a secondary mass $M^{\rm i}_{\rm 2}\sim 2-3M_{\odot}$, and an orbital period $P^{\rm i}\sim 1-4\,{\rm d}$. The WD binary results in SN Ia explosion with companion parameters ($T_{\rm eff}$ and $\log g$ actually) in the range of Tycho G. However, Figs 1 and 2 are for that at the moment of SN explosion, the distributions could be modified due to the explosion. Marietta, Burrows & Fryxell (2000) presented several high-resolution two-dimensional numerical simulations of the impacts of SN Ia explosion with companions. The impact make the companion in the WD+MS channel lose a mass of 0.15-0.17$M_{\odot}$, and receive a kick of 49 - 86${\rm km/s}$. Meng, Chen & Han (2007) adopted the simple analytic method of Wheeler, Lecar & McKee (1975), and calculated the impact to survey the influence of the initial parameters of the progenitor’s systems. With detailed stellar models and realistic separations that were obtained from binary evolution, they obtained an even lower ’stripped mass’, 0.03-0.13$M_{\odot}$, but a similar kick velocity 30-90${\rm km/s}$, which is perpendicular to the orbital velocity. A surviving companion star therefore has a mass lower by $\sim 0.1M_{\odot}$ and a space velocity larger by $\sim 10\%$ than that in Fig. 1. The companion stars are out of thermal equilibrium at the moment of SN explosion. The equilibrium radii are typically larger by $\sim 50\%$ than that at the moment of SN explosion. Therefore the surface gravity at equilibrium should be lower than that in Fig. 2. Podsiadlowski (2003) systematically explored the evolution and appearance of a typical companion star that has been stripped and heated by the supernova interaction during the post-impact re-equilibrium phase. Such a star may be significantly overluminous or underluminous. Fig. 2 could be a starting point for further studies of this kind. Fig. 3 is the distribution of orbital periods of the WD+MS systems at the moment of SN explosion. If I assume that the companion stars co-rotate with their orbits, I obtain their distributions of equatorial rotational velocities (see Fig. 4). We see that the surviving companion stars are fast rotators and their spectral lines should be broadened noticeably. Fig. 5 shows the distribution of masses lost during optically thick stellar wind phase. We see that a significant amount of mass is lost in the wind. Badenes et al. (2007) found that the wind with a velocity above $200\,{\rm km/s}$, which is believed to be reasonable, would excavate a large low-density cavity around its progenitor. However, the fundamental properties of seven young SN Ia remnants (Kepler, Tycho, SN 1006, 0509-67.5, 0519-69.0, N103B and SN 1885) obtained by Badenes et al. (2007) are incompatible with such large cavities. A lower wind velocity or the consideration of WD rotation could help to solve the problem (Badenes et al., 2007). Fig. 6 presents the distribution of mass transfer rates at the moment of SN explosion. The mass transfer rates can be converted to X-ray luminosities of the systems by $L_{\rm X}\sim\epsilon|\dot{M}|$, where $\epsilon=7\times 10^{18}{\rm erg/g}$ gives the approximate amount of energy obtained per gram of hydrogen burnt into helium or carbon/oxygen. The luminosity of the X-ray source close to the site of SN 2007on (4 years before the explosion) was estimated to be $(3.3\pm 1.5)\times 10^{37}\,{\rm erg/s}$ (Voss & Nelemans, 2008), corresponding to a mass accretion rate of $\sim 10^{-7}\,M_{\odot}/{\rm yr}$, which is consistent with Fig. 6. QU Carinae, a cataclysmic variable (CV), is suspected to be a SN Ia progenitor. It has a WD mass of $\sim 1.2M_{\odot}$, a once-reported orbital period of $0.45\,{\rm d}$, and more importantly, a very high mass transfer rate of $\sim 10^{-7}\,M_{\odot}/{\rm yr}$ (Kafka, Anderson & Honeycutt, 2008). These properties are consistent with Figs 3 and 6. However, BF Eridani, another CV with $M_{\rm WD}\sim 1.28M_{\odot}$, $M_{2}\sim 0.52M_{\odot}$ and $P\sim 0.27\,{\rm d}$ (Neustroev & Zharikov, 2008), appears not to be a SN Ia progenitor. Fig. 7 is the distribution of the surface nitrogen mass fraction of companion stars at the moment of explosion. We see that nitrogen can be significantly overabundant (with corresponding underabundance of carbon due to the CN- cycle). However, the surface can be seriously polluted by the ejecta of SN explosions. The simulation in this Letter was made with $\alpha_{\rm CE}=\alpha_{\rm th}=1.0$. If I adopt a lower value for $\alpha_{\rm th}$, say, 0.1, the birth rate of SNe Ia would be higher (a factor of 1.7) and the delay time from the star formation to SN explosion is shorter. This is because that binaries resulted from CE ejections tend to have shorter orbital periods for a small $\alpha_{\rm th}$ and are more likely to locate in the SN Ia production region (see Fig.3 of HP04). The companion stars with orbital velocity $V<110\,{\rm km/s}$ would be absent from Fig. 1, and the ones with $\log(g/{\rm cm\ s^{-2}})<3.1$ would be absent from Fig. 2. This is due to that WD binaries with long orbital periods are absent due to a small $\alpha_{\rm th}$. The distributions are snapshots at current epoch for a constant SFR. For a single star burst, most of the SN explosions occur between 0.1 and 1 Gyr after the burst (see Fig. 7 of HP04). The evolution of progenitor properties with time can be understood via Fig. 3 of HP04. A delay time from 0.1 to 1 Gyr corresponds to $M_{2}^{\rm i}$ of $\sim 3.2M_{\odot}$ to $\sim 1.8M_{\odot}$, and to $M_{\rm WD}^{\rm i}$ of $\sim 1.2M_{\odot}$ to $\sim 0.67M_{\odot}$ for the WD+MS system, respectively. As seen from Fig. 3 of HP04, the range of the orbital periods becomes narrower from a delay time of 0.1 to 1.0 Gyr. Those WD binaries result in SN Ia explosions via RLOF. Consequently, the range of progenitor properties at the moment of SN Ia, e.g. the orbital velocities of companion stars, the surface gravities, the equatorial rotational velocities, the mass transfer rates, becomes narrower with time. The mass transfer rate would be smaller with time as $M^{\rm i}_{2}$ becomes smaller. I thank an anonymous referee for his/her comments which help to improve the paper. I thank Ph. Podsiadlowski for stimulating discussions. This work was in part supported by the Natural Science Foundation of China under Grant Nos 10433030, 10521001 and 2007CB815406. ## References * Badenes et al. (2007) Badenes C., Hughes J.P., Bravo E., Langer N., 2007, ApJ, 662, 472 * Bergeron (2003) Bergeron P., ApJ, 586, 201 * Canal, Méndez & Ruiz-Lapuente (2001) Canal R., Méndez J., Ruiz-Lapuente R., ApJ, 550, L53 * Dewi & Tauris (2000) Dewi J.D.M., Tauris T.M., 2000, A&A, 360, 1043 * Fuhrmann (2005) Fuhrmann K., 2005, MNRAS, 359, L35 * Hachisu, Kato & Nomoto (1999) Hachisu I., Kato M., Nomoto K., 1999, ApJ, 522, 487 * Han (1998) Han Z., 1998, MNRAS, 296, 1019 * Han et al. (1995) Han Z., Eggleton P.P., Podsiadlowski Ph., Tout C.A., 1995, MNRAS, 277, 1443 * Han, Podsiadlowski & Eggleton (1995) Han Z., Podsiadlowski Ph., Eggleton P.P., 1995, MNRAS, 272, 800 * Han et al. (2002) Han Z., Podsiadlowski Ph., Maxted P.F.L., Marsh T.R., Ivanova N., 2002, MNRAS, 336, 449 * Han et al. (2003) Han Z., Podsiadlowski Ph., Maxted P.F.L., Marsh T.R., 2003, MNRAS, 341, 669 * Han & Podsiadlowski (2004) Han Z., Podsiadlowski Ph., 2004, MNRAS, 350, 1301 (HP04) * Han & Podsiadlowski (2006) Han Z., Podsiadlowski Ph., 2006, MNRAS, 368, 1095 * Hansen (2003) Hansen B.M.S., ApJ, 582, 915 * Iben & Tutukov (1984) Iben I.Jr., Tutukov A.V., 1984, ApJS, 54, 335 * Justham et al. (2008) Justham S., Wolf C., Podsiadlowski Ph., Han Z., 2008, A&A, submitted * Kafka, Anderson & Honeycutt (2008) Kafka S., Anderson R., Honeycutt R.K., 2008, ApJ, submitted (astro-ph/0801.3638) * Langer et al. (2000) Langer N., Deutschmann A., Wellstein S., Höflich P., 2000, A&A, 362, 1046 * Li & van den Heuvel (1997) Li X.D., van den Heuvel E.P.J., 1997, A&A, 322, L9 * Mannucci, Della Valle & Panagia (2006) Mannucci F., Della Valle M., Panagia N., 2006, MNRAS, 370, 773 * Marietta, Burrows & Fryxell (2000) Marietta E., Burrows A., Fryxell B., 2000, ApJS, 128, 615 * Meng, Chen & Han (2007) Meng X., Chen X., Han Z., 2007, PASJ, 59, 835 * Maoz & Mannucci (2008) Maoz D., Mannucci F., 2008, MNRAS, submitted (astro-ph/0801.2898) * Miller & Scalo (1979) Miller G.E., Scalo J.M., 1979, ApJS, 41, 513 * Neustroev & Zharikov (2008) Neustroev V.V., Zharikov S., 2008, MNRAS, submitted (astro-ph/0801.1082) * Nomoto, Thielemann & Yokoi (1984) Nomoto K., Thielemann F.,Yokoi K., 1984, ApJ, 286, 644 * Oppenheimer et al. (2001) Oppenheimer B.R., Hambly N.C., Digby A.P., Hodgkin S.T., Saumon D., 2001, Science, 292, 698 * Paczyński (1976) Paczyński B., 1976, in Eggleton P.P., Mitton S., Whelan J., eds, Structure and Evolution of Close Binaries. Kluwer, Dordrecht, p. 75 * Perlmutter et al. (1999) Perlmutter S. et al., 1999, ApJ, 517, 565 * Podsiadlowski (2003) Podsiadlowski Ph., 2003, ArXiv:astro-ph/0303660 * Podsiadlowski, Rappaport & Han (2003) Podsiadlowski Ph., Rappaport S., Han Z., 2003, MNRAS, 341, 385 * Reynolds et al. (2007) Reynolds S.P., et al., 2007, ApJ, 668, L135 * Roelofs et al. (2008) Roelofs G., Bassa C., Voss R., Nelemans G., 2008, MNRAS, submitted (astro-ph/0802.2097) * Riess et al. (1998) Riess A. et al., 1998, AJ, 116, 1009 * Ruiz-Lapuente et al. (2004) Ruiz-Lapuente P., et al., Nature, 431, 1069 * Voss & Nelemans (2008) Voss R., Nelemans G., 2008, Nature, 451, 802 * Webbink & Iben (1987) Webbink R.F., Iben I.Jr., 1987, in Philipp A.G.D., Hayes D.S., Liebert J.W., eds, IAU Colloq. No. 95, Second Conference on Faint Blue Stars. Davis Press, Schenectady, p.445 * Wheeler, Lecar & McKee (1975) Wheeler J.C., Lecar M., McKee C.F., 1975, ApJ, 200, 145 * Wolf (2005) Wolf C., 2005, A&A, 444, L49 Figure 1: A snapshot distribution of companion stars in the plane of ($V$, $M_{2}^{\rm SN}$) at current epoch, where $V$ is the orbital velocity and $M_{2}^{\rm SN}$ the mass at the moment of SN explosion. The number density decreases from inner regions to outer regions. Regions, from inside to outside with corresponding gradational grey or color in the legend (from bottom to top), together with the regions with higher number densities contain 50.0%, 68.3%, 95.4%, and 99.7% of all the systems, respectively. Figure 2: Similar to Fig. 1, but in the plane of ($\log T_{\rm eff}$, $\log g$), where $T_{\rm eff}$ is the effective temperature of companion stars at the moment of SN explosion, $\log g$ the surface gravity. The error bars denote the location of Tycho G (Ruiz-Lapuente et al., 2004). Figure 3: Similar to Fig. 1, but in the plane of ($\log P^{\rm SN}$, $M_{2}^{\rm SN}$), where $P^{\rm SN}$ is the orbital period at the moment of SN explosion. Figure 4: Similar to Fig. 1, but in the plane of ($V_{\rm rot}$, $M_{2}^{\rm SN}$), where $V_{\rm rot}$ is the equatorial rotational velocity of companion stars at the moment of SN explosion. Figure 5: Similar to Fig. 1, but in the plane of ($\Delta M_{\rm wind}$, $M_{2}^{\rm SN}$), where $\Delta M_{\rm wind}$ is the mass lost during the optically thick stellar wind phase of the WD binaries. Figure 6: Similar to Fig. 1, but in the plane of ($\log(-\dot{M}_{2})$, $M_{2}^{\rm SN}$), where $-\dot{M}_{2}$ is the mass transfer rate at the moment of SN explosion. Figure 7: Similar to Fig. 1, but in the plane of ($X_{\rm N}$, $M_{2}^{\rm SN}$), where $X_{\rm N}$ is the nitrogen mass fraction at the surface of companion stars at the moment of SN explosion.

arxiv-papers

2024-09-04T02:48:54.329205

0803.2037

# Optimal spatial transportation networks where link-costs are sublinear in link-capacity D J Aldous Department of Statistics 367 Evans Hall # 3860 U.C. Berkeley CA 94720 aldous@stat.berkeley.edu www.stat.berkeley.edu/users/aldous Research supported by N.S.F Grant DMS-0704159 ###### Abstract Consider designing a transportation network on $n$ vertices in the plane, with traffic demand uniform over all source-destination pairs. Suppose the cost of a link of length $\ell$ and capacity $c$ scales as $\ell c^{\beta}$ for fixed $0<\beta<1$. Under appropriate standardization, the cost of the minimum cost Gilbert network grows essentially as $n^{\alpha(\beta)}$, where $\alpha(\beta)=1-\frac{\beta}{2}$ on $0<\beta\leq\frac{1}{2}$ and $\alpha(\beta)=\frac{1}{2}+\frac{\beta}{2}$ on $\frac{1}{2}\leq\beta<1$. This quantity is an upper bound in the worst case (of vertex positions), and a lower bound under mild regularity assumptions. Essentially the same bounds hold if we constrain the network to be _efficient_ in the sense that average route-length is only $1+o(1)$ times average straight line length. The transition at $\beta=\frac{1}{2}$ corresponds to the dominant cost contribution changing from short links to long links. The upper bounds arise in the following type of hierarchical networks, which are therefore optimal in an order of magnitude sense. On the large scale, use a sparse Poisson line process to provide long-range links. On the medium scale, use hierachical routing on the square lattice. On the small scale, link vertices directly to medium-grid points. We discuss one of many possible variant models, in which links also have a designed maximum speed $s$ and the cost becomes $\ell c^{\beta}s^{\gamma}$. ## 1 Introduction To design a transportation network linking specified points (visualized as cities) in the plane, one might specify a cost functional and a benefit functional on all possible networks, and then consider networks which are optimal in the sense of minimizing cost for a given level of benefit. This paper addresses one particular choice of functionals, but our broader purpose (see section 1.1) is to draw the attention of statistical physicists to this class of problem. We study a simple model involving the “economy of scale” idea > One link of length $\ell$ and capacity $2c$ is less than twice as expensive > as two links of length $\ell$ and capacity $c$. We capture this idea by specifying that the cost of a link of length $\ell$ and capacity $c$ scales as $\ell c^{\beta}$ for some $0<\beta<1$. In the real world, network designers do not know in advance what traffic demand will be. We simplify by assuming that traffic demand is known (and uniform over all source-destination pairs) and routes are controlled, so that the volume $f(e)$ of flow across an edge (link) $e$ can be determined by the designers, and the corresponding link-capacity built. (Visualize links as roads, and flow-volume $f(e)$ as “number of vehicles per hour”. We are ignoring stochastic fluctuations in traffic). Thus our cost structure is $\mbox{cost of network }=\sum_{e}\ell(e)f^{\beta}(e)$ (1) where $\ell(e)=$ length of link $e$. To define the model carefully, write ${\mathbf{x}}^{n}=\\{x_{1},x_{2},\ldots,x_{n}\\}$ for a configuration of $n$ vertices in the square $[0,n^{1/2}]^{2}$ of area $n$. So $x_{i}$ is the position of vertex $i$. Create a connected network $G({\mathbf{x}}^{n})$ by adding links: links are line-segments with their natural Euclidean lengths, and links may meet at places not in the given vertex-set ${\mathbf{x}}^{n}$. To make the distinction clear let us refer to the given $n$ vertices as cities and any meeting places (which depend on our choice of network) as junctions. Between each source-destination pair $(i,j)$ of cities, flow of volume $n^{-3/2}$ (this scaling is explained below) is routed through the network. Define $\mathrm{cost}(G({\mathbf{x}}^{n}))$, the cost of the network, via (1). This setting specializes a setting considered by Gilbert [1], and we call the minimum-cost network the Gilbert network $\mathrm{Gil}({\mathbf{x}}^{n})$. See [2] for general properties of, and heuristic algorithms for, Gilbert networks over deterministic points. Gilbert networks may be optimal from a network operator viewpoint, but what about a network user? Write $\ell(x_{i},x_{j})$ for route-length, and $|x_{j}-x_{i}|$ for straight-line distance, between cities $i$ and $j$. For a typical configuration, the average distance $\mathrm{ave}_{i,j}|x_{j}-x_{i}|$ will be order $n^{1/2}$. The kind of “benefit to users” we have in mind is that the network provides routes almost as short as possible. So we call the sequence of networks $(G({\mathbf{x}}^{n}))$ _modestly efficient_ if $\mathrm{ave}_{i,j}(\ell(x_{i},x_{j})-|x_{j}-x_{i}|)=o(n^{1/2}).$ (2) The name reflects the remarkable fact [3] that there exist _extremely efficient_ networks for which this average is $O(\log n)$ while their length is only $1+o(1)$ times the minimum length of any connected network; such results pay no attention to flow-volumes or capacities, and so constitute the $\beta=0$ case of the present model. The problem we address in this paper is: > given the sequence $({\mathbf{x}}^{n})$, how small can we make > $\mathrm{cost}(G({\mathbf{x}}^{n}))$ subject to the _modestly efficient_ > constraint (2)? In the $\beta=0$ case just mentioned, we can make $\mathrm{cost}(G({\mathbf{x}}^{n}))$ be asymptotically the length of the Steiner tree (minimum length connected network) on ${\mathbf{x}}^{n}$, which is well known to be $O(n)$ in the worst case and in the typical case. Recall that $a_{n}=O(b_{n})$ means that $a_{n}/b_{n}$ is bounded as $n\to\infty$. It is often convenient to write the converse relationship $b_{n}=O(a_{n})$ as $a_{n}=\Omega(b_{n})$; if both $a_{n}=O(b_{n})$ and $a_{n}=\Omega(b_{n})$ then we write $a_{n}=\Theta(b_{n})$. In the case $\beta=1$ there is no “economy of scale” and so the minimum-cost network is just the complete graph, that is a direct link between each pair of cities. The associated cost is $\sum_{i}\sum_{j}n^{-3/2}|x_{i}-x_{j}|=n\times\frac{\mathrm{ave}_{i,j}|x_{i}-x_{j}|}{n^{1/2}}$ which is $O(n)$ in the worst case and in the typical case. Recall that the Gilbert network $\mathrm{Gil}({\mathbf{x}}^{n})$ is the minimum-cost network when there is no extra “modestly efficient” constraint. Theorem 1 shows that imposing the “modestly efficient” constraint makes little difference in an order of magnitude sense: in either case the optimal cost grows roughly as order $n^{\alpha(\beta)}$. ###### Theorem 1 Fix $0<\beta<1$. Define $\displaystyle\alpha(\beta)$ $\displaystyle=$ $\displaystyle 1-{\textstyle\frac{\beta}{2}},\quad 0<\beta\leq{\textstyle\frac{1}{2}}$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1+\beta}{2}},\quad{\textstyle\frac{1}{2}}\leq\beta<1.$ Let ${\mathbf{x}}^{n}$ be a configuration of $n$ cities in the square $[0,n^{1/2}]^{2}$. (a) Case $0<\beta\leq{\textstyle\frac{1}{2}}$. There exist modestly efficient networks for which $\mathrm{cost}(G({\mathbf{x}}^{n}))=O(n^{\alpha(\beta)})$ (except for $\beta=1/2$ the bound is $O(n^{3/4}\log n)$). Under the technical assumption (7) there do _not_ exist connected networks for which $\mathrm{cost}(G({\mathbf{x}}^{n}))=o(n^{\alpha(\beta)})$. So under (7) we have $\mathrm{cost}(\mathrm{Gil}({\mathbf{x}}^{n}))=\Theta(n^{\alpha(\beta)})$ for $\beta<{\textstyle\frac{1}{2}}$. (b) Case ${\textstyle\frac{1}{2}}<\beta<1$. Here $\mathrm{cost}(\mathrm{Gil}({\mathbf{x}}^{n}))=O(n^{\alpha(\beta)})$. Given $\omega_{n}\to\infty$ arbitrarily slowly, there exist modestly efficient networks for which $\mathrm{cost}(G({\mathbf{x}}^{n}))=O(\omega_{n}n^{\alpha(\beta)})$. Under the technical assumption (8), $\mathrm{cost}(\mathrm{Gil}({\mathbf{x}}^{n}))=\Theta(n^{\alpha(\beta)})$, but there do _not_ exist modestly efficient networks for which $\mathrm{cost}(G({\mathbf{x}}^{n}))=O(n^{\alpha(\beta)})$. Our discussion above of the cases $\beta=0$ and $\beta=1$ implies corresponding results in these cases with $\alpha(0)=1$ and $\alpha(1)=1$. The transition at $\beta=\frac{1}{2}$ corresponds to the dominant cost contribution changing from short links to long links, as we will explain in section 2.5. The technical regularity assumptions that we need to impose to obtain lower bounds reflect this transition: for $\beta<1/2$ we need to assume that nearest-neighbor distances are not atypically small, whereas for $\beta>1/2$ we assume a large-scale equidistribution of the city configuration. (We defer statements of these assumptions until the place they are actually used in the proof, to avoid interrupting the conceptual discussion here.) We show (section 2) that the upper bounds arise in the following type of hierarchical networks, which are therefore optimal in an order of magnitude sense. On the large scale, use a sparse Poisson line process to provide long-range links. On the medium scale, use hierachical routing on the square lattice. On the small scale, link cities directly to medium-grid points. It is perhaps counter-intuitive that one can use the same network for the whole range of $\beta$; the point is that only the medium- small scale structure really matters for $\beta<1/2$ and only the large scale structure really matters for $\beta>1/2$. Our arguments implicitly imply some weak properties of the exactly optimal networks. Undestanding in detail the structure of the Gilbert network (or the asymptotically optimal modestly efficient network) over random points in the critical case $\beta=1/2$ is a challenging problem, interesting because one expects the network to have some scale-free stucture, in the (correct) sense of invariance under spatial and flow-volume rescaling. One can imagine many variant models in which extra structure is incorporated. In section 4 we briefly discuss the case where links have designed speed $s$ and where the cost of a link becomes $\ell c^{\beta}s^{\gamma}$; in this case an analog of Theorem 1 remains true. ### 1.1 Optimal spatial network design methodology This paper contributes to a general program concerning networks linking points in the plane: > for mathematically simple cost/benefit functionals, study the properties > (geometry, cost and benefit values) of optimal networks as the number $n$ of > points tends to infinity. Network design problems arise in many applied fields, but serious real-world modelling leads to more complicated functionals tuned to specific applications than we have in mind. As complementary work, [3] gives a detailed treatment of the extremely efficient networks mentioned above that minimize average route length subject to total network length; and [4] analyzes a model (for e.g. passenger air travel or package delivery) where there is a substantial cost to transfer from one link to another. In the latter model, theory predicts that hub-and-spoke networks (as seen in the real world) are near-optimal and that, constraining the average number of transfers to be say $2$, the length of the shortest possible network scales as $n^{13/10}$. The methodological feature we want to emphasize concerns models for the position of $n$ cities (assumed for simplicity in a square of area $n$). In each problem we have studied one gets the same order of magnitude for optimal network cost for worst-case positions as one gets for arbitrary positions (under mild assumptions) and in particular the same as for random positions or for regular (e.g. lattice) positions. The bulk of statistical physics literature on spatial networks (surveyed in [5]) analyzes networks built according to some specific probability model which combines ingredients such as (a) geometric random graphs (link probability depends on inter-vertex distance); (b) proportional attachment probabilities for arriving vertices; (c) prescribed power law distribution of lattice vertex degrees; (d) networks based on recursive partitioning of space. This theoretical literature makes passing reference to optimality, but we have not seen analytic results demonstrating optimality over all possible networks in the spatial context (see [6] for non-spatial results, and [7, 8] for assumptions under which optimal networks are trees). For interesting empirical work see [9]. Our scaling conventions (a square of area $n$; flow-volume $n^{-3/2}$ between each source-destination pair) may seem arbitrary, but are chosen to fit the following standardizations: (i) cities have density $1$ per unit area; (ii) flow volume across unit area is order $1$. ## 2 The construction A network satisfying the requirements of Theorem 1 will be constructed in section 2.3 using mathematical ingredients described in sections 2.1 and 2.2. Figure 1 illustrates the construction. $s_{n}$$2^{M_{n}}$small cells$\sigma_{n}$$\triangleleft$$\triangledown$$\triangleleft$$\triangledown$$\triangleleft$$\sigma_{n}$$\theta_{n}$large cells$n^{1/2}$ Figure 1. Ingredients of the construction. Left: the hierarchical routing lattice, with higher-type edges indicated by thicker lines, and a typical route shown. Right: the large-scale grid and the Poisson line process. ### 2.1 Hierarchical routing on the square lattice Fix $M$ and consider the square grid on vertices $\\{0,1,2,\ldots,2^{M}-1\\}^{2}$. Declare lines (and their edges) to be of some type $0,1,2,\ldots,M$ according to the rule: the horizontal lines $\\{(x,y):y=(2j-1)2^{m}\\},\quad j=1,2,\ldots$ are type $m$ the boundary line $\\{(x,0)\\}$ is type $M$; and similarly for vertical lines. For each vertex $(x,y)$, define a route from $(x,y)$ to $(0,0)$ using only downward and leftward edges as follows. First choose the edge at $(x,y)$ of higher type (breaking ties arbitrarily). Then repeat the rule > Follow the current edge until it crosses an edge of strictly higher type, > then transfer to that edge until reaching $(0,0)$. See Figure 1, left side. It is elementary to verify ###### Lemma 2 For each $0\leq m\leq M$, the number of type-$m$ edges traversed by the route is at most $2^{m+1}$. ### 2.2 The Poisson line process A line in the plane may be parametrized by the point $z$ on the line which is closest to the origin (so the line segment from the origin to $z$ is orthogonal to the line); then write $z$ in radial coordinates as $(r,\theta)$. Recall [10] the notion of a Poisson line process (PLP) of intensity $\eta>0$, which makes precise the notion of “completely random” lines in the plane. Parametrizing lines by by their closest points $(r,\theta)$, this PLP has intensity $\eta$ with respect to Lebesgue measure on parameter space $(0,\infty)\times(0,2\pi)$. The PLP distribution is invariant under Euclidean transformations, and for a fixed set $A$ ${\mathbb{E}}(\mbox{length of line segments intersecting }A)=\pi\eta\times\mbox{area}(A).$ (3) (We write ${\mathbb{E}}$ for expectation and ${\mathbb{P}}$ for probability). The next result shows how the PLP is useful in constructing spatial networks. See Figure 1, right side. ###### Lemma 3 Let $n^{1/2}/\sigma_{n}$ be an integer. Construct a network as the superposition of the rectangular grid with cell side-length $\sigma_{n}$ and the Poisson line process of intensity $\eta$, intersected with the square $[0,n^{1/2}]^{2}$. Let $v_{i},v_{j}$ be vertices of the grid. Then ${\mathbb{E}}(\mbox{route-length $v_{i}$ to $v_{j}$})\leq|v_{i}-v_{j}|+C_{2}{\textstyle\frac{1}{\eta}}\log(\eta\sqrt{2n})$ for an absolute constant $C_{2}$. Lemma 3 is proved in [3], Lemma 11, and we will not repeat the argument here. (In essence, one analyzes the natural routing algorithm: move to a nearby line of the PLP, move along that line in the direction closer to the direction of the destination city, and when encountering another line of the PLP, switch to that line if its direction is closer to the destination city direction). Using the PLP gives us random networks, but a typical realization will have costs and lengths of the same order as the expectations in our formulas. ### 2.3 Construction of the networks We now describe how the ingredients above (hierarchical routing on the square lattice, the PLP) are used in a network construction. Recall ${\mathbf{x}}^{n}$ denotes the given configuration of $n$ cities. Take integers $\theta_{n}\uparrow\infty$ slowly and define $\sigma_{n}=n^{1/2}/\theta_{n}.$ Let $M_{n}$ be the integer such that $\sigma_{n}/2<2^{M_{n}}\leq\sigma_{n}.$ Define $s_{n}=\sigma_{n}/2^{M_{n}},\quad\quad\mbox{ (so $1\leq s_{n}<2$)}.$ Construct a network $G({\mathbf{x}}^{n})$ as follows. (i) Take the large-scale network in Lemma 3, with $\eta_{n}=\theta_{n}n^{-1/2}$. This network contains _large cells_ of side- length $\sigma_{n}$. (ii) Inside each large cell put a copy of the hierarchical routing lattice of section 2.1, with $M=M_{n}$, and scaled so that the basic _small cell_ of this lattice has side-length $s_{n}$. (iii) Link each city $x\in{\mathbf{x}}^{n}$ via a straight edge to the bottom left corner vertex $v(x)$ of its small cell. Figure 1 illustrates (i) and (ii). There is a natural way to define a route from $x_{i}$ to $x_{j}$ in this network. From $x_{i}$ take the link to $v(x_{i})$, then follow the section 2.1 routing scheme to the lower left corner $V(x_{i})$ of the large cell; navigate from $V(x_{i})$ to $V(x_{j})$ via the shortest route in the Lemma 3 graph. Note that in addition to the given $n$ cities, this network has several different kinds of junctions: the vertices of the grid, and places where lines of the PLP cross each other or cross the grid lines or cross the short stage (iii) links. In our model there is no cost associated with creating a junction or with routes using junctions; the costs involve only link lengths and route lengths. So the exact number of junctions is unimportant. ### 2.4 Analysis of the networks Clearly $\ell(x_{i},x_{j})\leq\ell(V(x_{i}),V(x_{j}))+2^{3/2}\sigma_{n}$ and so by Lemma 3 ${\mathbb{E}}\ell(x_{i},x_{j})\leq|x_{i}-x_{j}|+2^{3/2}\sigma_{n}+C_{2}{\textstyle\frac{1}{\eta_{n}}}\log(\eta_{n}\sqrt{2n}).$ From the definitions of $\sigma_{n},\eta_{n}$ we see ${\mathbb{E}}(\ell(x_{i},x_{j})-|x_{i}-x_{j}|)=o(n^{1/2})$ establishing the modestly efficient property. To analyze costs, we treat stages (i)-(iii) separately, and check that each stage cost is less than the bounds stated in Theorem 1. Stage (iii). There are $n$ links of the form $(x,v(x))$, each carrying flow volume $2(1-\frac{1}{n})n^{-1/2}$, and each having length at most $s_{n}\sqrt{2}$, and so the total cost of stage (iii) links is $O(n^{1-\frac{\beta}{2}})$. (4) Stage (ii). Now let $\mbox{${\mathcal{E}}$}_{m}$ be the set of type-$m$ edges. The number of such edges is $\\#\mbox{${\mathcal{E}}$}_{m}=O(n2^{-m})$. Recall that Hölder’s inequality shows that for any edge-set ${\mathcal{E}}$ $\sum_{e\in\mbox{${\mathcal{E}}$}}f^{\beta}(e)\leq(\\#\mbox{${\mathcal{E}}$})^{1-\beta}\left(\sum_{e\in\mbox{${\mathcal{E}}$}}f(e)\right)^{\beta}.$ Now $\sum_{e\in\mbox{${\mathcal{E}}$}_{m}}f(e)=2n^{-1/2}\sum_{x\in{\mathbf{x}}^{n}}\\#\\{\mbox{ type-$m$ edges in route $v(x)$ to $V(x)$}\\}\leq 2^{m+2}n^{1/2}$ using Lemma 2. Thus $\sum_{e\in\mbox{${\mathcal{E}}$}_{m}}f^{\beta}(e)=O\left((n2^{-m})^{1-\beta}\ (2^{m}n^{1/2})^{\beta}\right)=O\left(n^{1-\frac{\beta}{2}}2^{m(2\beta-1)}\right).$ (5) Writing $\mbox{${\mathcal{E}}$}_{med}$ for all edges in the copies of the hierarchical routing lattice, we find after summing over $0\leq m\leq M$ $\displaystyle\sum_{e\in\mbox{${\mathcal{E}}$}_{med}}f^{\beta}(e)$ $\displaystyle=$ $\displaystyle O\left(n^{1-\frac{\beta}{2}}\right),\quad 0<\beta<{\textstyle\frac{1}{2}}$ $\displaystyle=$ $\displaystyle O\left(n^{3/4}\log n\right),\quad\beta={\textstyle\frac{1}{2}}$ $\displaystyle=$ $\displaystyle O\left(n^{1-\frac{\beta}{2}}2^{M(2\beta-1)}\right)=O\left(n^{\frac{1}{2}+\frac{\beta}{2}}\right),\quad{\textstyle\frac{1}{2}}<\beta<1$ using $2^{M}<n^{1/2}$. Because edge-lengths here are $s_{n}<2$, these are bounds for the costs associated with stage (ii). Stage (iii). Write $\mbox{${\mathcal{E}}$}_{large}$ for the set of links of the large-scale network, that is the large-scale grid and the PLP lines. Flow along the route from $V(x_{i})$ to $V(x_{j})$ contributes $n^{-3/2}\ell(V(x_{i}),V(x_{j}))$ to the “flow $\times$ distance” measure, and so $\int_{\mbox{${\mathcal{E}}$}_{large}}f(e)de=n^{-3/2}\sum_{i}\sum_{j}\ell(V(x_{i}),V(x_{j}))$ where the left side denotes integrating along all links of the large-scale network. By the already-established modestly efficient property, $\sum_{i}\sum_{j}\ell(V(x_{i}),V(x_{j}))=(1+o(1))\sum_{i}\sum_{j}|x_{i}-x_{j}|=O(n^{5/2})$ and so $\int_{\mbox{${\mathcal{E}}$}_{large}}f(e)de=O(n).$ The total length $L_{n}$ of $\mbox{${\mathcal{E}}$}_{large}$ is the sum of $O(n^{1/2}\theta_{n})$ ($=$ contribution from large-scale grid) and $O(\eta_{n}n)$ ($=$ contribution from the PLP, using (3)), and so $L_{n}=O(n^{1/2}\theta_{n})$. The integral form of Hölder’s inequality now shows that the cost associated with $\mbox{${\mathcal{E}}$}_{large}$ is : $\int_{\mbox{${\mathcal{E}}$}_{large}}f^{\beta}(e)de\leq L_{n}^{1-\beta}\times\left(\int_{\mbox{${\mathcal{E}}$}_{large}}f(e)de\right)^{\beta}=O\left(\theta_{n}^{1-\beta}n^{(1+\beta)/2}\right).$ (6) Examining the cost of each stage, we check that the modestly efficient network we have constructed has its cost bounded as stated in Theorem 1. Moreover, if we eliminate the “modestly efficient” constraint then we can eliminate Stage (iii) of the construction (take $\theta_{n}=1$) and get the stated $O(n^{\alpha(\beta)})$ upper bound. ### 2.5 The transition at $\beta=1/2$ To summarize, the costs associated with the constructed networks arising from short, medium and large-scale links are bounded by expressions (4,5,6) respectively. By examining the exponents of $n$ we see that the transition at $\beta=\frac{1}{2}$ corresponds to the dominant cost contribution changing from short links to long links. The arguments we give below for the lower bound show this is a genuine effect (no alternate networks can do essentially better), not an artifact of the particular networks contructed above. ## 3 The lower bound In the settings of [3, 4] the lower bounds require some effort to prove, but in the present setting the proofs are short. ### 3.1 The case $0<\beta\leq 1/2$ Consider first the case $0<\beta\leq 1/2$. Impose the condition: there exists some small $\delta>0$ such that for at least $\delta n$ of the cities of ${\mathbf{x}}^{n}$, the distance to the nearest neighbor is at least $\delta$. (7) Consider a city $x\in{\mathbf{x}}^{n}$ satisfying this condition, and consider the link-segments of an arbitrary connected network within distance $\delta/2$ from $x$. Because flow of volume $2n^{-1/2}$ must enter or leave $x$, the cost associated with these link-segments (which by concavity of $f\to f^{\beta}$ is minimized when there is a single link-segment) is at least $\delta/2\times(2n^{-1/2})^{\beta}$. Summing over all (there are at least $\delta n$) such cities $x$, noting the link-segments are distinct as $x$ varies, the network cost is at least $\delta n\times\delta/2\times(2n^{-1/2})^{\beta}=\Omega(n^{1-\frac{\beta}{2}})$. ### 3.2 The case $1/2<\beta<1$ In the case $1/2<\beta<1$ we impose the classical equidistribution property for the configuration ${\mathbf{x}}^{n}=(x^{n}_{i},1\leq i\leq n)$ rescaled back to the unit square: the empirical distribution of $\\{n^{-1/2}x^{n}_{i},1\leq i\leq n\\}$ converges $\displaystyle\mbox{ in distribution to the uniform distribution on $[0,1]^{2}$}.$ (8) Our standardization conventions imply that the total volume of flow through the network is $\Theta(n^{1/2})$ and so assertion (a) below is obvious. ###### Lemma 4 (a) In the Gilbert network $\mathrm{Gil}({\mathbf{x}}^{n})$, the maximum edge- flow is bounded as $\max_{e}f(e)=O(n^{1/2}).$ (b) For any modestly efficient network $(G({\mathbf{x}}^{n}))$ on configurations satisfying the equidistribution condition (8), the maximum edge-flow is bounded as $\max_{e}f(e)=o(n^{1/2}).$ Granted this result, use the fact $\sum_{e}\ell(e)f(e)\geq n^{-3/2}\sum_{i}\sum_{j}|x_{i}-x_{j}|=\Theta(n)\mbox{ by equidistribution }$ and the general inequality $\mathrm{cost}(G({\mathbf{x}}^{n}))=\sum_{e}\ell(e)f^{\beta}(e)\geq\frac{\sum_{e}\ell(e)f(e)}{(\max_{e}f(e))^{1-\beta}}$ to deduce that $\mathrm{cost}(G({\mathbf{x}}^{n}))$ grows strictly faster than $n/n^{(1-\beta)/2}=n^{\alpha(\beta)}$ for any modestly efficient network, and no slower than order $n^{\alpha(\beta)}$ for the Gilbert network. Proof of Lemma 4(b). We first quote an easy fact from geometry. ###### Lemma 5 Let $Z_{1},Z_{2}$ be two independent uniform random points in the unit square $[0,1]^{2}$. There exists a constant $C$ such that for all $x\in[0,1]^{2}$ and all $\delta>0$ ${\mathbb{P}}(|Z_{1}-x|+|Z_{2}-x|\leq|Z_{1}-Z_{2}|+\delta)\leq C\delta^{1/2}.$ Now fix $\delta>0$. Write $X_{1},X_{2}$ for two uniform random picks from the set ${\mathbf{x}}^{n}$ of cities. The modestly efficient assumption implies ${\mathbb{P}}(\ell(X_{1},X_{2})\geq|X_{1}-X_{2}|+\delta n^{1/2})\to 0\mbox{ as }n\to\infty.$ Lemma 5 and the equidistribution assumption (8) imply ${\mathbb{P}}(|X_{1}-x|+|X_{2}-x|\leq|X_{1}-X_{2}|+\delta\quad\mbox{ for all }x)\leq C\delta^{1/2}+o(1).$ In order for the route from $X_{1}$ to $X_{2}$ to pass through point $x$, one of the two inequalities above must hold, and so $\sup_{x}{\mathbb{P}}(\mbox{ route $X_{1}$ to $X_{2}$ passes through $x$ })\leq C\delta^{1/2}+o(1).$ But $\delta$ is arbitrary, so this probability is $o(1)$, and the flow volume is exactly $n^{1/2}$ times this probability. ## 4 Associating speeds with links The main feature of our model – that the cost of building a link is sublinear in link capacity – is just one of many realistic features one might want to incorporate into a model. By focussing on route lengths, we have implicitly assumed that users travel at constant speed. A notable feature of real road or rail networks is that different links permit different speeds. In this section we state and briefly discuss a variant model in which links can be designed to permit different speeds. Suppose a link with length $\ell$, nominal capacity $c_{0}$ and nominal speed $s_{0}$ costs $\ell c_{0}^{\beta}s_{0}^{\gamma}$, for fixed $0<\gamma<\infty$. On such a link, traffic moves with speed $s_{0}$ provided the flow-volume $f$ is at most $c_{0}$; for larger flow-volumes, congestion causes the speed to drop, reaching speed zero (jammed) at volume $\sigma c_{0}$ for a constant $\sigma$. So $\sigma c_{0}$ is the maximum capacity. Precisely, $\mbox{ speed at flow-volume }f=s_{0}G(f/c_{0})$ where $G(u)=1$ for $0\leq u\leq 1$ and $G(u)$ decreases from $1$ to $0$ as $u$ increases from $1$ to $\sigma$. Otherwise the model is the same as before: we are given a configuration of $n$ cities in the square of area $n$, and we are required to route flow of volume $n^{-3/2}$ between each source-destination pair. For any network and feasible routing, define average speed as $\overline{\mathrm{speed}}=\frac{\mathrm{ave}_{i,j}|x_{i}-x_{j}|}{\mathrm{ave}_{i,j}t(x_{i},x_{j})}$ where $t(x_{i},x_{j})$ is the time taken to travel from $x_{i}$ to $x_{j}$. For this model, we ask > What is the minimum cost for a network on a given configuration > ${\mathbf{x}}^{n}$ of cities that allows $\overline{\mathrm{speed}}=s$? The answer is that, under the regularity assumptions of Theorem 1 (which are needed only for lower bounds), and ignoring $O(\log n)$ terms. minimum cost grows as order $s^{\gamma}n^{\alpha^{*}(\beta,\gamma)}$, where (9) $\displaystyle\alpha^{*}(\beta,\gamma)$ $\displaystyle=$ $\displaystyle 1-{\textstyle\frac{\beta}{2}}-{\textstyle\frac{\gamma}{2}},\quad 0<2\beta+\gamma\leq 1$ $\displaystyle=$ $\displaystyle{\textstyle\frac{1+\beta}{2}},\quad 1\leq 2\beta+\gamma.$ Let us briefly indicate how the previous analysis is adapted to this setting. Because costs scale with design speed $s_{0}$ as $s_{0}^{\gamma}$, it is enough to consider the case $\overline{\mathrm{speed}}=1$, and show that minimum cost grows as order $n^{\alpha^{*}(\beta,\gamma)}$. To construct a network, use the networks constructed previously and assign design speeds as follows. For links of the large-scale network, which routes will use for a distance of order $n^{1/2}$, design speed of order $1$. For type $m$ edges in the hierarchical routing lattice, which routes will use for a distance of order $2^{m}$, design speed of order $2^{m}n^{-1/2}\log n$. For the local links of the form $(x,v(x))$, which routes will use for distance $O(1)$, design speed of order $n^{-1/2}$. This ensures the typical times $t(x_{i},x_{j})$ are of order $n^{1/2}$ as required. To calculate the cost, we simply combine the previous estimates (4,5,6) of costs of providing flow- volumes of different links with the costs of the design speeds stipulated above; the total cost is of order $n^{1-\frac{\beta}{2}}\times n^{-\frac{\gamma}{2}}\ +\ \sum_{m=0}^{M}n^{1-\frac{\beta}{2}}2^{m(2\beta-1)}\times(2^{m}n^{-1/2}\log n)^{\gamma}\ +\ n^{(1+\beta)/2}\times 1$ and this works out to be of the form $n^{\alpha^{*}(\beta,\gamma)}$ stated. ## References ## References * [1] E.N. Gilbert. Minimum cost communication networks. Bell System Tech. J., 46:2209–2227, 1967. * [2] D. A. Thomas and J. F. Weng. Minimum cost flow-dependent communication networks. Networks, 48:39–46, 2006. * [3] D.J. Aldous and W.S. Kendall. Short-length routes in low-cost networks _via_ Poisson line patterns. http://front.math.ucdavis.edu/math.PR/0701140. To appear in Adv. Applied. Probability, 2007. * [4] D.J. Aldous. Asymptotics and optimality for hub and spoke models in spatial transportation networks. http://arxiv.org/abs/cond-mat/0702502. To appear in Math. Proc. Cambridge Philos. Soc., 2007. * [5] Y. Hayashi and J. Matsukubo. A review of recent studies of geographical scale-free networks. IPSJ Trans., 47:776, 2006. http://xxx.arXiv.org:physics/0512011. * [6] L. Donetti, F. Neri, and M. A. Muñoz. Optimal network topologies: expanders, cages, Ramanujan graphs, entangled networks and all that. J. Stat. Mech. Theory Exp., 2006:P08007, 2006. * [7] A. Bejan. Shape and Structure, from Engineering to Nature. Cambridge University Press, 2000. * [8] M. Barthélemy and A. Flammini. Optimal traffic networks. J. Stat. Mech. Theory Exp., 2006:L07002, 2006. * [9] M.T. Gastner and M.E.J. Newman. Shape and efficiency in spatial distribution networks. J. Stat. Mech. Theory Exp., pages P01015, 9 pp., (electronic), 2006\. * [10] D. Stoyan, W. S. Kendall, and J. Mecke. Stochastic Geometry and its Applications. Wiley Series in Probability and Mathematical Statistics: Applied Probability and Statistics. John Wiley & Sons Ltd., Chichester, 2nd edition, 1995\.

arxiv-papers

2024-09-04T02:48:54.333834

0803.2179

# Heat Conduction Process on Community Networks as a Recommendation Model Yi-Cheng Zhang11footnotemark: 1 and Marcel Blattner Physics Department, University of Fribourg, 1700 Fribourg, Switzerland and Physics Department, Renmin University, Beijing, China Yi-Kuo Yu222Corresponding author, email: yyu@ncbi.nlm.nih.gov 1 ∗email: yi-cheng.zhang@unifr.ch National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bethesda, MD 20894, USA (May 3rd, 2007) ###### Abstract Using heat conduction mechanism on a social network we develop a systematic method to predict missing values as recommendations. This method can treat very large matrices that are typical of internet communities. In particular, with an innovative, exact formulation that accommodates arbitrary boundary condition, our method is easy to use in real applications. The performance is assessed by comparing with traditional recommendation methods using real data. ###### pacs: 44.10.+i, 89.70.+c, 89.20.Hh With the advent of the internet, there sprout many web sites that enable large communities to aggregate and interact. For example livejournal.com allows its 3 million members to share interests and life experiences; del.icio.us is a social bookmark service for people to share their findings on the World Wide Web. Thousands of such web sites are built by web entrepreneurs and activists for the public, and their number is growing ever faster. This brings about massive amount of accessible information, more than each individual is able or willing to process. Information search, filtering, and recommendation thus become indispensable in internet era. Ideally speaking, a good recommendation mechanism should be able to “guess” what a person may want to select based on what he or she already selected Maslov and Y.C. Zhang (2001); Blattner et al. (2006). Many such mechanisms are in actual use (like www.amazon.com proposing its readers with new books), however, jury is still out as to what is the best model. For a review of current techniques, see Adomavicius and Tuzhilin (2005). Based on the heat conduction (or diffusion) process, we propose a recommendation model capable of handling individualized boundary conditions (BC). To better explain our model, we first illustrate using the friendship network of $N$ people: each person (member) is a node, and a pair of nodes is connected by an edge provided they are mutual friends. The collection of these information forms the symmetric adjacency matrix $A$: element $A_{ij}=1$, or $0$ depending on whether people $i$ and $j$ are mutual friends (1) or not (0). Although it is possible to consider asymmetric connection, this generalization will not be studied here. To recommend friends to any individual member, we first set (Dirichlet) BC: to set the values on the directly connected nodes as 1 and some remote nodes (will be further specified) as 0. Values on all other nodes are treated as variables to be determined. These values can be interpreted as the probabilities that these nodes might be selected as friends. We now describe an efficient and effective strategy to solve the proposed heat conduction problem. From $A$, we first construct a propagator matrix $P=D^{-1}A$, where $D$ is the diagonal degree matrix. Denote $H$ as the temperature vector of $N$ components: the source-components are high temperature nodes with temperature $1$; the sink-components are low temperature nodes with temperature $0$. Our task is to find, through thermal equilibrium, the temperatures associated with the remaining nodes that are neither sinks nor sources. The discrete Laplace operator, analog of $-\nabla^{2}$, on this network is $L=I-P$, where $I$ is the identity matrix. We only need to solve $LH=f$ (1) where $f$ is the external flux vector. Note that this is the discrete analog of $-\kappa\nabla^{2}T(\vec{r})=\nabla\cdot{\vec{J}}(\vec{r})$ with $H(i)$ plays the role of $\kappa T(\vec{r})$ and $f(i)$ plays the role of $\nabla\cdot{\vec{J}}(\vec{r})$. Because Laplace operator conserves total heat and tend to spread heat from high temperature region to low temperature region, the only way to maintain the fixed temperature values at the sources and sinks is to apply external heat flux (inflow at sources and outflow at sinks). For the rest of the nodes, the equilibrium condition demands that no net heat flux should occur. Therefore, the only allowed nonzero components of $f$ are source- and sink- components. The computation of the temperature vector is straightforward. It is convenient to group the source and sink components together into a block $H_{1}$, and the rest free variables another block $H_{2}$. That is $H=\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)\;.$ (2) Likewise, we group the Laplace operator in a similar fashion and eq. (1) may be expressed as $\left(\begin{array}[]{ll}L_{11}&L_{12}\\\ L_{21}&L_{22}\end{array}\right)\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)=\left(\begin{array}[]{c}f\\\ 0\end{array}\right)\;.$ (3) All we need to solve is the homogeneous equation $L_{21}H_{1}+L_{22}H_{2}=0\;,$ (4) without the need to know $f$. Fixing the values of $H_{1}$, $H_{2}$ can be readily found using standard iterative methods Press et al. (1992). The above approach, although straightforward, represents a daunting challenge: for each individual, we must solve the huge matrix problem once – a prohibitively expensive task for a typical internet community having millions of members. The standard way to get around this dilemma is to resort to the Green’s function method. Starting from eq.(1) we would like to have a Green’s function $\Omega^{\prime}$ such that eq.(1) can be inverted: $\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)=\Omega^{\prime}\left(\begin{array}[]{c}f\\\ 0\end{array}\right)\;$ (5) to get $H_{2}=\Omega^{\prime}_{21}{\Omega^{\prime}_{11}}^{-1}H_{1}$. However, $\Omega^{\prime}=L^{-1}=(I-P)^{-1}$ is divergent: the Laplace operator has a zero eigenvalue and the inverse $L^{-1}$ is meaningful only if $(H_{1},H_{2})^{T}$ is in the subspace that is orthogonal to the eigenvector of zero eigenvalue. A fortunate scenario like this has occurred in the studies of random resistor networks Korniss et al. (2006); Wu (2004). To simultaneously deal with all possible BC, we lose the freedom to limit the solution to a certain subspace. Nevertheless, we have a good understanding regarding this divergence. Basically, the $P$ matrix has an eigenvalue one with the right eigenvector being a column of $1$s $|u^{0}\rangle=\left(1,1,\cdots,1\right)^{T}$ and with left eigenvector being $\langle v^{0}|=\left({d_{1}\over d},{d_{2}\over d},\ldots,{d_{N}\over d}\right)$ where $d_{i}$ denotes the degree of node $i$ and $d=\sum_{i}d_{i}$ being the sum of degrees. Note that with this notation, we have $\langle v^{0}|u^{0}\rangle=1$. We may then decompose $P$ into $P=Q+|u^{0}\rangle\langle v^{0}|$ with $Q|u^{0}\rangle\langle v^{0}|=|u^{0}\rangle\langle v^{0}|Q=0$. Further, the spectral radius of $Q$ is now guaranteed to be smaller than $1$ and thus $(I-Q)$ is invertible with $(I-Q)^{-1}=\sum_{n=0}^{\infty}Q^{n}$. We may then rewrite the eq.(3) as $\displaystyle(I-Q)\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)$ $\displaystyle=$ $\displaystyle\left(\begin{array}[]{c}f\\\ 0\end{array}\right)+|u^{0}\rangle\langle v^{0}|\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)$ (12) $\displaystyle=$ $\displaystyle\left(\begin{array}[]{c}f\\\ 0\end{array}\right)+c(H)|u^{0}\rangle$ (15) where the $H$-dependent constant may be written as $c(H)=\langle v_{1}^{0}|H_{1}\rangle+\langle v_{2}^{0}|H_{2}\rangle$. We need to explain the notation further. Basically $|u^{0}_{1}\rangle$, represents a column vector whose components are obtained from the column vector $|u^{0}\rangle$ with component labels corresponding to that of the sources and the sinks. On the other hand, $|u^{0}_{2}\rangle$ represents a column vector that is the remainder of $|u^{0}\rangle$ after removing the components whose labels correspond to the sources and sinks. Similarly, we define $\langle v^{0}_{1}|$ to be a row vector whose components are obtained from the row vector $\langle v^{0}|$ with component labels corresponding to that of the sources and the sinks; while $\langle v^{0}_{2}|$ represents a row vector that is the remainder of $\langle v^{0}|$ after removing the components whose labels correspond to the sources and sinks. To simplify the notation, we will represent $c(H)$ by $c$ without explicitly showing its $H$ dependence. Note that since $Q|u^{0}\rangle=0$, upon multiplying $\Omega\equiv(I-Q)^{-1}$ to both side of eq.(15) we have $\left(\begin{array}[]{c}H_{1}\\\ H_{2}\end{array}\right)=\left(\begin{array}[]{cc}\Omega_{11}&\Omega_{12}\\\ \Omega_{21}&\Omega_{22}\end{array}\right)\left(\begin{array}[]{c}f\\\ 0\end{array}\right)+c|u_{0}\rangle$ (16) or equivalently $\left(\begin{array}[]{c}H_{1}-cu^{0}_{1}\\\ H_{2}-cu^{0}_{2}\end{array}\right)=\left(\begin{array}[]{cc}\Omega_{11}&\Omega_{12}\\\ \Omega_{21}&\Omega_{22}\end{array}\right)\left(\begin{array}[]{c}f\\\ 0\end{array}\right)\;.$ (17) Consequently, we may write $H_{2}$ in the following form $|H_{2}\rangle=c\,|u^{0}_{2}\rangle+\Omega_{21}\Omega_{11}^{-1}|H_{1}\rangle-c\,\Omega_{21}\Omega_{11}^{-1}|u^{0}_{1}\rangle\;.$ (18) Using the definition that $c=\langle v^{0}_{1}|H_{1}\rangle+\langle v^{0}_{2}|H_{2}\rangle$, we obtain $c=\langle v^{0}_{1}|H_{1}\rangle+\langle v^{0}_{2}|\Omega_{21}\Omega_{11}^{-1}|H_{1}\rangle+c\left[\langle v_{2}^{0}|u^{0}_{2}\rangle-\langle v^{0}_{2}|\Omega_{21}\Omega_{11}^{-1}|u^{0}_{1}\rangle\right],$ or equivalently $c={\langle v^{0}_{1}|H_{1}\rangle+\langle v^{0}_{2}|\Omega_{21}\Omega_{11}^{-1}|H_{1}\rangle\over 1-\left[\langle v_{2}^{0}|u^{0}_{2}\rangle-\langle v^{0}_{2}|\Omega_{21}\Omega_{11}^{-1}|u^{0}_{1}\rangle\right]}$ (19) Substituting this result back to eq. (18), we obtain $H_{2}$ with computational complexity solely depending on $\Omega_{21}\Omega_{11}^{-1}$. Note that we only needs to invert the matrix $(I-Q)$ once and for all. Upon specifying the boundary nodes, one needs to reshuffle the rows and columns of the matrix as well as vectors – a relatively efficient operation. This operation groups the source nodes and sink nodes in one block to make easy the computation of $\Omega_{11}^{-1}$. Let us emphasize that our final expression is written in a rather general setting that it can be applied to cases when $P$ is either row-normalized or column-normalized. In the case of column-normalized $P$, we will have $|u_{\rm col.~{}norm.}^{0}\rangle=(\langle v^{0}_{\rm row~{}norm.}|)^{T}$ and $\langle v_{\rm col.~{}norm.}^{0}|=(|u^{0}_{\rm row~{}norm.}\rangle)^{T}$. The solution structures (18-19), however, does not change. Although an exact Green’s function method with Dirichlet boundary condition using spectral analysis (eigenvalues and eigenvectors) has been established by Chung and Yau Chung and Yau (2000), we find our method more convenient for computational purpose. With our method, the Greens function $\Omega$ is computed once and can be used for all different BC. This is immensely more efficient than finding all the eigenvalues and eigenvectors for every BC needed for each individual. Furthermore, it would not be practical to find all the eigenvectors of matrices resulting from networks of millions of nodes. To apply our method, one may either choose to fully invert $(I-Q)$ or take its approximate form. The direct inversion of $(I-Q)$ may still be computationally challenging for a matrix of size millions by millions. In terms of approximations, we find the use of $(I-Q)^{-1}\equiv\lim_{M\to\infty}\Omega(M)$ particularly useful, with $\Omega(M)\equiv\left[I+P+\cdots+P^{M}-M|u_{0}\rangle\langle v_{0}|\right]\;.$ (20) This approximation gets better for larger $M$. This is because the larger $M$ is, the smaller the difference between $P^{M}$ and $|u_{0}\rangle\langle v_{0}|$. One may then use $\Omega_{21}(M){\Omega_{11}}^{\\!\\!\\!\\!-1}(M)$ in place of $\Omega_{21}{\Omega_{11}}^{\\!\\!\\!\\!-1}$. The quality of this approximation may be be verified comparing the two models: the exact solution(18-19) versus the approximate one (ie. replacing $\Omega_{21}{\Omega_{11}}^{\\!\\!\\!\\!-1}$ by $\Omega_{21}(M){\Omega_{11}}^{\\!\\!\\!\\!-1}(M)$ in the exact solution). The convergence of the approximate solution to the exact solution (eqs.(18-19)) was first tested on an artificially generated random network of $100$ nodes. Aside from the condition that the nodes do not form disjoint clusters, a pair of nodes has probability $p=0.1$ to be connected. One then randomly selects a sink node and a source node that are not directly linked. We expect to get very similar shape of the temperature-profile as in the exact case. This is because for the row-normalized matrix, the $|u^{0}\rangle$ vector being a column vector with $1$ in each entry may induce a small but uniform offset in the approximate solution. In Fig. 1, we plot the “temperature-profile” of the $15$ hottest nodes from the exact solution and the “temperature-profile” of the same nodes using our approximation solution of various $M$. A good agreement between the exact solution and the approximate solution is reached at about $M=10$. Figure 1: Comparison between the exact solution (bold line) eqs.(18-19) and our approximation. For both cases we plot the “hottest” nodes. For better visualization we shifted the profiles such that the first node coincide in the graph. We observe a good agreement between the exact solution and the approximation for $M=10$ in our artificial network. To test the usability of our approach in real world, we use the movielens database. MovieLens (movielens.umn.edu; grouplens.org) ratings are recorded on a five stars scale and contain additional information, such as the time at which an evaluation was made. The data set we downloaded contains $N=6040$ users $\times$ $M=3952$ movies. However, only a fraction $\xi_{M}=0.041$ of all possible votes were actually expressed. To be able to perform the calculation in reasonable time, we decide to further reduce the data size in each dimension by roughly $50$%. To preserve the statistical properties of the original data, the pruning is done randomly without bias. In particular, we tried to maintain the probability distribution of the number of votes per users, as well as the sparsity and the $N/M$ ratio. We want to stress that this is crucial when testing the performance of predictive algorithms on real data in an objective way. In fact, many recommender systems can be found in the literature that rely on dense voting matrices Goldberg et al. (2001); Waern (2004), at least in the traning data set. Typically, users who have judged too few items are struck out, as well as items that have received too few votes. We did not comply to such convention and made an effort to keep the filtering level as low as possible, although this makes predictions much more difficult. Figure 2: Prediction performance on movielens database. The heat conduction model outperforms the mean predictor and the Pearson correlation based method as well. $\xi$ denotes the fraction of possible votes in the matrix. The vertical line, corresponding approximately to the giant cluster formation threshold in the movie – movie network, has vote density $\xi\approx 2N^{-1/2}M^{-1/6}$ Bollobás (2001), where $N$ is the number of users, $M$ is the number of movies. Once filtered, we cast the data set in a vote matrix ${\mathbf{V}}$, with number of users $N=3020$ and number of movies $M=1976$. In this reduced vote matrix, the matrix element $V_{\alpha,i}$ represents the number of stars assigned to movie $j$ by user $\alpha$ and is set to zero for unexpressed votes. The total filling fraction of ${\mathbf{V}}$ is $\xi_{M}=0.0468$. The votes in ${\mathbf{V}}$ are then sorted according to their relative timestamps. The last $n_{\rm test}=10^{4}$ expressed votes are collected to form our test set, while the rest of the expressed votes form our training set. We denote by ${\mathbf{V}}(t)$ the vote matrix information up to time $t$. That is, in ${\mathbf{V}}(t)$ all the unexpressed votes up to time $t$ are set to have zero star. For the purpose of rating prediction, one will need a movie – movie network. To accomplish this task, one may compute the correlation coefficient $C_{ij}(t)$ between movie $i$ and movie $j$ using the expressed votes up to a certain time $t$ in the training set. Specifically, we denote $\mu_{i}(t)\equiv{1\over N}\sum_{\alpha=1}^{N}V_{\alpha,i}(t)$ and $\sigma_{i}^{2}(t)\equiv{1\over N}\sum_{\alpha=1}^{N}[V_{\alpha,i}(t)-\mu_{i}(t)]^{2}$. The correlation coefficient reads $C_{ij}(t)\equiv{\sum_{\alpha}[V_{\alpha,i}(t)-\mu_{i}(t)][V_{\alpha,j}(t)-\mu_{j}(t)]\over\sigma_{i}(t)\sigma_{j}(t)}\;.$ (21) With a specified cutoff $C_{\rm cut}$, one obtains an adjacency matrix $A(t)$, with $A_{ij}(t)=\theta(C_{ij}(t)-C_{\rm cut}(t))$. The value of $C_{\rm cut}(t)$ is set so that the average degree per node $k(t)$ for the movie – movie network has the same number of non-zero entries as $[{\mathbf{V}}(t)]^{T}[{\mathbf{V}}(t)]$. Keeping the test set data fixed, we progressively fill the vote matrix the training set data over time (using the relative time stamps), say up to time $t$. We then use $A(t)$ to construct the the propagator $D(t)$ based on the information accumulated up to $t$. For each viewer (user), the BC is simply given by the votes expressed by the user up to time $t$. In the event that a user only has one vote (or none) up to time $t$, the BC for that user is given by randomly choosing one (or two) movie(s) and use the average rating(s) of the movie(s) up to that time as the boundary values Bollobás (2001). We then use our algorithm to make predictions on the entire test set. This test protocol is intended to reproduce real application tasks, where one aims to predict future votes –which is, of course, much harder than predicting randomly picked evaluations. It is somewhat less realistic to fix the test set once and for all, but this has the advantage to allow for more objective comparisons of the results. Many different accuracy metrics have been proposed to assess the quality of recommendations (see ref. Herlocker et al. (2004)), we choose the Root Square Mean Error: $RSME=\sqrt{\sum_{(\beta,j)\in\rm test}(V_{\beta,j}^{{}^{\prime}}-V_{\beta,j})^{2}/n_{\rm test}},$ (22) where $V^{\prime}_{\beta,j}$ represents the predicted vote from our algorithm, $V_{\beta,j}$ represents the actual vote (rated by user $\beta$ on movie $j$) in the test set, and the sum runs over all expressed votes in the test set. In our experiments, the RSME is calculated, at different sparsity values $\xi$, on a unique test set. Fig. 2 summarizes the performance comparison of our model with the mean predictor (the prediction is simply given by the objects mean value) and the widely used Pearson correlation based method Resnick et al. (1994); Herlocker et al. (2000). Our model outperforms both after enough votes (of the order of $N^{1/2}M^{5/6}$) have been expressed. Since the dimensions of the vote matrix ${\mathbf{V}}$ is known in a real application, given the number of expressed votes, it is relatively easy to see where one stands in terms of information content and whether our method will perform well using the given partial information. In summary, we have devised a recommendation mechanism using analog to heat conduction. The innovation of our method is its capability to compute the Green’s function needed just once to accommodate all possible BC. In terms of generalization, it is apparent that our method can be applied to network with weighted edges, with $A_{ij}=w_{ij}\geq 0$. Whether such a generalization will improve the performance will be investigated in a separate publication. Finally, we stress that our study is not aimed to extract statistical properties out of networks through constructing model networks mimicking the real world networks Newman (2003c); Park and Newman (2004); nor are we pursuing analysis of slowly decaying eigenmodes Eriksen et al. (2003) in the absence of boundary condtitions. Instead, our goal is to provide a framework that is capable of providing individualized information extraction from a real world network. YCZ and MB were partially supported by Swiss National Science Foundation grant 205120-113842. YCZ acknowledges hospitality at Management School, UESTC, China, where part of the work is done. The research of YKY was supported by the Intramural Research Program of the National Library of Medicine at the NIH. ## References * Maslov and Y.C. Zhang (2001) S. Maslov and Y.C. Zhang, Phys. Rev. Lett. 87, 248701 (2001). * Blattner et al. (2006) M. Blattner, Y.C. Zhang, and S. Maslov, Physica A 373, 753 (2006). * Adomavicius and Tuzhilin (2005) G. Adomavicius and A. Tuzhilin, IEEE Transactions on Knowledge and Data Engineering 17, 734 (2005), ISSN 1041-4347. * Press et al. (1992) W. Press, S. Teukolsky, B. Flannery, and V. Vetterling, _Numerical Recipes in C_ (Cambridge University Press, New York, USA, 1992). * Korniss et al. (2006) G. Korniss, M. Hastings, K. Bassler, M. Berryman, B. Kozma, and D. Abbott, Phys. Lett. A 350, 324 (2006), ISSN 1046-8188. * Wu (2004) F. Wu, J. Phys. A 37, 6653 (2004). * Chung and Yau (2000) F. Chung and S. Yau, Journal of Combinatorial Theory(A) pp. 141–214 (2000). * Goldberg et al. (2001) K. Goldberg, T. Roeder, D. Guptra, and C. Perkins, Information Retrieval 4, 133 (2001). * Waern (2004) A. Waern, User Modeling and User-Adapted Interaction 14, 201 (2004). * Bollobás (2001) Assuming that the vote matrix is filled randomly, one can show that the density needs to be $\xi\geq N^{-1/2}M^{-1/6}$ to have in the movie–movie network a linking probability $p\geq M^{-1/3}$, which marks the onset of giant cluster formation. See B. Bollobás, Random Graphs, chap. 6 (Cambridge University Press, New York, USA, 2001). * Bollobás (2001) This is to avoid the artifact of null information retrieval: e.g. assume only one boundary node with a specified temperature, all nodes will reach the same temperature upon thermal equilibrium. * Herlocker et al. (2004) J. Herlocker, J. Konstan, L. Terveen, and J. Riedl, ACM Trans. Inf. Syst. 22, 5 (2004), ISSN 1046-8188. * Resnick et al. (1994) P. Resnick, N. Iacovou, M. Suchak, P. Bergstorm, and J. Riedl, in _Proceedings of ACM 1994 Conference on Computer Supported Cooperative Work_ (ACM, Chapel Hill, North Carolina, 1994), pp. 175–186. * Herlocker et al. (2000) J. Herlocker, J. Konstan, and J. Riedl, in _Computer Supported Cooperative Work_ (2000), pp. 241–250. * Newman (2003c) M. Newman, SIAM Review 45, 167 (2003c). * Park and Newman (2004) J. Park and M. Newman, Phys. Rev. E 70, 066117 (2004). * Eriksen et al. (2003) K. A. Eriksen, I. Simonsen, S. Maslov, and K. Sneppen, Phys. Rev. Lett. 90, 148701 (2003).

arxiv-papers

2024-09-04T02:48:54.339334

0803.2208

# Double-exciton component of the cyclotron spin-flip mode in a quantum Hall ferromagnet S. Dickmann and V.M. Zhilin Institute for Solid State Physics of RAS, Chernogolovka 142432, Moscow District, Russia. ###### Abstract We report on the calculation of the cyclotron spin-flip excitation (CSFE) in a spin-polarized quantum Hall system at unit filling. This mode has a double- exciton component which contributes to the CSFE correlation energy but can not be found by means of a mean field (MF) approach. The result is compared with available experimental data. PACS numbers 73.21.Fg, 73.43.Lp, 78.67.De A two-dimensional electron gas (2DEG) in a high perpendicular magnetic field possesses many remarkable features.an82 In particular, it presents a rare case of strongly correlated system governed by real Coulomb interaction (not by a model Hamiltonian!) where, nevertheless, some solutions of the quantum many-body problem can be found exactly. Indeed, under the conditions of integer quantum Hall effect (when the filling factor is $\nu=1,2,3,...$), the one-cyclotron magnetoplasma and the lowest spin-flip modes are calculated analytically to the leading order in the parameter $r_{\rm c}\\!=\\!E_{\rm C}/\hbar\omega_{c}$. by81 ; by83 ; ka84 ; di05 [$\omega_{c}$ is the cyclotron frequency; $E_{\rm C}\\!=\alpha e^{2}/\kappa l_{B}$ is the characteristic interaction energy, $\alpha$ being the average form-factor related to the finite thickness of the 2DEG ($0.3\,\raise 1.72218pt\hbox{$<$}\kern-8.00003pt\lower 3.01385pt\hbox{$\sim$}\,\alpha\\!<\\!1$); $l_{B}$ is the magnetic length.] This astounding property is the feature of either filled or half-filled highest-occupied Landau level (LL) where the simplest-type excitations are single excitons or superposition of single-exciton modes. The many-body problem is thereby reduced to the two-body one, i.e. to the interaction of electron with an effective hole. Being quite in the context of similar studies, the present letter concerns however the case which can not be reduced to a single-exciton problem. We remind that 2DEG excitons are characterized by sublevels $a\\!=\\!(n_{a},\sigma_{a})$ and $b\\!=\\!(n_{b},\sigma_{b})$ where electron is promoted from the $n_{a}$-th LL with spin-component $S_{z}\\!=\\!\sigma_{a}$ to the $n_{b}$-th LL with $S_{z}\\!=\\!\sigma_{b}$. The relevant quantum numbers are $\delta n\\!=\\!n_{b}\\!-\\!n_{a}$, $\delta S_{z}\\!=\\!\sigma_{b}\\!-\\!\sigma_{a}$, and the two-dimensional (2D) wave vector ${\bf q}$. The single exciton problem is exactly solvable in the following cases: (i) at odd filling $\nu$ when $\delta n\\!=\\!1$ and $\delta S_{z}\\!=\\!0$ (magnetoplasmon) or $\delta n\\!=\\!0$ and $\delta S_{z}\\!=\\!-1$ (spin wave);by81 ; ka84 ; dizh05 (ii) at even $\nu$ when $\delta n\\!=\\!1$ and $\delta S_{z}\\!=\\!0,\pm 1$ (magnetoplasmon and spin- flip triplet).ka84 ; di05 ; dizh05 At the same time the two-body problem may be discussed within a MF approach (in some publications called ‘time-dependent Hartree-Fock’ approximation mc85 ; lo93 ) which excludes any quantum fluctuations from a single exciton to double- or many-exciton states. For the above simplest cases of $\delta n$ and $\delta S_{z}$, the MF calculation gives an asymptotically exact result which may be found perturbatively to the first order in $r_{\rm c}$,foot1 because these $(\delta n,\,\delta S_{z})$ sets can not correspond to any states except single-exciton modes. Any complication of $(\delta n,\delta S_{z})$ makes the calculations substantially more difficult due to the necessary expansion of the basis to the entire continuous set of many-exciton states with the same total numbers $\delta n$, $\delta S_{z}$, and ${\bf q}$. For example, the double-cyclotron plasmon with $\delta n\\!=\\!2$, $\delta S_{z}\\!=\\!0$ and with given ${\bf q}$ ‘dissociates’ into double-exciton states consisting of one-cyclotron plasmon’s pairs with the total momentum equal to ${\bf q}$.ka84 At odd $\nu$, a similar ‘dissociation’ occurs for the CSFE, where $\delta n\\!=\\!-\delta S_{z}\\!=\\!1$. The proper double-exciton states are pairs of a magnetoplasmon ($\delta n\\!=\\!1,\,\delta S_{z}\\!=\\!0$) and a spin wave ($\delta n\\!=\\!0,\,\delta S_{z}\\!=\\!-1$). The problem thus changes from the two- body case to the four-body one, and the correct solution should be presented in the form of combination of the single-exciton mode and continuous set of double-exciton states.dizh05 It is important that in both cases the desired solution corresponds to a discrete line against the background of a continuous spectrum of free exciton pairs. The technique of correct solution has to be of essentially non-Hartree-Fock (non-HF) type. Actually this letter concerns the fundamental question of consistency of the MF approach. By considering the case of unit filling factor where the number of electrons is equal to the number of magnetic flux quanta $N_{\phi}$, now we report on a study of the CSFE with ${\bf q}\\!=\\!0$. This state is optically active and identified in the ILS experiments.pi92 ; va06 Besides, it is exactly this spin-flip magnetoplasma mode which is the key component of the elementary perturbation used in the microscopic approach to the skyrmionic problem.di02 The calculation is performed in ‘quasi-analytical’ way which should, in principle, lead to the result which is exact in the leading approximation in $r_{\rm c}$. In our case the envelope function determining the combination of the double-exciton states is one-dimensional — i.e., it only depends on the modulus of the excitons’ relative momentum. This function is chosen in the form of expansion over infinite orthogonal basis, where every basis vector obeys a specific symmetry condition necessary for the total envelope function. Even to the first-order approximation in $r_{\rm c}$, we obtain a non-HF correction to the former HF result pi92 ; lo93 for the CSFE energy. As a technique, we use the excitonic representation (ER) which is a convenient tool for description of the 2DEG in a perpendicular magnetic field.dz83 ; di05 ; dizh05 When acting on the vacuum $|{\rm 0}\rangle$ (in our case $|{\rm 0}\rangle\\!=\\!|\overbrace{\uparrow,\uparrow,...\uparrow}^{\displaystyle{\vspace{-5mm}\mbox{\tiny{$\;N_{\phi}$}}^{\vphantom{\int^{\infty}}}}}\,\rangle$), the exciton operators produce a set of basis states which diagonalize the single-particle term of the Hamiltonian and some part ${\hat{H}}_{\rm ED}$ of the interaction Hamiltonian.dizh05 ; di02 Exciton states are classified by ${\bf q}$, and it is essential that in this basis the LL degeneracy is lifted. So, the generic Hamiltonian is ${\hat{H}}\\!=\\!{\hat{H}}_{1}\\!+\\!{\hat{H}}_{\rm int}$ where $\begin{array}[]{l}{\hat{H}}_{1}=\sum\limits_{\sigma}\int\\!d{\bf r}\,{\hat{\Psi}}_{\sigma}^{\dagger}({\bf r})\left[\frac{1}{2m^{*}}\left(i{\vec{\nabla}}-{e\vec{A}/c}\right)^{2}\\!+\\!g\mu_{B}B{\hat{S}}_{z}\right]\\!{\hat{\Psi}}_{\sigma}({\bf r})\quad\mbox{and}\\\ {\hat{H}}_{\rm int}=\frac{1}{2}\sum\limits_{\sigma_{1},\sigma_{2}}\int\\!d{\bf r}_{1}d{\bf r}_{2}\,{\hat{\Psi}}_{\sigma_{2}}^{\dagger}({\bf r}_{2}){\hat{\Psi}}_{\sigma_{1}}^{\dagger}({\bf r}_{1})U({\bf r}_{1}\\!-\\!{\bf r}_{2}){\hat{\Psi}}_{\sigma_{1}}({\bf r}_{1}){\hat{\Psi}}_{\sigma_{2}}({\bf r}_{2}).\end{array}$ $None$ Choosing, e.g, the Landau gauge and substituting for the Schrödinger operator ${\hat{\Psi}}^{\dagger}_{\sigma}\\!\\!=\\!\\!\sum_{np}a^{\dagger}_{np\sigma}\psi_{np\sigma}^{*}$ (indexes $n,p,\sigma$ label the LL number, intra-LL state, and spin sublevel), one can express the Hamiltonian (1) in terms of combinations of various components of the density-matrix operators.di05 ; dizh05 ; di02 These are exciton operators defined as dz83 ; di05 ; dizh05 ; di02 ${{\cal Q}}_{ab{\bf q}}^{{\dagger}}={N_{\phi}}^{-1/2}\sum_{p}\,e^{-iq_{x}p}b_{p+\frac{q_{y}}{2}}^{{\dagger}}\,a_{p-\frac{q_{y}}{2}}\quad\mbox{and}\quad{{\cal Q}}_{ab{\bf q}}={{\cal Q}}_{ba\,-{\bf q}}^{{\dagger}}$ $None$ and obeying the commutation algebra dizh05 $\left[{\cal Q}_{cd\,{\bf q}_{1}}^{{\dagger}},{\cal Q}_{ab\,{\bf q}_{2}}^{{\dagger}}\right]\\!\equiv\\!N_{\phi}^{-1/2}\left(e^{-i({\bf q}_{1}\\!\times{\bf q}_{2})_{z}/2}\delta_{b,c}{\cal Q}_{ad\,{\bf q}_{1}\\!+\\!{\bf q}_{2}}^{{\dagger}}-e^{i({\bf q}_{1}\\!\times{\bf q}_{2})_{z}/2}\delta_{a,d}{\cal Q}_{cb\,{\bf q}_{1}\\!+\\!{\bf q}_{2}}^{{\dagger}}\right)$ $None$ (in our units $l_{B}\\!=\\!\sqrt{c\hbar/eB}\\!=\\!1$). Here $a,b,c,...$ are binary indexes (see above), which means that $a_{p}^{\dagger}\\!=\\!a_{n_{a}p\,\sigma_{a}}^{\dagger}$, $b_{p}^{\dagger}\\!=\\!{a}^{\dagger}_{n_{b}p\,\sigma_{b}}$… We will also employ for binary indexes the notations $n\\!=\\!(n,\uparrow)$ and ${\overline{n}}\\!=\\!(n,\downarrow)$, so that the single-mode component of the CSFE is defined as ${\cal Q}^{\dagger}_{0\overline{1}{\bf q}}|0\rangle$. The interaction Hamiltonian can be presented as ${\hat{H}}_{\rm int}\\!=\\!{\hat{H}}_{\rm ED}\\!+{\hat{H}}^{\prime}$ where ${\hat{H}}_{\rm ED}$, if applied to the state ${{\cal Q}}_{ab{\bf q}}^{{\dagger}}|0\rangle$, yields a combination of single-exciton states with the same numbers $\delta n$, $\delta S_{z}$, and ${\bf q}$ (see Refs. dizh05, ; di02, and therein ${\hat{H}}_{\rm ED}$ expressed in terms of exciton operators). In the framework of the above HF approximation, the CSFE correlation energy pi92 ; lo93 is obtained from the equation ${\cal E}_{0\overline{1}}({q})\\!=\\!\langle{\mbox{\rule{0.0pt}{11.38109pt}}}0|{\cal Q}_{0\overline{1}{\bf q}}[{\hat{H}}_{\rm int},{\cal Q}_{0\overline{1}{\bf q}}^{{\dagger}}]|0{\mbox{\rule{0.0pt}{11.38109pt}}}\rangle$ where only the ${\hat{H}}_{\rm ED}$ part of the interaction Hamiltonian contributes to the expectation. In the following, we need this so-called HF value at $q=0$, namely ${\cal E}_{0\overline{1}}(0)\\!\equiv\\!{\cal E}_{\rm HF}\\!=\\!\frac{1}{2}\int_{0}^{\infty}{p^{3}dp}V(p)e^{-p^{2}/2}$ where $2\pi V(q)$ is the Fourier component of the effective Coulomb vertex in the layer.pi92 (In the strictly 2D limit $\alpha\\!\to\\!1$, and $V(q)\\!\to e^{2}/\kappa l_{B}q$.) The problem arises due to the ‘troublesome’ part ${\hat{H}}^{\prime}$ of the interaction Hamiltonian which can not be diagonalized in terms of single- exciton states. For our task we keep in ${\hat{H}}^{\prime}$ only the terms contributing to $\left[{\hat{H}}^{\prime},{\cal Q}_{0\overline{1}{\bf q}}^{{\dagger}}\right]|0\rangle$ and besides preserving the cyclotron part of the total energy (i.e. commuting with ${\hat{H}}_{1}$). In terms of the ER these are di05 ; dizh05 ${\hat{H}}_{0{\overline{1}}}^{\prime}=\sum_{\bf q}\frac{q^{2}}{2}V(q)e^{-q^{2}/2}{{\cal Q}}_{01{\bf q}}^{{\dagger}}{{\cal Q}}_{\overline{0}\,\overline{1}{\bf q}}\;+\;\mbox{H.c.}$ $None$ Using Eqs. (3) and identities ${\cal Q}^{\dagger}_{aa{\bf q}}|0\rangle\\!\equiv\\!N_{\phi}^{-1/2}\delta_{{\bf q},{\bf 0}}|0\rangle$ if $a\\!=\\!(0,\uparrow)$ and ${\cal Q}^{\dagger}_{aa{\bf q}}|0\rangle\\!\equiv\\!0$ if $a\\!\not=\\!(0,\uparrow)$, one can find that the operation of ${\hat{H}}_{0{\overline{1}}}^{\prime}$ on vector ${\cal Q}_{0\overline{1}{\bf q}}^{{\dagger}}|0\rangle$ results in a combination of states of the type of ${N_{\phi}}^{-1/2}\sum_{\bf s}f({\bf s}){\cal Q}_{0\overline{0}\,{\bf q}/2\\!-\\!{\bf s}}^{\dagger}{\cal Q}_{01\,{\bf q}/2\\!+\\!{\bf s}}^{\dagger}|0\rangle$ with a certain regular and square integrable envelope function, $\int\\!|f({\bf s})|^{2}d{\bf s}\\!\sim\\!1$. The norm of this combination is not small as compared to $\langle 0|{\cal Q}_{0\overline{1}{\bf q}}{\cal Q}^{\dagger}_{0\overline{1}{\bf q}}|0\rangle\\!\equiv\\!1$, and the terms (4) must be taken into account when calculating the CSFE energy. On the other hand, if the set of double-exciton states $|{\bf s},{\bf q}\rangle\\!=\\!{\cal Q}_{0\overline{0}\,{\bf q}/2\\!-\\!{\bf s}}^{{\dagger}}{\cal Q}_{01\,{\bf q}/2\\!+\\!{\bf s}}^{{\dagger}}|0\rangle$ is considered, then one finds that they, first, are not exactly but ‘almost’ orthogonal: $\langle{\bf q}_{1},{\bf s}_{1}|{\bf s}_{2},{\bf q}_{2}\rangle\\!=\\!\delta_{{\bf q}_{1},{\bf q}_{2}}\left\\{\delta_{{\bf s}_{1},{\bf s}_{2}}\right\\}$, where $\left\\{\delta_{{\bf s}_{1},{\bf s}_{2}}\right\\}\\!\equiv\\!\delta_{{\bf s}_{1},{\bf s}_{2}}\\!-e^{i({\bf s}_{1}\\!\times{\bf s}_{2})_{z}}\\!/N_{\phi}$; and, second, $|{\bf s},{\bf q}\rangle$ satisfies the equation $\left[{\hat{H}}_{\rm int},{\cal Q}_{0\overline{0}\,{\bf q}/2\\!-\\!{\bf s}}^{\dagger}{\cal Q}_{01\,{\bf q}/2\\!+\\!{\bf s}}^{{\dagger}}\right]|0\rangle=\left[{\cal E}_{\rm sw}(|{\bf q}/2\\!+\\!{\bf s}|)+{\cal E}_{\rm mp}(|{\bf q}/2\\!-\\!{\bf s}|)\right]|{\bf s},{\bf q}\rangle+|{\tilde{\varepsilon}}\rangle,$ $None$ where the state $|{\tilde{\varepsilon}}\rangle$ has a negligibly small norm: $\langle{\tilde{\varepsilon}}|{\tilde{\varepsilon}}\rangle\sim E_{\rm C}^{2}/N_{\phi}$. Therefore the double-exciton state $|{\bf s},{\bf q}\rangle$ in the thermodynamic limit actually corresponds to free noninteracting excitons: one of them is a spin exciton (spin wave) with energy $|g\mu_{B}B|\\!+\\!{\cal E}_{\rm sw}$ where ${\cal E}_{\rm sw}(q)\\!=\\!\int_{0}^{\infty}{pdp}V(p)e^{-p^{2}/2}\left[1\\!-\\!J_{0}(pq)\right],$ $None$ while the other is a magnetoplasmon with energy $\hbar\omega_{c}\\!+\\!{\cal E}_{\rm mp}$ where ${\cal E}_{\rm mp}(q)=\frac{q^{2}}{2}V(q)e^{-q^{2}/2}\\!+\\!\int_{0}^{\infty}\\!{pdp}e^{-p^{2}/2}V(p)\left(1-\frac{p^{2}}{2}\right)\left[1\\!-\\!J_{0}(pq)\right]$ $None$ [$J_{0}$ is the Bessel function (cf. Refs. by81, ; ka84, )]. Thus we try for the CSFE state the vector $|X_{\bf q}\rangle\\!=\\!{\hat{X}}_{\bf q}|0\rangle$ where ${\hat{X}}_{\bf q}$ is a combined operator ${\hat{X}}_{\bf q}={\cal Q}^{\dagger}_{0\overline{1}\,{\bf q}}+\frac{1}{\sqrt{2N_{\phi}}}\sum_{\bf s}\varphi_{q}({\bf s}){\cal Q}_{0\overline{0}\,{\bf q}/2\\!-\\!{\bf s}}^{\dagger}{\cal Q}_{01\,{\bf q}/2\\!+\\!{\bf s}}^{{\dagger}}\,.$ $None$ Actually only a certain ‘antisymmetrized’ part $\\{\varphi_{q}\\}$ of the envelope functions contributes to the double-exciton combination in $|X_{\bf q}\rangle$.by83 ; di05 ; dizh05 In our case the antisymmetry transform is $\\{\varphi_{q}\\}=\left[\varphi_{q}({\bf s})-\frac{1}{N_{\phi}}\sum_{{\bf s}^{\prime}}e^{i({\bf s}\times{\bf s}^{\prime})_{z}}\varphi_{q}({\bf s}^{\prime})\right]$. Such a specific feature originates from the generic permutation antisymmetry of the Fermi wave function of our many-electron system. We may therefore consider only ‘antisymmetric’ functions for which $\varphi_{q}\\!=\\!\\{\varphi_{q}\\}/2\,.$ $None$ Our task is to find the energy of the eigenvector $|X_{\bf q}\rangle$ and the ‘wave function’ $\varphi_{q}({\bf s})$, assuming that the latter is regular and square integrable. If $E_{q}$ is the correlation part of the total CSFE energy (namely, $E_{\rm CSFE}\\!=\\!E_{\rm vac}\\!+\\!|g\mu_{B}B|\\!+\\!\hbar\omega_{c}\\!+\\!E_{q}$), then $E_{q}$ is found from foot2 $\left[{\hat{H}}_{\rm ED}\\!+\\!{\hat{H}}_{0\overline{1}}^{\prime}\,,{\hat{X}}_{\bf q}\right]|0\rangle=E_{q}|X_{\bf q}\rangle.$ $None$ Now we project this equation onto two basis states $|{\bf p},{\bf q}\rangle$ and ${\cal Q}^{\dagger}_{0\overline{1}{\bf q}}|0\rangle$, and obtain two closed coupled equations $\begin{array}[]{l}\left(2N_{\phi}\right)^{1/2}\left\langle{\bf q},{\bf p}|\left[{\hat{H}}_{01}^{\prime},{\cal Q}^{\dagger}_{0\overline{1}{\bf 0q}}\right]|0\right\rangle{}{}\\\ {}\qquad{}\qquad+\sum\limits_{\bf s}\varphi_{q}({\bf s})\left\langle{\bf q},{\bf p}|\left[{\hat{H}}_{ED},{\cal Q}_{0\overline{0}\,\\!{\bf q}/2\\!-\\!{\bf s}}^{\dagger}{\cal Q}_{01\,\\!{\bf q}/2\\!+\\!{\bf s}}^{\dagger}\right]|0\right\rangle=E_{q}\varphi_{q}({\bf p})\end{array}$ $None$ and ${\cal E}_{0\overline{1}}(q)+(2N_{\phi})^{-1/2}\sum_{\bf s}\varphi_{q}({\bf s})\left\langle 0|{\cal Q}_{0\overline{1}{\bf q}}\left[{\hat{H}}_{01}^{\prime},{\cal Q}_{0\overline{0}\,\\!{\bf q}/2\\!-\\!{\bf s}}^{\dagger}{\cal Q}_{01\,{\bf q}/2\\!+\\!{\bf s}}^{\dagger}\right]|0\right\rangle=E_{q}$ $None$ for $E_{q}$ and $\varphi_{q}({\bf p})$. Next step is a routine treatment of Eqs. (11) and (12) in terms of calculation of commutators guided by commutation rules (3). In the ${\bf q}\\!=\\!0$ case, which we immediately consider, the function $\varphi_{0}({\bf p})$ depends only on the modulus of ${\bf p}$. As a result we obtain foot3 $\begin{array}[]{r}\left[E-{\cal E}_{\rm sw}(q)-{\cal E}_{\rm mp}(q)\right]\varphi_{0}(q)+\displaystyle{\int_{0}^{\infty}\\!\\!sds}\left[K_{1}(s,q)\varphi_{0}(s)\vphantom{\displaystyle{\frac{K_{2}(s)}{\pi}\\!\int_{0}^{\pi}}}\right.{}{}{}\\\ {}\qquad{}\qquad{}\qquad{}+\left.\displaystyle{\frac{K_{2}(s)}{\pi}\\!\int_{0}^{\pi}\\!d\phi}\left(1\\!-\\!\cos[{\bf s}\times{\bf q}]\right)\varphi_{0}(|{\bf q}\\!+\\!{\bf s}|)\right]=g(q)\end{array}$ $None$ and $E-{\cal E}_{\rm HF}=\frac{1}{\sqrt{2}}\int_{0}^{\infty}\\!\\!dpp^{3}V(p)e^{-p^{2}/2}\varphi_{0}(p)$ $None$ (we omit subscript 0 in $E_{0}$), where $g(q)=\frac{q^{2}}{2\sqrt{2}}V(q)e^{-q^{2}/2}-\frac{1}{2\sqrt{2}}\int_{0}^{\infty}p^{3}V(p)e^{-p^{2}/2}J_{0}(pq)dp,$ $None$ $K_{1}(q,s)=\frac{s^{2}}{2}e^{-s^{2}/2}V(s)J_{0}(qs),\quad\mbox{and}\quad{}K_{2}(s)=\left(2\\!-\\!\frac{s^{2}}{2}\right)V(s)e^{-s^{2}/2}$ $None$ ($\phi$ in Eq. (13) is the angle between ${\bf s}$ and ${\bf q}$). The problem has thus been integrable to yield in the thermodynamic limit a pair of coupled integral equations for one-dimensional function $\varphi_{0}(q)$ and the eigenvalue $E$. In order to solve this system we employ the method of expansion in orthogonal functions $\varphi_{0}(q)=\sum_{n=1,3,5,...}^{2N-1}A_{n}\psi_{n}(q)\,.$ $None$ These $\psi_{n}\\!=\\!\sqrt{2}L_{n}(q^{2})e^{-q^{2}/2}$ with odd indexes of the Laguerre polynomials ($\int_{0}^{\infty}qdq\psi_{m}\psi_{n}\\!=\\!\delta_{m,n}$) are chosen as a natural basis satisfying: (i) the property of integrability and expected analytic and asymptotic features of $\varphi_{0}(q)$; (ii) the antisymmetry condition (9). In other words, we change from the basis formed by the set of nonorthogonal double-exciton states $|{\bf s},{0}\rangle\\!\equiv\\!{\cal Q}_{0\overline{0}-\\!{\bf s}}^{\dagger}{\cal Q}_{01{\bf s}}^{\dagger}|0\rangle$ to a new set of basis states $|{\rm DX},n\rangle\\!=\\!(2N_{\phi})^{-1/2}\sum_{\bf s}\psi_{n}(s)|{\bf s},{0}\rangle$ which are strictly orthogonal. Indeed, one can check by employing Eq. (3) and identity $\frac{1}{N_{\phi}}\sum_{\bf s}e^{i({\bf q}\times{\bf s})_{z}}\psi_{n}(q)\\!\equiv\\!\int_{0}^{\infty}sdsJ_{0}(qs)\psi_{n}(s)\\!\equiv\\!-\psi_{n}(q)$ that $\langle m,{\rm DX}|{\rm DX},n\rangle\\!\equiv\\!\delta_{m,n}$. The integer number $N$ is dimensionality of this new double-exciton basis. After substitution of Eq. (17) into Eq. (14) the latter takes the form: $E=F$, where $F={\cal E}_{\rm HF}+\frac{1}{\sqrt{2}}\sum_{n=1,3,5,...}^{2N-1}A_{n}\int_{0}^{\infty}dpp^{3}V(p)e^{-p^{2}/2}\psi_{n}(p).$ $None$ Let us consider the ideal 2D case where $V(q)=1/q$. (Here and below energy is measured in units of $e^{2}/\kappa l_{B}$.) After substitution of the expansion (18) into Eq. (13), further multiplication by basis functions $\psi_{m}(q)$ and integration ($\int...qdq$) lead to the set of $N$ linear algebraic equations with respect to $A_{n}$. Finding $A_{n}$ for a given $E$ and substituting them into Eq. (18), we obtain $F(E)$. The condition $F(E)\\!=\\!E$ yields the desired result $E=E_{\rm SF}$. Figure 1: Graphical solution of Eqs. (13) and (14). Intersection of the $F\\!=\\!E$ straight line with the dotted line corresponds to the CSFE energy, $E_{\rm SF}\\!\approx 0.71\\!$. See text for details. Fig. 1 shows the result of calculations for $N=50$. The lines which are restricted by vertical asymptotes reflect the result of calculation of $F(E)$. Points of singularity $E^{(i)}$, at which $F$ goes to infinity, are roots of the equation $D_{N}(E)=0$ where $D_{N}$ is the determinant corresponding to the “left-side” of the set of equations for $A_{n}$. By increasing $N$ we increase the order of equation $D_{N}(E)=0$, so that this has up to $N$ real roots. Indeed, when observing the evolution of $F(E)$ with increasing $N$, one finds that the number of singular points grows, and they become more densely placed. For $N\to\infty$ one could expect that a singular point appears within an arbitrarily small vicinity of every value $E$. Since all the vertical asymptotes $E=E^{(i)}$ are crossed by the straight line $F=E$ (see Fig. 1), we come to the conclusion that for any $E$ there is a singular solution of Eqs. (13) and (14). Such solutions with singular functions $\varphi(q)$ form a band. The physical meaning of this result is quite transparent. Namely, the band corresponds to energy ${\cal E}_{\rm sw}(q)\\!+\\!{\cal E}_{\rm mp}(q)$ of unbound exciton pairs. Now we only consider the solution $E=F(E)$, where the $F\\!=\\!E$ line crosses a conventional envelope curve tracing the regions of regularity of $\varphi_{0}$ determined by Eq. (17). Such regions at a finite $N$ should be as distant as possible from the points of singularity, and we simply define them as the vicinities of “middle” points $\overline{E}^{(i)}=\frac{1}{2}(E^{(i)}+E^{(i+1)})$. The envelope curve may obviously be defined as the line passing through the points $[\overline{E}^{(i)},\,F(\overline{E}^{(i)})]$. The intersection with the straight line $F=E$ occurs at the only point stable with respect to evolution of this picture at $N\to\infty$. This intersection point is readily seen in Fig. 1. Fig. 1 shows the build-up of singular points (vertical lines) with vanishing $E$ and vice versa a certain rarefication of singularities in the vicinity of $E_{\rm SF}$. The former reflects growth of the density of states at the bottom of the exciton band whereas the latter is a usual effect of the “levels’ repulsion”. Note that the non-Hartree-Fock shift for the CSFE level is positive as compared to the value $E_{\rm HF}\\!=\\!0.627$. This is expected because the repulsion of the CSFE from the lower-lying crowded states of unbound excitons should be stronger than from the upper states having comparatively low density. At the same time, one can also see in Fig. 1 some trend towards the concentration of singularity points $E^{(i)}$ at higher energies $E$. This is evidently a consequence of the density of states growth at the top of the exciton band. In general, the larger is $N$ the more accurate is the calculation of $\varphi_{0}(q)$ and $E$, i.e. the envelope curve in Fig. 1 becomes discernible and may be drawn only at considerable $N$. At the same time the analysis reveals that the intersection point with the $F\\!=\\!E$ line is rather stable and only weakly depends on $N$. This feature prompts us to consider the case $N\\!=\\!1$ where double-exciton states mixed with ${\cal Q}^{\dagger}_{0\overline{1}{\bf q}}|0\rangle$ are modelled by a single vector $|{\rm DX},1\rangle$. Actually the $N\\!=\\!1$ approximation for the problem determined by Eqs. (13), (14) and (17) is equivalent to a variational procedure for the trial double-mode state $|{X}_{0}^{\rm DM}\rangle\\!=\\!{\cal Q}^{\dagger}_{0\overline{1}{\bf 0}}|0\rangle\\!+\\!A_{1}|{\rm DX},1\rangle$, where the correlation part of the excitation energy is found from equation $E=\mathop{\rm min}\nolimits\limits_{A_{1}}\\!\left(\frac{\langle{X}_{0}^{\rm DM}|{\hat{H}}_{\rm int}|{X}_{0}^{\rm DM}\rangle}{\langle{X}_{0}^{\rm DM}|{X}_{0}^{\rm DM}\rangle}\right)\\!-\\!{E}_{\rm vac}^{\rm int}\,$ $None$ (${E}_{\rm vac}^{\rm int}$ denotes the correlation part of the ground-state energy). After minor manipulations we find that this simple double-mode approximation (DMA) reduces our problem to the secular equation $\mbox{det}\\!\left|(E-{\EuScript E}_{i})\delta_{ik}+(1\\!-\\!\delta_{ik}){\EuScript D}_{ik}\right|\\!=\\!0$ (indexes $i$ and $k$ are 1 or 2), where ${\EuScript E}_{1}\\!=\\!\int_{0}^{\infty}\\!qdqV(q)\epsilon(q),\;$ ${\EuScript E}_{2}\\!=\\!{\cal E}_{\rm HF},\;$ and ${}\;{\EuScript D}_{12}\\!\equiv\\!{\EuScript D}_{21}\\!=\\!\int_{0}^{\infty}\\!qdqV(q)d(q)$ with $\epsilon\\!=\\!2q^{2}(1\\!-\\!q^{2})^{2}e^{-3q^{2}/2}\\!+\\!\frac{1}{2}(q^{2}\\!-\\!5q^{2}\\!+\\!q^{4})e^{-q^{2}}\\!-\\!\frac{1}{16}(q^{2}\\!-\\!4)^{3}e^{-3q^{2}/4}\\!+\\!({q^{2}}/{2}\\!-\\!2)e^{-q^{2}/2}$ and $d\\!=\\!q^{2}(q^{2}\\!-\\!1)e^{-q^{2}}.$ Only the largest root of this secular equation has physical meaning. In the ideal 2D case we easily obtain the DMA correlation energy of the CSFE: $E_{\rm SF}=0.766$. Comparing this result with Fig. 1 we conclude that even the DMA works rather well. Figure 2: Main picture: DMA and HF shifts in dimensionless units against the form-factor parameter $b$. Inset: DMA shift against the magnetic field when $b\\!=\\!5.45\,B^{-1/2}$ ($b\\!=\\!0.213\,l_{B}\\!/\mbox{nm},\;$ $l_{B}$ in nm’s, $B$ in Teslas); symbols are experimental data for the $25\,$nm quantum wells.va06 Fig. 2 shows the CSFE correlation energy calculated within the DMA and employing the HF approximation, if the vertex $V$ for a real 2DEG is defined as $V\\!=\\!F_{b}(q)/q$ with the formfactor $F_{b}(q)\\!=\\!\frac{1}{8}\left(1\\!+\\!\frac{q}{b}\right)^{-3}\left[8\\!+\\!9\frac{q}{b}\\!+\\!3\left(\frac{q}{b}\right)^{2}\right].$an82 ; lo93 Here $b=b_{0}l_{B}$ is a dimensionless parameter corresponding to dimensionless $q$. ($b_{0}$ is considered to be independent of the magnetic field.) It is seen that the non-HF shift of the CSFE energy, being about $15\%$ in the strict 2D limit (i.e., in the $b\\!\to\\!\infty$ case), becomes smaller ($\sim\\!5-6\%$) in real samples. This difference is not observable experimentally.va06 Meanwhile, the DMA results are in good agreement with experimental data where the CSFE correlation energy is measured as a function of magnetic field, see inset in Fig. 2. The chosen value, $b_{0}\\!=\\!0.213/$nm, is quite consistent with the available wide quantum wells.va06 In conclusion, we note that preliminary analysis indicates that the non-HF shift should be more substantial in the case of a fractional filling, e.g. at $\nu\\!=\\!1/3$. Moreover, contrary to the single-mode approximation lo93 shifting the energy to lower values as compared to the HF result, the approach taking into account the double-exciton component should lead to a considerable positive shift in the CSFE correlation energy. The authors acknowledge support of the RFBR and hospitality of the Max Planck Institute for Physics of Complex Systems (Dresden) where partly this work was carried out. The authors also thank I.V. Kukushkin, L.V. Kulik and A.B. Van’kov for the discussion. ## References * (1) T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437 (1982). The Quantum Hall Effect, Ed. by R.R. Prange and S.M. Girvin, 2nd Ed. (Springer, New York, 1990). * (2) Yu.A. Bychkov, S.V. Iordanskii, and G.M. Eliashberg, JETP Lett. 33, 143 (1981). * (3) Yu.A. Bychkov and E.I. Rashba, Sov. Phys. JETP 58, 1062 (1983). * (4) C. Kallin and B.I. Halperin, Phys. Rev. B 30, 5655 (1984). * (5) S. Dickmann, I.V. Kukushkin. Phys. Rev. B 71, 241310(R) (2005). * (6) S.M. Dickmann, V.M. Zhilin, and D.V. Kulakovskii, JETP 101, 892 (2005). * (7) A.H. MacDonald J. Phys. C 18, 1003 (1985). * (8) J.P. Longo and C. Kallin, Phys. Rev. B 47, 4429 (1993). * (9) The first-order calculations as far as the MF approach for the ${\bf q}\\!=\\!0$ cyclotron spin-flip plasmon at even $\nu$ give zero energy shift from the cyclotron gap $\hbar\omega_{c}$. Actually the negative shift may be found exactly by performing full perturbative calculation to the second order in $r_{\rm c}$.di05 * (10) A. Pinczuk, B.S. Dennis, D. Heiman, C. Kallin, L. Brey, C. Tejedor, S. Schmitt-Rink, L.N. Pfeiffer, and K.W. West, Phys. Rev. Lett. 68, 3623 (1992). * (11) A.B. Van’kov, L.V. Kulik, I.V. Kukushkin, V.E. Kirpichev, S. Dickmann, V.M. Zhilin, J.H. Smet, K. von Klitzing, and W. Wegscheider. Phys. Rev. Lett., 97, 246801 (2006). * (12) S. Dickmann, Phys. Rev. B 65, 195310 (2002). * (13) A. B. Dzyubenko and Yu. E. Lozovik, Sov. Phys. Solid State 25, 874 (1983); ibid 26, 938 (1984); J. Phys. A 24, 415 (1991). * (14) If using ER, the relevant operators in our case are ${\hat{H}}_{\rm ED}\\!=\\!\sum\limits_{a=0,\overline{0},1}\\!\\!\\!{\hat{H}}_{a}\\!+\\!\\!\\!\\!\\!\sum\limits_{ab=0\overline{0},01,\overline{0}1}\\!\\!\\!{\hat{H}}_{ab}$, where $\begin{array}[]{l}{\hat{H}}_{a}\\!=\\!\frac{1}{2}\sum_{\bf q}\\!V(q)h_{aa}^{2}({\bf q})\left({\cal Q}_{aa{\bf q}}^{\dagger}{\cal Q}_{aa{\bf q}}\\!-\\!N_{\phi}^{-1/2}{\cal Q}_{aa{\bf 0}}^{\dagger}\right),\\\ {\hat{H}}_{ab}\\!=\\!\sum_{\bf q}\\!V(q)\left[h_{aa}({\bf q})h_{bb}({\bf q}){\cal Q}_{aa{\bf q}}^{\dagger}{\cal Q}_{bb{\bf q}}\\!+\\!|h_{ab}({\bf q})|^{2}\delta_{\sigma_{a},\sigma_{b}}\left({\cal Q}_{ab{\bf q}}^{\dagger}{\cal Q}_{ab{\bf q}}\\!-\\!N_{\phi}^{-1/2}{\cal Q}_{bb{\bf 0}}^{\dagger}\right)\right]\\\ \mbox{and}\quad h_{ab}\\!=\\!\left(\frac{n_{a}!}{n_{b}!}\right)^{1/2}\\!\\!\left(\frac{iq_{x}\\!+\\!q_{y}}{\sqrt{2}}\right)^{n_{b}\\!-\\!n_{a}}\\!\\!L_{n_{a}}^{n_{b}\\!-\\!n_{a}}(q^{2}/2)e^{-q^{2}/4}\quad(L^{j}_{i}\;\;\mbox{is Laguerre polinomial}).\end{array}$ * (15) For reference we write out Eqs. (11) and (12) in the ${\bf q}\not=0$ case: $\\!\\!\begin{array}[]{l}\left[E_{q}-{\cal E}_{\rm sw}(|{{\bf q}}/2\\!-\\!{{\bf p}}|)-{\cal E}_{\rm mp}(|{{\bf q}}/2\\!+\\!{{\bf p}}|)\right]\varphi_{q}({\bf p})\\!-\\!\frac{1}{\sqrt{2}}\left\\{g_{q}({\bf p})\right\\}{}{}{}{}\\\ =(2\pi)^{-1}\int d{{\bf s}}\,\varphi_{q}({{\bf s}})\left[{\vphantom{-{\tilde{U}}_{01}(|{{\bf q}}/2+{{\bf s}}|)}}\left[U_{00}(|{{\bf p}}-{{\bf s}}|)-{\tilde{U}}_{01}(|{\bf q}/2+{\bf s}|)\right]e^{i({\bf p}\times{\bf s})_{z}}\right.{}{}\\\ \left.+U_{01}(|{{\bf p}}-{{\bf s}}|)e^{i({\bf s}\times{\bf p})_{z}}-U_{00}(|{\bf p}-{\bf s}|)e^{i\left({\bf q}\times({\bf s}-{\bf p})\right)_{z}/2}\right.-\\!\left.U_{01}(|{\bf p}-{\bf s}|)e^{i\left({\bf q}\times({\bf p}-{\bf s})\right)_{z}/2}{\vphantom{-{\tilde{U}}_{01}(|{{\bf q}}/2+{{\bf s}}|)}}\right]\end{array}$ and ${}\qquad{}\qquad{}\qquad{}\qquad E_{q}={\cal E}_{0\overline{1}}(q)+\frac{1}{\pi\sqrt{2}}\int d{{\bf p}}\,g^{*}_{q}({\bf p})\varphi_{q}({\mbox{\boldmath$p$}})$ with the ‘free’ term ${}\;\frac{1}{\sqrt{2}}\left\\{g_{q}({\bf p})\right\\},\;$ where $\;\,g_{q}({{\bf p}})={\tilde{U}}_{01}(|{\bf p}+{\bf q}/2|)e^{i\left({\bf p}\times{\bf q}\right)_{z}/2},\;\,$ ${\tilde{U}}_{01}\\!=\\!V(q)|h_{01}({\bf q})|^{2}\;\,$ and $\;U_{n_{a}n_{b}}\\!=\\!V(q)h^{2}_{ab}({\bf q})$ (see notations of Ref. foot2, ). If ${\bf q}$ is chosen parallel to ${\hat{y}}$, then $\varphi_{q}({\bf p})$ is an even function with respect to the replacement $p_{x}\\!\to\\!-p_{x}$. The HF result ${\cal E}_{0\overline{1}}(q)$ was calculated in Ref. lo93, .

arxiv-papers

2024-09-04T02:48:54.343517

0803.2303

# The nontrivial zeros of the Zeta Function lie on the Critical Line Pedro Geraldo111In Memorian to: G. F. B. Riemann (1826 - 1866). (pegeraldo@luz.edu.ve) Departamento de Matemáticas Facultad de Ingeniería Universidad del Zulia Núcleo COL Cabimas (4013), Venezuela (10/24/2008) ###### Abstract In this paper is stablished a characterization of the solutions of the equation: $\zeta(z)=0$. Then such a characterization is used to give a proof for Riemann’s Conjecture. Classification Subject: 30B40 & 11M26 Key words: Riemann’s zeta function. Analytic continuation. Critical line. Riemann’s conjecture. ## 1 The Riemann Zeta Function Let $t\in\mathbb{R}^{+}$ and $\log{t}$ be its real value, then: $\forall\,n\in\mathbb{Z}\wedge n\geq 1:\,\Big{|}\frac{1}{n^{z}}\Big{|}=\frac{1}{n^{\mathrm{Re}z}}=\frac{1}{e^{\mathrm{Re}z\cdot\log{n}}}$ is a well defined function for every $z\in\mathbb{C}$. Let $\delta>0$ be an arbitrary real number. For $\mathrm{Re}z\geq 1+\delta$, we have: $\Big{|}\frac{1}{n^{z}}\Big{|}=\frac{1}{n^{\mathrm{Re}z}}\leq\frac{1}{n^{1+\delta}}$ The $p$-series $\displaystyle\sum_{n=1}^{\infty}\frac{1}{n^{1+\delta}}$ is convergent and Weirstrass criterion says that the series $\displaystyle\sum_{n=1}^{\infty}\frac{1}{n^{z}}$ is also absolutely convergent for $\mathrm{Re}z>1$.The Riemann Zeta Function is defined as follow: $\zeta(z)=\sum_{n=1}^{\infty}\frac{1}{n^{z}}\,\,\,\hbox{ for }\,\,\,\mathrm{Re}z>1$ $\zeta$ is analytic in the half-plane $\mathrm{Re}z>1$ and uniformly convergent in every compact set contained in that half-plane $\mathrm{Re}z>1$. ###### Definition 1. Let $E$ and $F$ be sets. Suppose $P\subset E$ and $g:P\to F$ be an application. The application $f:E\to F$ is said to be an extension of $g$ over $E$ relative to $F$ if $f|_{P}=g$. In general such an application is not unique see [9]. However, any Analytic continuations (extension) if they exist are unique, see [19] and [33]. ###### Theorem 1. $\zeta$ can be continued across the boundary $\mathrm{Re}z=1$ of the half- plane $\mathrm{Re}>1$, and proves to be a Meromorphic function having the single pole $z=1$ with the principal part $\frac{1}{z-1}$; i.e., $z=1$ is a simple pole with residue $+1$. ###### Proof. See [21] ∎ From Theorem 1, we get that: 1. $(i)$ The analytic Continuation of $\zeta$ up to the boundary $\mathrm{Re}z=0$ is given by $\zeta(z)=1+\frac{1}{z-1}-z\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z+1}}$ (1) 2. $(ii)$ The analytic Continuation of $\zeta$ up to $\mathrm{Re}z=-1$ is given by $\zeta(z)=1+\frac{1}{z-1}-\frac{z}{2!}\Big{[}\zeta(z+1)-1\Big{]}-\frac{z(z+1)}{2!}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{t^{2}dt}{(n+t)^{z+2}}$ 3. $(iii)$ The analytic Continuation of $\zeta$ up to $\mathrm{Re}z=-2$ is given by $\begin{split}\zeta(z)=&1+\frac{1}{z-1}-\frac{z}{2!}\Big{[}\zeta(z+1)-1\Big{]}-\frac{z(z+1)}{3!}\Big{[}\zeta(z+2)-1\Big{]}\\\ &-\frac{z(z+1)(z+2)}{3!}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{t^{3}dt}{(n+t)^{z+3}}\end{split}$ and so forth by induction. The most important issue here is that by definition 1, it is enough to proof the Riemann Conjeture for $\zeta(z)=1+\frac{1}{z-1}-z\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z+1}}$ see also [33] prop. 16.10. Remember that: $\zeta(z)=0$ for $z=-2,-4,-6,\ldots$ which can be deduced from Riemann’s functional equation: $\zeta(z)=2(2\pi)^{z-1}\Gamma(1-z)\zeta(1-z)\sin(\frac{1}{2}\pi z)|\text{ for }z\not=1$ (2) $-1<Rez<1$ We also know that $\zeta(0)\not=0$ and $\zeta(1)\not=0$, similar reasoning gives that $\zeta$ has no other zeros outside the Critical strip $\overline{B}$ than the trivials: $\\{-2,-4,\ldots\\}$. ###### Definition 2. The points $z=-2,-4,-6,\ldots$ are called the trivials zeros of $\zeta$. Let us define the following sets: 1. $(i)$ $F=\\{z\in\mathbb{C}:\mathrm{Re}z=\frac{1}{2}\\}$ called the critical line 2. $(ii)$ $B_{1}=\\{z\in\mathbb{C}:0<\mathrm{Re}z<\frac{1}{2}\\}$ 3. $(iii)$ $B_{2}=\\{z\in\mathbb{C}:\frac{1}{2}<\mathrm{Re}z<1\\}$ 4. $(iv)$ $B=B_{1}\cup B_{2}$ 5. $(v)$ $\overline{B}=\\{z\in\mathbb{C}:0\leq\mathrm{Re}z\leq 1\\}$ called the critical strip. See more about this in [13]. The Riemann Hypothesis is equivalent to say that $\zeta$ has no zeros in $B$. ###### Lemma 1. If $z_{0}\in(\mathbb{C}\setminus\\{0,1\\})$, then: $z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=1\Rightarrow\zeta(z_{0})\neq 0$ ###### Proof. $\begin{array}[]{l}\displaystyle\zeta(z_{0})=1+\dfrac{1}{z_{0}-1}-z_{0}\displaystyle{\sum_{n=1}^{\infty}\int_{0}^{1}\dfrac{tdt}{(n+t)^{z_{0}+1}}=1+\frac{1}{z_{0}-1}-1=\dfrac{1}{z_{0}-1}},\\\ \\\ \displaystyle{\zeta(z_{0})=\dfrac{1}{z_{0}-1}\Rightarrow(z_{0}-1)\zeta(z_{0})=1\Rightarrow\zeta(z_{0})\neq 0}\end{array}$ ∎ ###### Corollary 1. $\displaystyle\zeta(z_{0})=0\Rightarrow z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}\neq 1$ ###### Theorem 2. If $z_{0}\in(\mathbb{C}-\\{0,1\\})$. Then, $\zeta(z_{0})=0\Leftrightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=1$ ###### Proof. * (i) $\begin{array}[]{rcl}0=\zeta(z_{0})&\Rightarrow&\displaystyle{0=1+\frac{1}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}}\\\ &=&\displaystyle{\frac{z_{0}-1+1}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}}\\\ &=&\displaystyle{\frac{z_{0}}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=\frac{z_{0}-z_{0}(z_{0}-1)\displaystyle{\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}-1}}}}{z_{0}-1}}\\\ &\Rightarrow&\displaystyle{0=\frac{z_{0}-z_{0}(z_{0}-1)\displaystyle{\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}}}{z_{0}-1}}\\\ &\Rightarrow&\displaystyle{0\cdot(z_{0}-1)=z_{0}-z_{0}(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}}\\\ &\Rightarrow&\displaystyle{0=z_{0}\Big{[}1-(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}\Big{]}}\\\ &\Rightarrow&\displaystyle{1-(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=0}\\\ &\Rightarrow&\displaystyle{(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+t}}=1}.\\\ \end{array}$ * (ii) $\displaystyle(z_{0}-1)\sum_{1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=1\Rightarrow\zeta(z_{0})=0$ $\begin{array}[]{l}\displaystyle{\zeta(z_{0})=1+\frac{1}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\frac{tdt}{(n+t)^{z_{0}+1}}}=\\\ \\\ \displaystyle{(z_{0}-1)\sum_{1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}+\frac{1}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\frac{tdt}{(n+t)^{z_{0}+1}}}=\\\ \\\ \displaystyle{z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}+\frac{1}{z_{0}-1}-z_{0}\sum_{n=1}^{\infty}\frac{tdt}{(n+t)^{z_{0}+1}}}=\\\ \\\ \displaystyle{-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}+\frac{1}{z_{0}-1}=\frac{-(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}}}+1}{z_{0}-1}=\frac{-1+1}{z_{0}-1}=\frac{0}{z_{0}-1}=0}\\\ \\\ \displaystyle{\Rightarrow\zeta(z_{0})=0}\\\ \\\ \end{array}$ ∎ ###### Lemma 2. For $B=\\{z=x+iy|0<x<\frac{1}{2}\veebar\frac{1}{2}<x<1\wedge y\in\mathbb{R}\\}$. Then $\forall\alpha\neq 0\wedge\forall x:\,0<x<\frac{1}{2}\veebar\frac{1}{2}<x<1$ we have that: $1\neq\alpha(x+iy)[(x+iy)-1]$ ###### Proof. Let’s suppose that $\exists\alpha_{1}\neq 0\wedge\exists(x_{1}+iy)$ such that $0<x_{1}<\frac{1}{2}\veebar\frac{1}{2}<x_{1}<1$ and $1=\alpha_{1}[(x_{1}+iy)^{2}-(x_{1}+iy)]$ then, derivating with respect to y we find that: $0=\alpha_{1}[2(x_{1}+iy)i-i]=\alpha_{1}[2(x_{1}+iy)-1]i$ $\Rightarrow 2(x_{1}+iy)-1=0$ $\Rightarrow(x_{1}+iy)=\frac{1}{2}$ $\Rightarrow x_{1}+iy=\frac{1}{2}+0i$ $\Rightarrow x_{1}=\frac{1}{2}\,\text{ This is Absurd!}$ Therefore, $\forall\alpha\neq 0\wedge\forall x:0<x<\frac{1}{2}\veebar\frac{1}{2}<x<1$ we have that: $1\neq\alpha(x+iy)[(x+iy)-1]$ in particular if $x_{0}+iy_{0}=z_{0}\in B$, we have $1\neq\alpha z_{0}(z_{0}-1)\,\,\forall\alpha\neq 0$ ∎ ###### Theorem 3 (The Riemman’s conjeture). $\forall z\in B:\zeta(z)\neq 0$ ###### Proof. Let’s suppose that: $\exists z_{0}\in B:\zeta(z_{0})=0$ $\zeta(z_{0})=0\wedge(Theorem2)\Rightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=1$ (3) $(2)\wedge\zeta(z_{0})=0\Rightarrow\zeta(1-z_{0})=0$ (4) $\zeta(1-z_{0})=0\wedge(Theorem2)\Rightarrow- z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}=1$ (5) $(5)\Rightarrow\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}\neq 0$ (6) $\text{Claim!}\,\,\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}\neq-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ Let’s suppose: $\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ $\Rightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ $\Rightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ (7) $(3)\wedge(5)\wedge(7)\Rightarrow 1=1+\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ $\Rightarrow\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}=0\,\text{ This is Absurd!}\,\,\,\text{ (By (6))}$ Then: $\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}\neq-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}$ (8) $(8)\Rightarrow\exists!\alpha_{1}\neq 0,\alpha_{1}\in\mathbb{C}\text{ such that }\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}$ (9) $(9)\Rightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}(z_{0}-1)$ $\Rightarrow(z_{0}-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z_{0}+1}}=-z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}(z_{0}-1)$ (10) $(3)\wedge(5)\wedge(10)\Rightarrow 1=1+\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}(z_{0}-1)$ $\Rightarrow\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}(z_{0}-1)=0$ $\Rightarrow- z_{0}\Big{[}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}+\alpha_{1}(z_{0}-1)\Big{]}=0$ $\Rightarrow- z_{0}\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{2-z_{0}}}-\alpha_{1}z_{o}(z_{0}-1)=0$ (11) $(5)\wedge(11)\Rightarrow 1-\alpha_{1}z_{0}(z_{0}-1)=0\Rightarrow 1=\alpha_{1}z_{0}(z_{0}-1)\,\,\text{This is Absurd!}\,\text{ (By lemma 2).}$ Then the proposition “$\exists z_{0}\in B:\zeta(z_{0})=0$” is false. Therefore: $\forall z\in B:\zeta(z)\neq 0$ ∎ ## 2 Conclusion of the Saga. It is known that $\zeta(z)=0$ for some $z\in F$. See for example: [11], [25], [35] or [38]. Not every $z\in F$ is solution for $\zeta(z)=0$, for example $\frac{1}{2}=z_{0}\in F$ and it is not difficult to prove that $\zeta(z_{0})\neq 0$. We can say now that: $R=\\{z\in\overline{B}:\zeta(z)=0\\}=\\{z\in F:(z-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z+1}}=1\\}$ Now we know that the non-trivial zeros of $\zeta(z)=0$ are on the critical line. Therefore: To find non-trivial solutions for $\zeta(z)=0$;$z\in F$, we can try the system: $\left\\{\begin{array}[]{rcl}\displaystyle{(z-1)\sum_{n=1}^{\infty}\int_{0}^{1}\frac{tdt}{(n+t)^{z+1}}}&=&1\\\ z&=&\frac{1}{2}+iy\end{array}\right.$ for $n$ big enough could be useful to try the system $\left\\{\begin{array}[]{rcr}\displaystyle\int_{0}^{1}\frac{tdt}{(n+t)^{z+1}}&\approx&\frac{1}{n(n+1)}\\\ z&=&\frac{1}{2}+iy\end{array}\right.$ ###### Corollary 2. If every statement of the type “$RH\Leftrightarrow A=B$” is true. Then $A=B$. Where $A=B$ means a relation betwen $A$ and $B$. See below. ###### Proof. That “RH” is true follows from Theorem 3. Then: $\text{``}RH\text{''}\Rightarrow\text{``}A=B\text{''}$ $\underline{\text{``}RH\text{''}\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }\text{ }}$ $\text{``}A=B\text{''}$ See [5] pags. 13–16. ∎ ## 3 Applications. $1)$ Redheffer (1977) $R.H.\Leftrightarrow\forall\,\varepsilon>0,\exists\,\,C(\varepsilon)>0:|\det{(A(n))}|<C(\varepsilon)n^{\frac{1}{2}+\varepsilon}$ where $A(n)$ is the $n\times n$ matrix of $0$’s and $1$’s defined by $A(i,j)=\begin{cases}1&\hbox{ if }j=1\hbox{ or if }i|j\\\ 0&\hbox{ otherwise }\end{cases}$ This is an important result for linear analysis for example. See [12] $2)$ Lagarias (2002) Let $\sigma(n)$ denotes the sum of the positive divisors of $n$. Then $R.H.\Leftrightarrow\forall\,n:\,\sigma(n)\leq H_{n}+\exp{(H_{n})}\log{H_{n}}$ where $H_{n}=1+\frac{1}{2}+\frac{1}{3}+\cdots+\frac{1}{n}$. This is an important result for number theory for example. See [12]. $3)$ Nyman-Beurling $R.H.\Leftrightarrow Span_{L^{2}(0,1)}=\\{\mathcal{N}_{\alpha}:0<\alpha<1\\}=L^{2}(0,1)$ where $\mathcal{N}_{\alpha}(t)=\left\\{\frac{\alpha}{t}\right\\}-\alpha\left\\{\frac{1}{t}\right\\}$ and $\\{x\\}=x-[x]$ is the fractional part of $x$ this is an important result for Real and Functional Analysis for example. See [12] Others results like these can be seen in [12]. I believe that one of the must important result to be studied after this one is the paper of Andre Weil. See [12]. ###### Corollary 3. If $\chi=\chi_{{}_{1}}$ is the principal character $mod\,k$ then: $\forall s\in B:\,L(s,\chi_{{}_{1}})\neq 0$ ###### Proof. See [2] Theorem 11.7 and then use theorem 3. ∎ ## 4 Open Questions 1. 1. Are simple the zeros of the $\zeta$ Riemann Function? 2. 2. It is known that of all imaginary quadratic field $Q(\sqrt{-d})$ with class number $h$, we have $d<Ch^{2}\log{h}^{2}$, except for at most one exceptional field, for which $d$ may be Larger. Does there exist such an Exceptional Field? Hint.- See [28] and [29]. ## References * [1] L.V. Alphors. Complex Analysis. Mc Graw Hill Book Company. Second Edition. Tokyo 1966\. * [2] T. M. Apostol. Introduction to Analytic Number Theory. Springer- Verlag. New York Inc. 1980\. * [3] R. Bellman. A Brief Introduction to Theta function. Holt Rinehart and Wistons. USA 1961\. * [4] R. Bellman. A Collection of Modern Mathematical Classics. Analysis. Dover. New York 1961\. * [5] M.L. Bittinger. Proof, Logic an Sets. Addison Wesley Publishing Company. Reading Massachusetts. USA. 1982. * [6] E. Bombieri. Problems of the Millenium the Riemann Hypothesis. http://www.claymath.org/prizeproblems/riemann.htm. * [7] H. Cartan. Theory elementaire des Fonctions Analytiques D’une ou Plusieurs Variables Complexes. Hermann Editeurs Des Sciences Et Des Arts. Paris. 1985. * [8] K. Chandrasekharan. Introduction to Analytic Number Theory. Springer Verlag. New York. 1968. * [9] L. Chambadal. Dictionarie Des Mathematiques Modernes. Libraie Larousse. Paris. 1969. * [10] H. Cohn. Advanced Number Theory. Dover Publishing, Inc. New York. 1962\. * [11] J.B. Conrey. More Than Two Fifths of the zeros of the Riemann Zeta Functions are on the Critical Line. J. Reine Angew. MAth 399 (1989) 1-26. * [12] J.B. Conrey. The Riemann Hypothesis. Notices of the American Mathematical Society. Volumn 50. Number 3. (March 2003) 341353. * [13] J.B. Conway. Function of one Complex Variable. Springer Verlag. N.Y. 1973\. * [14] L.E. Dickson. History of the Theory of Numbers. Chelsea New York. 1952\. * [15] M.H. Edwards. Riemann’s Zeta Function. Academic Press, New York- London. 1974. * [16] L. Flatto. Advanced Calculus. The Williams and Wilkins Company. Baltimore 1976. USA. * [17] E. Gentile, Notas de Algebra. EDEBA. Buenos Aires. 1976. * [18] HArdy, G.H. and Wright, E.M. An Introduction to the Theory of Numbers. $4^{th}$ Ed. Clarendon Press. Oxford. 1960. * [19] S.T. Hu. Introduction to General Topology. Holden-Day, Inc. San Francisco. USA 1966. * [20] A.A. Karatsuba. Fundamentos de la Teoría analítica de los Números. Editorial MIR. Moscu. 1979. * [21] K. Knopp. Theory of Functions. Parts I and II. Dover Publications. New York. 1947. * [22] S. Lang. Complex Analysis. Addison Wesley. Reading Mass. USA. 1976. * [23] Theory of Numbers. Spriger-Verlag. New York. USA 1980 * [24] D. Laugwitz, A. Sheinitzer. Bernhard Riemann 1826-1866. Birkhauser. Boston. Baser. Berlin 1998. * [25] N. Levison. More than One third of zeros of Riemann’s Zeta Function are on $\sigma=\frac{1}{2}$. Advances Math. 13. 383-486 (1974). * [26] S. MacLane. Symbolic Logic. American Mathematical Montly. Vol. 46. P. 289\. (1939). * [27] A. Markushevich. Teoría de las Funciones Analíticas.. Tomos I & II. Editorial MIR. Moscú 1970. * [28] H.L. Montgomery, The pair correlation of zeros of the zeta function, Proc. sympos. Pure Math., vol. 24, amer. Math. soc., Providence, R.I., 1973, pp. 181-193. * [29] H.L. Montgomery, and P.J. Weinberger, Notes on small class numbers, Acta Arith. 24 (1974), 329-342. * [30] H.L. Montgomery, Distribution of Zeros of the Riemann Zeta Function, Proc. Intr. Congrss of Math. Vancouver, 1974, pp. 379-381. * [31] R. Narashiman, Y. Nievergelt. Complex Analysis in one Variable. Second Edition. Birkhauser, Boston 2000. * [32] A.M. Odlyzko. http://www.dtc.umn.edu/~odlizko/. * [33] W. Rudin. Real and Complex Analysis. Second Edition. Mc. Graw Hill. Series in Higher Math. * [34] I.E.Segal & R.A.Kunze. Integrals and operators. Mc. Graw Hill. company N.Y. 1968. * [35] A. Selberg. On the zeros of the Zeta Function of Riemann. College Papers. Springer Verlag. New York. 1989 Vol I, 156-159. * [36] A. Selberg. Old and New Conjetures and Results About A Class of Dirichlet Series. Vol II. With A Foreword By K. Chandrasekharan. * [37] C.L. Siegel. Analytic Number Theory. (Lectures Notes By B. Riemann) New York University 1945. * [38] E.C. Titchmarsh. The theory of the Riemann Zeta-Function. Claredon Press. Oxford 1951\. * [39] I.Vinogradov. Fundamentos de la Teoría de los Números. Editorial MIR. Moscú. 1971. Prof: Pedro J, Geraldo C. Home address: Calle principal 130. Delicias Nuevas Cabimas (4013). Edo Zulia Venezuela. E-mail: pegeraldo@luz.edu.ve & pegeraldo@yahoo.com Home phone: 0264-2513221

arxiv-papers

2024-09-04T02:48:54.348959

0803.2330

11institutetext: Tianshu Luo, Yimu Guo 22institutetext: Institute of Solid Mechanics, Department of Applied Mechanics, Zhejiang University, Hangzhou, Zhejiang, 310027, P.R.China 22email: ltsmechanic@zju.edu.cn 33institutetext: Yimu Guo 44institutetext: Institute of Solid Mechanics, Department of Applied Mechanics, Zhejiang University, Hangzhou, Zhejiang, 310027, P.R.China 44email: guoyimu@zju.edu.cn # A Sort of Relation Between a Dissipative Mechanical System and Conservative Ones Tianshu Luo Yimu Guo (Received: date / Accepted: date) ###### Abstract In this paper we proposed a proposition: for any nonconservative classical mechanical system and any initial condition, there exists a conservative one; the two systems share one and only one common phase curve; the Hamiltonian of the conservative system is the sum of the total energy of the nonconservative system on the aforementioned phase curve and a constant depending on the initial condition. Hence, this approach entails substituting an infinite number of conservative systems for a dissipative mechanical system corresponding to varied initial conditions. One key way we use to demonstrate these viewpoints is that by the Newton-Laplace principle the nonconservative force can be reasonably assumed to be equal to a function of a component of generalized coordinates $q_{i}$ along a phase curve, such that a nonconservative mechanical system can be reformulated as countless conservative systems. Utilizing the proposition, one can apply the method of Hamiltonian mechanics or Lagrangian mechanics to dissipative mechanical system. The advantage of this approach is that there is no need to change the definition of canonical momentum and the motion is identical to that of the original system. ###### Keywords: Hamiltonian, dissipation, non-conservative system, damping, symplectic algorithm ††journal: China Science Bulletein ## 1 Introduction In general, Hamiltonian mechanics and Lagrangian mechanics are applied to conservative classical mechanical system or conservative quantum-mechanical system. In this paper we attempt to find a sort of relationship between a dissipative classical mechanical system between nonconservative classical mechanical ones, then we might apply some methods derived from symplectic geometry to dissipative classical mechanical system. Some researchers attempt to represent a dissipative system as Hamiltonian formalism or Lagrangian formalism. For instances, about half a century ago, CalirolaPCalirola1941 ,Kana1948PThPh…3..440K adopted the Hamiltonian $H_{ck}(q,p)=\frac{1}{2}\left(e^{-2\eta t}p^{2}+e^{2\eta t}\omega^{2}q^{2}\right),$ (1) which leads exactly to the classical equation of motion of a damped harmonic oscillator, $\ddot{x}+2\eta\dot{x}+\omega^{2}x^{2}=0,\ \ \eta>0$ (2) In this Hamiltonian-description, the canonical momentum is defined as $p_{ck}=e^{2\eta t}p$ In 1940s Morse and Feshbachbook3 gave an example of an artificial Hamiltonian for a damped oscillator based on a “mirror-image” trick, incorporating a second oscillator with negative friction. The resulting Hamiltonian is unphysical: it is unbounded from below and under time reversal the oscillator is transformed into its “mirror-image”. By this arbitrary trick dissipative systems can be handled as though they were conservative. BatemanPhysRev.38.815 proposed a similar approach. For the system (2), we have $\displaystyle\ddot{x}+2\eta\dot{x}+\omega^{2}x^{2}=0\ \ (original)$ (3) $\displaystyle\ddot{y}-2\eta\dot{x}+\omega^{2}x^{2}=0\ \ (mirror-image).$ (4) Correspondingly, there is Bateman(-Morse-Feshbach) Lagrangian: $L_{B}(x,\dot{x},y,\dot{y})=\dot{x}\dot{y}+\eta(x\dot{y}-\dot{x}y)-m\omega^{2}xy$ (5) Rajeevart7 considered that a large class of dissipative systems can be brought to a canonical form by introducing complex coordinates in phase space and a complex-valued Hamiltonian. Rajeevart7 indicated that Eq.(2) can be brought to diagonal form by a linear transformation: $z=A\left[-\mathrm{i}(p+\eta x)+\omega_{1}x\right],\frac{\mathrm{d}z}{\mathrm{d}t}=\left[-\gamma+\mathrm{i}\omega_{1}\right]z,$ (6) where $\omega_{1}=\sqrt{\omega_{2}-\gamma^{2}},$ (7) and the constant $A=1/\sqrt{2\omega_{1}}$. Then art7 defined the complex- valued function as Hamiltonian $\mathcal{H}=(\omega_{1}+\mathrm{i}\eta)zz^{*},$ (8) which satisfied $\frac{\mathrm{d}z}{\mathrm{d}t}=\left\\{\mathcal{H},z\right\\},\ \ \frac{\mathrm{d}z^{*}}{\mathrm{d}t}=\left\\{\mathcal{H},z^{*}\right\\}$ By reviewing the works of PCalirola1941 1948PThPh…3..440K PhysRev.38.815 book3 art7 , we can find that they attempt to transform a dissipative system into a conservative system entirely and these approaches might be suitable for Hamiltonian representation of one-dimensional damped oscillators (weak non- Lagrangian systems) and quantization. Because by observing Eq.(1), Eq.(4), Eq.(5) and the transformation (6), one can find that the damping coefficient is independent of other particles, and art7 had wrote: ’ These complex coordinates are the natural variable(normal modes) of the system. ’ In area of quantum mechanics, PhysRevA.81.022112 ; Kochan2010219 attempts to quantize dissipative forces in terms of the two form $\Omega$ ( an analog of $\mathrm{d}p\wedge\mathrm{d}q-\mathrm{d}H\wedge\mathrm{d}t$), avoiding to obtain Hamiltonian or Lagrangian formulation of non-Lagrangian system. Marsden Marsden2007 and other researchers applied the equations as below to the problem of stability of dissipative systems $\displaystyle\dot{p}_{i}$ $\displaystyle=$ $\displaystyle-\frac{\partial H}{\partial q_{i}}+\bm{F}\left(\frac{\partial{r}}{\partial q_{i}}\right)$ $\displaystyle\dot{q}_{i}$ $\displaystyle=$ $\displaystyle\frac{\partial H}{\partial p_{i}},$ (9) where the position vector $r$ depends on the canonical variable $\\{q,p\\}$, i.e. $r(q,p)$, $H$ denotes Hamiltonian, and $\bm{F}(\partial{r}/\partial{q_{i}})$ denotes a generalized force in the direction $i$, $i=1,\dots,n$. Marsden considered that Eqs.(9) was composed of a conservative part and a non-conservative part. Eq.(9) apparently is a representation of dissipative mechanical systems in the phase space. Although one can utilize the approaches discussed in some papersPCalirola1941 1948PThPh…3..440K PhysRev.38.815 book3 art7 to convert Eq.(9) into a conservative system, one must first change the definition of the canonical momentum of the system. If one uses numerical algorithms to solve the Hamiltonian system, the numerical solution will lose the physical characteristics of the original system, because the phase flow of the original system is different from that of the new system. We need a Hamiltonian system that shares common phase flow or solution with the original system. But this demand cannot be satisfied, because it conflicts with Louisville’s theorem. Therefore, we would have to attempt to find other relationship between dissipative systems and conservative ones. Based on Eq.(9), in this paper we will attempt to demonstrate that a dissipative mechanical system shares a single common phase curve with a conservative system. In the light of this property, we will propose an approach to substitute a group of conservative systems for a dissipative mechanical system. In the following section, we will illustrate the relationship between a dissipative mechanical system and a conservative one. ## 2 Relationship between a Dissipative Mechanical System and a Conservative One ### 2.1 A Proposition Under general circumstances, the force $\bm{F}$ is a damping force that depends on the variable set $q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n}$. $F_{i}$ denotes the components of the generalized force $\bm{F}$. $F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})=\bm{F}\left(\frac{\partial{r}}{\partial q_{i}}\right).$ (10) Thus we can reformulate Eq.(9) as follows: $\displaystyle\dot{p}_{i}$ $\displaystyle=$ $\displaystyle-\frac{\partial H}{\partial q_{i}}+F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})$ $\displaystyle\dot{q}_{i}$ $\displaystyle=$ $\displaystyle\frac{\partial H}{\partial p_{i}}.$ (11) Suppose the Hamiltonian quantity of a conservative system without damping is $\hat{H}$. Thus we may write a Hamilton’s equation of the conservative system : $\displaystyle\dot{p}_{i}$ $\displaystyle=$ $\displaystyle-\frac{\partial{\hat{H}}}{\partial q_{i}}$ $\displaystyle\dot{q}_{i}$ $\displaystyle=$ $\displaystyle\frac{\partial\hat{H}}{\partial p_{i}}.$ (12) We do not intend to change the definition of momentum in classical mechanics, but we do require that a special solution of Eq.(12) is the same as that of Eq.(11). We may therefore assume a phase curve $\gamma$ of Eq.(11) coincides with that of Eq.(12). The phase curve $\gamma$ corresponds to an initial condition $q_{i0},p_{i0}$. Consequently by comparing Eq.(11) and Eq.(12), we have $\displaystyle\left.\frac{\partial{\hat{H}}}{\partial{q_{i}}}\right|_{\gamma}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial H}{\partial q_{i}}\right|_{\gamma}-\left.F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})\right|_{\gamma}$ $\displaystyle\left.\frac{\partial{\hat{H}}}{\partial{p_{i}}}\right|_{\gamma}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial H}{\partial p_{i}}\right|_{\gamma},$ (13) where $\left.\frac{\partial{\hat{H}}}{\partial{q_{i}}}\right|_{\gamma},\left.\frac{\partial H}{\partial q_{i}}\right|_{\gamma},\left.\frac{\partial{\hat{H}}}{\partial{p_{i}}}\right|_{\gamma}and\left.\frac{\partial H}{\partial p_{i}}\right|_{\gamma}$ denote the values of these partial derivatives on the phase curve $\gamma$ and $\left.F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})\right|_{\gamma}$ denotes the value of the force $F_{i}$ on the phase curve $\gamma$. In classical mechanics the Hamiltonian $H$ of a conservative mechanical system is mechanical energy and can be written as: $H=\int_{\gamma}\left(\frac{\partial{H}}{\partial{q_{i}}}\right)\mathrm{d}q_{i}+\int_{\gamma}\left(\frac{\partial H}{\partial p_{i}}\right)\mathrm{d}p_{i}+const_{1},$ (14) where $const_{1}$ is a constant that depends on the initial condition described above. If $q_{i}=0,p_{i}=0$, then $const_{1}=0$. The mechanical energy $H$ of the system (11) can be evaluated via Eq. (14) too. The Einstein summation convention has been used this section. Thus an attempt has been made to find $\left.\hat{H}\right|_{\gamma}$ through line integral along the phase curve $\gamma$ of the dissipative system $\displaystyle\int_{\gamma}\frac{\partial{\hat{H}}}{\partial{q_{i}}}\mathrm{d}q_{i}$ $\displaystyle=$ $\displaystyle\int_{\gamma}\left[\frac{\partial H}{\partial q_{i}}-F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})\right]\ \mathrm{d}q_{i}$ $\displaystyle\int_{\gamma}\frac{\partial\hat{H}}{\partial p_{i}}\mathrm{d}p_{i}$ $\displaystyle=$ $\displaystyle\int_{\gamma}\frac{\partial H}{\partial p_{i}}\mathrm{d}p_{i}.$ (15) Analogous to Eq.(14), we have $\left.\hat{H}\right|_{\gamma}=\int_{\gamma}\frac{\partial{\hat{H}}}{\partial{q_{i}}}\mathrm{d}q_{i}+\int_{\gamma}\frac{\partial{\hat{H}}}{\partial p_{i}}\mathrm{d}p_{i}+const_{2},$ (16) where $const_{2}$ is a constant which depends on the initial condition. Substituting Eq.(14)(15) into Eq.(16), we have $\left.\hat{H}\right|_{\gamma}=H-\int_{\gamma}F_{i}(q_{1},\cdots,q_{n},\dot{q}_{1},\cdots,\dot{q}_{n})\mathrm{d}q_{i}+const.$ (17) where $const=const_{2}-const_{1}$, and $H=\left.H\right|_{\gamma}$ because $H$ is mechanical energy of the nonconservative system(11). According to the physical meaning of Hamiltonian, $const_{1}$, $const_{2}$ and $const$ are added into Eq.(14)(16)(17) respectively such that the integral constant vanishes in the Hamiltonian quantity. ArnoldArnold1997 had presented the Newton-Laplace principle of determinacy as, ’This principle asserts that the state of a mechanical system at any fixed moment of time uniquely determines all of its (future and past) motion.’ In other words, in the phase space the position variable and the velocity variable are determined only by the time $t$. Therefore, we can assume that we have already a solution of Eq.(11) $\displaystyle q_{i}$ $\displaystyle=$ $\displaystyle q_{i}(t)$ $\displaystyle\dot{q_{i}}$ $\displaystyle=$ $\displaystyle\dot{q_{i}}(t),$ (18) where the solution satisfies the initial condition. We can divide the whole time domain into a group of sufficiently small domains and in these domains $q_{i}$ is monotone, and hence we can assume an inverse function $t=t(q_{i})$. If $t=t(q_{i})$ is substituted into the nonconservative force $\left.F_{i}\right|_{\gamma}$, we can assume that: $\left.F_{i}(q_{1}(t(q_{i})),\cdots,q_{n}(t(q_{i})),\dot{q}_{1}(t(q_{i})),\cdots,\dot{q}_{n}(t(q_{i})))\right|_{\gamma}=\mathcal{F}_{i}(q_{i}),$ (19) where $\mathcal{F}_{i}$ is a function of $q_{i}$ alone. In Eq.(19) the function $F_{i}$ is restricted on the curve $\gamma$, such that a new function $\mathcal{F}_{i}(q_{i})$ yields. Thus we have $\displaystyle\int_{\gamma}F_{i}\mathrm{d}q_{i}$ $\displaystyle=$ $\displaystyle\int_{q_{i0}}^{q_{i}}\mathcal{F}_{i}(q_{i})\mathrm{d}q_{i}=W_{i}(q_{i})-W_{i}(q_{i0}).$ (20) According to Eq.(20) the function $\mathcal{F}_{i}$ is path independent, and therefore $\mathcal{F}_{i}$ can be regarded as a conservative force. For that Eq.(19) represents an identity map from the nonconservative force $F_{i}$ on the curve $\gamma$ to the conservative force $\mathcal{F}_{i}$ which is distinct from $F_{i}$. It must be noted, that Eq.(19) is tenable only on the phase curve $\gamma$. Consequently the function form of $\mathcal{F}_{i}$ depends on the aforementioned initial condition; from other initial conditions $\mathcal{F}_{i}$ with different function forms will yield. According to the physical meaning of Hamiltonian, $const$ is added to Eq.(17) such that the integral constant vanishes in Hamiltonian quantity. Hence $const=-W_{i}(q_{i0})$. Substituting Eq.(20) and $const=-W_{i}(q_{i0})$ into Eq.(17), we have $\left.\hat{H}\right|_{\gamma}=H-W_{i}(q_{i})$ (21) where $-W_{i}(q_{i})$ denotes the potential of the conservative force $\mathcal{F}_{i}$ and $W_{i}(q_{i})$ is equal to the sum of the work done by the nonconservative force $F$ and $const$. In Eq.(21) $\hat{H}$ and $H$ are both functions of $q_{i},p_{i}$ and $W_{i}(q_{i})$ a function of $q_{i}$. Eq.(21) and Eq.(17) can be thought of as a map from the total energy of the dissipative system(11) to the Hamiltonian of the conservative system(12). Indeed, $\left.\hat{H}\right|_{\gamma}$ and the total energy differ in the constant $const=-W_{i}(q_{i0})$. When the conservative system takes a different initial condition, if one does not change the function form of $\left.\hat{H}\right|_{\gamma}$, one can consider $\left.\hat{H}\right|_{\gamma}$ as a Hamiltonian quantity $\hat{H}$, $\hat{H}=\left.\hat{H}\right|_{\gamma}=H-W_{i}(q_{i})$ (22) and the conservative system(12) can be thought of as an entirely new conservative system. Based on the above, the following proposition is made: ###### Proposition 1 For any nonconservative classical mechanical system and any initial condition, there exists a conservative one; the two systems share one and only one common phase curve; the value of the Hamiltonian of the conservative system is equal to the sum of the total energy of the nonconservative system on the aforementioned phase curve and a constant depending on the initial condition. ###### Proof First we must prove the first part of the Proposition 1, i.e. that a conservative system with Hamiltonian presented by Eq.(22) shares a common phase curve with the nonconservative system represented by Eq.(11). In other words the Hamiltonian quantity presented by Eq.(22) satisfies Eq.(13) under the same initial condition. Substituting Eq.(22) into the left side of Eq.(13), we have $\displaystyle\frac{\partial{\hat{H}(q_{i},p_{i})}}{\partial{q_{i}}}$ $\displaystyle=$ $\displaystyle\frac{\partial H(q_{i},p_{i})}{\partial{q_{i}}}-\frac{\partial{W_{j}(q_{j})}}{\partial{q_{i}}}$ $\displaystyle\frac{\partial{\hat{H}(q_{i},p_{i})}}{\partial{p_{i}}}$ $\displaystyle=$ $\displaystyle\frac{\partial H(q_{i},p_{i})}{\partial{p_{i}}}-\frac{\partial{W_{j}(q_{j})}}{\partial{p_{i}}}.$ (23) It must be noted that although $q_{i}$ and $p_{i}$ are considered as distinct variables in Hamilton’s mechanics, we can consider $q_{i}$ and $\dot{q_{i}}$ as dependent variables in the process of constructing of $\hat{H}$. At the trajectory $\gamma$ we have $\displaystyle\frac{\partial{{W_{j}(q_{j})}}}{\partial{q_{i}}}$ $\displaystyle=$ $\displaystyle\frac{\partial{(\int_{q_{j0}}^{q_{j}}\mathcal{F}_{j}(q_{j})\mathrm{d}q_{j}+W_{i}(q_{i0}))}}{\partial{q_{i}}}=\mathcal{F}_{i}(q_{i})$ $\displaystyle\frac{\partial{{W_{j}(q_{j})}}}{\partial{p_{i}}}$ $\displaystyle=0,$ (24) where $\mathcal{F}_{i}(q_{i})$ is equal to the damping force $F_{i}$ on the phase curve $\gamma$. Hence under the initial condition $q_{0},p_{0}$, Eq.(13) is satisfied. As a result, we can state that the phase curve of Eq.(12) coincides with that of Eq.(11) under the initial condition; and $\hat{H}$ represented by Eq.(22) is the Hamiltonian of the conservative system represented by Eq.(12). Then we must prove the second part of Proposition 1: the uniqueness of the common phase curve. We assume that eq.(12) shares two common phase curves, $\gamma_{1}$ and $\gamma_{2}$, with eq.(11). Let a point of $\gamma_{1}$ at the time $t$ be $z_{1}$, a point of $\gamma_{2}$ at the time $t$ $z_{2}$, and $g^{t}$ the Hamiltonian phase flow of eq.(12). Suppose a domain $\Omega$ at $t$ which contains only points $z_{1}$ and $z_{2}$, and $\Omega$ is not only a subset of the phase space of the nonconservative system(11) but also that of the phase space of the conservative system(12). Hence there exists a phase flow $\hat{g}^{t}$ composed of $\gamma_{1}$ and $\gamma_{2}$, and $\hat{g}^{t}$ is the phase flow of eq.(11) restricted by $\Omega$. According to the following Louisville’s theoremArnold1978 : ###### Theorem 2.1 The phase flow of Hamilton’s equations preserves volume: for any region $D$ we have $volume\ of\ g^{t}D=volume\ of\ D$ where $g^{t}$ is the one-parameter group of transformations of phase space $g^{t}:(p(0),q(0))\longmapsto:(p(t),q(t))$ $g^{t}$ preserves the volume of $\Omega$. This implies that the phase flow of eq.(11) $\hat{g}^{t}$ preserves the volume of $\Omega$ too. But the system (11) is not conservative, which conflicts with Louisville’s theorem; hence only a phase curve of eq.(12) coincides with that of eq.(11). ∎∎ In the next section three examples is given to demonstrate Proposition 1. ## 3 Examples In this section, first two simple analytical examples are given, then a pro forma example is given. ### 3.1 One-dimensional Analytical Example Consider a special one-dimensional simple mechanical system: $\ddot{x}+c\dot{x}=0,$ (25) where $c$ is a constant. The exact solution of the equation above is $x=A_{1}+A_{2}e^{-ct},$ (26) where $A_{1},A_{2}$ are constants. From the equation above, we derived the velocity: $\dot{x}=-cA_{2}e^{-ct}.$ (27) From the initial condition $x_{0},\dot{x}_{0}$, we find $A_{1}=x_{0}+\dot{x}_{0}/c,A_{2}=-\dot{x}_{0}/c$. From Eq.(26) $t=-\frac{1}{c}\ln\frac{x-A_{1}}{A_{2}}$ (28) Substituting the equation above into Eq.(27), such we have $\dot{x}=-c(x-A_{1})=-c(x-A_{1})$ (29) The dissipative force $F$ in the dissipative system (25) is $F=c\dot{x}.$ (30) Substituting Eq.(29) into Eq.(30), such we have the conservative force $\mathcal{F}$ $\mathcal{F}=-c^{2}(x-A_{1});$ (31) Clearly the conservative force $\mathcal{F}$ depends on the initial condition of the dissipative system (25), in other words an initial condition determine a conservative force. Consequently a new conservative system yields $\ddot{x}+\mathcal{F}=0\rightarrow\ddot{x}-c^{2}(x-A_{1})=0.$ (32) The stiffness coefficient of the equation above must be negative. One can readily verify that the particular solution (26) of the dissipative system can satisfy the conservative one (32). This point agrees with Proposition (1). The potential of the conservative system（32）is $V=\int_{0}^{x}\left[-c^{2}(x-A_{1})\right]=-\frac{c^{2}}{2}x^{2}+c^{2}A_{1}x$ If $t\rightarrow\infty$，$x\rightarrow A_{1}$ and $\dot{x}\rightarrow 0$. This implies that the kinetic energy of the corresponding conservative system would tend to $0$ and the potential a constant $C^{2}A_{1}^{2}/2$ which is equal to the energy loss of the original system. Both the mechanical energy of the conservative system (32) at initial instance and $t\rightarrow\infty$ are $c^{2}A_{1}^{2}/2$. ### 3.2 Two-dimensional Analytical Example Let us consider a special two-dimensional mechanical system $\displaystyle\ddot{x}+\dot{x}-\dot{y}$ $\displaystyle=$ $\displaystyle 0$ $\displaystyle\ddot{y}-\dot{x}+\dot{y}$ $\displaystyle=$ $\displaystyle 0.$ (33) The exact solution of the equation above with initial initial condition $x_{0},y_{0},\dot{x}_{0},\dot{y}_{0}$ is $\displaystyle x(t)$ $\displaystyle=$ $\displaystyle-\frac{\dot{y}_{0}-\dot{x}_{0}-4x_{0}}{4}+\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{4}+\frac{t(\dot{y}_{0})}{2}+\frac{t(\dot{x}_{0})}{2}$ $\displaystyle y(t)$ $\displaystyle=$ $\displaystyle\frac{\dot{y}_{0}-\dot{x}_{0}+4y_{0}}{4}-\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{4}+\frac{t(\dot{y}_{0})}{2}+\frac{t(\dot{x}_{0})}{2}$ (34) For convenience to obtain $t=t(x),t=t(y)$，let $\dot{x}_{0}+\dot{y}_{0}=0$, then simplify the particular solution above to $\displaystyle x(t)=-\frac{\dot{y}_{0}-\dot{x}_{0}-4x_{0}}{4}+\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{4}$ $\displaystyle y(t)=\frac{\dot{y}_{0}-\dot{x}_{0}+4y_{0}}{4}-\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{4},$ (35) From the equation above, we derived the velocity: $\displaystyle\dot{x}$ $\displaystyle=$ $\displaystyle-\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{2},$ (36) $\displaystyle\dot{y}$ $\displaystyle=$ $\displaystyle\frac{{e}^{-2t}(\dot{y}_{0}-\dot{x}_{0})}{2}$ (37) Let the phase curve be denoted as $\gamma$. From Eq.(35), we obtain the inverse functions $\displaystyle t$ $\displaystyle=$ $\displaystyle-\frac{1}{2}\ln\left[\frac{4}{\dot{y}_{0}-\dot{x}_{0}}(x-x_{0})+1\right]$ (38) $\displaystyle t$ $\displaystyle=$ $\displaystyle-\frac{1}{2}\ln\left[-\frac{4}{\dot{y}_{0}-\dot{x}_{0}}(y-y_{0})+1\right]$ (39) Substituting Eq.(38)(39) into Eq.(36), we have the map at $\gamma$ from $x,y$ to $\dot{x}$: $\displaystyle\dot{x}(x)$ $\displaystyle=$ $\displaystyle-2x-\frac{\dot{y}_{0}-\dot{x}_{0}}{2}+2x_{0}$ (40) $\displaystyle\dot{x}(y)$ $\displaystyle=$ $\displaystyle 2y-\frac{\dot{y}_{0}-\dot{x}_{0}}{2}-2y_{0}$ (41) Substituting Eq.(38)(39) into Eq.(37), we have the map at $\gamma$ from $x,y$ to $\dot{y}$: $\displaystyle\dot{y}(y)$ $\displaystyle=$ $\displaystyle-2y+\frac{\dot{y}_{0}-\dot{x}_{0}}{2}+2y_{0}$ (42) $\displaystyle\dot{y}(x)$ $\displaystyle=$ $\displaystyle 2x+\frac{\dot{y}_{0}-\dot{x}_{0}}{2}-2x_{0}$ (43) The components of nonconservative $\bm{F}$ in the system (33) are $\displaystyle F_{1}$ $\displaystyle=$ $\displaystyle\dot{x}-\dot{y}$ (44) $\displaystyle F_{2}$ $\displaystyle=$ $\displaystyle-\dot{x}+\dot{y}$ (45) Substituting Eq.(40)(43) into $F_{1}$(44), then take the quantity as the first component the conservative force $\mathcal{F}$: $\mathcal{F}_{1}(x)=-4x-(\dot{y}_{0}-\dot{x}_{0})+4x_{0}.$ (46) Substituting Eq.(41)(42) into $F_{2}$(45), then take the quantity as the second component the conservative force $\mathcal{F}$: $\mathcal{F}_{2}(y)=-4y+(\dot{y}_{0}-\dot{x}_{0})+4y_{0}$ (47) Since $\partial\mathcal{F}_{1}/\partial y=\partial\mathcal{F}_{2}/\partial x=0$, $\mathcal{F}$ must be conservative. Consequently we obtain a new conservative system: $\displaystyle\ddot{x}$ $\displaystyle=$ $\displaystyle-\mathcal{F}_{1}$ $\displaystyle=$ $\displaystyle 4x+(\dot{y}_{0}-\dot{x}_{0})-4x_{0}$ $\displaystyle\ddot{y}$ $\displaystyle=$ $\displaystyle-\mathcal{F}_{2}$ (48) $\displaystyle=$ $\displaystyle 4y-(\dot{y}_{0}-\dot{x}_{0})-4y_{0}.$ We can readily prove that the particular solution (35) can satisfy Eq.(48) too. In this case, this point agrees with Proposition 1 too. ### 3.3 A Formell Example in Vibration Mechanics Take an $n$-dimensional oscillator with damping as an example, the governing equation of which is as below: $\ddot{\bm{q}}+\mathsfsl{C}\dot{\bm{q}}+\mathsfsl{K}\bm{q}=0,$ (49) where $\bm{q}=\left[q_{1},\dots,q_{n}\right]^{T}$, superscript $T$ denotes a matrix transpose, $\mathsfsl{C}=\left[\begin{array}[]{ccc}C_{11}&\dots&C_{1n}\\\ \vdots&\ddots&\vdots\\\ C_{n1}&\dots&C_{nn}\end{array}\right],\mathsfsl{K}=\left[\begin{array}[]{ccc}K_{11}&\dots&K_{12}\\\ \vdots&\ddots&\vdots\\\ K_{21}&\dots&K_{22}\end{array}\right]$ , and $C_{ij}$ and $K_{ij}$ are constants. It is complicated to solve Eq.(49). If Eq.(49) is higher dimensional, it is almost impossible to solve Eq.(49) analytically. Therefore we assume that a solution exists already. $\bm{q}=\bm{q}(t)=\left[q_{1}(t),\dots,q_{n}(t)\right].$ (50) Suppose a group of inverse functions $t=t(q_{1}),\dots,t=t(q_{n}).$ (51) As in Eq.(19) we can consider that the damping forces are equal to some conservative force under an initial condition $\begin{array}[]{ccc}c_{11}\dot{q}_{1}=\varrho_{11}(q_{1})&\dots&c_{1n}\dot{q}_{n}=\varrho_{1n}(q_{1})\\\ \vdots&\ddots&\vdots\\\ c_{n1}\dot{q}_{1}=\varrho_{21}(q_{n})&\dots&c_{nn}\dot{q}_{n}=\varrho_{nn}(q_{n}),\end{array}$ (52) where $\varrho_{ij}(q_{i})$ is a function of $q_{i}$. For convenience, these conservative forces can be defined as functions which are analogous to elastic restoring forces: $\begin{array}[]{ccc}\varrho_{11}(q_{1})=\kappa_{11}(q_{1})q_{1}&\dots&\varrho_{1n}(q_{1})=\kappa_{1n}(q_{1})q_{1}\\\ \vdots&\ddots&\vdots\\\ \varrho_{n1}(q_{1})=\kappa_{n1}(q_{n})q_{n}&\dots&\varrho_{nn}(q_{n})=\kappa_{nn}(q_{n})q_{n},\end{array}$ (53) where $\kappa_{ij}(q_{i})$ is a function of $q_{i}$. An equivalent stiffness matrix $\mathsfsl{\tilde{K}}$ is obtained, which is a diagonal matrix $\mathsfsl{\tilde{K}}_{ii}=\sum_{l=1}^{n}\kappa_{il}(q_{l}).$ (54) Consequently an $n$-dimensional conservative system is obtained $\bm{\ddot{q}}+(\mathsfsl{K}+\mathsfsl{\tilde{K}})\bm{q}=0$ (55) which shares a common phase curve with the $n$-dimensional damping system(49). The Hamiltonian of Eqs.(55) is $\hat{H}=\frac{1}{2}\bm{p}^{T}\bm{p}+\frac{1}{2}\bm{q}^{T}\mathsfsl{K}\bm{q}+\int_{\bm{0}}^{\bm{q}}(\tilde{\mathsfsl{K}}\bm{q})^{T}\mathrm{d}\bm{q},$ (56) where $\bm{0}$ is a zero vector, $\bm{p}=\dot{\bm{q}}$. $\hat{H}$ in Eq.(56) is the mechanical energy of the conservative system(55), because $\int_{\bm{0}}^{\bm{q}}(\tilde{\mathsfsl{K}}\bm{q})^{T}\mathrm{d}\bm{q}$ is a potential function such that $\hat{H}$ doest not depend on any path. ### 3.4 Discussion Based on the above, we can outline the relationship between a dissipative mechanical system and a group of conservative systems by means of Fig. 1. The relationship can be stated from two perspectives: Figure 1: A Dissipative Mechanical System and Conservative Systems If one explains the relationship from a geometrical perspective, one can obtain Proposition 1. In this paper the conservative systems (12) and (55) are called the substituting systems. Although a substituting system shares a common phase curve with the original system, under other initial conditions the substituting system exhibits different phase curves. Therefore the phase flow of the substituting system differs from that of the original system, it follows that the substituting systems is not equal to the original system. According to Louisville’s theorem (2.1), the phase flow of the original dissipative system Eq.(11) certainly does not preserve its phase volume, but the phase flow of the substituting conservative Eq.(12) does. One also could explain the relationship from a mechanical perspective. It is known that there are non-conservative forces in a nonconservative system. The total energy of the nonconservative system consists of the work done by nonconservative forces. Hence the function form of the total energy depends on a phase curve i.e. under an initial condition. If one constrains the total energy function to a phase curve $\gamma$, the total energy function can be converted into a function of $q,p$. One take $\hat{H}$ consisting of this new function and a constant as a Hamiltonian quantity, such that a Hamilton’s system (i.e., a conservative system) is obtained. Under the initial condition mentioned above, the solution curve of the conservative system is the same as that of the original nonconservative system; under other initial conditions the solution curve of the conservative is different from that of the original nonconservative system. Since one defines the forces(19,52,53,54) in the new system, the Hamiltonian quantity of the conservative can be thought of as the mechanical energy of the new conservative system as Eq.(56). One might doubt that the orbit of a dissipative dynamical system must be asymptotic, can the asymptotic orbit coincide with one of a conservative mechanical system. In some literatureSunyishui_book_e_2008 , a conservative system defined a system with the behavior of the preservation of phase volume. HasselblattHasselblatt_Katok2003 had explained the question: ’A key to understanding this difference is given by a property that is not directly observed by looking at individual orbits but by considering the evolution of large sets of initial conditions simultaneously, the preservation of phase volume.’ This point agrees with the second part of the proof of the Proposition 1. The Hamiltonians of the new conservative systems in general are not analytically integrable, unless the original mechanical system is integrable. The reason is that the work done by damping force depends on the phase curve. If the system is integrable, then the phase curve can be explicitly written out, the system has an analytical solution, and therefore the work done by damping force can be explicitly integrated. Subsequently, the Hamiltonian $\hat{H}$ can be explicitly expressed. Most systems do not have an analytical solution. Despite this, the Hamilton quantity, coordinates and momentum must satisfy Eq.(12) under a certain initial condition. Why had KleinKlein1928 written, ”Physicists can make use of these theories only very little, an engineers nothing at all”? The answer: when one is seeking an analytical solution to a classical mechanics problem by utilizing Hamiltonian formalism, in fact one must inevitably convert the problem back to Newtonian formalism. This means that an explicit form of Hamiltonian quantity is not necessary for classical mechanics. What is important is the relationship between $q,p$ and the Hamiltonian quantity embodied in the Hamilton’s Equation. ## 4 conclusions We can conclude that a dissipative mechanical system has such properties: for any nonconservative classical mechanical system and any initial condition, there exists a conservative one, the two systems share one and only one common phase curve; the Hamiltonian of the conservative system is the sum of the total energy of the nonconservative system on the aforementioned phase curve and a constant depending on the initial condition. We can further conclude, that a dissipative problem can be reformulated as an infinite number of non- dissipative problems, one corresponding to each phase curve of the dissipative problem. One can avoid having to change the definition of the canonical momentum in the Hamilton formalism, because under a certain initial condition the motion of one of the group of conservative systems is the same as the original dissipative system. ## References * (1) Arnold., V.I.: Mathematical Methods of classical Mechanics, second edition. Springer-Verlag, Berlin (1978) * (2) Arnold., V.I.: Mathematical aspects of classical and celestial mechanics. Springer-Verlag, Berlin (1997) * (3) Bateman, H.: On dissipative systems and related variational principles. Phys. Rev. 38(4), 815–819 (1931). DOI 10.1103/PhysRev.38.815 * (4) Caldirola, P.: Forze non conservative della meccanica quantistica. Nuovo Cim 18, 393–400 (1941) * (5) F.Klein: Entwickelung der Mathematik im 19 Jahrhundert. Teubner (1928) * (6) Hasselblatt, B., Katok, A.: A FIRST COURSE IN DYNAMICS with a Panorama of Recent Developments. AMBRIDGE UNIVERSITY PRESS (2003) * (7) Kanai, E.: On the Quantization of the Dissipative Systems. Progress of Theoretical Physics 3, 440–442 (1948) * (8) Kochan, D.: Functional integral for non-lagrangian systems. Phys. Rev. A 81(2), 022,112 (2010). DOI 10.1103/PhysRevA.81.022112 * (9) Kochan, D.: How to quantize forces (?): An academic essay on how the strings could enter classical mechanics. Journal of Geometry and Physics 60(2), 219 – 229 (2010). DOI DOI: 10.1016/j.geomphys.2009.09.014. URL http://www.sciencedirect.com/science/article/B6TJ8-4XDCHN8-1/2/ed88d72d5e7e026557c477b9416a744e * (10) Krechetnikov, R., Marsden, J.E.: Dissipation-induced instabilities in finite dimensions. Reviews of Modern Physics 79, 519–553 (2007). DOI 10.1103/RevModPhys.79.519 * (11) P.Morse, Feshbach, H.: Methods of Theoretical Physics. McGraw-Hill, New York (1953) * (12) Rajeev, S.: A canonical formulation of dissipative mechanics using complex-valuedhamiltonians. ANNALS of PHYSICS 322(3), 1541–1555 (2007) * (13) Sun, Y., Zhou, Y.: Introduction to Modern Celestial Mechanics. Higher Education Press (2008)

arxiv-papers

2024-09-04T02:48:54.353033

0803.2395

# RAId_DbS: A Mass-Spectrometry Based Peptide Identification Web Server with Knowledge Integration Gelio Alves Aleksey Ogurtsov and Yi-Kuo Yu111to whom correspondence should be addressed: yyu@ncbi.nlm.nih.gov National Center for Biotechnology Information, National Library of Medicine, NIH, Bethesda, MD 20894 (2008; 2008) ###### Abstract ## 1 Summary: In anticipation of the individualized proteomics era and the need to integrate knowledge from disease studies, we have augmented our peptide identification software RAId_DbS to take into account annotated single amino acid polymorphisms, post-translational modifications, and their documented disease associations while analyzing a tandem mass spectrum. To facilitate new discoveries, RAId_DbS allows users to conduct searches permitting novel polymorphisms. ## 2 Availability: The webserver link is http://www.ncbi.nlm.nih.gov/ /CBBResearch/qmbp/raid_dbs/index.html. The relevant databases and binaries of RAId_DbS for Linux, Windows, and Mac OS X are available from the same web page. ## 3 Contact: yyu@ncbi.nlm.nih.govyyu@ncbi.nlm.nih.gov ### Introduction Like single nucleotide polymorphisms (SNPs) that occur roughly every $300$ base pairs (Collins et al., 1998), single amino acid polymorphisms (SAPs) also differentiate individuals from one another. In addition to from nonsynonymous SNPs, SAPs may result from post-transcriptional regulations such as mRNA editing. SAPs together with post-translational modifications (PTMs) often distinguish healthy/diseased forms of proteins. Integration of this annotated, disease-related knowledge with data analysis facilitates speedy, dynamic information retrieval that may significantly benefit clinical laboratory studies. To incorporate knowlege information within peptide searches, we start by constructing a human protein database where information about annotated SNPs, SAPs, PTMs, and their disease associations (if any) are integrated. We have also modified our peptide identification software RAId_DbS (Alves et al., 2007a) to take into account this additional information while performing peptide searches. Consequently, part of our work may be considered an improvement over that of Schandorff et al. (2007) who extended the human protein database to include only SAPs but without PTMs and without integration of disease information. Besides using our web server, a user may also download a standalone executable to be installed on her/his local machines. Once a user chooses to do so, she/he will find an important feature of the standalone version: the flexibility for users to add their own SAP and/or PTM information to various proteins they are interested in and even to add new protein sequences to the database. ### Implementation Summary In addition to giving a brief introduction to our software RAId_DbS and its augmentation, we focus in this section on explaining how we accommodate the SAPs, PTMs, and their disease associations in our database. Appropriate comparison to existing approaches will also be discussed. Prior to database construction, we perform a information-preserved protein clustering (see supplementary information). #### Database Construction To minimize inclusion of less confident annotations, we only keep the SAPs and PTMs that are consistently documented in more than one source. For example, for proteins with Swiss-Prot accession number, we only keep the SAPs and PTMs that are annotated both by Swiss-Prot and GeneBank. For proteins without Swiss-Prot accession numbers, the retentions of SAPs and PTMs are described in the supplementary information. A typical sequence in our augmented human protein database carries with it annotated SAPs and PTMs in a simple format, see Figure 1 and its caption. Our data format minimizes redundancy. For example, if a single site contains two SAPs, construction method proposed by Schandorff et al. (2007) will demand two almost identical partial sequences, each may be several tens of amino acids in length, be appended after the primary sequence, while in our case it only takes up a few additional bytes. The compactness of our database becomes obvious when incorporating the information of two nearby sites, each containing several annotated SAPs and PTMs, into the database. In our construction, we only need a few additional bytes. But in other approaches, it may introduce a combinatorial expansion due to including/excluding and pairing of different variations at both sites along with the flanking peptides. Another key difference between our method and other database methods is that we do not need to limit the number of enzymatic miscleavages. When needed, users of RAId_DbS may modify the database, add new sequences, or even create their own databases following the same format. There is a separate information file that contains the protein accession numbers, detailed SAP and PTM information, and disease associations. If one wishes to add additional SAPs or PTMs, one simply updates both the ASCII database file as well as the information file. When reporting a hit with annotated SAPs or PTMs, RAId_DbS automatically reports the corresponding detailed information and disease association if it exists. Example search results of augmented RAId_DbS. (a) $E$-value $P$-value Peptide Mol. Wt. Protein ID Novel SAP Disease 1.184e-01 1.744e-05 RTKLKDC…KIAR 2897.500 (NP_114412;…;Q9H2L5) disabled ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 4.084e+00 9.345e-03 KQQELAA…VSSR 2898.520 (NP_072096;…;O75420) disabled (b) $E$-value $P$-value Peptide Mol. Wt. Protein ID Novel SAP Disease 3.977e-07 1.834e-10 KsVEEYANCHLAR 1448.650 (NP_001054;…;P02787) disabled 4.779e-01 2.205e-04 KsVqEYANCHLAR 1447.670 (NP_001054;…;P02787) disabled ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 7.524e-01 3.470e-04 R$\ell$MNAsMVWAQAAR 1448.720 (NP_000337;…;P48436) disabled {($\ell$;2;108;Campomelic dysplasia (CMD1) [MIM:114290]) (s;6;112;Campomelic dysplasia (CMD1) [MIM:114290]) } RAId_DbS by default perform searches considering the parent ion to have charge +1, +2, +3. The search results are then pooled together to form a single result ranked by $E$-values. This is why for the same spectrum RAId_DbS may report peptide hits with very different masses. In part (a) ((b)), searches were done with annotated SAPs and PTMs turned off (on). The lowercase letters in the peptide indicate SAPs. A novel SAP, if present and enabled in the searches, will be specified in the column headed by Novel SAP. Note that in the disease related annotation, there are four fields separated by three semicolons. The first field, a lower case amino acid letter, indicates the SAP; second field, an integer, indexes the SAP position in the peptide; the third field, an integer, indexes the SAP position in the protein; the fourth field shows the annotated disease association. #### RAId_DbS Augmentation Taking into account the finite sample effect and skewness, the asymptotic score statistics ($P$-values) of RAId_DbS (Alves et al., 2007a) is derived theoretically. The final $E$-value for each peptide hit, however, is obtained by multiplying the peptide’s $P$-value by the number of peptides of its category. RAId_DbS then ranks peptide hits according to $E$-values, not $P$-values. This avoids overstating the significance of a hit from a larger effective database and is particularly helpful in reducing false positives when we allow SAPs and PTMs in the searches (see the supplementary information for details). In addition to performing database searches that consider annotated SAPs and PTMs, we also allow users to include for consideration one novel SAP per peptide. This feature, helpful for scrutinizing otherwise unidentifiable spectra, will increase the effective peptide database size. However, it does not cause harm since the $E$-values associated with peptide hits containing novel SAPs are obtained by multiplying their $P$-values with a much larger number than that for peptide hits without novel SAPs (see supplementary information for details). MLLATLLLLLLGGALAHPDRIIFPNHACEDPPAVLLEVQGTLQRPLVR$\langle${W00}$\rangle$D SRTSPAN$\langle$(N08,N09,N10,N11,N12)$\rangle$CTWLILGSKEQTVTIRFQKLHLACGSERL TLRSPLQPLISLCEAPPSPLQLPGGN$\langle$(N08,N09,N10,N11,N12)$\rangle$VTITYSYAGA RAPMGQGFLLSYSQDWLM$\langle${V00}$\rangle$CLQEEFQCLNHRCVSAVQR…………[ Figure 1: Protein sequence (NP_054764) used as an example to demonstrate our database structure. A “[” character is always inserted after the last amino acid of each protein to serve as a separator. Annotated SAPs and PTMs associated with an amino acid are included in a pair of angular brackets following that amino acid. SAPs are further enclosed by a pair of curly brackets while PTMs are further enclosed by a pair of round brackets. Amino acid followed by two zeros indicates an annotated SAP. Every annotated PTM has a two-digit positive integer that is used to distinguish different modifications. ### Example Using a tandem mass (MS2) spectrum taken from the profile dataset described earlier (Alves et al., 2007b), we illustrate in Table 1 two search results in the human protein database with the annotated SAPs and PTMs turned off (a) and on (b) respectively. In case (a), the best hit is a false positive with $E$-value about $0.11$ implying that one probably ends up declaring no significant peptide hit for this spectrum. In case (b), however, the best hit is a true positive (a peptide from human transferin with an annotated SAP) with $E$-value about $4.0\times 10^{-7}$. This example shows that if properly used, allowing SAPs/PTMs may increase peptide identification rate. That is, it may be fruitful to turn on the SAPs/PTMs when a regular search returns no significant hit. Blindly turn on SAPs/PTMs, however, may cause loss of sensitivity due to the increase of search space. In supplementary information, using the $54$ training spectra of PEAKS, we compare RAId_DbS’s peptide identifications with and without SAPs/PTMs. The purpose is to study the degree of loss in senstivity when turning on the SAPs/PTMs. Although for the data set tested there is no obvious loss in sensitivity (perhaps due to the statistical accuracy of RAId_DbS), we recommend running searches with SAPs/PTMs on only when a regular search returns no significant hit. ### Conclusion To enable speedy information retrieval and to enhance the protein coverage while analyzing MS2 peptide spectra, we have augmented the capability of RAId_DbS and integrated with protein database additional information such as SAPs, PTMs, and disease annotations. Incorporation of known SAPs and PTMs during initial searches may enhance the peptide identification rate. Integration of disease knowledge and information may be crucial in many time- pressed clinical uses. We are currently investigating the possibility of combining various isoforms of proteins into a single entry in addition to clustering almost identical proteins. We are experimenting with keeping the longest form of the protein and marking at the beginning of the sequence possible deletions. Once achieved, this enhancement will further reduce redundant searches which should result in a shorter run time. Another objective is to cover more organisms. Currently, we have finished database construction of $17$ organisms, including Homo sapiens, Drosophila melanogaster, Saccharomyces cerevisiae etc. (see supplementary information for details). We will provide more organismal databases on our web server once constructed. ### Acknowledgement This work was supported by the Intramural Research Program of the National Library of Medicine at National Institutes of Health/DHHS. Funding to pay the Open Access publication charges for this article was provided by the NIH. ## References * Alves et al. (2007a) Alves, G., Ogurtsov A.Y. and Yu, Y.-K. (2007a) RAId_DbS: Peptide Identification using Database Searches with Realistic Statistics. Biology Direct, 2:25. * Alves et al. (2007b) Alves, G., Ogurtsov A.Y., Wu, W., Wang, G., Shen, R.-F. and Yu, Y.-K. (2007b) Calibrating E-values for MS2 database search methods. Biology Direct, 2:26. * Collins et al. (1998) Collins, F.S., Brooks, L.D. and Chakravarti, A. (1998) A DNA polymorphism discovery resource for research on human genetic variation. Genome Res., 8, 1229-31. * Schandorff et al. (2007) Schandorff, S., Olsen, J.V., Bunkenborg, J., Blagoev, B., Zhang, Y., Andersen, J.S. and Mann, M. (2007) A mass spectrometry-friendly database for cSNP identification. Nat. Methods, 4, 465-466. ## Supplementary Information ### Information-preserved Protein Clustering and Database Construction We extract $34,197$ human protein sequences with a total of $16,814,674$ amino acids from the file (last updated 09/05/2006) ftp://ftp.ncbi.nlm.nih.gov/genomes/H_sapiens/protein/protein.gbk.gz. Each protein sequence is accompanied by a list of annotated SAPs and PTMs. Out of the $34,197$ proteins, we found $29,979$ unique proteins with a total of $15,324,913$ amino acids. To avoid having multiple copies of identical or almost identical proteins in the database, we first cluster the $34,197$ sequences by running an all-against-all BLAST. Two sequences with identical lengths and aligned gaplessly with less than $2\%$ mismatches are clustered together, and each sequence is called a qualified hit of the other. Any other sequence that satisfies this condition with a member of an existing cluster is assigned to that existing cluster. All the annotations in the same cluster are then merged. We find it possible for every given cluster to choose a consensus sequence that will make all other members its polymorphous forms. Hence, we only retain one protein sequence for each of the $29,272$ clusters. The total number of amino acids associated with these $27,272$ consensus proteins is $15,001,326$. Although we only retain one sequence (the consensus sequence) per cluster, the information of other member sequences are still kept. For example, when a member sequence and the consensus sequence disagree at two sites, the presence of the member sequence is documented by introducing two cluster-induced SAPs at the two sites of the consensus sequence. The originally annotated SAPs and PTMs of the member sequence are also merged into those of the consensus sequence. Figure 2 illustrates how this is process is done iteratively. In our information file, each SAP or PTM is documented with its origin. SAPs arising from clustering are easily distinguished from annotated SAPs. For member sequences that are identical to the consensus sequence, the accession number of those member sequences are also recorded with their SAPs/PTMs annotations merged into the consensus sequence. When a user selects not to have annotated SAPs, RAId_DbS still allows for cluster-induced SAPs resulting in an effective search of the original databases but with minimum redundancy. The strategy employed by RAId_DbS to search for SAPs and PTMs will be briefly described in the RAId_DbS section below. consensus seq. …DPR… … …LQRLVADN$\langle$(N08)$\rangle$GSE … member seq. …DPR$\langle${W00}$\rangle$…LKRLVVDN$\langle$(N11)$\rangle$GSE … updated consensus seq. …DPR$\langle${W00}$\rangle$…LQ$\langle${K00}$\rangle$RLVA$\langle${V00}$\rangle$DN$\langle$(N08,N11)$\rangle$GSE… Figure 2: Information-preserved protein clustering example. Once a consensus sequence is selected, members of the clusters are merged into the consensus one-by-one. This figure illustrates how the information of a member sequence is merged into the consensus sequence. The difference in the primary sequences between a member and the consensus introduces cluster-induced SAPs. In this example, the residues Q and A (in red) in the consensus are different from the residues K and V (in blue) in the member sequence. As a consequence, K becomes a cluster-induced SAP associated with Q and V becomes a cluster-induced SAP associated with A at these respective sites of the consensus. The annotated SAP, {W00}, associated with residue R in the member sequence is merged into the consensus sequence, see the updated consensus sequence in the figure. Note that the annotated PTM, $\langle$(N11)$\rangle$, associated with N in the member sequence is merged with a different annotated PTM, $\langle$(N08)$\rangle$, at the same site of the consensus sequence. As mentioned earlier, although the SAPs, PTMs are merged, each annotation’s origin and disease associations are kept in the information file allowing for faithful information retrieval at the final reporting stage of the RAId_DbS program. The consensus protein in a given cluster is then used as a query to BLAST against the NCBI’s nr database to retrieve its RefSeq accession number and its corresponding Swiss-Prot (http://ca.expasy.org/sprot/) accession number, if it exists, from the best qualified hit. It is possible for a cluster to have more than one accession number. This happens when there is a tie in the qualified best hits and when a protein sequence in nr actually is documented with more than one accession number. To minimize inclusion of less confident annotations, we only keep the SAPs and PTMs that are consistently documented in more than one source. For example, for proteins with Swiss-Prot accession number, we only keep the SAPs and PTMs that are annotated both by Swiss-Prot and GeneBank. For proteins without Swiss-Prot accession numbers, the retentions of SAPs and PTMs are described below. The PTM annotations are kept only if they are present in the gzipped document HPRD_FLAT_FILES_090107.tar.gz of the Human Protein Reference Database: http://www.hprd.org/download. The SAP annotations are kept only if they are in agreement with the master table, SNP_mRNA_pos.bcp.gz (last updated 01/10/2007), of dbSNP: ftp://ftp.ncbi.nlm.nih.gov/snp/organisms/human_9606/ /database/organism_data. ### RAId_DbS Taking into account the finite sample effect and skewness, the form of asymptotic score statistics ($P$-values) of RAId_DbS (Alves et al., 2007a) is derived theoretically. Since the skewness varies per spectrum, the parameters for our theoretical distribution are spectrum-specific. For each spectrum considered, our theoretical distribution (used to compute $P$-value) mostly agrees well with the score histogram accumulated. The final $E$-value for each peptide hit, however, is obtained by multiplying the peptide’s $P$-value by the number of peptides of its category. As a specific example, when Trypsin is used as the digesting enzyme, RAId_DbS allows for incorrect N-terminal cleavages. RAId_DbS has internal counters, $C_{c}$ and $C_{inc}$, counting respectively the number of scored peptides with correct and incorrect N-terminal cleavage. In general, $C_{inc}\gg C_{c}$. When calculating the $E$-value of a peptide with correct N-terminal cleavage, RAId_DbS multiplies the peptide’s $P$-value by $C_{c}$. However, the $E$-value of a peptide with incorrect N-terminal cleavage will be obtained by multiplying the peptide’s $P$-value by $C_{c}+C_{inc}$ (Alves et al., 2007a). In line with the Bonferroni correction, our approach avoids overstating the significance of a hit from a larger effective database (the pool of peptides regardless of whether the N-terminal cleavage is correct) versus a hit from a smaller effective database (the pool of peptides with correct N-terminal cleavage only). The same idea is used in the augmented RAId_DbS. That is, different counters are set up to record the number of scored peptides in different categories. As a specific example, when novel SAPs are allowed, RAId_DbS creates a new counter, $C_{novel\\_sap}$, to record the number of scored peptides with a novel SAP. This is in general a much larger number than other counters. When one calculates the $E$-value associated with a peptide hit that contains a novel SAP, one will multiply the peptide’s $P$-value by the sum of a number of coutners with $C_{novel\\_sap}$ included. However, in the same search, for a peptide without novel SAP, its $E$-value is obtained by multiplying the peptide’s $P$-value by the sum of a number of counters excluding $C_{novel\\_sap}$. The same approach is applied to PTMs and other annotations. Below we briefly sketch how RAId_DbS deals with the presence of annotated SAPs, PTMs as well as novel SAPs. In our database format, annotated SAPs and PTMs are inserted right after the site of variation. When searching the database for peptides with parent ion mass 1500 Da, RAId_DbS sums the masses of amino acids within each possible peptide to see if the total mass is within 3 Da of $1500$ Da. At this stage, sites with variations will have, instead of a fixed mass, several possible masses depending on the number of SAPs/PTMs are annotated at these sites. Each peptide fragmnent covering some of those sites will therefore have several effective masses, each corresponding to a specific arrangement of SAPs/PTMs. If some of these masses happens to be within 3 Da of $1500$ Da, RAId_DbS will score this peptide with corresponding annotated SAPs/PTMs that give rise to the proper masses. If none of these masses are within the allowed molecular mass range, that peptide will not be scored. Note that this approach is computationally efficient in terms of mass selection. For example, if a peptide contains a site with annotated SAPs/PTMs, one computes the mass of this peptide by summing once the amino acid masses of other sites. It is then a simple matter to see whether the addition of this sum to the list of masses associated with the site with SAPs/PTMs may fall in the desriable mass range. This approach is particularly powerful when there are more than one site with SAPs/PTMs in the peptide considered. The combinatorics associated with two sites with SAPs/PTMs only result in a longer list of possible masses to be added to the mass sum of unvaried sites. This should be constrasted with methods that incorporate SAPs via appending polymorphous peptides to the end of the primary sequence. In the latter approach, the program needs to do the mass sum multiple times, repeating the mass sum of unvaried sites, and thus may significantly slow down the searches. Despite RAId_DbS’s strategic advantage, introduction of SAPs/PTMs does increase the complexity of the algorithm. Therefore, we limit per peptide the maximum number of annotated SAPs to be $2$ and the maximum number of annotated PTMs to be $5$. To facilitate discovery, RAId_DbS also permits novel SAPs, but limited to one novel SAP per not-yet-annotated peptide, meaning peptides that do not contain any annotated SAPs/PTMs. This is because the introduction of novel SAP largely expand the search space, and if one allows novel SAPs on peptides already documented with SAPs/PTMs, the search space expansion will be even larger and may render the search intractable. Currently, the novel SAP is expedited via a pre-computed list of amino acid mass difference. As an example, assume that one is searching for a peptide with parent ion mass $1500$ Da, and a not-yet-annotated candidate peptide has mass $1477$ Da, $23$ Da smaller than the target mass. It happens that $23$ Da is also the mass difference between Tryptophan and Tyrosine, and if the candidate peptide contains a Tyrosine, RAId_DbS will replace that Tyrosine with a Tryptophan and score the new peptide. If the candidate peptide contains two Tyrosines, RAId_DbS will replace one Tyrosine at a time with a Tryptophan and score both the new peptides. It is evident that the complexity grows fast if one were to allow for two novel SAPs per petpide. It is commonly believed that when searching in a larger database, one is bound to loose sensitivity. This may be true if the $E$-value for every hit is obtained by multiplying the peptide’s $P$-value by the same number regardless of the category that peptide belongs to. As we have explained earlier, RAId_DbS does not do that. It uses a method equivalent to Bonferroni correction. We use $E$-values to rank peptide hits and each peptide’s $E$-value is obtained by multiplying its $P$-value by the corresponding size of the effective database that the peptide belongs to. Consequently, peptide hits falling in a category that has a large effective database size essentially need to have smaller $P$-values than those of peptide hits falling in a category that has a small effective database size. In Figure 3, we show the Receiver Operating Characteristic (ROC) curves when analyzing the training $54$ spectra of PEAKS. In this data set, there are $17$ spectra from yeast, $23$ spectra from bovine, and $14$ spectra from horse. The true positive proteins are already provided by PEAKS. We search the spectra generated by proteins of yeast, bovine, and horse respectively in the databases fo yeast, bovine, and horse. Since the true positive proteins are already known, it is relatively easy to perform the ROC analysis using the search results from the $54$ spectra. There are three ROC curves shown in Figure 3, one for searches without SAPs/PTMs, one for searches allowing annotated SAPs/PTMs, and one for searches allowing both annotated SAPs/PTMs as well as novel SAPs. Figure 3: ROC curves for three different search strategies employed when running RAId_DbS. ### Summary of Organismal Databases Constructed So far, we have finished constructing databses for $17$ organisms. We summarize these database in Table Summary of Organismal Databases Constructed below. Note that the disease information is only available for the human database. For human database, we have $123,464$ SAPs and $81,984$ PTMs. Out of those SAPs and PTMs, $15,787$ of them have disease associations. Summary of Augmented Organismal Databases Searchable by RAId_DbS. Organism DB_name SAPs included PTMs included DB_size (byte) Homo sapiens hsa.seq 123464 81984 16,292,193 Anopheles gambiae angam.seq 350 50 6,042,277 Arabidopsis thaliana artha.seq 5207 11977 12,318,213 Bos taurus botau.seq 3295 15810 11,188,490 Caenorhabditis elegans caele.seq 1045 7756 10,050,609 Canis familiaris cafam.seq 2766 4196 18,458,474 Danio rerio darer.seq 7358 3841 14,477,794 Drosophila melanogaster drmel.seq 5611 9290 9,796,785 Equus caballus eqcab.seq 485 1045 9,404,150 Gallus gallus gagal.seq 1109 6522 8,728,501 Macaca mulatta mamul.seq 1370 1262 14,498,187 Mus musculus mumus.seq 27614 61684 14,363,491 Oryza sativa orsat.seq 1291 2182 10,679,924 Pan troglodytes patro.seq 5201 3734 20,227,873 Plasmodium falciparum plfal.seq 56 184 3,995,386 Rattus norvegicus ranor.seq 9297 33240 15,879,569 Saccharomyces cerevisiae sacer.seq 5507 13220 2,927,330

arxiv-papers

2024-09-04T02:48:54.357902

0803.2401

# Rotation numbers of invariant manifolds around unstable periodic orbits for the diamagnetic Kepler problem Zuo-Bing Wu State Key Laboratory of Nonlinear Mechanics, Institute of Mechanics, Chinese Academy of Sciences, Beijing 100080, China ###### Abstract In this paper, a method to construct topological template in terms of symbolic dynamics for the diamagnetic Kepler problem is proposed. To confirm the topological template, rotation numbers of invariant manifolds around unstable periodic orbits in a phase space are taken as an object of comparison. The rotation numbers are determined from the definition and connected with symbolic sequences encoding the periodic orbits in a reduced Poincaré section. Only symbolic codes with inverse ordering in the forward mapping can contribute to the rotation of invariant manifolds around the periodic orbits. By using symbolic ordering, the reduced Poincaré section is constricted along stable manifolds and a topological template, which preserves the ordering of forward sequences and can be used to extract the rotation numbers, is established. The rotation numbers computed from the topological template are the same as those computed from their original definition. ## 1 Introduction Many interesting nonlinear systems in experiments are wells described by low- dimensional dynamical models. Their dynamical processes include of transition of periodic orbits from stable to unstable and bifurcation to chaos. To make a global understanding of the system, some topological and geometrical methods based on periodic orbits are developed[1]. Symbolic dynamics, as a coarse- grained description of the dynamics, provides an effective tool to depict the topological dynamics[2]. In special, for a chaotic system, two curve families of stable and unstable manifolds intersect each other and decompose a Poincaré section[3,4]. Enumeration and existence of unstable periodic orbits (UPOs) in the Poincaré section are determined[5,6] and some numerical methods such as finding UPOs are proposed[7]. Besides the features of dynamics in a Poincaré section, evolution of manifolds in a phase space, which possibly gains a geometric insight into the dynamics, is another important characteristic of the system. For example, bifurcation to chaos can be identified by tracing the evolution of stable and unstable manifolds in the phase space with parameters. Recently, some algorithms for computing two-dimensional stable and unstable manifolds in a three-dimensional phase space are proposed[8,9,10] and applied to visualize the structure of chaos[11]. Since UPOs are closely related to invariant manifolds, a basic issue in their relations is describing how local invariant manifolds rotate around UPOs in the phase space. Moreover, how reducing their topological relation into a two-dimensional template is important for understanding the global organization of UPOs in chaos. To motivate the visualization of rotation of invariant manifolds around UPOs and construction of topological template in reduction of stable manifolds, we consider the model for diamagnetic Kepler problem (DKP)[12]. In our previous works, two coordinate axes are chosen as a Poincaré section to form an annulus in a lifted space. The dynamics on the annulus can be reduced by considering the symmetry of system. In view of stretching and wrapping in the lifted space, symbolic dynamics without involving bounces has been established[13]. Due to the ordering of stable and unstable manifolds in the minimal domain (a reduced Poincaré section), a method to extract UPOs corresponding to short symbolic strings is proposed. A one to one correspondence between UPOs and symbolic sequences is shown under the system symmetry decomposition[14]. Although we only focused on the case of zero scaled energy, in our numerical experiences, the methods can be still used at the scaled energy $\pm 0.1$. In this paper, for the DKP, we calculate rotation numbers of invariant manifolds around UPOs and set up a connection between rotation numbers and symbolic sequences. Using symbolic dynamics, we reduce stable manifolds to construct a topological template, which preserves the topological relation of UPOs. ## 2 Model system and Poincaré section The Hamiltonian of a hydrogenic electron (with zero angular momentum) in a uniform magnetic field ${\bf B}$ directed along the $z$-axis is given by $H=(1/2m)(p_{\rho}^{2}+p_{z}^{2})-e^{2}/(\rho^{2}+z^{2})^{1/2}+\frac{1}{2}m\omega^{2}\rho^{2},$ (1) where $\omega=eB/2mc$ is half the cyclotron frequency. Converting to atomic units and transforming the cylindrical coordinates to semiparabolic ones, the Hamiltonian becomes $h=\frac{p_{\mu}^{2}}{2}+\frac{p_{\nu}^{2}}{2}-\epsilon(\mu^{2}+\nu^{2})+\frac{1}{8}\mu^{2}\nu^{2}(\mu^{2}+\nu^{2})\equiv 2,$ (2) where $\epsilon=E\gamma^{-2/3}$ is the scaled energy depending on the energy $E$ and the dimensionless field strength parameter $\gamma=2\omega$. The symmetry group of $h$ consists of the identity $e$, two reflections $\sigma_{\mu}$, $\sigma_{\nu}$ across the $\mu$, $\nu$ axes, two diagonal reflections $\sigma_{13}$, $\sigma_{24}$, and three rotations $C_{4}$, $C_{2}$ and $C_{4}^{3}$ by $\pi/2$, $\pi$ and $3\pi/2$ around the center, respectively[15]. In the following, the symmetries $C_{4}$, $C_{2}$ and $C_{4}^{3}$ are denoted by $\rho$, $\pi$ and $\bar{\rho}$, respectively. The time-reversal symmetry is denoted by $T$. Figure 1 displays an orbit and boundary of the transformed potential for $\epsilon=0$. A Poincaré section is chosen as follows. Imagine that the $\mu$ and $\nu$ axes are both of a finite width and length. A counter-clockwise contour is taken along the perimeter of the area forming by the two crossing imaginary rectangles. The Poincaré section is then obtained by recording the position and the tangent component of the momentum along the contour, i.e., the Birkhoff canonical coordinates[16] at intersecting points with the contour where an orbit enter the inside of the contour. The length of the contour is infinite. It is more convenient to transform the contour to one with a finite length. For example, in the first quadrant, the transformations $s=-\mu/(1+\mu)$ along the positive $\mu$ axis and $s=\nu/(1+\nu)$ along the positive $\nu$ axis convert the segment of the original contour in the first quadrant to interval of length 2 parametrized with $s\in[-1,1)$. The variable corresponding to the momentum is taken as $v=-p_{\mu}/p$ at the positive $\mu$ axis, and $v=p_{\nu}/p$ at the positive $\nu$ axis, where $p=\sqrt{p_{\mu}^{2}+p_{\nu}^{2}}$. In this way we may parametrize the whole contour with $s\in[-1,7)$ and define corresponding $v$. The rotational symmetry under $\rho$, $\pi$ and $\bar{\rho}$ in the original configurational space becomes the translational symmetry of shifting $s$ by a multiple of 2 in the $s-v$ plane. The dynamics on the Poincaré surface is then represented by a map on the annulus $s\in[-1,7)$ and $v\in[-1,1]$, which is taken as a fundamental domain (FD). When consider an image and preimage of the FD, we need extend to its lifted space. The partial image and preimage of the FD in the lifted space is given in Fig. 2. For example, in Fig. 2(a), zones 1 (2) and 1’ (2’) in the strip 1 (2) are mapped forward into zones +3 (+0) and +2 (+1), respectively. In the same way, the backward mapping of the strip 1 is given in Fig. 2(b). Since the rotational symmetry of the Hamiltonian (2) corresponds to the translational symmetry in lifted space, the annulus ($s\in[-1,7)$, $v\in[-1,1]$) on the Poincaré section can be reduced to a domain ($s\in[0,2)$, $v\in[-1,1]$). In the conservative system, classical dynamics preserves an invariant volume in the phase space under constricting, stretching and folding. This behavior can be displayed in the Poincaré section. In Fig. 3(a), we draw the 9 lines ($s\in(0,1)$, $v\in[-1,1]$) in the reduced domain (RD) and their forward mapping in the lifted space. In the mapping, the original zone in the RD is stretched and folded, as well as wrapped. In order to display ordering of the lines in the forward mapping, we also plot connecting lines between two different strips in the lifted space. In the lifted space ($s\in(2,3)$, $v\in[-1,1]$), the ordering of lines in the top-left part preserves the same as the original one and the ordering in the bottom-right part is in reverse. In another lifted space ($s\in(1,2)$, $v\in[-1,1]$), the ordering of lines in the whole part is in reverse. It is clear that the wrapping of lines in the forward mapping is clockwise, if the lines in the strip ($s\in(1,2)$, $v\in[-1,1]$) are stuck to those in the strip ($s\in(2,3)$, $v\in[-1,1]$) in terms of their ordering. In the same way, in Fig. 3(b), the similar result can be obtained from the 9 lines ($s\in(1,2)$, $v\in[-1,1]$) in the RD and their forward mapping in the lifted space. So, the forward mapping illustrates the rotation of RD in the clockwise direction. ## 3 Stable and unstable invariant manifolds In general, the invariant manifolds as a subset are contained in manifolds. The method for calculating stable and unstable manifolds (dynamical foliations) used in through the rest of our works is detailed in [2]. Using the same method, we can thus generate the stable and unstable invariant manifolds through unstable periodic points in a two-dimensional Poincaré section. Here we present a short introduction of the method. (i) Unstable manifolds: Taking a circle around $n$ steps backward mapping ($x_{-n}$, $y_{-n}$) of an unstable periodic point ($x_{0}$, $y_{0}$), we get an ellipse centered at ($x_{0}$, $y_{0}$) after the same steps forward mapping. Its long axis points to the most stretching direction. When we fix the point ($x_{0}$, $y_{0}$) and increase $n$, the ellipse is stretched and rotated, as well as its most stretching direction changes slightly. When $n\rightarrow\infty$, the most stretching direction approaches a limit. This direction is the most stretching direction of the point ($x_{0}$, $y_{0}$). After going a short distance along the direction, we get a new point ($x_{0}$, $y_{0}$). Repeating the above process, we can get the new most stretching direction. Finally, an unstable invariant manifold is generated by connecting the points ($x_{0}$, $y_{0}$). In fact, the deformation of ellipse is closely related to the dynamical matrix of Poincaré mapping. (ii) Stable manifolds: Taking a circle around $n$ steps forward mapping ($x_{n}$, $y_{n}$) of an unstable periodic point ($x_{0}$, $y_{0}$), we get an ellipse centered at ($x_{0}$, $y_{0}$) after the same steps backward mapping. Following the above similar process, we can get the most stable direction of the point ($x_{0}$, $y_{0}$) and a stable invariant manifold. ## 4 Evolution of unstable manifolds in rotated Poincaré sections In order to display the evolution of invariant manifolds around an UPO, we rotate counter-clockwise the ($\mu$, $\nu$) coordinates to the ($\mu_{\phi}$, $\nu_{\phi}$) coordinates with an angle $\phi\in[0,\pi/2]$. By using the same transformations in Section II, the Poincaré map ($s_{\phi}$, $v_{\phi}$) is obtained from the ($\mu_{\phi}$, $\nu_{\phi}$) coordinates and reduced to a domain ($s_{\phi}\in[0,2)$, $v_{\phi}\in[-1,1]$). Since $h$ has $C_{4v}$ and time-reversal symmetries, we take 4 UPOs (4)(5)(14)(15) with different symmetries in the Table I as examples to investigate the evolution of invariant manifolds. Their plots in the configuration space with 10 rotation coordinate axes and periodic points with unstable invariant manifolds in the RDs are drown in Figs. 4(a)-(d), respectively. From the initial point in the $+\nu_{\phi=0}$ coordinate axis corresponding to the point 1 in the RD with $\phi=0$, each UPO goes into the second quadrant as displayed in the configuration space by an arrow. Its time process is recorded in the RDs with 10 rotation angles $\phi$. So, the figure in the RD with $\phi=\pi/2$ is the same as that in the RD with $\phi=0$ besides the first point in the former RD is the second point in the later one. In a periodic process, we will calculate the advanced phase $\theta$ of an unstable manifold in rotation around the periodic orbits to determine rotation number $\theta/2\pi$. The phase $\theta$ is counted positive (negative) when the rotation of unstable manifold around the periodic orbits is counter-clockwise (clockwise). The UPOs with 4 different symmetries are described as follows: (i) The UPO (4) displayed in the configuration space of Fig. 4(a) has $\sigma_{\mu}$ and $T$ symmetries, but not $\rho$, $\pi$, $\bar{\rho}$, $\sigma_{\nu}$, $\sigma_{13}$ and $\sigma_{24}$ symmetries. In the second quadrant of configuration space, the orbit starting from the $+\nu_{\phi=0}$ coordinate axis goes to the $-\mu_{\phi=0}$ ($+\nu_{\phi=\pi/2}$) coordinate axis. It corresponds to that the point 1 moves to 2 in the RD with $\phi=0$, i.e. 1 in the RD with $\phi=\pi/2$. In the process, the orbit passes through the $+\nu_{\phi=\pi/18}$, $+\nu_{\phi=\pi/9}$, $\cdots$, $+\nu_{\phi=4\pi/9}$ coordinate axes, while the point 1 moves in the RDs with $\phi=\pi/18$, $\phi=\pi/9$, $\cdots$, $\phi=4\pi/9$. At the same time, an unstable manifold passing through the point 1 evolves in the RDs. The phase of unstable manifold advances by an angle close to $-\pi$. In the third quadrant of configuration space, the orbit starting from the $-\mu_{\phi=0}$ coordinate axis goes to the $-\nu_{\phi=0}$ ($-\mu_{\phi=\pi/2}$) coordinate axis. It corresponds to that the point 2 moves to 3 in the RD with $\phi=0$, i.e. 2 in the RD with $\phi=\pi/2$. In the process, the orbit passes through the $-\mu_{\phi=\pi/18}$, $-\mu_{\phi=\pi/9}$, $\cdots$, $-\mu_{\phi=4\pi/9}$ coordinate axes, while the point 2 moves in the RDs with $\phi=\pi/18$, $\phi=\pi/9$, $\cdots$, $\phi=4\pi/9$. The phase of unstable manifold advances by an angle close to $-\pi$. The orbit starting from the $-\nu_{\phi=0}$ coordinate axis goes into the fourth quadrant of configuration space and back to the $-\nu_{\phi=0}$ coordinate axis. It corresponds to that the point 3 moves to 4 in the RD with $\phi=0$. In the process, the orbit passes through the $-\nu_{\phi=\pi/18}$ coordinate axis two times, while the point 3 goes into the RD with $\phi=\pi/18$, then leaps to the point 4 and comes back the RD with $\phi=0$. In the moving processes of the point 3 from the RD with $\phi=0$ into the RD with $\phi=\pi/18$ and of the point 4 from the RD with $\phi=\pi/18$ into the RD with $\phi=0$, the directions of unstable manifolds are almost invariant. However, in the leaping process of the orbit from the point 3 to 4 in the RD with $\phi=\pi/18$, an unstable direction indicated as the arrow from the periodic point to its neighboring point on the unstable manifold is approximately reversed. The phase of unstable manifold advances by an angle close to $-\pi$. In the third quadrant of configuration space, the orbit starting from the $-\nu_{\phi=0}$ ($-\mu_{\phi=\pi/2}$) coordinate axis goes to the $-\mu_{\phi=0}$ coordinate axis. It corresponds to that the point 4 moves to 5 in the RD with $\phi=0$, i.e., the point 3 in the RD with $\phi=\pi/2$ moves to 5 in the RD with $\phi=0$. In the process, the orbit passes through the $-\mu_{\phi=4\pi/9}$, $-\mu_{\phi=7\pi/18}$, $\cdots$, $-\mu_{\phi=\pi/18}$ coordinate axes, while the point 3 moves in the RDs with $\phi=4\pi/9$, $\phi=7\pi/18$, $\cdots$, $\phi=\pi/9$ and then the point 5 moves in the RD with $\phi=\pi/18$. Since the $-\nu_{\phi=\pi/18}$ coordinate axis intersects the orbit in the fourth quadrant, two points 3 and 4 are added in the RD with $\phi=\pi/18$. The phase of unstable manifold advances by an angle close to $-\pi$. In the second quadrant of configuration space, the orbit starting from the $-\mu_{\phi=0}$ ($+\nu_{\phi=\pi/2}$) coordinate axis goes to the $+\nu_{\phi=0}$ coordinate axis. It corresponds to that the point 5 moves to 6 in the RD with $\phi=0$, i.e., the point 4 in the RD with $\phi=\pi/2$ moves to 6 in the RD with $\phi=0$. In the process, the orbit passes through the $+\nu_{\phi=4\pi/9}$, $+\nu_{\phi=7\pi/18}$, $\cdots$, $+\nu_{\phi=\pi/18}$ coordinate axes, while the point 4 moves in the RDs with $\phi=4\pi/9$, $\phi=7\pi/18$, $\cdots$, $\phi=\pi/9$ and then the point 6 moves in the RD with $\phi=\pi/18$. The phase of unstable manifold advances by an angle close to $-\pi$. The orbit starting from the $+\nu_{\phi=0}$ ($+\mu_{\phi=\pi/2}$) coordinate axis goes into the first quadrant of configuration space and back to the $+\nu_{\phi=0}$ ($+\mu_{\phi=\pi/2}$) coordinate axis. It corresponds to that the point 6 moves to 1 in the RD with $\phi=0$, i.e., the point 5 moves to 6 in the RD with $\phi=\pi/2$. In the process, the orbit passes through the $+\mu_{\phi=4\pi/9}$ coordinate axis two times, while the point 5 goes into the RD with $\phi=4\pi/9$, then leaps to 6 and comes back the RD with $\phi=\pi/2$. In the moving processes of the point 5 from the RD with $\phi=\pi/2$ into the RD with $\phi=4\pi/9$ and of the point 6 from the RD with $\phi=4\pi/9$ into the RD with $\phi=\pi/2$, the directions of unstable manifolds are almost invariant. However, in the leaping process from the point 5 to 6 in the RD with $\phi=4\pi/9$, an unstable direction is approximately reversed. The phase of unstable manifold advances by an angle close to $-\pi$. So, in the periodic process, the unstable manifold returns to its original position, as well as the phase of unstable manifold advances by -6$\pi$. The rotation number of UPO (4) is -3. (ii) The UPO (5) displayed in the configuration space of Fig. 4(b) passes through the origin. Its right limit orbit has $\sigma_{13}$ symmetry, but not $\rho$, $\pi$, $\bar{\rho}$, $T$, $\sigma_{\mu}$, $\sigma_{\nu}$ and $\sigma_{24}$ symmetries. Similarly, in the periodic process, the phase of unstable manifold advances by -6$\pi$. The rotation number of UPO (5) is -3. (iii) The UPO (14) displayed in the configuration space of Fig. 4(c) has $\sigma_{\nu}$ symmetry, but not $\rho$, $\pi$, $\bar{\rho}$, $T$, $\sigma_{\mu}$, $\sigma_{13}$ and $\sigma_{24}$ symmetries. Similarly, in the periodic process, the phase of unstable manifold advances by -8$\pi$. The rotation number of UPO (14) is -4. (iv) The UPO (15) displayed in the configuration space of Fig. 4(d) has $\rho$, $\pi$, $\bar{\rho}$ symmetries, but not $T$, $\sigma_{\mu}$, $\sigma_{\nu}$, $\sigma_{13}$ and $\sigma_{24}$ symmetries. Similarly, in the periodic process, the phase of unstable manifold advances by -16$\pi$. The rotation number of UPO (15) is -8. Thus, in the rotation of unstable invariant manifolds around UPOs, we have determined rotation numbers. At the same time, we have also obtained that the advanced phase $|\theta|$ of unstable invariant manifold in a Poincaré mapping does not exceed $\pi$. In the Sect. V, we will present a method to calculate the rotation numbers in a Poincaré section. ## 5 Rotation of unstable manifolds in a Poincaré section According to the natural ordering in the lifted space and the occurrence of tangencies of manifolds, we have the region partition in the RD with symbols ($L_{0}$, $R_{0}$, $R_{1}$, $R_{2}$ and $L_{2}$) and the ordering for forward sequences[13] $\bullet L_{0}<\bullet R_{0}<\bullet R_{1}<\bullet R_{2}<\bullet L_{2}.$ (3) The forward mapping preserves the ordering in regions of $\bullet L_{0}$ and $\bullet L_{2}$, but reverses the ordering in regions of $\bullet R_{0}$, $\bullet R_{1}$ and $\bullet R_{2}$. In the RD, some symmetries of the Hamiltonian (2) are reduced, it can be reflected by the relation of orbit periods to sequence ones. So, we can firstly calculate the advanced phases of unstable directions in rotation around UPOs in the RD and then add the contribution of symmetries to determine rotation numbers. The 4 UPOs with different symmetries in the Sect. IV are still taken as examples. (i) In Fig. 5(a), we draw the periodic points encoded by $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ and stable and unstable invariant manifolds passing through the points. The periodic points are denoted by circles. In order to illustrate the evolution of unstable direction around the periodic points, we take another initial point near the periodic point 1 on the unstable invariant manifold. The forward mapping of the point is also drawn in the figure and denoted by crosses. The arrows from periodic points to their neighboring points on the unstable invariant manifolds display unstable directions. In the forward mapping from the periodic point 1 to 2, the symbolic sequence $\bullet R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ is shifted to $\bullet R_{0}R_{1}R_{2}^{2}R_{1}R_{0}$ and the original unstable direction is approximately reversed. Since the rotation is clockwise, we obtain $-\pi$ rotation of the unstable direction. In the forward mapping from the periodic point 2 to 3, the symbolic sequence $\bullet R_{0}R_{1}R_{2}^{2}R_{1}R_{0}$ is shifted to $\bullet R_{1}R_{2}^{2}R_{1}R_{0}^{2}$ and the unstable direction is approximately reversed. We also obtain $-\pi$ rotation of the unstable direction, i.e. $-2\pi$ rotation of the original unstable direction. In the same way, $-\pi$ rotation of the unstable direction is obtained in the forward mapping from the periodic point 3 to 4. In the periodic point 4, we take another neighboring point denoted by a triangle to replace the point denoted by a cross. In the same way, $3\times(-\pi)$ rotation of the unstable direction is obtained in the forward mapping from the periodic point 4 to 5, from the periodic point 5 to 6 and from the periodic point 6 to 1. Thus, during the mapping in the sequence period, the original unstable direction goes back and the total advance of phase is $-6\pi$. Since the orbit period is equal to the sequence one, i.e., the UPO (4) has not the $\rho$, $\pi$, $\bar{\rho}$ symmetries, the rotation number of UPO encoded by $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ is -3. (ii) In Fig. 5(b), the periodic points with the right limit encoded by $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ and the stable and unstable invariant manifolds passing through the points are drawn. In the forward mapping from the periodic point 1 to 2, the symbolic sequence $\bullet L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ is shifted to $\bullet R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}L_{0}$ and the original unstable direction is approximately preserved. In the forward mapping from the periodic point 5 to 6, the same result is obtained. In other forward mappings, $6\times(-\pi)$ rotation of the unstable direction is added. Thus, during the mapping in the sequence period, the original unstable direction goes back and the total advance of phase is $-6\pi$. Since the orbit period is equal to the sequence one, the rotation number of UPO (5) encoded by $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ is -3. (iii)In Fig. 5(c), the periodic points encoded by $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ and stable and unstable invariant manifolds passing through the points are drawn. In the forward mapping from the periodic point 1 to 2, the symbolic sequence $\bullet L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ is shifted to $\bullet R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}L_{0}$ and the original unstable direction is approximately preserved. In the forward mapping from the periodic point 6 to 7, the same result is obtained. In other forward mappings, $8\times(-\pi)$ rotation of the unstable direction is added. Thus, during the mapping in the sequence period, the original unstable direction goes back and the total advance of phase is $-8\pi$. Since the orbit period is equal to the sequence one, the rotation number of UPO (14) encoded by $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ is -4. (iv) In Fig. 5(d), the periodic points encoded by $L_{0}R_{1}^{2}R_{0}^{2}$ and stable and unstable invariant manifolds passing through the points are drawn. In the forward mapping from the periodic point 1 to 2, the symbolic sequence $\bullet L_{0}R_{1}^{2}R_{0}^{2}$ is shifted to $\bullet R_{1}^{2}R_{0}^{2}L_{0}$ and the original unstable direction is approximately preserved. In other forward mappings, $4\times(-\pi)$ rotation of the unstable direction is added. Thus, during the mapping in the sequence period, the original unstable direction goes back and the total advance of phase is $-4\pi$. Since the orbit period is 4 times of the sequence one, i.e., the UPO (15) has the $\rho$, $\pi$, $\bar{\rho}$ symmetries, the rotation number of the UPO encoded by $L_{0}R_{1}^{2}R_{0}^{2}$ is -8. In the above examples describing the rotation of unstable directions around periodic points, the forward map corresponding to the shift with $L_{0}$ or $L_{2}$ ($R_{0}$ or $R_{1}$ or $R_{2}$) approximately preserves (reserves) the original unstable direction. So, we can multiply the numbers of $R_{0}$, $R_{1}$ and $R_{2}$ in 5-letter symbolic sequences by one half of the ratios of orbit periods to sequence ones to determine rotation numbers of UPOs. Since the RD has the $\pi$-rotation symmetry, the 5-letter symbolic dynamics can be reduced to the 3-letter one in the minimal domain (MD) ($s\in[0,1)$, $v\in[-1,1]$)[13]. The MD is partitioned and denoted by symbols $L_{0}$, $R_{0}$ and $R_{1}$. The correspondence of 5-letter symbolic sequences with 3-letter ones is $L_{0}\rightarrow L_{0}$, $R_{0}\rightarrow R_{0}$, $R_{1}\rightarrow R_{1}$, $R_{2}\rightarrow R_{0}$ and $L_{2}\rightarrow L_{0}$. In general, the number of $R_{0}$, $R_{1}$ and $R_{2}$ in 5-letter symbolic sequences is twice of the number of $R_{0}$ and $R_{1}$ in 3-letter ones. The ratios of orbit periods to sequence ones for the former are one half of those for the later. Of course, the simple repeating of 3-letter symbolic sequences in 5-letter ones will be removed. For example, the 3-letter symbolic sequences $R_{0}^{2}R_{1}$ and $L_{0}R_{0}^{2}R_{1}$ correspond to the 5-letter ones $R_{0}(R_{2})R_{0}(R_{2})R_{1}R_{2}(R_{0})R_{2}(R_{0})R_{1}$ and $L_{0}(L_{2})R_{0}(R_{2})R_{0}(R_{2})R_{1}L_{2}(L_{0})R_{2}(R_{0})R_{2}(R_{0})R_{1}$, respectively. Rotation numbers of two UPOs can be determined by calculating total numbers of letters $R_{0}$ and $R_{1}$ in 3-letter symbolic sequences and multiplying them by one half of the ratios of orbit periods to sequence ones. The same rotation numbers of the UPOs can be also obtained by using the method for 5-letter symbolic sequences. Thus, using the method, we extract rotation numbers of 38 UPOs from symbolic sequences as given in Table I. ## 6 Topological Template After the region partition and symbolic ordering are introduced, the families of stable and unstable manifolds constitute curve coordinates in the RD. Each stable (unstable) manifold has the same forward (backward) symbolic sequence. The ordering on stable (unstable) manifolds is described by that of forward (backward) symbolic sequences[14]. In Fig. 6, two families of sub-manifolds divided by the partition line $\bullet C_{0}$ or $\bullet C_{2}$ have the opposite ordering. The ordering of stable (unstable) manifolds increases monotonically from the left-bottom (left-top) to right-top (right-bottom) along each unstable (stable) manifold. Along each stable manifold in zones $\bullet L_{0}$, $\bullet R_{0}$ and $\bullet R_{1}$ of Fig. 6, we constrict all points in the curve to a point. The point preserves the forward symbolic sequence and the ordering of stable manifold. So, in the left region of Fig. 6, the points in three zones are reduced to three lines. Connecting the three lines, we obtain a belt partitioned by the symbols $\bullet C_{0}$ and $\bullet B_{0}$, and denoted by the symbols $\bullet L_{0}$, $\bullet R_{0}$ and $\bullet R_{1}$ as given in the top of Fig. 7(a). From the left to right along the belt, the ordering of forward sequences increases monotonically. The forward mapping of the left region ($s\in(0,1)$ and $v\in[-1,1]$) in Fig. 6, i.e. the right region ($s\in(2,3)$, $v\in[-1,1]$) in Fig. 3(a), can be reduced in the RD. So, in the forward mapping, the zones $\bullet L_{0}$ and $\bullet R_{0}$ still keep in the region ($s\in(0,1)$, $v\in[-1,1]$) encoded by $L_{0}\bullet$ and $R_{0}\bullet$, respectively, but the zone $\bullet R_{1}$ moves in the region ($s\in(1,2)$, $v\in[-1,1]$) encoded by $R_{1}\bullet$. In Fig. 6, we again partition the RD and encode it by corresponding backward symbols. The two zones $L_{0}\bullet$ and $R_{0}\bullet$ are partitioned to five zones by the lines $\bullet B_{0}$ and $\bullet C_{0}$, as well as the zone $R_{1}\bullet$ is partitioned to three zones by the lines $\bullet B_{2}$ and $\bullet C_{2}$. In the stretching and folding processes of forward mapping, the original three zones are mapped into the eight zones. We still constrict all points along stable manifolds in each zone. Thus, the points in eight zones are reduced to five lines in the left region and three lines in the right region. Connecting the eight lines, we obtain a belt partitioned by the symbols $L_{0}\bullet B_{0}$, $C_{0}\bullet R_{1}$, $R_{0}\bullet B_{0}$, $R_{0}\bullet C_{0}$, $R_{1}\bullet C_{2}$ and $R_{1}\bullet B_{2}$, and encoded by the strings $L_{0}\bullet R_{0}$, $L_{0}\bullet R_{1}$, $R_{0}\bullet R_{1}$, $R_{0}\bullet R_{0}$, $R_{0}\bullet L_{0}$, $R_{1}\bullet L_{2}$, $R_{1}\bullet R_{2}$ and $R_{1}\bullet R_{1}$ in the bottom of Fig. 7(a). The ordering of the bottom belt is the same as that of the top one. So, the forward mapping of the region ($s\in(0,1),v\in[-1,1]$) can be described by a twisting part of topological template as given in Fig. 7(a). In the same way, along each stable manifold in zones $\bullet R_{1}$, $\bullet R_{2}$ and $\bullet L_{2}$ of Fig. 6, all points in the curve are constricted to a point. A belt containing the point is partitioned by the symbols $\bullet B_{2}$ and $\bullet L_{2}$, and encoded by the symbols $\bullet R_{1}$, $\bullet R_{2}$ and $\bullet L_{2}$ as given in the top of Fig. 7(b). From the left to right along the belt, the ordering of forward sequences increases monotonically. The forward mapping of the right region ($s\in(1,2)$ and $v\in[-1,1]$) in Fig. 6, i.e. the left region ($s\in(-1,0)$, $v\in[-1,1]$) in Fig. 3(b), can be reduced in the RD. So, in the forward mapping, the zones $\bullet L_{2}$ and $\bullet R_{2}$ still keep in the region ($s\in(1,2)$, $v\in[-1,1]$) encoded by $L_{2}\bullet$ and $R_{2}\bullet$, but the zone $\bullet R_{1}$ moves in the region ($s\in(0,1)$, $v\in[-1,1]$) encoded by $R_{1}\bullet$. We still constrict all points along stable manifolds in each zone and obtain the belt partitioned by the symbols $R_{1}\bullet B_{0}$, $R_{1}\bullet C_{0}$, $R_{2}\bullet C_{2}$, $R_{2}\bullet B_{2}$, $C_{2}\bullet R_{1}$ and $L_{2}\bullet B_{2}$, and encoded by the strings $R_{1}\bullet R_{1}$, $R_{1}\bullet R_{0}$, $R_{1}\bullet L_{0}$, $R_{2}\bullet L_{2}$, $R_{2}\bullet R_{2}$, $R_{2}\bullet R_{1}$, $L_{2}\bullet R_{1}$ and $L_{2}\bullet R_{2}$ in the bottom of Fig. 7(b). So, the forward mapping of the region ($s\in(1,2)$, $v\in[-1,1]$) can be described by a twisting part of topological template as given in Fig. 7(b). In the two twisting parts of topological template, the belts reflect approximately the direction of unstable manifolds. Using the twisting parts of topological templates, we can easily calculate rotation numbers for given symbolic sequences. In the same way, the 4 UPOs with different symmetries in Sect. II are still taken as examples. For the sequence $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ encoding the UPO(4), firstly, $R_{0}\bullet R_{0}R_{1}R_{2}^{2}R_{1}$ is obtained by shifting $R_{1}\bullet R_{0}^{2}R_{1}R_{2}^{2}$. When an arrow is put on the $\bullet R_{0}$ zone of top belt in Fig. 7(a), after a forward mapping, the arrow is moved on the $R_{0}\bullet R_{0}$ zone of bottom belt in Fig. 7(a). Since the arrow rotates clockwise to its opposite direction, we count the process as -1. Then, $R_{0}\bullet R_{1}R_{2}^{2}R_{1}R_{0}$ is obtained by shifting $R_{0}\bullet R_{0}R_{1}R_{2}^{2}R_{1}$. Since an arrow on the $\bullet R_{0}$ zone of top belt in Fig. 7(a) rotates clockwise to its opposite direction, we count the forward mapping as -1. Repeating the above process, we get the total number -6 counting the forward mapping in sequence period. Since the orbit period is equal to the sequence one, we can thus obtain rotation number of the UPO encoded by $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ is -3. For the sequence $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ encoding the UPO(5), after a forward mapping for $\bullet L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$, an arrow on the top belt in Fig. 7(a) moves parallelly on the zone $L_{0}\bullet R_{0}$ of bottom belt. We count the forward mapping as 0. Following the same process, we get the total number -6 counting the forward mapping in the sequence period. Since the orbit period is equal to the sequence one, we can thus obtain rotation number of the UPO encoded by $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ is -3. Similarly, for the sequence $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ ($L_{0}R_{1}^{2}R_{0}^{2}$) encoding the UPO(14) (UPO(15)), we get the total number -8 (-4) counting the forward mapping in the sequence period. Since the orbit period is equal to (4 times of) the sequence one. we can thus obtain rotation number of the UPO encoded by $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ ($L_{0}R_{1}^{2}R_{0}^{2}$) is -4 (-8). By comparing with the former computation for the 4 UPOs from the definition and in a Poincaré section, the same results are extracted from the topological template. By combining the two twisting parts in Figs. 7(a)(b), suspension of the Poincaré mapping, which displays the relative position of zones in the forward mapping from the top belt to bottom one, is obtained. A global topological template of the RD is constructed by connecting the suspension with a flow corresponding to the Poincaré mapping in Fig. 8. The template preserves the ordering of forward sequences encoding stable manifolds in a belt and the same or inverse ordering of symbolic encoding in the forward mapping of all parts in the belt. ## 7 Conclusion and discussion In summary, we have presented the systematic study of the evolution of invariant manifolds around unstable periodic orbits and the reduction of them to construct a topological template in terms of symbolic dynamics for the diamagnetic Kepler problem. To confirm the topological template, rotation numbers of invariant manifolds around unstable periodic orbits in a phase space, which quantify the evolution, are determined from the definition and connected with symbolic sequences encoding the periodic orbits. Only symbolic codes, which correspond to the forward mapping with inverse ordering, can contribute to the rotation of invariant manifolds. By using symbolic ordering, the reduced Poincaré section is constricted along stable manifolds and a topological template, which preserves the ordering of forward sequences and can be used to extract the rotation numbers, is established. The rotation numbers computed from the topological template are the same as those computed from the original definition. Since unstable periodic orbits in phase space are the skeleton of the chaotic system, the local evolution of manifolds near unstable periodic orbits can present basic features of the global evolution of manifolds in phase space. One of the basic features can be quantified by the rotation number and embedded in the topological template. In the semiclassical Green’s function, the phase correction is related to Maslov indices of the UPOs[17], which has been connected with symbolic sequences of unstable periodic orbits due to boundary coding[18]. The relation of Maslov indices to rotation numbers of unstable periodic orbits remains to be determined. ## 8 REFERENCES [1] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systems and Bifurcation of Vector Fields, Springer, New York, 1983. [2] B.-L. Hao and W.-M. Zheng, Applied Symbolic Dynamics and Chaos, World Scientific, Singapore, 1998. [3] P. Grassberger and H. Kantz, Phys. Lett. A 113 (1985) 235. [4] Y. Gu, Phys. Lett. A 124, (1987) 340. [5] P. Cvitanović, G. H. Gunaratne and I. Procaccia, Phys. Rev. A 38, (1988) 1503. [6] P. Grassberger, H. Kantz and U. Moenig, J. Phys. A 22, (1989) 5217. [7] K. T. Hansen, Phys. Rev. E 52, (1995) 2388. [8] M. E. Johnson, M. S. Jolly, and I. G. Kevrekidis, Numer. Algorithms 74, (1997) 125. [9] M. Dellnitz and A. Hohmann, Numer. Algorithms 75, (1997) 293. [10] B. Krauskopf and H. Osinga, Chaos 9, (1999) 768. [11] H. M. Osinga and B. Krauskopf, Computers and Graphics 26, (2002) 815. [12] H. Friedrich and D. Wintgen, Phys. Repts. 183, (1989) 37. [13] Z.-B. Wu and W.-M. Zheng, Physica Scripta 59, (1999) 266 [$chao- dyn/9907016$]. [14] Z.-B. Wu and J.-Y. Zeng, Physica Scripta 61, (2000) 406 [$nlin/0004004$]. [15] P. Cvitanović and B. Eckhardt, Nonlinearity 6, (1993) 277. [16] G. D. Birkhoff, Acta Mathematica 50, (1927) 359. [17] J. B. Delos, Advan. in Chem. Phys. 65, 161 (1986). [18] B. Eckhardt and D. Wintgen, J. Phys. B 23, 355 (1990). Table I. Rotation numbers and symbolic sequences of UPOs for the diamagnetic Kepler problem at $\epsilon=0$. No 3-lett. Seq. & its Period 5-lett. Seq. & its Period Orb. Period Rot. Num. 1 $R_{0}$ 1 $R_{0}$ 1 4 -2 2 $L_{0}R_{1}$ 2 $L_{0}R_{1}L_{2}R_{1}$ 4 4 -1 3 $R_{0}R_{1}$ 2 $R_{0}R_{1}R_{2}R_{1}$ 4 4 -2 4 $R_{0}^{2}R_{1}$ 3 $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$ 6 6 -3 5 $L_{0}R_{0}^{2}R_{1}$ 4 $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$ 8 8 -3 6 $L_{0}R_{1}R_{0}^{2}$ 4 $L_{0}R_{1}R_{2}^{2}L_{2}R_{1}R_{0}^{2}$ 8 8 -3 7 $L_{0}R_{1}R_{0}R_{1}$ 4 $L_{0}R_{1}R_{2}R_{1}$ 4 4 -1.5 8 $R_{0}^{3}R_{1}$ 4 $R_{0}^{3}R_{1}R_{2}^{3}R_{1}$ 8 8 -4 9 $R_{0}^{2}R_{1}^{2}$ 4 $R_{0}^{2}R_{1}^{2}$ 4 8 -4 10 $L_{0}R_{0}L_{0}R_{1}^{2}$ 5 $L_{0}R_{0}L_{0}R_{1}^{2}$ 5 20 -6 11 $L_{0}R_{0}^{2}R_{1}^{2}$ 5 $L_{0}R_{0}^{2}R_{1}^{2}$ 5 20 -8 12 $L_{0}R_{1}L_{0}R_{1}^{2}$ 5 $L_{0}R_{1}L_{2}R_{1}^{2}L_{2}R_{1}L_{0}R_{1}^{2}$ 10 10 -3 13 $L_{0}R_{1}R_{0}^{2}R_{1}$ 5 $L_{0}R_{1}R_{2}^{2}R_{1}$ 5 20 -8 14 $L_{0}R_{1}R_{0}R_{1}^{2}$ 5 $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$ 10 10 -4 15 $L_{0}R_{1}^{2}R_{0}^{2}$ 5 $L_{0}R_{1}^{2}R_{0}^{2}$ 5 20 -8 16 $L_{0}R_{1}^{2}R_{0}R_{1}$ 5 $L_{0}R_{1}^{2}R_{0}R_{1}L_{2}R_{1}^{2}R_{2}R_{1}$ 10 10 -4 17 $R_{0}^{4}R_{1}$ 5 $R_{0}^{4}R_{1}R_{2}^{4}R_{1}$ 10 10 -5 18 $R_{0}^{3}R_{1}^{2}$ 5 $R_{0}^{3}R_{1}^{2}$ 5 20 -10 19 $R_{0}^{2}R_{1}R_{0}R_{1}$ 5 $R_{0}^{2}R_{1}R_{2}R_{1}$ 5 20 -10 20 $R_{0}^{2}R_{1}^{3}$ 5 $R_{0}^{2}R_{1}^{3}R_{2}^{2}R_{1}^{3}$ 10 10 -5 21 $R_{0}R_{1}R_{0}R_{1}^{2}$ 5 $R_{0}R_{1}R_{2}R_{1}^{2}R_{2}R_{1}R_{0}R_{1}^{2}$ 10 10 -5 22 $L_{0}R_{0}L_{0}R_{1}R_{0}R_{1}$ 6 $L_{0}R_{0}L_{0}R_{1}R_{2}R_{1}$ 6 12 -4 23 $L_{0}R_{0}L_{0}R_{1}^{3}$ 6 $L_{0}R_{0}L_{0}R_{1}^{3}L_{2}R_{2}L_{2}R_{1}^{3}$ 12 12 -4 24 $L_{0}R_{0}^{2}L_{0}R_{1}^{2}$ 6 $L_{0}R_{0}^{2}L_{0}R_{1}^{2}$ 6 6 -2 25 $L_{0}R_{0}^{4}R_{1}$ 6 $L_{0}R_{0}^{4}R_{1}L_{2}R_{2}^{4}R_{1}$ 12 12 -5 26 $L_{0}R_{0}^{3}R_{1}^{2}$ 6 $L_{0}R_{0}^{3}R_{1}^{2}$ 6 6 -2.5 27 $L_{0}R_{0}^{2}R_{1}^{3}$ 6 $L_{0}R_{0}^{2}R_{1}^{3}L_{2}R_{2}^{2}R_{1}^{3}$ 12 12 -5 28 $L_{0}R_{1}L_{0}R_{1}R_{0}R_{1}$ 6 $L_{0}R_{1}L_{2}R_{1}R_{0}R_{1}L_{2}R_{1}L_{0}R_{1}R_{2}R_{1}$ 12 12 -4 29 $L_{0}R_{1}R_{0}^{4}$ 6 $L_{0}R_{1}R_{2}^{4}L_{2}R_{1}R_{0}^{4}$ 12 12 -5 30 $L_{0}R_{1}R_{0}^{3}R_{1}$ 6 $L_{0}R_{1}R_{2}^{3}R_{1}$ 6 12 -5 31 $L_{0}R_{1}R_{0}R_{1}R_{0}R_{1}$ 6 $L_{0}R_{1}R_{2}R_{1}R_{0}R_{1}L_{2}R_{1}R_{0}R_{1}R_{2}R_{1}$ 12 12 -5 32 $L_{0}R_{1}^{2}R_{0}^{3}$ 6 $L_{0}R_{1}^{2}R_{0}^{3}$ 6 6 -2.5 33 $L_{0}R_{1}^{3}R_{0}^{2}$ 6 $L_{0}R_{1}^{3}R_{2}^{2}L_{2}R_{1}^{3}R_{0}^{2}$ 12 12 -5 34 $R_{0}^{5}R_{1}$ 6 $R_{0}^{5}R_{1}R_{2}^{5}R_{1}$ 12 12 -6 35 $R_{0}^{4}R_{1}^{2}$ 6 $R_{0}^{4}R_{1}^{2}$ 6 12 -6 36 $R_{0}^{3}R_{1}R_{0}R_{1}$ 6 $R_{0}^{3}R_{1}R_{2}R_{1}$ 6 12 -6 37 $R_{0}^{3}R_{1}^{3}$ 6 $R_{0}^{3}R_{1}^{3}R_{2}^{3}R_{1}^{3}$ 12 12 -6 38 $R_{0}^{2}R_{1}^{4}$ 6 $R_{0}^{2}R_{1}^{4}$ 6 12 -6 ## 9 FIGURE CAPTION Fig. 1. A typical orbit and boundary of the transformed potential for the diamagnetic Kepler problem at $\epsilon=0$. Fig. 2. A image (a) and preimage (b) of the strips 1 and 2 of the fundamental domain ($s\in[-1,7)$, $v\in[-1,1]$) in the lifted space. Fig. 3. 9 lines in (a) ($s\in(0,1)$, $v\in[-1,1]$) or (b) ($s\in(1,2)$, $v\in[-1,1]$) of the reduced domain and their forward mapping in the correspondent lifted space. The different types of lines display the relative changes between original positions and their forward mappings along the $s$ coordinate axis. Fig. 4. UPOs with different symmetry in the configuration space and periodic points with unstable invariant manifolds in rotation Poincaré sections: (a) the UPO (4); (b) the UPO (5); (c) the UPO (14); (d) the UPO (15). Fig. 5. Periodic points encoded by (a) $R_{0}^{2}R_{1}R_{2}^{2}R_{1}$; (b) $L_{0}R_{0}^{2}R_{1}L_{2}R_{2}^{2}R_{1}$; (c) $L_{0}R_{1}R_{2}R_{1}^{2}L_{2}R_{1}R_{0}R_{1}^{2}$; (d) $L_{0}R_{1}^{2}R_{0}^{2}$ in the 5-letter encoding and their stable and unstable invariant manifolds. Fig. 6. Stable and unstable manifolds with partition lines in the reduced domain. Fig. 7. Twisting parts of topological template describing the forward mapping on the regions (a) ($s\in[0,1)$, $v\in[-1,1]$) and (b) ($s\in[1,2)$, $v\in[-1,1]$). For crossing of two lines in suspension of the Poincaré mapping, the front (back) one is denoted by solid lines (short dashes or the combination of solid lines and short dashes). The top and bottom belts are denoted by solid lines or dashes depending on their positions. The connecting lines of two parts of the broken belt in the forward mapping are denoted by long dashes. Fig. 8. A flow of topological template of the reduced domain. The notions for belts are the same as Fig. 7, except that projection of several parts in the bottom belt on one belt in terms of forward symbolic codings is connected by short dashes. The flow is denoted by short dashes.

arxiv-papers

2024-09-04T02:48:54.362607

0803.2408

# Anisotropic spin transport in two-terminal mesoscopic rings: the Rashba and Dresselhaus spin-orbit interactions Miao Wang Kai Chang kchang@red.semi.ac.cn SKLSM, Institute of Semiconductors, Chinese Academy of Sciences, P. O. Box 912, Beijing 100083, China ###### Abstract We investigate theoretically the spin transport in two-terminal mesoscopic rings in the presence of both the Rashba spin-orbit interaction (RSOI) and the Dresselhaus spin-orbit interaction (DSOI). We find that the interplay between the RSOI and DSOI breaks the original cylindric symmetry of mesoscopic ring and consequently leads to the anisotropic spin transport, i.e., the conductance is sensitive to the positions of the incoming and outgoing leads. The anisotropic spin transport can survive even in the presence of disorder caused by impurity elastic scattering in a realistic system. ###### pacs: 73.23.-b ## I Introduction In recent years, the spin-orbit interaction (SOI) in low-dimensional semiconductor structures has attracted considerable attention because of its potential application in all-electrical controlled spintronic devices. wolf ; Tsitsishvili There are two types of SOI in conventional semiconductors. One is the Rashba spin-orbit interaction(RSOI) induced by structure inversion asymmetry, Rashba ; Bychkov and the other is the Dresselhaus spin-orbit interaction(DSOI) induced by bulk inversion asymmetry Dresselhaus . The strength of the RSOI can be tuned by external gate voltages or asymmetric doping. In thin quantum wells, the strength of the DSOI is comparable to that of the RSOI. Lommer The interplay between the RSOI and DSOI leads to a significant change in the transport property. There are a few works on the effects of the competition between these two types of SOI on the transport properties of 2DEG, Ganichev ; Chang ; Yang especially in mesoscopic rings Vasilo . The circular photogalvanic effect can be used to separate the contribution of the RSOI and DSOI, and the relative strengths of the RSOI and DSOI can be extracted from the photocurrent. Ganichev The RSOI and DSOI can interfere in such a way that the spin dependent features disappear even though the individual SOI is still strong, e.g., vanishing spin splitting in the presence of the equal-strength RSOI and DSOI. Ganichev This cancellation results in extremely long spin relaxation time in specific crystallographic directions, and the disappearance of the beating pattern in SdH oscillation. Yang Recently, advanced growth techniques have made it possible to fabricate high quality semiconductor rings, Fuhrer which have attracted considerable attention due to the intriguing quantum interference phenomenon arising from their unique topological geometry. The Aharonov-Bohm (AB) and the Aharonov- Casher (AC) effects are typical examples of quantum mechanical phase interference, which have been demonstrated experimentally Tonomura ; Kong and theoretically Balatsky on semiconductor rings. The quantum transport properties through semiconductor ring structures with the RSOI alone have attracted considerable interest. Molnar ; SoumaBK ; Entin ; Cao ; Foldi ; Nitta SOIs in semiconductors behave like an in-plane momentum-dependent magnetic field and lead to a lifting of spin degeneracy of energy bands. This effective magnetic field induces a wave phase difference between the upper arm and lower arm, resulting in the oscillation of the conductance. wolf ; Nitta ; Diego Therefore, the conductance oscillates with increasing the strength of the RSOI.Molnar ; SoumaBK The ring subjected to the DSOI alone shows the exact same oscillation, since the Hamiltonian of the RSOI alone is mathematically equivalent to that of the DSOI alone by a unitary transformation. Sheng The interplay between the RSOI and DSOI results in a periodic potential in an isolated ring, producing the gap in the energy spectrum, suppressing the persistent currents, Sheng and breaking the cylindrical symmetry of mesoscopic rings. This interesting feature leads to the anisotropic spin transport and could be detected using the transport property in an open two-terminal mesoscopic ring. This anisotropic spin transport is a new result, is dominant difference between our work and the previous studies, Molnar ; SoumaBK ; Entin ; Cao ; Foldi and should be important for the potential application of spintronic devices. In this paper, we investigate theoretically the spin transport in two-terminal mesoscopic rings in the presence of both the RSOI and DSOI. We find that the interplay between the RSOI and DSOI leads to a significant change in the transmission, the localization of electrons, and the spin polarization of the current. This interplay weakens and smoothens the oscillation of the conductance, and breaks the original cylindrical symmetry, leading to the anisotropic spin transport. The paper is organized as follows, in Sec. II, we present the theoretical model and formulation. The numerical results and discussions are given in Sec. III. Finally, the conclusion is given in Sec. IV ## II THEORETICAL MODEL Figure 1: Schematic diagram of a 1D semiconductor mesoscopic ring with two leads. Electrons are injected from the left lead, pass through the ring, and exit from the right lead. SOI only exists in the ring. A semiconductor mesoscopic ring (see Fig. 1) in the presence of the RSOI and DSOI can be described by the single-particle effective mass Hamiltonian $\displaystyle\hat{H}$ $\displaystyle=\frac{-\hbar^{2}k^{2}}{2m^{\ast}}+\alpha(\sigma_{x}k_{y}-\sigma_{y}k_{x}){}$ $\displaystyle+\beta(\sigma_{x}k_{x}-\sigma_{y}k_{y})+V(r),$ (1) where the $x$ axis is along the [100] direction, $k=-i\nabla$ is the electron wave vector, $m^{\ast}$ is the electron effective mass, $\sigma_{i}(i=x,y,z)$ are the Pauli matrices, $\alpha$ is the strength of the RSOI, and $\beta$ is the strength of the DSOI. $V(r)$ is the radial confining potential, which is neglected hereafter since we consider that electrons only occupy the lowest subband in a ring with narrow width. The one-dimensional Hamiltonian of a ring in a dimensionless form in lattice representation is Souma $\displaystyle\hat{H}_{ring}$ $\displaystyle=\sum^{N}_{n=1}\sum_{\sigma=\uparrow,\downarrow}\varepsilon_{n}\hat{c}^{{\dagger}}_{n,\sigma}\hat{c}_{n,\sigma}$ $\displaystyle-\sum^{N}_{n=1}\sum_{\sigma,\sigma^{{}^{\prime}}=\uparrow,\downarrow}[t^{n,n+1;\sigma,\sigma^{{}^{\prime}}}_{\phi}\hat{c}^{{\dagger}}_{n;\sigma}\hat{c}_{n+1;\sigma^{{}^{\prime}}}+h.c.],$ (2) where the hopping energies are given in the $2\times 2$ matrix form as: $\displaystyle t^{n,n+1}_{\phi}$ $\displaystyle=t\hat{I}_{s}-i\frac{\alpha}{2a}(cos\phi_{n,n+1}\sigma_{x}+sin\phi_{n,n+1}\sigma_{y})$ $\displaystyle-i\frac{\beta}{2a}(cos\phi_{n,n+1}\sigma_{y}-sin\phi_{n,n+1}\sigma_{x}),$ (3) where $\phi$ is the angular coordinate and $\varepsilon_{n}$ is the on-site potential energy. The operator $\hat{c}_{n,\sigma}(\hat{c}^{{\dagger}}_{n,\sigma})$ annihilates (creates) a spin $\sigma$ electron at the site $n$ of the ring. $\phi_{n,n+1}$ is the angle between the $n$-th site and the $n+1$-th site. $t=\hbar^{2}/2m^{\ast}a^{2}$, with $a$ being the lattice spacing constant, is the nearest-neighbor hopping term in the lead. The spin-resolved conductance of a two-terminal device can be obtained by using the Landauer-Büttiker’s formula BK : $\mathbf{G}=\left(\begin{array}[]{cc}G_{\uparrow\uparrow}&G_{\uparrow\downarrow}\\\ G_{\downarrow\uparrow}&G_{\downarrow\downarrow}\end{array}\right)=\frac{e^{2}}{h}\sum_{p,p^{{}^{\prime}}=1}^{M}\left(\begin{array}[]{cc}|\mathbf{t}_{pp^{{}^{\prime}},\uparrow\uparrow}|^{2}&|\mathbf{t}_{pp^{{}^{\prime}},\uparrow\downarrow}|^{2}\\\ |\mathbf{t}_{pp^{{}^{\prime}},\downarrow\uparrow}|^{2}&|\mathbf{t}_{pp^{{}^{\prime}},\downarrow\downarrow}|^{2}\end{array}\right),$ (4) where $M$ is the number of conducting channels, the transmission matrix elements $\mathbf{t}=2\sqrt{-\text{Im}\sum^{r}_{L}\otimes I_{s}}\cdot G^{r}_{1N}\cdot\sqrt{-\text{Im}\sum^{r}_{R}\otimes I_{s}}$ and $|\mathbf{t}_{nn^{{}^{\prime}},\sigma\sigma^{{}^{\prime}}}|^{2}$ represents the probability for a spin-$\sigma$ electron incoming from the left lead in the orbital state $|n\rangle$ to appear as a spin-$\sigma^{{}^{\prime}}$ electron in the orbital channel $|n^{{}^{\prime}}\rangle$ in the right lead. We can calculate the conductance from lead $p$ to lead $q$ by using the Fisher-Lee relation Fisher . The detailed formula can be found in the Ref. Green, : $\mathbf{G}^{R}=[EI-H_{c}-\Sigma^{R}]^{-1},$ (5) $\overline{\mathbf{T}}_{pq}=Tr[\Gamma_{p}G^{R}\Gamma_{q}G^{A}],$ (6) where $H_{c}$ is the Hamiltonian of the 1D isolated ring. $\Gamma_{p}(i,j)=\sum_{m}\chi_{m}(p_{i})\frac{\hbar v_{m}}{a}\chi_{m}(p_{j})$ describes the coupling of the ring conductor to the leads. We assume the RSOI and DSOI only exist in the ring, and are absent in the leads. The self-energy $\Sigma^{R}=\sum_{p=1,2}\Sigma_{p}^{R}$, where $\Sigma_{p}^{R}(i,j)=t^{2}g_{p}^{R}(p_{i},p_{j})$, describes the effect of the external leads on the ring. The Green’s function between two points along the leads is given by $g_{p}^{R}(p_{i},p_{j})=-\frac{1}{t}\sum_{m}\chi_{m}(p_{i})exp[ik_{m}a]\chi_{m}(p_{j})$. The function $\chi_{m}(p_{i})$ describes the $m$-th mode in lead $i$. In this paper, we take $a$ as the length unit and $E_{0}=\hbar^{2}/2m^{\ast}a^{2}$ as the energy unit. The local density of electron states is Green : $\rho(r,E)=\frac{1}{2\pi}A(r,r;E)=-\frac{1}{\pi}\mathbf{Im}[G^{R}(r,r;E)],$ (7) where $A\equiv i[G^{R}-G^{A}]$ is the spectral function, which can also be written: $\displaystyle\rho(r,E)$ $\displaystyle\sim\sum_{n}\frac{1}{2\pi}\frac{\gamma_{n}\psi_{n}(r)\phi^{*}_{n}(r)}{(E-\varepsilon_{n0}+\Delta_{n})^{2}+(\gamma_{n}/2)^{2}}$ $\displaystyle\rightarrow\sum_{n}\delta(E-\varepsilon_{n0})|\psi_{n}(r)|^{2}\quad\mathbf{as}\quad\gamma_{n}\rightarrow 0,$ (8) where $\hbar/2\gamma_{n}$ represents the lifetime of an electron remaining in state $n$ before it escapes into the leads, $\varepsilon_{n0}$ is the eigenenergy of the isolated conductor, and $\psi$ ($\phi$) is the eigenstates of the effective Hamiltonian [$H_{c}+\Sigma^{R}$] ([$H_{c}+\Sigma^{A}$]) Green . ## III RESULTS AND DISCUSSIONS ### III.1 1D ring with both RSOI and DSOI Many previous works investigating the spin transport through a 1D ring account only for the RSOI. Molnar The RSOI behaves like an effective in-plane momentum-dependent magnetic field. This effective magnetic field induces a phase difference between the electrons traveling clockwise and counterclockwise along the ring’s upper and lower arms. Therefore, the conductance of a 1D ring in the presence of the RSOI oscillates quasi- periodically with changing the strength of the RSOI and the Fermi energy $E_{F}$. We study the transport through a mesoscopic ring in the presence of both the RSOI and DSOI. First, we consider the ballistic transport through the mesoscopic ring in the presence of the RSOI(DSOI) alone. In Fig. 2, we plot the conductance through a 1D ring as a function of the strength of the RSOI $Q_{r}$. This figure shows that the conductances are exactly same when the right lead is located at symmetric positions, e.g., $\phi=\pm\frac{1}{4}\pi,\pm\frac{1}{2}\pi$, and $\pm\frac{3}{4}\pi$. The RSOI or DSOI alone in the ring does not break the cylindrical symmetry and the transport is still isotropic when the outgoing leads are located at symmetric positions with respect to the $x$-axis (see the dashed lines in the insets of Fig. 2). The quantum interference between the alternation paths, the spin-up or spin-down clockwise and anticlockwise, is responsible for the oscillation of the conductance. Figure 2: The conductance through a 1D ring in the presence of the RSOI or DSOI alone as a function of the strength of the RSOI $Q_{r}\equiv\alpha N/2ta\pi$, $E_{F}$=-0.1, and the outgoing lead is located at $\pm\frac{1}{2}\pi,\pm\frac{1}{4}\pi,\pm\frac{3}{4}\pi$, respectively (see the insets). Figure 3: The conductance through a 1D ring in the presence of the RSOI and DSOI as a function of the strength of the RSOI and DSOI, $Q_{d}\equiv\beta N/2ta\pi$, $Q_{r}=Q_{d}$, $E_{F}$=-0.1. The outgoing lead is located at $\pm\frac{1}{2}\pi,\pm\frac{1}{4}\pi,\pm\frac{3}{4}\pi$ respectively. When the 1D mesoscopic ring is subjected to both the RSOI and DSOI, as shown in Fig. 3, the conductances become asymmetric when the outgoing lead is located at symmetric positions, e.g., $\phi=\pm\frac{1}{4}\pi,\pm\frac{1}{2}\pi$, and $\pm\frac{3}{4}\pi$. The anisotropy of the conductance is induced by the interplay of the RSOI and DSOI, which leads to a periodic potential $\frac{\alpha\beta}{2}\sin{2\phi}$. Sheng The height of the periodic potential is determined by the product of the strengths of the RSOI and DSOI, and the periodicity of the potential is fixed at $\pi$. The potential exhibits barriers at $\phi=\frac{1}{4}\pi,-\frac{3}{4}\pi$, and the valleys at $\phi=-\frac{1}{4}\pi,\frac{3}{4}\pi$. Thus, the conductance displays asymmetric features for the symmetric positions of the outgoing leads. If the incoming lead locates at $\phi=\frac{3}{4}\pi$ (see Fig. 4), we find the transmission becomes symmetric for the outgoing lead locating at the symmetric positions respect to the new incoming lead. In Fig. 4, we plot the conductance of a 1D ring with the incoming lead located at $\phi=\frac{3}{4}\pi$. The conductance becomes symmetric again with respect to the straight line $\phi=\frac{3}{4}\pi$ and $\phi=-\frac{1}{4}\pi$ (the dashed lines in the insets of Fig. 4). The periodic potential $\frac{\alpha\beta}{2}\sin{2\phi}$ induced by the interplay between the RSOI and DSOI Sheng results in the maxima at $\phi=\frac{1}{4}\pi,-\frac{3}{4}\pi$, and the minima at $\phi=\frac{3}{4}\pi,-\frac{1}{4}\pi$. Figure 4: Same as Fig. 3, but the incoming lead is located at $\phi=\frac{3}{4}\pi$, and the outgoing lead is located at $\phi=0,\pm\frac{1}{2}\pi,\frac{1}{4}\pi,-\frac{3}{4}\pi,\pi$, respectively. In order to describe the magnitude of the anisotropy of the conductance induced by the interplay of the RSOI and DSOI, we define the ratio $\eta$ as: $\eta(\phi,-\phi)=\frac{G_{\phi}-G_{-\phi}}{(G_{\phi}+G_{-\phi})/2}\ ,$ (9) where $G_{\pm\phi}$ is the conductance when the right lead is located at the positions with an angle $\pm\phi$ with respect to the $x$ axis. In Fig. 5, we plot $\eta(\pi/4,-\pi/4)$ as a function of the strength of the RSOI and DSOI when the left lead is located at the position of $\phi=\pi$. $\eta$ oscillates with the changing strength of the RSOI and DSOI. The maximum of the anisotropy of the conductance can approach $20\%$. Figure 5: (Color online) The ratio $\eta$ as a function of the strength of the RSOI $Q_{r}$ and DSOI $Q_{d}$, when $E_{F}=-0.1$. The incoming lead is located at $\phi=\pi$, and the outgoing lead is located at $\phi=\pi/4,-\pi/4$, respectively. This anisotropic transport can be interpreted as follows. The interplay between the RSOI and DSOI leads to an effective periodic potential $\frac{\alpha\beta}{2}\sin{2\phi}$. Sheng The potential height is related to the strength of the RSOI and DSOI. $\eta(\phi,-\phi)=0$ when the ring subjected to the DSOI alone because the periodic potential $\frac{\alpha\beta}{2}\sin{2\phi}$ disappears when $\alpha=0$. This effective periodic potential exhibits the maxima at $\phi=\frac{1}{4}\pi$, and $-\frac{3}{4}\pi$, and the minima at $\phi=-\frac{1}{4}\pi$, and $\frac{3}{4}\pi$. Therefore, the interplay between RSOI and DSOI breaks the cylindrical symmetry of the ring (see Fig. 7). In order to clarify the effect of the invasive role of the lead on the anisotropy of the spin transport, we consider different strengths between the ring and leads (as shown in Fig. 6). We find that the conductance decreases with decreasing the coupling strength, but the anisotropy ratios are almost same as before. We believe that the anisotropic spin transport property is caused by the interplay between the Rashba and Dresselhaus spin-orbit interactions. Figure 6: (Color online) The conductance of 1D ring as a function of the strength of equal RSOI and DSOI, when $E_{F}=-0.1$ for different coupling strengths $t_{0}=1,0.6,0.4$. Fig. 7 describes how the conductance varies with the variation of the strengths of the RSOI and DSOI. The conductance oscillates quasiperiodically as the strengths of the RSOI and DSOI increase, and is symmetric with respect to the straight line $\alpha=\beta$, since the Hamiltonian of the RSOI and that of the DSOI are equivalent and can be transferred by the $SU(2)$ unitary transformation. The contribution from the RSOI and DSOI to the spin splitting of electrons cancel each other, Sheng which results in the disappearance of the oscillation along $\alpha=\beta$. This feature provides a possible way to detect the strength of the DSOI since the strength of the RSOI can be tuned by the external electric fields. Figure 7: (Color online) The conductance of a 1D ring as a function of the strength of the RSOI $Q_{r}$ and DSOI $Q_{d}$, when $E_{F}=-0.1$. The incoming lead is located at $\phi=\pi$, while the outgoing lead is located at $\phi=0$. Below, we demonstrate that the interplay between the RSOI and DSOI also results in the variation of the local density of electrons in the ring. In Fig. 8, we plot the local density of electrons in the ring from Eq. 7 with and without the SOI. Fig. 8(a) and (b), shows that the local density of electrons shows slow and very rapid oscillations. The fast oscillation comes from the contribution of each site of the lattice, while the slow variation of the envelope corresponds to the bound (quasibound) states in the isolated (open) ring. This feature is analogous to the situation of the effective mass theory, where the electron wave function can be expressed as the product of two parts: the band-edge Bloch function and the slow varying envelope function. The former denotes the contribution from the atomic wave function, and the latter describes the bound (quasibound) state from the external potential, e.g., the quantum well potential. Similar results can be found in Ref. Li, . There is only a slight difference between the local densities of electron states with and without the RSOI, but a significant change in the presence of both the RSOI and DSOI (see Fig. 8(c)). The local density of electrons exhibit maxima at $\phi=-\frac{1}{4}\pi,\frac{3}{4}\pi$. This characteristic is also caused by the periodic potential induced by the interplay between the RSOI and DSOI. The positions of $\phi=\frac{1}{4}\pi,-\frac{3}{4}\pi$ ($\phi=-\frac{1}{4}\pi,\frac{3}{4}\pi$) correspond to a potential barrier (well), where the local density of electron states is smaller (larger). The interplay between the RSOI and DSOI induces periodic potential and breaks the original cylindrical symmetry of the ring, consequently changing the local density of electron states. Figure 8: The local density of electrons along the ring $\phi$ when $E_{F}=0.1$ (a) without the RSOI and DSOI; (b) with the RSOI alone; (c) with equal RSOI and DSOI ($Q_{r}=Q_{d}=11.3$). The above analysis assumes perfectly clean 1D systems, in which there is no elastic or inelastic scattering at $T=0$. In a realistic system, there will be many impurities in the sample. Disorder could be incorporated by the fluctuation of the on-site energies, which distribute randomly within the range width $w[\varepsilon_{n}\rightarrow\varepsilon_{n}+w_{n}$ with $-w/2<w_{n}<w/2]$. In Fig. 9(a), we plot the conductance as a function of Fermi energy $E_{F}$ without RSOI. The ratio $\eta(\frac{1}{4}\pi,-\frac{1}{4}\pi)$ is negligible for (weak and strong) different disorders $w=0.1,0.3$ when the system is without the RSOI. Fig. 9(b) plots the conductance of a 1D ring as a function of the strength of RSOI and DSOI, when $Q_{r}=Q_{d}$, for the various random widths $w=0.1,0.3,1$ ($w=1$ for inset). It can be clearly seen that the disorder-averaged conductance for the strong disorder case ($w=1$) shows almost the same anisotropy as that for the weak disorder case ($w=0.1,0.3$). (see Fig. 9(b)) Figure 9: (Color online)(a) The conductance of a 1D ring as a function of Fermi energy $E_{F}$ without SOIs for random width $w=0.1,0.3$; (b) The conductance and $\eta$ of a 1D ring as a function of the strength of the RSOI, for outgoing lead located at $\phi=\frac{1}{4}\pi,-\frac{1}{4}\pi$, and $Q_{r}=Q_{d}$, $E_{F}=0.1$, $w=0.1,0.3$. The inset shows the conductance and the anisotropic ratio $\eta$ when $w=1$. While the anisotropy of the 1D ring becomes significant as the strengths of the RSOI and DSOI increase, random disorder increases the scattering of the ring, and decreases conductance compared to that of a clean 1D ring. The anisotropic spin transport can still survive even in the presence of weak and strong disorder. ### III.2 The spin polarization of current The spin polarization vector of current $\boldsymbol{P}=(P_{x},P_{y},P_{z})$ can be evaluated as follows Souma ; BK : $\boldsymbol{P}^{\sigma}=Tr_{s}[\hat{\rho}^{\sigma}\mathbf{\hat{\sigma}}],$ (10) where the density matrix is given by: $\hat{\rho}^{\sigma}=\frac{e^{2}/h}{G^{\uparrow\sigma}+G^{\downarrow\sigma}}\sum_{p,p^{{}^{\prime}}=1}^{M}\left(\begin{array}[]{cc}|\mathbf{t}_{pp^{{}^{\prime}},\uparrow\sigma}|^{2}&\mathbf{t}_{pp^{{}^{\prime}},\uparrow\sigma}\mathbf{t}^{*}_{pp^{{}^{\prime}},\downarrow\sigma}\\\ \mathbf{t}_{pp^{{}^{\prime}},\downarrow\sigma}\mathbf{t}^{*}_{pp^{{}^{\prime}},\uparrow\sigma}&|\mathbf{t}_{pp^{{}^{\prime}},\downarrow\sigma}|^{2}\end{array}\right),$ (11) where $Tr_{s}$ denotes the trace in the spin Hilbert space. Then, the spin polarized vector $\boldsymbol{P}$ is Souma : $\displaystyle P^{\sigma}_{x}$ $\displaystyle=$ $\displaystyle\frac{G^{\uparrow\sigma}-G^{\downarrow\sigma}}{G^{\uparrow\sigma}+G^{\downarrow\sigma}},$ (12) $\displaystyle P^{\sigma}_{y}$ $\displaystyle=$ $\displaystyle\frac{2e^{2}/h}{G^{\uparrow\sigma}+G^{\downarrow\sigma}}\sum_{p,p^{{}^{\prime}}=1}^{M}\mathbf{Re}[\mathbf{t}_{pp^{{}^{\prime}},\uparrow\sigma}\mathbf{t}^{*}_{pp^{{}^{\prime}},\downarrow\sigma}],$ (13) $\displaystyle P^{\sigma}_{z}$ $\displaystyle=$ $\displaystyle\frac{2e^{2}/h}{G^{\uparrow\sigma}+G^{\downarrow\sigma}}\sum_{p,p^{{}^{\prime}}=1}^{M}\mathbf{Im}[\mathbf{t}_{pp^{{}^{\prime}},\uparrow\sigma}\mathbf{t}^{*}_{pp^{{}^{\prime}},\downarrow\sigma}],$ (14) where the $x$-axis is chosen as the spin-quantized axis, $\hat{\sigma}_{x}|\uparrow\rangle=+|\uparrow\rangle$ and $\hat{\sigma}_{x}|\downarrow\rangle=-|\downarrow\rangle$, so that Pauli spin matrix has the following form: $\hat{\sigma}_{x}=\left(\begin{array}[]{cc}1&0\\\ 0&-1\end{array}\right),\ \hat{\sigma}_{y}=\left(\begin{array}[]{cc}0&1\\\ 1&0\end{array}\right),\ \hat{\sigma}_{z}=\left(\begin{array}[]{cc}0&i\\\ -i&0\end{array}\right).$ (15) For the spin polarized injection, i.e., $P_{x}=1$, the magnitude of the spin polarization $P$ in the outgoing lead will not change, i.e., $|P|=1$ since there is no other orbit channel to interact with the spin.Zhai Fig. 10 depicts the current spin polarization $P_{i}(i=x,y,z)$ of a 1D ring as a function of the strength of the RSOI $Q_{r}$ and the positions of the outgoing lead. The RSOI behaves like an effective in-plane momentum-dependent magnetic field, and the fully spin-up polarized current in the incoming lead will be changed to the spin-down current in the outgoing lead at large RSOI. The three components of the outgoing polarization vector also show cylindrical symmetry for the RSOI or DSOI alone, since the RSOI or DSOI alone does not break the cylindrical symmetry of a 1D ring. The spin polarization $P_{x}$ decreases rapidly from $P_{x}=1$ to $P_{x}\approx-1$ as the strength of the RSOI increases when the outgoing lead is located at the position near $\phi=0$, while the spin polarization $P_{y}$ and $P_{z}$ oscillate and decrease to zero. When the outgoing lead locates away from the $x$-axis, i.e.,$\phi=0$, $P_{y}$ and $P_{z}$ oscillate quickly with increasing $Q_{r}$. Figure 10: (Color online) The contour plot of the spin polarization of current as a function of the strength of the RSOI $Q_{r}$ alone and the position of the right lead in the absence of the DSOI, $E_{F}=-0.1$, $Q_{d}=0$. (a) for $P_{x}$; (b) for $P_{y}$; (c) for $P_{z}$. The spin- quantized axis is the $x$-axis. In Fig. 11, we show how the spin polarizations $P_{i}(i=x,y,z)$ vary with the strength of the SOIs and the position of the outgoing lead $\phi$ in the presence of equal-strength RSOI and DSOI, i.e., $Q_{r}=Q_{d}$. All three components $P_{x}$, $P_{y}$, and $P_{z}$ oscillate regularly as the strengths of the RSOI and DSOI increase, and show significant anisotropy of spin polarization with respect to the position of the outgoing lead. This feature can also be understood from the interplay between the effective periodic potential induced by the SOIs and the quantum interference. For a fixed strength of the SOI, the asymmetric characteristic of the polarization $\mathbf{P}$ as a function of the angle $\phi$ arises from the cylinder symmetry breaking induced by the effective potential $\frac{\alpha\beta}{2}\sin{2\phi}$. The quantum interference between the spin -up and -down electrons traveling clockwise and/or counterclockwise along the ring’s upper and lower arms leads to the oscillation of the polarization $\mathbf{P}$ as a function of the strengths of the SOIs at a fixed angle $\phi$. Compared to Fig. 10, the spin polarization $P_{x}$ will decrease to $0$ instead of $-1$ as the strengths of the SOIs increase. This is because the DSOI behaves like a twisted in-plane magnetic field, while the effective magnetic field induced by the RSOI always points along the radial of the ring. Figure 11: (Color online) The same as Fig. 10, but includes the DSOI. ## IV Conclusion We investigate theoretically the spin transport through a two-terminal mesoscopic ring in the presence of both the RSOI and DSOI. We find that the interplay between the RSOI and DSOI leads to the anisotropic transport through a two-terminal cylindrical mesoscopic ring, i.e., breaks the cylindrical symmetry. This interesting feature arises from the periodic potential along the ring caused by the interplay between the RSOI and DSOI. This interplay also results in a significant variation in electron density and the spin polarization of current. The anisotropy of the spin transport through the mesoscopic ring induced by the interplay between the RSOI and DSOI can survive even in the presence of the disorder effect. Furthermore, the anisotropy of the spin transport should play an important role in the potential application of all-electrical spintronic devices. ###### Acknowledgements. This work is partly supported by NSFC Grant No. 62525405 and the knowledge innovation project of CAS. ## References * (1) S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Molnár, M. L. Roukes, A. Y. Chtchelkanova, and D. M. Treger, Science 294, 1488 (2001). * (2) E. Tsitsishvili, G. S. Lozano, and A. O. Gogolin, Phys. Rev. B 70, 115316 (2004). * (3) E. I. Rashba, Sov. Phys. Solid State 2, 1109 (1960). * (4) Y. A. Bychkov and E. I. Rashba, J. Phys. C 17, 6039 (1984). * (5) G. Dresselhaus, Phys. Rev. 100, 580 (1955). * (6) G. Lommer, F. Malcher, and U. Rössler, Phys. Rev. Lett. 60, 728 (1988). * (7) S. D. Ganichev, V. V. Bel’kov, L. E. Golub, E. L. Ivchenko, P. Schneider, S. Giglberger, J. Eroms, J. De Boeck, G. Borghs, W. Wegscheider, D. Weiss, and W. Prettl, Phys. Rev. Lett. 92, 256601 (2004). * (8) M. C. Chang, Phys. Rev. B 71, 085315 (2005). * (9) W. Yang and K. Chang, Phys. Rev. B 73, 045303 (2006). * (10) X. F. Wang and P. Vasilopoulos, Phys. Rev. B 72, 165336 (2005). * (11) A. Fuhrer, S. Lüescher, T. Ihn, T. Heinzel, K. Ensslin, W. Wegscheider, and M. Bichler, Nature 413, 822 (2001). * (12) A. Tonomura, N. Osakabe, T. Matsuda, T. Kawasaki, J. Endo, S. Yano, and H. Yamada, Phys. Rev. Lett. 56, 792 (1986). * (13) M. König, A. Tschetschetkin, E. M. Hankiewicz, J. Sinova, V. Hock, V. Daumer, M. Schäfer, C. R. Becker, H. Buhmann, and L. W. Molenkamp, Phys. Rev. Lett. 96, 076804 (2006). * (14) A. V. Balatsky and B. L. Altshuler. Phys. Rev. Lett. 70, 001678 (1993). * (15) B. Molnár, F. M. Peeters, and P. Vasilopoulos, Phys. Rev. B 69, 155335 (2004). * (16) S. Souma and B. K. Nikolić, Phys. Rev. Lett. 94, 106602 (2005). * (17) O. Entin-Wohlman, Y. Gefen, Y. Meir, and Y. Oreg, Phys. Rev. B 45, 11890 (1992). * (18) B. H. Wu and J. C. Cao, Phys. Rev. B 74, 115313 (2006). * (19) P. Földi, B. Molnár, M. G. Benedict, and F. M. Peeters, Phys. Rev. B 71, 033309 (2005). * (20) J. Nitta, F. E. Meijer, and H. Takayanagi, Appl. Phys. Lett. 75, 695 (1999). * (21) D. Frustaglia and K. Richter, Phys. Rev. B 69, 235310 (2004). * (22) J. S. Sheng and K. Chang, Phys. Rev. B 74, 235315 (2006). * (23) S. Souma and B. K. Nikolić, Phys. Rev. B 70, 195346 (2004). * (24) B. K. Nikolić and S. Souma, Phys. Rev. B 71, 195328 (2005). * (25) D. S. Fisher and P. A. Lee, Phys. Rev. B 23, 6851 (1981). * (26) S. Datta, Electronic transport in mesoscopic systems _(Cambridge University Press, New York, 1997)._ * (27) L. Yang, M. L. Cohen, and S. G. Louie, Nano. Lett. 7, 3112 (2007). * (28) F. Zhai and H. Q. Xu, Phys. Rev. Lett. 94, 246601 (2005).

arxiv-papers

2024-09-04T02:48:54.367788

0803.2566

Convergence properties of fixed-point search with general but equal phase shifts for any number of iterations 111The paper was supported by NSFC(Grants No. 60433050 and 60673034), the basic research fund of Tsinghua university NO: JC2003043. Dafa Lia222email address:dli@math.tsinghua.edu.cn, Xiangrong Lib, Hongtao Huangc, Xinxin Lid a Department of mathematical sciences, Tsinghua University, Beijing 100084 CHINA b Department of Mathematics, University of California, Irvine, CA 92697-3875, USA c Electrical Engineering and Computer Science Department University of Michigan, Ann Arbor, MI 48109, USA d Department of computer science, Wayne State University, Detroit, MI 48202, USA The correspondence Author Dafa Li, Phone Number is (8610)62773561 Fax No. is (8610) 62785847 Abstract Grover presented the fixed-point search by replacing the selective inversions by selective phase shifts of $\pi/3$. In this paper, we investigate the convergence behavior of the fixed-point search algorithm with general but equal phase shifts for any number of iterations. PACS number: 03.67.Lx Keywords: Amplitude amplification, the fixed-point search, quantum computing. ## 1 Introduction Grover’s quantum search algorithm is used to find a target state in an unsorted database of size $N$[1][2]. The Grover’s quantum search algorithm can be considered as a rotation of the state vectors in two-dimensional Hilbert space generated by the start ($s$) and target ($t$) vectors[2]. The amplitude of the target state increases monotonically towards its maximum and decreases monotonically after reaching the maximum [3]. This search algorithm is called the amplitude amplification algorithm. For the size $N=2^{n}$ of the database, quantum search algorithm requires $O(\sqrt{N})$ steps to find the target state. As mentioned in [4] [5], unless we stop when it is right at the target state, it will drift away. A fixed-point search algorithm was presented in [4] to avoid drifting away from the target state. The fixed-point search algorithm obtained by replacing the selective inversions by selective phase shifts of $\pi/3$, converges to the target state irrespective of the number of iterations. The main advantage of the fixed-point search with equal phase shifts of $\pi/3$ is that it performs well for small but unknown initial error probability and the fixed-point behavior leads to robust quantum search algorithms [4]. However, the target state is the limit state when the number of iterations tends to the infinite. For readability, we introduce the fixed-point search algorithm as follows. In [4] the transformation $UR_{s}^{\pi/3}U^{+}R_{t}^{\pi/3}U$, where $U$ is any unitary operator, was applied to the start state $|s\rangle$, $\displaystyle R_{s}^{\pi/3}$ $\displaystyle=$ $\displaystyle I-[1-e^{i\frac{\pi}{3}}]|s\rangle\langle s|,$ $\displaystyle R_{t}^{\pi/3}$ $\displaystyle=$ $\displaystyle I-[1-e^{i\frac{\pi}{3}}]|t\rangle\langle t|,$ (1) where $|t\rangle$ stands for the target state. The transformation $UR_{s}^{\pi/3}U^{+}R_{t}^{\pi/3}U$ is denoted as Grover’s the Phase-$\pi/3$ search algorithm in [6]. Let us consider the fixed-point search algorithm with general but equal phase shifts as follows. $\displaystyle R_{s}^{\theta}$ $\displaystyle=$ $\displaystyle I-[1-e^{i\theta}]|s\rangle\langle s|,$ $\displaystyle R_{t}^{\theta}$ $\displaystyle=$ $\displaystyle I-[1-e^{i\theta}]|t\rangle\langle t|.$ (2) The transformation $UR_{s}^{\theta}U^{+}R_{t}^{\theta}U$ was called as the Phase-$\theta$ search algorithm and studied in [7]. It is enough to let $\theta$ be in $[0,\pi]$. Note that if we apply $U$ to the start state $|s\rangle$, then the amplitude of reaching the target state $|t\rangle$ is $U_{ts}$[2], where $\left|\left|U_{ts}\right|\right|^{2}=1-\epsilon$. As indicated in [2], in the case of database search, $|U_{ts}|$ is almost $1/\sqrt{N}$, where $N$ is the size of the database. Thus, $\epsilon$ is almost $1-1/N$ and $\epsilon$ is close to $1$ for the large size of database. Apply the operations $U$, $R_{s}^{\theta}$, $U^{+}$, $R_{t}^{\theta}$, and $U$ to the start $|s\rangle$ and let $D(\theta)$ be the deviation of the state $UR_{s}^{\theta}U^{+}R_{t}^{\theta}U|s\rangle$ from the $t$ state for any phase shifts of $\theta$. The deviation $D(\theta)$ was reduced in [7] and is rewritten as follows. $D(\theta)=4(1-\cos\theta)^{2}\epsilon(\epsilon-d)^{2},$ (3) where $d=\frac{1-2\cos\theta}{2(1-\cos\theta)}$. It was shown that $D(\theta)$ is between $0$ and $1$ in [7]. For the Phase-$\pi/3$ search algorithm, $D(\pi/3)=\epsilon^{3}$[4]. In [8], we explored the performance of the fixed-point search with general but different phase shifts for one iteration. In [7], we discussed the performance of the fixed-point search with general but equal phase shifts for one iteration. In this paper, we investigate convergence behavior of the fixed-point search with general but equal phase shifts for any number of iterations. It is useful for designing fixed-point search algorithms for different choices of the phase shift parameter $\theta$. The following results are established in Section 2. (1). The fixed-point search with equal phase shifts of $\theta\leq\pi/2$ converges to the target state. (2). The fixed-point search with equal phase shifts of $\theta$, where $\pi/2<\theta\leq\arccos(-1/4)$, converges the target state with the probability of at least $80\%.$ (3). The fixed-point search with equal phase shifts of $\theta$, where $\arccos(-1/4)<\theta\leq 2\pi/3$, converges the target state with the probability of among $66.6\%$ and $80\%.$ (4). The fixed-point search with equal phase shifts of $\theta$, where $2\pi/3<\theta\leq\pi$, does not converge. In section 3, we analyze the convergence rate for different values of $\theta$. It is demonstrated that the Phase-$\pi/3$ is not always optimal and the convergence rate can be improved by choosing $\theta>\pi/3$. In section 4, we show that for the size $N=2^{n}$ of the database, $O(n)$ iterations of the Phase-$\theta$ search can find the target state. However, as indicated in [4], $O(n)$ iterations of the Phase-$\theta$ search involve the exponential queries. ## 2 Convergence performance of the Phase-$\theta$ search for any number of iterations Let $\epsilon_{0}=\epsilon$ and $0<\epsilon<1$. Then, from Eq. (3) one can obtain the following iteration equation $\epsilon_{m+1}=4(1-\cos\theta)^{2}\epsilon_{m}(\epsilon_{m}-d)^{2}.$ (4) In this section, we discuss the convergence behavior of the Phase-$\theta$ search for any number of iterations. For the Phase-$\pi/3$ search, after recursive application of the basic iteration for $m$ times, the failure probability $\epsilon_{m}=$ $\epsilon^{3^{m}}$ and the success probability $\left|U_{m,ts}\right|=1-\epsilon^{3^{m}}$[4]. The $\epsilon_{m}$ in Eq. (4) is the failure probability of the Phase-$\theta$ search algorithm after $m$ iterations. From inference [11] in [7], Eq. (4) has the following fixed-points: $0$, $1$ ($\theta\neq 0$), $a$, where $a=\cos\theta/(\cos\theta-1)$ ($\theta\neq 0$). In other words, if the sequence $\\{\epsilon_{m}\\}$ in Eq. (4) has a limit then the limit must be $0$, $1$ or $a$. Clearly $a<d$. To study the convergence performance for any number of iterations, we need the following results which are listed in the following paragraphs (A), (B), and (C). (A). From Eq. (4), we obtain the following, $\epsilon_{m}-\epsilon_{m-1}=4\epsilon_{m-1}\left(\cos\theta-1\right)^{2}\left(1-\epsilon_{m-1}\right)(a-\epsilon_{m-1}).$ (5) Eqs. (4) and (5) imply the following convergence property. Property 1. (1.1) If $\epsilon_{m}=d$, then $\epsilon_{m+l}=0$, for any $l>0$. (1.2). if $\epsilon_{m-1}>a$ and $\epsilon_{m-1}\neq 0$, $\epsilon_{m}<\epsilon_{m-1}$; (1.3). If $\epsilon_{m-1}<a$ and $\epsilon_{m-1}\neq 0$, $\epsilon_{m}>\epsilon_{m-1}$. (B). When $\pi/2\leq\theta\leq\pi$, we have the following equation. $\epsilon_{m}-a=4(\cos\theta-1)^{2}(\epsilon_{m-1}-a)(\epsilon_{m-1}-b)(\epsilon_{m-1}-c),$ (6) where $b=\frac{1}{2}-\frac{\sqrt{-\cos\theta(2-\cos\theta)}}{2(1-\cos\theta)}$, and $c=\frac{1}{2}+\frac{\sqrt{-\cos\theta(2-\cos\theta)}}{2(1-\cos\theta)}$. When $\epsilon_{i}=a$, $b$ or $c$, $\epsilon_{i+l}=a$ for any $l>0$. Note that $b<d$, $a<d$, and $d<c$. (C). Let $f(x)=4(1-\cos\theta)^{2}x(x-d)^{2}.$ (7) Then, the derivative of $f(x)$ is $f^{\prime}(x)=12(1-\cos\theta)^{2}(x-d)(x-d/3).$ (8) From Eq. (8), (1). $f^{\prime}(x)=0$ at $r$ and $d$, where $r=d/3$; (2).$\ f^{\prime}(x)<0$ when $r<x<d$; (3). $f^{\prime}(x)>0$ when $x<r$ or $x>d$; (4). When $\theta>\pi/3$, $f(x)$ has a relative maximum $g=\frac{2(1-2\cos\theta)^{3}}{27(1-\cos\theta)}$ at $r$ and a relative minimum $0$ at $d$. ### 2.1 When $0<\theta\leq\pi/2$, for any $\epsilon_{0}\in(0,1)$, the Phase-$\theta$ search converges to the target state. Note that $0$ is an attractive fixed-point when $0<\theta<\pi/2$ and $0$ is also a semi-attractive fixed-point when $\theta=\pi/2$. See inference [11] in [7]. (1). $0<\theta\leq\pi/3$ For this case, $d\leq 0$ and $a<0$. In Eq. (4), $d=0$ means $\epsilon_{m+1}=\epsilon_{m}^{3}$, which is Grover’s Phase-$\pi/3$ search. From $d<0$ and Eq. (4), $\epsilon_{m}>0$. By property (1.2), always $\epsilon_{m}<\epsilon_{m-1}$ when $0<\theta\leq\pi/3$. That is, the sequence $\\{\epsilon_{m}\\}$ in Eq. (4) decreases monotonically.Therefore, for any $\epsilon_{0}$ in $(0$, $1)$ $\lim_{m\rightarrow\infty}\epsilon_{m}=0$. (2). $\pi/3<\theta\leq\pi/2$ For this case, $a\leq 0$, $0<d\leq 1/2$. Hence, from Eq. (4) $0\leq\epsilon_{i}<1$. By property (1.2), always $\epsilon_{m}\leq\epsilon_{m-1}$ when $\pi/3<\theta\leq\pi/2$. That is, the sequence $\\{\epsilon_{m}\\}$ in Eq. (4) decreases. Factually, the sequence $\\{\epsilon_{m}\\}$ in Eq. (4) decreases monotonically and $\epsilon_{m}>0$, or is of the form $\epsilon_{0}>\epsilon_{1}>...>\epsilon_{k}=0$ and $\epsilon_{l}=0$ for any $l>k$. Therefore, for any $\epsilon_{0}$ in $(0$, $1)$ $\lim_{m\rightarrow\infty}\epsilon_{m}=0$. Example 1. For the Phase-$\pi/2$ search, $\epsilon_{m}=\epsilon_{m-1}(2\epsilon_{m-1}-1)^{2}$. Let $\epsilon_{0}=0.99999$. See Fig. 1. $\epsilon_{1}=\allowbreak 0.999\,95$, $\ \ \epsilon_{2}=\allowbreak 0.999\,75$, $\epsilon_{3}=\allowbreak 0.998\,75$, $\ \ \epsilon_{4}=\allowbreak 0.993\,76$, $\epsilon_{5}=\allowbreak 0.969\,11$, $\ \ \epsilon_{6}=\allowbreak 0.853\,07$, $\epsilon_{7}=\allowbreak 0.425\,37$, $\ \ \epsilon_{8}=\allowbreak 9.\,\allowbreak 476\,6\times 10^{-3}$. ### 2.2 When $\pi/2<\theta<\arccos(-1/4)$, the Phase-$\theta$ search converges the target state with the success probability of $(1-a)>80\%$. For the Phase-$\theta$ search, $a<g<r<b<d<c$. Note that $a$ is an attractive fixed-point. See inference [11] in [7]. From Eq. (6), we have the following property. Property 2. (2.1). $a<\epsilon_{m}\leq g$ whenever $a<\epsilon_{m-1}<b$; (2.2). $0\leq\epsilon_{m}<a$ whenever $b<\epsilon_{m-1}<c$ or $\epsilon_{m-1}<a$. The convergence region of the Phase-$\theta$ search (A). When $\epsilon_{0}\in(0,c]$ and $\epsilon_{0}\neq d$, the deviation from the target state converges to the fixed-point $a$. There are four cases. The argument is the following. Case 1. When $\epsilon_{0}=a$ or $b$ or $c$, it is trivial by Eq. (6). Case 2. $\epsilon_{0}<a$. By property (2.2), $0<\epsilon_{m}<a$ for any $m$. By property 1, the sequence $\\{\epsilon_{m}\\}$ increases monotonically. Hence, the sequence $\\{\epsilon_{m}\\}$ converges to $a$ from below. Case 3. $a<\epsilon_{0}<b$. By property (2.1), always $a<\epsilon_{m}\leq g$ for any $m>0$, and by property 1, the sequence $\\{\epsilon_{m}\\}$ decreases monotonically. Hence, the sequence $\\{\epsilon_{m}\\}$ converges to $a$ from above. Case 4. $b<\epsilon_{0}<c$ and $\epsilon_{0}\neq d$. By property (2.2), $0<\epsilon_{1}<a$. Then, it turns to case 2. Conclusively, when $\epsilon_{0}\in(0,c]$ and $\epsilon_{0}\neq d$, from the above four cases, $\epsilon_{m}\neq d$, hence $\lim_{m\rightarrow\infty}\epsilon_{m}=a$. (B). When $\epsilon_{0}\in(c,1)$, the deviation from the target state converges to the fixed-points $a$ or $0$. By property (1.2), $\epsilon_{0}>...>\epsilon_{j^{\ast}-1}(>c)>\epsilon_{j^{\ast}\text{ }}(\leq c)$. If $\epsilon_{j^{\ast}\text{ }}=d$, then $\epsilon_{m}=0$ for any $m>j^{\ast}$. Otherwise, $\lim_{m\rightarrow\infty}\epsilon_{m}=a$ by the above (A). ### 2.3 Phase-$\arccos(-1/4)$ search converges the target state with the success probability of $80\%$. For the Phase-$\arccos(-1/4)$ search, $a=1/5$ is an attractive fixed-point, see inference [11] in [7]. $b=a=1/5$, $d=3/5$, and $c=4/5$. The iteration equation is $\epsilon_{m}=\epsilon_{m-1}(\allowbreak 5\epsilon_{m-1}-3)^{2}/4$. Eq. (6) becomes the following. $\epsilon_{m}-1/5=\allowbreak\frac{25}{4}\left(\epsilon_{m-1}-4/5\right)\left(\epsilon_{m-1}-1/5\right)^{2}.$ (9) From Eq. (9) we have the following property. Property 3. (3.1). $\epsilon_{m}<1/5$ when $\epsilon_{m-1}<4/5$ and $\epsilon_{m-1}\neq 1/5$. (3.2). $\epsilon_{m}>1/5$ when $\epsilon_{m-1}>4/5$. The convergence region of the Phase-$\arccos(-1/4)$ search (A). When $\epsilon_{0}\in(0,4/5]$ and $\epsilon_{0}\neq 3/5$, the deviation from the target state converges to the fixed-point $1/5$. When $\epsilon_{0}=1/5$ or $4/5$, it is trivial by Eq. (6). When $\epsilon_{0}\in(0,4/5)$ and $\epsilon_{0}\neq 1/5$, always $\epsilon_{m}<1/5$ for $m>0$ by property (3.1) and the sequence $\\{\epsilon_{m}\\}$ increases monotonically from $m>0$ by property (1.3). Therefore, the sequence $\\{\epsilon_{m}\\}$ converges to $1/5$ from below. (B) When $\epsilon_{0}\in(4/5,1)$, the deviation from the target state converges to the fixed-points $1/5$ or $0$. By property (1.2), $\epsilon_{0}>\epsilon_{1}>....>\epsilon_{m}(\leq 4/5)$. Case 1. If $\epsilon_{m}=3/5$, then $\epsilon_{i}=0$ for any $i>m$. Case 2\. Otherwise, by the above (A), $\lim_{m\rightarrow\infty}\epsilon_{m}=1/5$. Example 2. Let $\epsilon_{0}=0.9999;$ $\epsilon_{1}=\allowbreak 0.999\,4$, $\ \ \epsilon_{2}=\allowbreak 0.996\,4$, $\ \ \ \ \ \ \ \ \epsilon_{3}=\allowbreak 0.978\,55$, $\ \epsilon_{4}=\allowbreak 0.876\,41$, $\epsilon_{5}=\allowbreak 0.418\,50$, $\ \epsilon_{6}=\allowbreak 8.\,\allowbreak 616\,5\times 10^{-2}$, $\ \epsilon_{7}=\allowbreak 0.142\,19$, $\ \epsilon_{8}=\allowbreak 0.186\,26$, $\epsilon_{9}=\allowbreak 0.199\,28$, $\ \epsilon_{10}=\allowbreak 0.2$. ### 2.4 When $\arccos(-1/4)<\theta\leq 2\pi/3$, the Phase-$\theta$ search converges the target state with the success probability of $(1-a)$, where $66\%\leq(1-a)<80\%$. For the Phase-$\theta$ search, $b<r<a<g<d<c$. Note that $a$ is an attractive fixed-point when $\arccos(-1/4)<\theta<2\pi/3$ and $1/3$ is a semi-attractive fixed-point when $\theta=2\pi/3$. See inference [11] in [7]. From Eq. (6), we have the following property. Property 4. (4.1). $a<\epsilon_{m}\leq g$ whenever $b<\epsilon_{m-1}<a$; (4.2). $0\leq\epsilon_{m}<a$ whenever $a<\epsilon_{m-1}<c$ or $\epsilon_{m-1}<b$. The convergence region of the Phase-$\theta$ search (A). When $\epsilon_{0}\in(0,c]$ and $\epsilon_{0}\neq d$, the deviation from the target state converges to the fixed-point $a$. There are seven cases. We argue them as follows. Case 1. If $\epsilon_{0}=a$ or $b$ or $c$, then it is trivial. Case 2. $a<\epsilon_{0}\leq g$. The proof is put in Appendix A. Case 3. $\epsilon_{0}<b$. By property (1.3), $\epsilon_{j}$ increases monotonically from $\epsilon_{0}$ until $\epsilon_{j^{\ast}-1}<b$ and $b\leq\epsilon_{j^{\ast}}<f(b)=a$ since $f^{\prime}(x)>0$ when $x<b$. If $\epsilon_{j^{\ast}}=b$, it is trivial. Otherwise, by property (4.1), $\epsilon_{j^{\ast}+1}$ is in $(a,g]$. Now it turns to case 2. Case 4. $b<\epsilon_{0}\leq r$. When $\epsilon_{0}=r$, $\epsilon_{m}=f^{(m-1)}(g)$. From the proof of case 2, $\lim_{m\rightarrow\infty}f^{(m-1)}(g)=a$. Next consider that $b<\epsilon_{0}<r$. Since $f^{\prime}(x)>0$ when $b<x<r$, $f(b)<f(\epsilon_{0})<f(r)$. That is, $a<\epsilon_{1}<g$. It turns to case 2. Case 5. $r<\epsilon_{0}<a$. Since $f^{\prime}(x)<0$ when $r<x<a$, $a<\epsilon_{1}<g$. It turns to case 2. Case 6. $g<\epsilon_{0}<d$. Since $f^{\prime}(x)<0$ when $g<x<d$ and $a<g$, $0<\epsilon_{1}<f(g)<a$. Then, it turns to cases 1, 3, 4, 5. Case 7. $d<\epsilon_{0}<c$. Since $f^{\prime}(x)>0$ when $d<x<c$, $0<\epsilon_{1}<a$. Then, it turns to cases 1, 3, 4, 5. (B). When $\epsilon_{0}\in(c,1)$, the deviation from the target state converges to the fixed-points $a$ or $0$. When $\epsilon_{0}>c$, by property (1.2) the sequence $\\{\epsilon_{i}\\}$ decreases monotonically from $\epsilon_{0}$ to $\epsilon_{i^{\ast}\text{ }}\leq c$. Case 1, if $\epsilon_{i^{\ast}\text{ }}=d$, then $\epsilon_{i}=0$ for any $i>i^{\ast}$. Case 2. Otherwise, by the above (A) $\lim_{m\rightarrow\infty}\epsilon_{m}=a$. Example 3. For the Phase-$2\pi/3$ search, $a=1/3$. The iteration equation becomes $\epsilon_{m}=\epsilon_{m-1}(3\epsilon_{m-1}-2)^{2}$. Let $\epsilon_{0}=0.99999$. We have the following iterations. See Fig. 1. $\epsilon_{1}=0.999\,93$, $\ \epsilon_{2}=0.999\,51$, $\epsilon_{3}=0.996\,57$, $\ \epsilon_{4}=0.976\,17$, $\epsilon_{5}=0.841\,59$, $\ \epsilon_{6}=0.231\,76$, $\ \ \epsilon_{7}=\allowbreak 0.394\,52$, $\ \epsilon_{8}=\allowbreak 0.263\,50$, $\epsilon_{9}=\allowbreak 0.385\,47$, $\ \epsilon_{10}=\allowbreak 0.274\,32$, $\epsilon_{11}=\allowbreak 0.380\,05$, $\ \epsilon_{12}=\allowbreak 0.280\,99$, $\epsilon_{13}=\allowbreak 0.376\,17$. ### 2.5 $2\pi/3<\theta\leq\pi$, the Phase-$\theta$ search does not converge. For the Phase-$\theta$ search, $b<r<a<d<c$. From Eq. (6), we have the following property. Property 5 (5.1). $a<\epsilon_{m}\leq g$ whenever $b<\epsilon_{m-1}<a$; (5.2). $0\leq\epsilon_{m}<a$ whenever $a<\epsilon_{m-1}<c$ or $\epsilon_{m-1}<b$. For large $\epsilon$, by property (1.2), the sequence $\\{\epsilon_{i}\\}$ decreases monotonically from $\epsilon_{0}$ to $\epsilon_{i^{\ast}}(\leq c)$. If $\epsilon_{i^{\ast}}=d$, then $\epsilon_{i}=0$ for any $i>i^{\ast}$. If $\epsilon_{i^{\ast}}=a$, $b$, or $c$, then $\epsilon_{i}=a$ when $i>i^{\ast}$. Otherwise, when $i>i^{\ast}$, $\epsilon_{i}$ oscillate around the fixed point $a$ by property 1. However, the sequence $\\{\epsilon_{i}\\}$ does not converges because $a$, $0$ and $1$ are repulsive fixed-points. Example 4. For the Phase-$\pi$ search, the iteration equation becomes $\epsilon_{m}=\epsilon_{m-1}(4\epsilon_{m-1}-3)^{2}$, $a=1/2$. Let $\epsilon=0.99999$. We have the following iterations. See Fig. 1. $\epsilon_{1}==0.999\,91$, $\ \ \epsilon_{2}=0.999\,19,$ $\ \epsilon_{3}=0.992\,73$, $\ \epsilon_{4}=0.935\,83,$ $\epsilon_{5}=0.517\,07$, $\ \ \ \epsilon_{6}=0.448\,87$, $\epsilon_{7}=0.651\,25$, $\ \epsilon_{8}=0.101\,61,$ $\epsilon_{9}=0.683\,49$, $\ \ \ \epsilon_{10}=4.\,837\,6\times 10^{-2}$. Clearly, the sequence $\\{\epsilon_{i}\\}$ monotonically decreases from $\epsilon_{0}$ to $\epsilon_{6}$. Note that after the sixth iteration, $\epsilon_{m}$ oscillate around the fixed point $1/2$. ## 3 A comparison of rates of convergence after any number of iterations For the Phase-$\pi/3$ search, let the iteration equation be $\epsilon_{m}(\pi/3)=(\epsilon_{m-1}(\pi/3))^{3}$, where $\epsilon_{m}(\pi/3)$ is the failure probability of the Phase-$\pi/3$ search algorithm after $m$ iterations. For the Phase-$\theta$ search, we can rewrite Eq. (4) as $\epsilon_{m}(\theta)=4(1-\cos\theta)^{2}\epsilon_{m-1}(\theta)(\epsilon_{m-1}(\theta)-d)^{2}$, where the $\epsilon_{m}(\theta)$ is the failure probability of the Phase-$\theta$ search algorithm after $m$ iterations. We want to compare the failure probability of the Phase-$\theta$ ($\neq\pi/3$) search algorithm with the one of the Phase-$\pi/3$ search after $m$ iterations. It is known that the less the failure probability is, the faster the algorithm converges. By factoring, $\displaystyle\epsilon_{m}(\theta)-\epsilon_{m}(\pi/3)$ $\displaystyle=$ (10) $\displaystyle\epsilon_{m-1}(\theta)(2\cos\theta-1)(1-\epsilon_{m-1}(\theta))(3-2\cos\theta)\ast$ $\displaystyle(\epsilon_{m-1}(\theta)-\frac{1-2\cos\theta}{3-2\cos\theta})+\epsilon_{m-1}^{3}(\theta)-\epsilon_{m-1}^{3}(\pi/3).$ We have the following results. (1). $\pi/3<\theta\leq\pi$ Case 1. For large $\epsilon$, the Phase-$\theta$ search converges faster than the Phase-$\pi/3$ search for $m$ iterations until $\epsilon_{m-1}(\theta)<$ $\frac{1-2\cos\theta}{3-2\cos\theta}$. In [7], we show if $\epsilon_{0}(\theta)=\epsilon_{0}(\pi/3)=\epsilon>\frac{1-2\cos\theta}{3-2\cos\theta}$ then $\epsilon_{1}(\theta)<\epsilon_{1}(\pi/3)=\epsilon^{3}$. If $\epsilon_{m-1}(\theta)>\frac{1-2\cos\theta}{3-2\cos\theta}$ and $\epsilon_{m-1}(\theta)<\epsilon_{m-1}(\pi/3)$, then by Eq. (10)$\ \epsilon_{m}(\theta)<\epsilon_{m}(\pi/3)$. Thus, $\epsilon_{i}(\theta)<\epsilon_{i}(\pi/3)$, where $i=1$, $2$, …, $m-1$, until $\epsilon_{m-1}(\theta)<$ $\frac{1-2\cos\theta}{3-2\cos\theta}$. It says that after $m$ iterations, the failure probability of the Phase-$\theta$ search is less than the one of the Phase-$\pi/3$ search until $\epsilon_{m-1}(\theta)<$ $\frac{1-2\cos\theta}{3-2\cos\theta}$. It suggests us first to use the fixed- point search with large phase shifts for the large size of database. Case 2. For small $\epsilon$, the Phase-$\pi/3$ search converges faster than the Phase-$\theta$ search for $m$ iterations until $\epsilon_{m-1}(\theta)>\frac{1-2\cos\theta}{3-2\cos\theta}$. In [7], we show if $\epsilon_{0}(\theta)=\epsilon_{0}(\pi/3)=\epsilon<\frac{1-2\cos\theta}{3-2\cos\theta}$ then $\epsilon_{1}(\theta)>\epsilon_{1}(\pi/3)=\epsilon^{3}$. If $\epsilon_{m-1}(\theta)<\frac{1-2\cos\theta}{3-2\cos\theta}$ and $\epsilon_{m-1}(\theta)>\epsilon_{m-1}(\pi/3)$, then by Eq. (10)$\ \epsilon_{m}(\theta)>\epsilon_{m}(\pi/3)$. (2). When $0<\theta<\pi/3$, the Phase-$\pi/3$ search converges faster than the Phase-$\theta$ search for any $\epsilon$ for any number of iterations. When $\epsilon_{0}(\theta)=\epsilon_{0}(\pi/3)=\epsilon$, in [7] we show $\epsilon_{1}(\theta)>\epsilon_{1}(\pi/3)$. Assume that $\epsilon_{m-1}(\theta)>\epsilon_{m-1}(\pi/3)$. From Eq. (10), it is easy to see that $\epsilon_{m}(\theta)>\epsilon_{m}(\pi/3)$. Therefore, $\epsilon_{m}(\theta)>\epsilon_{m}(\pi/3)$ for any $m$. Hence, when $0<\theta<\pi/3$, the Phase-$\pi/3$ search converges faster than the Phase-$\theta$ search for any $\epsilon$ for any number of iterations. ## 4 For any known $\epsilon$, $O(n)$ iterations can find the target state. Assume that a database has $N=2^{n}$ states (items). Then a state (an item) is found with the probability of $1/N$[2]. In other words, the failure probability $\epsilon=1-1/N$. It is known that the Phase-$\pi/3$ search converges the target state. In this section, we investigate how to use the fixed-point search to find the target state in a database when $\epsilon$ is known. As discussed in [4], the fixed-point search is a recursive algorithm, therefore the number of queries grows exponentially with the number of recursion levels. For example, the Phase-$\pi/3$ search at $i$-level recursion involves $q_{i}=(3^{i}-1)/2$ queries [6]. This implies that $O(n)$ iterations of the Phase-$\theta$ search involve the exponential queries. ### 4.1 When $\epsilon\leq 3/4$, only one iteration is needed to find the target state. When $0\leq\epsilon\leq\frac{3}{4}$, $\left|1-\frac{1}{2(1-\epsilon)}\right|\leq 1$. Let $\cos\theta=1-\frac{1}{2(1-\epsilon)}$. Then $D(\theta)=0$. Therefore, if $\epsilon$ is fixed and $0\leq\epsilon\leq\frac{3}{4}$, then we choose $\theta=\arccos[1-\frac{1}{2(1-\epsilon)}],$ which is in $(\pi/3$ ,$\pi]$, as phase shifts. The Phase-$\arccos[1-\frac{1}{2(1-\epsilon)}]$ search will obviously make the deviation vanish. It means that one iteration will reach $t$ state if the $\theta$ is chosen as phase shifts. Ref. [7]. ### 4.2 When $\epsilon>3/4$, $O(n)$ iterations can find the target state. #### 4.2.1 First use the Phase-$\pi/3$ search For the Phase-$\pi/3$ search, $\epsilon_{n}=\epsilon^{3^{n}}$. There exists the least natural number $n^{\ast}$ such that $\epsilon^{3^{n^{\ast}}}\leq 3/4$. By calculating, $n^{\ast}=\lceil(\ln\ln\frac{4}{3}-\ln\ln\frac{1}{\epsilon})/\ln 3\rceil$. Lemma 1. For the Phase-$\pi/3$ search, $n^{\ast}=$ $O(n)$. Proof. In the case of database search, Let $N=2^{n}$. Then $\epsilon=1-2^{-n}$, and $\lim_{n\rightarrow+\infty}\frac{n^{\ast}}{n}=\frac{\ln 2}{\ln 3}$. Almost $\frac{n\ln 2}{\ln 3}\approx\allowbreak\lceil 0.63n\rceil$. Let $N=10^{n}$. Then $\epsilon=1-10^{-n}$, and $\lim_{n\rightarrow+\infty}\frac{n^{\ast}}{n}=\frac{\ln 10}{\ln 3}=\frac{1}{\lg 3}$. Almost $\frac{n}{\lg 3}\approx 2n$. Thus, $n^{\ast}=O(n)$. Hence, when $\epsilon>3/4$, after $n^{\ast}$ iterations of the Phase-$\pi/3$ search the failure probability $\epsilon_{n^{\ast}}\leq$ $3/4$. Then, after one iteration of the Phase-$\arccos[1-\frac{1}{2(1-\epsilon_{n^{\ast}})}]$ search by using the result in section 4.1, it will reach $t$ state. Example 5. Let $N=10^{4}$. Then $\epsilon=1-10^{-4}$, $n^{\ast}=8$,$\ \epsilon_{7}=\allowbreak 0.803\,32$, $\epsilon_{8}=\allowbreak 0.518\,4$. See Fig.1. However, for this purpose, it only needs 4 iterations for the Phase-$\pi$ search. See example 7. Example 6. Let $N=2^{10}$. Then $\epsilon=1-2^{-10}$, $n^{\ast}=6$, $\epsilon_{5}=\allowbreak 0.788\,56$, $\epsilon_{6}=\allowbreak 0.490\,35$. #### 4.2.2 First use the Phase-$\theta$ ($\neq\pi/3$) search Let $\epsilon>3/4$. Then, by property (1.2), for the Phase-$\theta$ search, there exists the least natural number $m^{\ast}(\theta)$ such that $\ \epsilon_{0}>\epsilon_{1}>...>\epsilon_{m^{\ast}(\theta)-1}(>3/4)>\epsilon_{m^{\ast}(\theta)}(\leq 3/4)$…. Thus, after $m^{\ast}(\theta)$ iterations of the Phase-$\theta$ search, the failure probability $\epsilon_{m^{\ast}(\theta)}\leq 3/4$. Then, after one iteration for the Phase-$\arccos[1-\frac{1}{2(1-\epsilon_{m(\theta)^{\ast}})}]$ search by using the result in section 4.1, it will reach $t$ state. Next let us calculate $m^{\ast}(\theta)$. Let $\delta=1-\epsilon$, where $\delta$ is the success probability. When $\epsilon$ is close to $1$, $\delta$ is close to $0$. Then, for large $\epsilon$, by induction $\epsilon_{l}=1-[1+4(1-\cos\theta)]^{l}\delta+O(\delta^{2})$. Thus, $\epsilon_{l}\approx 1-[1+4(1-\cos\theta)]^{l}\delta$. By this approximate formula of $\epsilon_{l}$, $m^{\ast}(\theta)$ $\approx M^{\ast}(\theta)=\lceil\frac{-2\lg 2-\lg\delta}{\lg(1+4(1-\cos\theta))}\rceil$. In the case of database search, let $N=2^{n}$. Then $\epsilon=1-2^{-n}$, $\delta=2^{-n}$, and $m^{\ast}(\theta)$ $\approx$ $M^{\ast}(\theta)=\lceil\frac{(n-2)\lg 2}{\lg(1+4(1-\cos\theta))}\rceil$. For the Phase-$\pi/3$ search, $M^{\ast}(\pi/3)=\lceil\frac{n\ln 2}{\ln 3}-\frac{2\ln 2}{\ln 3}\rceil$. Note that $\frac{2\ln 2}{\ln 3}=\allowbreak 1.\,\allowbreak 261\,9$. Therefore, when $n$ is large enough $M^{\ast}(\pi/3)\approx$ $m^{\ast}(\pi/3)=n^{\ast}$. For the Phase-$\pi$ search, $m^{\ast}(\pi)\approx M^{\ast}(\pi)=\lceil(n-2)\lg 2/(2\lg 3)\rceil\approx(\lg 2)n$. See Table (I). Let $N=10^{n}$. Then $\epsilon=1-10^{-n}$, $\delta=10^{-n}$, and $m^{\ast}(\theta)\approx M^{\ast}(\theta)=\lceil\frac{n-2\lg 2}{\lg(1+4(1-\cos\theta))}\rceil$. For the Phase-$\pi/3$ search, $M^{\ast}(\pi/3)=\lceil\frac{n}{\lg 3}-\frac{2\lg 2}{\lg 3}\rceil$. Note that $\frac{2\lg 2}{\lg 3}=\frac{2\ln 2}{\ln 3}$. Therefore, when $n$ is large enough $M^{\ast}(\pi/3)\approx$ $m^{\ast}(\pi/3)=n^{\ast}$. For the Phase-$\pi$ search, $m^{\ast}(\pi)\approx M^{\ast}(\pi)=\lceil(n-2\lg 2)/(2\lg 3)\rceil\approx n$. See Example 7. Let $N=10^{4}$. Then $\epsilon=1-10^{-4}$, $M^{\ast}(\pi)=4$, $\epsilon_{4}=0.475\,32$. See Table (II). Lemma 2. For the Phase-$\theta$ $(\neq\pi/3)$ search, $m^{\ast}(\theta)=$ $O(n)$. Proof. When $\pi/3<\theta\leq\pi$, as discussed in case 1 of (1) in Sec. 3, $m^{\ast}(\theta)<n^{\ast}$. By lemma 1, this lemma holds. When $0<\theta<\pi/3$, from the approximate formula of $m^{\ast}(\theta)$, $m^{\ast}(\theta)=$ $O(n)$. Remark. $M^{\ast}(\theta)$ monotonically decreases as $\theta$ increases from $0$ to $\pi$, especially $\frac{M^{\ast}(\pi)}{n^{\ast}}\approx 1/2$. Therefore, we suggest first to use Phase-$\pi$ search for $m^{\ast}(\pi)$ times to get the failure probability $\epsilon_{m^{\ast}}\leq 3/4$. ## 5 Summary In this paper, we investigate convergence performance of the Phase-$\theta$ search for any number of iterations. We discuss the convergence region and rate of the Phase-$\theta$ search and study the convergence behavior of the Phase-$\theta$ search for different initial $\epsilon_{0}$. Acknowledgement We want to thank the reviewer of [7] for suggesting us to study the convergence behavior of the fixed-point search with general but equal phase shifts for any number of iterations. ## 6 Appendix A Proof. Since $f^{\prime}(x)<0$ when $r<x<d$, $f(g)\leq\epsilon_{1}<a$. Note that $r<f(g)$. Thus, $r<f(g)\leq\epsilon_{1}<a$. Let $f^{(k)}(x)=f(f^{(k-1)}(x))$. Since $f^{\prime}(x)<0$, $a<\epsilon_{2}\leq f^{(2)}(g)<f(r)=g$ and $r<f(g)<f^{(3)}(g)\leq\epsilon_{3}<a$. By induction, generally $a<\epsilon_{2k}\leq f^{(2k)}(g)<f^{(2k-2)}(g)<...f^{(2)}(g)<g$ and $r<f(g)<...<f^{(2k-1)}(g)<f^{(2k+1)}(g)\leq\epsilon_{2k+1}<a$. That is, $\epsilon_{i}$ oscillate around the fixed point $a$ by property 1 and between $f^{(2k)}(g)$ and $f^{(2k+1)}(g)$. It is plain that the sequence $\\{f^{(2k)}(g)\\}$ decreases monotonically as $k$ increases while the sequence $\\{f^{(2k+1)}(g)\\}$ increases monotonically as $k$ does. Hence, the sequences $\\{f^{(2k)}(g)\\}$ and $\\{f^{(2k+1)}(g)\\}$ have limits. Let $\lim_{k\rightarrow\infty}f^{(2k)}(g)=\alpha$ and $\lim_{k\rightarrow\infty}f^{(2k+1)}(g)=\beta$. Clearly, $\alpha$, $\beta<d$. From Eq. (4), $f^{(2k)}(g)=4(1-\cos\theta)^{2})f^{(2k-1)}(g)(f^{(2k-1)}(g)-d)^{2}$ and $f^{(2k+1)}(g)=4(1-\cos\theta)^{2})f^{(2k)}(g)(f^{(2k)}(g)-d)^{2}$. By taking the limits, we obtain $\alpha=4(1-\cos\theta)^{2})\beta(\beta-d)^{2}$ and $\beta=4(1-\cos\theta)^{2})\alpha(\alpha-d)^{2}$. By substituting, $\beta=[4(1-\cos\theta)^{2})]^{2}\beta(\beta-d)^{2}(\alpha-d)^{2}$. By cancelling, $[4(1-\cos\theta)^{2})]^{2}(\beta-d)^{2}(\alpha-d)^{2}=1$. Then, there are two cases. Case 1. $4(1-\cos\theta)^{2})(d-\beta)(d-\alpha)=1$. By solving this equation, $\alpha=\beta=1$ or $\alpha=\beta=a$. Since $\alpha$, $\beta<d<1$, then $\alpha=\beta=a$. Case 2. $4(1-\cos\theta)^{2})(d-\beta)(d-\alpha)=-1$. There is no solution because $\alpha$, $\beta<d$. Therefore, $\lim_{k\rightarrow\infty}f^{(2k)}(g)=\lim_{k\rightarrow\infty}f^{(2k+1)}(g)=a$. Then, $\lim_{k\rightarrow\infty}f^{(k)}(g)=a$, and also $\lim_{m\rightarrow\infty}\epsilon_{m}=a$. We finish the proof. ## References * [1] L.K.Grover, Phys. Rev. Lett. 79 (1997) 325. * [2] L.K.Grover, Phys. Rev. Lett. 80 (1998) 4329. * [3] D. Li et al., Theor. Math. Phys. 144(3) (2005) 1279-1287. * [4] L.K.Grover, Phys. Rev. Lett. 95 (2005) 150501. * [5] G. Brassard, Science 275 (1997) 627. * [6] T.Tulsi, L. Grover, and A. Patel, quant-ph/0505007. Also, Quant. Inform. and Comput. 6(6) (2006) 483–494. * [7] D. Li et al., Eur. Phys. J. D 45 (2007) 335-340. * [8] D. Li et al., Phys. Lett. A 362 (2007) 260-264. Also see quant-ph/0604062.

arxiv-papers

2024-09-04T02:48:54.374818

0803.2634

# Global well posedness and scattering for the elliptic and non-elliptic derivative nonlinear Schrödinger equations with small data Wang Baoxiang wbx@math.pku.edu.cn Corresponding author. LMAM, School of Mathematical Sciences, Peking University, Beijing 100871, People’s Republic of China (March 10, 2008) ###### Abstract We study the Cauchy problem for the generalized elliptic and non-elliptic derivative nonlinear Schrödinger equations, the existence of the scattering operators and the global well posedness of solutions with small data in Besov spaces $B^{s}_{2,1}(\mathbb{R}^{n})$ and in modulation spaces $M^{s}_{2,1}(\mathbb{R}^{n})$ are obtained. In one spatial dimension, we get the sharp well posedness result with small data in critical homogeneous Besov spaces $\dot{B}^{s}_{2,1}$. As a by-product, the existence of the scattering operators with small data is also shown. In order to show these results, the global versions of the estimates for the maximal functions on the elliptic and non-elliptic Schrödinger groups are established. ###### keywords: Derivative nonlinear Schrödinger equation, elliptic and non-elliptic cases, estimates for the maximal function, global well posedness, small data. MSC: 35 Q 55, 46 E 35, 47 D 08. ## 1 Introduction We consider the Cauchy problem for the generalized derivative nonlinear Schrödinger equation (gNLS) $\displaystyle{\rm i}u_{t}+\Delta_{\pm}u=F(u,\bar{u},\nabla u,\nabla\bar{u}),\quad u(0,x)=u_{0}(x),$ (1.1) where $u$ is a complex valued function of $(t,x)\in\mathbb{R}\times\mathbb{R}^{n}$, $\displaystyle\Delta_{\pm}u=\sum^{n}_{i=1}\varepsilon_{i}\partial^{2}_{x_{i}},\quad\varepsilon_{i}\in\\{1,\,-1\\},\quad i=1,...,n,$ (1.2) $\nabla=(\partial_{x_{1}},...,\partial_{x_{n}})$, $F:\mathbb{C}^{2n+2}\to\mathbb{C}$ is a polynomial, $\displaystyle F(z)=P(z_{1},...,z_{2n+2})=\sum_{m+1\leq|\beta|\leq M+1}c_{\beta}z^{\beta},\quad c_{\beta}\in\mathbb{C},$ (1.3) $m,M\in\mathbb{N}$ will be given below. There is a large literature which is devoted to the study of (1.1). Roughly speaking, three kinds of methods have been developed for the local and global well posedness of (1.1). The first one is the energy method, which is mainly useful to the elliptic case $\Delta_{\pm}=\Delta=\partial^{2}_{x_{1}}+...+\partial^{2}_{x_{n}}$, see Klainerman [21], Klainerman and Ponce [22], where the global classical solutions were obtained for the small Cauchy data with sufficient regularity and decay at infinity, $F$ is assumed to satisfy an energy structure condition ${\rm Re}\;\partial F/\partial(\nabla u)=0$. Chihara [6, 7] removed the condition ${\rm Re}\;\partial F/\partial(\nabla u)=0$ by using the smooth operators and the commutative estimates between the first order partial differential operators and ${\rm i}\partial_{t}+\Delta$, suitable decay conditions on the Cauchy data are still required in [6, 7]. Recently, Ozawa and Zhang [25] removed the assumptions on the decay at infinity of the initial data. They obtained that if $n\geq 3$, $s>n/2+2$, $u_{0}\in H^{s}$ is small enough, $F$ is a smooth function vanishing of the third order at origin with ${\rm Re}\;\partial F/\partial(\nabla u)=\nabla(\theta(|u|^{2}))$, $\theta\in C^{2},\;\theta(0)=0$, then (1.1) has a unique classical global solution $u\in(C_{w}\cap L^{\infty})(\mathbb{R},H^{s})\cap C(\mathbb{R},H^{s-1})\cap L^{2}(\mathbb{R};H^{s-1}_{2n/(n-2)})$. The main tools used in [25] are the gauge transform techniques, the energy method together with the endpoint Strichartz estimates. The second way consists in using the $X^{s,b}$-like spaces, see Bourgain [3] and it has been developed by many authors (see [2, 4, 15] and references therein). This method depends on both the dispersive property of the linear equation and the structure of the nonlinearities, which is very useful for the lower regularity initial data. The third method is to mainly use the dispersive smooth effects of the linear Schrödinger equation, see Kenig, Ponce and Vega [17, 18]. The crucial point is that the Schrödinger group has the following locally smooth effects ($n\geq 2$): $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\|e^{{\rm i}t\Delta}u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}\lesssim\|u_{0}\|_{\dot{H}^{-1/2}},$ (1.4) $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|\nabla\int^{t}_{0}e^{{\rm i}(t-s)\Delta}f(s)ds\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})},$ (1.5) where $Q_{\alpha}$ is the unit cube with center at $\alpha$. Estimate (1.5) contains one order smooth effect, which can be used to control the derivative terms in the nonlinearities. Such smooth effect estimates are also adapted to the non-elliptic Schrödinger group, i.e., (1.4) and (1.5) still hold if we replace $e^{{\rm i}t\Delta}$ by $e^{{\rm i}t\Delta_{\pm}}$. Some earlier estimates related to (1.4) were due to Constantin and Saut [5], Sjölin [26] and Vega [34]. In [17, 18], the local well posedness of (1.1) in both elliptic and non-elliptic cases was established for sufficiently smooth large Cauchy data ($m\geq 1$, $u_{0}\in H^{s}$ with $s>n/2$ large enough). Moreover, they showed that the solutions are almost global if the initial data are sufficiently small, i.e., the maximal existing time of solutions tends to infinity as initial data tends to 0. Recently, the local well posedness results have been generalized to the quasi-linear (ultrahyperbolic) Schrödinger equations, see [19, 20]. As far as the authors can see, the existence of the scattering operators for Eq. (1.1) and the global well posedness of (1.1) in the non-elliptic cases are unknown. ### 1.1 Main results In this paper, we mainly apply the third method to study the global well posedness and the existence of the scattering operators of (1.1) in both the elliptic and non-elliptic cases with small data in $B^{s}_{2,1}$, $s>3/2+n/2$. We now state our main results, the notations used in this paper can be found in Sections 1.3 and 1.4. ###### Theorem 1.1 Let $n\geq 2$ and $s>n/2+3/2$. Let $F(z)$ be as in (1.3) with $2+4/n\leq m\leq M<\infty$. We have the following results. (i) If $\|u_{0}\|_{B^{s}_{2,1}}\leq\delta$ for $n\geq 3$, and $\|u_{0}\|_{B^{s}_{2,1}\cap\dot{H}^{-1/2}}\leq\delta$ for $n=2$, where $\delta>0$ is a suitably small number, then (1.1) has a unique global solution $u\in C(\mathbb{R},\;B^{s}_{2,1})\cap X_{0},$ where $\displaystyle X_{0}=\left\\{u\;:\;\begin{array}[]{l}\|D^{\beta}u\|_{\ell^{1,s-1/2}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\ \ |\beta|\leq 1\\\ \|D^{\beta}u\|_{\ell^{1,s-1/2}_{\triangle}\ell^{2+4/n}_{\alpha}(L^{\infty}_{t,x}\cap(L^{2m}_{t}L^{\infty}_{x})(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\ \ |\beta|\leq 1\end{array}\right\\}.$ (1.8) Moreover, for $n\geq 3$, the scattering operator of Eq. (1.1) carries the ball $\\{u:\,\|u\|_{B^{s}_{2,1}}\leq\delta\\}$ into $B^{s}_{2,1}$. (ii) If $s+1/2\in\mathbb{N}$ and $\|u_{0}\|_{H^{s}}\leq\delta$ for $n\geq 3$, and $\|u_{0}\|_{H^{s}\cap\dot{H}^{-1/2}}\leq\delta$ for $n=2$, where $\delta>0$ is a suitably small number, then (1.1) has a unique global solution $u\in C(\mathbb{R},\;H^{s})\cap X,$ where $\displaystyle X=\left\\{u\;:\;\begin{array}[]{l}\|D^{\beta}u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq s+1/2\\\ \|D^{\beta}u\|_{\ell^{2+4/n}_{\alpha}(L^{\infty}_{t,x}\cap(L^{2m}_{t}L^{\infty}_{x})(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq 1\end{array}\right\\}.$ (1.11) Moreover, for $n\geq 3$, the scattering operator of Eq. (1.1) carries the ball $\\{u:\,\|u\|_{H^{s}}\leq\delta\\}$ into $H^{s}$. We now illustrate the proof of (ii) in Theorem 1.1. Let us consider the equivalent integral equation $\displaystyle u(t)=S(t)u_{0}-{\rm i}\mathscr{A}F(u,\bar{u},\nabla u,\nabla\bar{u}),$ (1.12) where $\displaystyle S(t):=e^{{\rm i}t\Delta_{\pm}},\ \ \ \mathscr{A}f:=\int^{t}_{0}e^{{\rm i}(t-s)\Delta_{\pm}}f(s)ds.$ (1.13) If one applies the local smooth effect estimate (1.5) to control the derivative terms in the nonlinearities, then the working space should contains the space $\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))$. For simplicity, we consider the case $F(u,\bar{u},\nabla u,\nabla\bar{u})$ $=(\partial_{x_{1}}u)^{\nu+1}$. By (1.4) and (1.5), we immediately have $\displaystyle\|\nabla u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{H^{1/2}}+\sum_{\alpha\in\mathbb{Z}^{n}}\|(\partial_{x_{1}}u)^{\nu+1}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{H^{1/2}}+\|\nabla u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\|\nabla u\|^{\nu}_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (1.14) Hence, one needs to control $\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$. In [17, 18], it was shown that for $\nu=2$, $\displaystyle\|S(t)u_{0}\|_{\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}\leq C(T)\|u_{0}\|_{H^{s}},\quad s>n/2+2.$ (1.15) In the elliptic case (1.15) holds for $s>n/2$. (1.15) is a time-local version which prevents us to get the global existence of solutions. So, it is natural to ask if there is a time-global version for the estimates of the maximal function. We can get the following $\displaystyle\|S(t)u_{0}\|_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\leq C\|u_{0}\|_{H^{s}},\quad s>n/2,\ \ \nu\geq 2+4/n.$ (1.16) Applying (1.16), we have for any $s>n/2$, $\displaystyle\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|\nabla u_{0}\|_{H^{s}}+\|\nabla(\partial_{x_{1}}u)^{1+\nu}\|_{L^{1}(\mathbb{R},H^{s}(\mathbb{R}^{n}))}.$ (1.17) One can get, say for $s=[n/2]+1$, $\displaystyle\|\nabla(\partial_{x_{1}}u)^{1+\nu}\|_{L^{1}(\mathbb{R},H^{s}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\sum_{|\beta|\leq s+2}\|D^{\beta}u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\|\nabla u\|^{\nu}_{\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ (1.18) Hence, we need to further estimate $\|D^{\beta}u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ for all $|\beta|\leq s+2$ and $\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}$. We can conjecture that a similar estimate to (1.14) holds: $\displaystyle\sum_{|\beta|\leq s+2}\|D^{\beta}u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{H^{s+3/2}}+\sum_{|\beta|\leq s+2}\|D^{\beta}u\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\|\nabla u\|^{\nu}_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (1.19) Finally, for the estimate of $\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}$, one needs the following $\displaystyle\|S(t)u_{0}\|_{\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}\leq C\|u_{0}\|_{H^{s-1/\nu}},\quad s>n/2,\ \ \nu\geq 2+4/n.$ (1.20) Using (1.20), the estimate of $\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}$ becomes easier than that of $\|\nabla u\|_{\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$. Hence, the solution has a self-contained behavior by using the spaces $\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})),\;\ell^{\nu}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))$ and $\ell^{\nu}_{\alpha}(L^{2\nu}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))$. We will give the details of the estimates (1.16) and (1.20) in Section 2. The nonlinear mapping estimates as in (1.18) and (1.19) will be given in Section 4. Next, we use the frequency-uniform decomposition method developed in [31, 32, 33] to consider the case of initial data in modulation spaces $M^{s}_{2,1}$, which is the low regularity version of Besov spaces $B^{n/2+s}_{2,1}$, i.e., $B^{n/2+s}_{2,1}\subset M^{s}_{2,1}$ is a sharp embedding and $M^{s}_{2,1}$ has only $s$-order derivative regularity (see [27, 29, 32], for the final result, see [33]). We have the following local well posedness result with small rough initial data: ###### Theorem 1.2 Let $n\geq 2$. Let $F(z)$ be as in (1.3) with $2\leq m\leq M<\infty$. Assume that $\|u_{0}\|_{M^{2}_{2,1}}\leq\delta$ for $n\geq 3$, and $\|u_{0}\|_{M^{2}_{2,1}\cap\dot{H}^{-1/2}}\leq\delta$ for $n=2$, where $\delta>0$ is sufficiently small. Then there exists a $T:=T(\delta)>0$ such that (1.1) has a unique local solution $u\in C([0,T],\;M^{2}_{2,1})\cap Y,$ where $\displaystyle Y=\left\\{u\;:\;\begin{array}[]{l}\|D^{\beta}u\|_{\ell^{1,3/2}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq 1\\\ \|D^{\beta}u\|_{\ell^{1}_{\Box}\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq 1\end{array}\right\\}.$ (1.23) Moreover, $\lim_{\delta\searrow 0}T(\delta)=\infty$. The following is a global well posedness result with Cauchy data in modulation spaces $M^{s}_{2,1}$: ###### Theorem 1.3 Let $n\geq 2$. Let $F(z)$ be as in (1.3) with $2+4/n\leq m\leq M<\infty$. Let $s>3/2+(n+2)/m$. Assume that $\|u_{0}\|_{M^{s}_{2,1}}\leq\delta$ for $n\geq 3$, and $\|u_{0}\|_{M^{s}_{2,1}\cap\dot{H}^{-1/2}}\leq\delta$ for $n=2$, where $\delta>0$ is a suitably small number. Then (1.1) has a unique global solution $u\in C(\mathbb{R},\;M^{s}_{2,1})\cap Z,$ where $\displaystyle Z=\left\\{u\;:\;\begin{array}[]{l}\|D^{\beta}u\|_{\ell^{1,s-1/2}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq 1\\\ \|D^{\beta}u\|_{\ell^{1}_{\Box}\ell^{m}_{\alpha}(L^{\infty}_{t,x}\cap(L^{2m}_{t}L^{\infty}_{x})(\mathbb{R}\times Q_{\alpha}))}\lesssim\delta,\;|\beta|\leq 1\end{array}\right\\}.$ (1.26) Moreover, for $n\geq 3$, the scattering operator of Eq. (1.1) carries the ball $\\{u:\,\|u\|_{M^{s}_{2,1}}\leq\delta\\}$ into $M^{s}_{2,1}$. Finally, we consider one spatial dimension case. Denote $\displaystyle s_{\kappa}=\frac{1}{2}-\frac{2}{\kappa},\quad\tilde{s}_{\nu}=\frac{1}{2}-\frac{1}{\nu}.$ (1.27) ###### Theorem 1.4 Let $n=1$, $M\geq m\geq 4$, $u_{0}\in\dot{B}^{1+\tilde{s}_{M}}_{2,1}\cap\dot{B}^{s_{m}}_{2,1}$. Assume that there exists a small $\delta>0$ such that $\|u_{0}\|_{\dot{B}^{1+\tilde{s}_{M}}_{2,1}\cap\dot{B}^{s_{m}}_{2,1}}\leq\delta.$ Then (1.1) has a unique global solution $u\in X=\\{u\in\mathscr{S}^{\prime}(\mathbb{R}^{1+1}):\|u\|_{X}\lesssim\delta\\}$, where $\displaystyle\|u\|_{X}$ $\displaystyle=\sup_{s_{m}\leq s\leq\tilde{s}_{M}}\sum_{i=0,1}\sum_{j\in\mathbb{Z}}|\\!|\\!|\partial^{i}_{x}\triangle_{j}u|\\!|\\!|_{s}\ \ for\ \ m>4,$ $\displaystyle\|u\|_{X}$ $\displaystyle=\sum_{i=0,1}\big{(}\|\partial^{i}_{x}u\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}+\sup_{\tilde{s}_{m}\leq s\leq\tilde{s}_{M}}\sum_{j\in\mathbb{Z}}|\\!|\\!|\partial^{i}_{x}\triangle_{j}u|\\!|\\!|_{s}\big{)}\ \ for\ \ m=4,$ $\displaystyle|\\!|\\!|\triangle_{j}v$ $\displaystyle|\\!|\\!|_{s}:=2^{sj}(\|\triangle_{j}v\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}+2^{j/2}\|\triangle_{j}v\|_{L^{\infty}_{x}L^{2}_{t}})$ $\displaystyle\ \ \ \ \ \ +2^{(s-\tilde{s}_{m})j}\|\triangle_{j}v\|_{L_{x}^{m}L_{t}^{\infty}}+2^{(s-\tilde{s}_{M})j}\|\triangle_{j}v\|_{L_{x}^{M}L_{t}^{\infty}}.$ (1.28) Recall that the norm on homogeneous Besov spaces $\dot{B}^{s}_{2,1}$ can be defined in the following way: $\displaystyle\|f\|_{\dot{B}^{s}_{2,1}}=\sum^{\infty}_{j=-\infty}2^{sj}\left(\int^{2^{j+1}}_{2^{j}}|\mathscr{F}f(\xi)|^{2}d\xi\right)^{1/2}.$ (1.29) ### 1.2 Remarks on main results It seems that the regularity assumptions on initial data are not optimal in Theorems 1.1–1.3, but Theorem 1.4 presents the sharp regularity condition to the initial data. To illustrate the relation between the regularity index and the nonlinear power, we consider a simple cases of (1.1): $\displaystyle{\rm i}u_{t}+\Delta_{\pm}u=u_{x_{1}}^{\nu},\ \ u(0)=\phi.$ (1.30) Eq. (1.30) is invariant under the scaling $u\to u_{\lambda}=\lambda^{(2-\nu)/(\nu-1)}u(\lambda^{2}t,\lambda x)$ and moreover, $\displaystyle\|\phi\|_{\dot{H}^{s}(\mathbb{R}^{n})}=\|u_{\lambda}(0,\cdot)\|_{\dot{H}^{s}(\mathbb{R}^{n})},\quad s=1+\tilde{s}_{\nu-1}:=1+n/2-1/(\nu-1).$ (1.31) From this point of view, we say that $s=1+\tilde{s}_{\nu-1}$ is the critical regularity index of (1.30). In [23], Molinet and Ribuad showed that (1.30) is ill-posed in one spatial dimension in the sense if $s_{1}\not=\tilde{s}_{\nu-1}+1$, the flow map of equation (1.30) $\phi\rightarrow u$ (if it exists) is not of class $C^{\nu}$ from $\dot{B}^{s_{1}}_{2,1}(\mathbb{R})$ to $C([0,\infty),\dot{B}^{s_{1}}_{2,1}(\mathbb{R}))$ at the origin $\phi=0$. For each term in the polynomial nonlinearity $F(u,\bar{u},\nabla u,\nabla\bar{u})$ as in (1.3), we easily see that the critical index $s$ can take any critical index between $s_{m}$ and $1+\tilde{s}_{M}$. So, our Theorem 1.4 give sharp result in the case $m\geq 4$. On the other hand, Christ [9] showed that in the case $\nu=2$, $n=1$, for any $s\in\mathbb{R}$, there exist initial data in $H^{s}$ with arbitrarily small norm, for which the solution attains arbitrarily large norm after an arbitrarily short time (see also [24]). From Christ’s result together with Theorems 1.4, we can expect that there exists $m_{0}>1$ (might be non-integer) so that for $\nu-1\geq m_{0}$, $s=1+\tilde{s}_{\nu-1}$ is the minimal regularity index to guarantee the well posedness of (1.30), at least for the local solutions and small data global solutions in $H^{s}$. However, it is not clear for us how to find the exact value of $m_{0}$ even in one spatial dimension. However, in higher spatial dimensions, it seems that $1/2+1/M$-order derivative regularity is lost in Theorem 1.1 and we do not know how to attain the regularity index $s\geq 1+\tilde{s}_{M}$. In two dimensional case, if $\Delta_{\pm}=\Delta$ and the initial value $u_{0}$ is a radial function, we can remove the condition $u_{0}\in\dot{H}^{-1/2}$, $\|u_{0}\|_{\dot{H}^{-1/2}}\leq\delta$ by using the endpoint Strichartz estimates as in the case $n\geq 3$. Considering the nonlinearity $F(u,\nabla u)=(1-|u|^{2})^{-1}|\nabla u|^{2k}u$, Theorem 1.2 holds for the case $k\geq 1$. Theorems 1.1 and 1.3 hold for the case $k\geq 2$. Since $(1-|u|^{2})^{-1}=\sum^{\infty}_{k=0}|u|^{2k}$, one easily sees that we can use the same way as in the proof of our main results to handle this kind of nonlinearity. ### 1.3 Notations Throughout this paper, we will always use the following notations. $\mathscr{S}(\mathbb{R}^{n})$ and $\mathscr{S}^{\prime}(\mathbb{R}^{n})$ stand for the Schwartz space and its dual space, respectively. We denote by $L^{p}(\mathbb{R}^{n})$ the Lebesgue space, $\|\cdot\|_{p}:=\|\cdot\|_{L^{p}(\mathbb{R}^{n})}$. The Bessel potential space is defined by $H^{s}_{p}(\mathbb{R}^{n}):=(I-\Delta)^{-s/2}L^{p}(\mathbb{R}^{n})$, $H^{s}(\mathbb{R}^{n})=H^{s}_{2}(\mathbb{R}^{n})$, $\dot{H}^{s}(\mathbb{R}^{n})=(-\Delta)^{-s/2}L^{2}(\mathbb{R}^{n})$.111$\mathbb{R}^{n}$ will be omitted in the definitions of various function spaces if there is no confusion. For any quasi-Banach space $X$, we denote by $X^{*}$ its dual space, by $L^{p}(I,X)$ the Lebesgue-Bochner space, $\|f\|_{L^{p}(I,X)}:=(\int_{I}\|f(t)\|^{p}_{X}dt)^{1/p}$. If $X=L^{r}(\Omega)$, then we write $L^{p}(I,L^{r}(\Omega))=L^{p}_{t}L^{r}_{x}(I\times\Omega)$ and $L^{p}_{t,x}(I\times\Omega)=L^{p}_{t}L^{p}_{x}(I\times\Omega)$. Let $Q_{\alpha}$ be the unit cube with center at $\alpha\in\mathbb{Z}^{n}$, i.e., $Q_{\alpha}=\alpha+Q_{0},Q_{0}=\\{x=(x_{1},...x_{n}):-1/2\leq x_{i}<1/2\\}.$ We also needs the function spaces $\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))$, $\|f\|_{\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))}:=\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|^{q}_{L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha})}\right)^{1/q}.$ We denote by $\mathscr{F}$ ($\mathscr{F}^{-1}$) the (inverse) Fourier transform for the spatial variables; by $\mathscr{F}_{t}$ ($\mathscr{F}^{-1}_{t}$) the (inverse) Fourier transform for the time variable and by $\mathscr{F}_{t,x}$ ($\mathscr{F}^{-1}_{t,x}$) the (inverse) Fourier transform for both time and spatial variables, respectively. If there is no explanation, we always denote by $\varphi_{k}(\cdot)$ the dyadic decomposition functions as in (1.36); and by $\sigma_{k}(\cdot)$ the uniform decomposition functions as in (1.38). $u\star v$ and $u*v$ will stand for the convolution on time and on spatial variables, respectively, i.e., $(u\star v)(t,x)=\int_{\mathbb{R}}u(t-\tau,x)v(\tau,x)d\tau,\ \ (u*v)(t,x)=\int_{\mathbb{R}^{n}}u(t,x-y)v(t,y)dy.$ $\mathbb{R},\mathbb{N}$ and $\mathbb{Z}$ will stand for the sets of reals, positive integers and integers, respectively. $c<1$, $C>1$ will denote positive universal constants, which can be different at different places. $a\lesssim b$ stands for $a\leq Cb$ for some constant $C>1$, $a\sim b$ means that $a\lesssim b$ and $b\lesssim a$. We denote by $p^{\prime}$ the dual number of $p\in[1,\infty]$, i.e., $1/p+1/p^{\prime}=1$. For any $a>0$, we denote by $[a]$ the minimal integer that is larger than or equals to $a$. $B(x,R)$ will denote the ball in $\mathbb{R}^{n}$ with center $x$ and radial $R$. ### 1.4 Besov and modulation spaces Let us recall that Besov spaces $B^{s}_{p,q}:=B^{s}_{p,q}(\mathbb{R}^{n})$ are defined as follows (cf. [1, 30]). Let $\psi:\mathbb{R}^{n}\to[0,1]$ be a smooth radial bump function adapted to the ball $B(0,2)$: $\displaystyle\psi(\xi)=\left\\{\begin{array}[]{ll}1,&|\xi|\leq 1,\\\ {\rm smooth},&|\xi|\in[1,2],\\\ 0,&|\xi|\geq 2.\end{array}\right.$ (1.35) We write $\delta(\cdot):=\psi(\cdot)-\psi(2\,\cdot)$ and $\displaystyle\varphi_{j}:=\delta(2^{-j}\cdot)\ \ {\rm for}\ \ j\geq 1;\quad\varphi_{0}:=1-\sum_{j\geq 1}\varphi_{j}.$ (1.36) We say that $\triangle_{j}:=\mathscr{F}^{-1}\varphi_{j}\mathscr{F},\quad j\in\mathbb{N}\cup\\{0\\}$ are the dyadic decomposition operators. Beove spaces $B^{s}_{p,q}=B^{s}_{p,q}(\mathbb{R}^{n})$ are defined in the following way: $\displaystyle B^{s}_{p,q}=\left\\{f\in\mathscr{S}^{\prime}(\mathbb{R}^{n}):\;\|f\|_{B^{s}_{p,q}}=\left(\sum^{\infty}_{j=0}2^{sjq}\|\,\triangle_{j}f\|^{q}_{p}\right)^{1/q}<\infty\right\\}.$ (1.37) Now we recall the definition of modulation spaces (see [12, 13, 31, 32, 33]). Here we adopt an equivalent norm by using the uniform decomposition to the frequency space. Let $\rho\in\mathscr{S}(\mathbb{R}^{n})$ and $\rho:\,\mathbb{R}^{n}\to[0,1]$ be a smooth radial bump function adapted to the ball $B(0,\sqrt{n})$, say $\rho(\xi)=1$ as $|\xi|\leq\sqrt{n}/2$, and $\rho(\xi)=0$ as $|\xi|\geq\sqrt{n}$. Let $\rho_{k}$ be a translation of $\rho$: $\rho_{k}(\xi)=\rho(\xi-k),\;k\in\mathbb{Z}^{n}$. We write $\displaystyle\sigma_{k}(\xi)=\rho_{k}(\xi)\left(\sum_{k\in\mathbb{Z}^{n}}\rho_{k}(\xi)\right)^{-1},\quad k\in\mathbb{Z}^{n}.$ (1.38) Denote $\displaystyle\Box_{k}:=\mathscr{F}^{-1}\sigma_{k}\mathscr{F},\quad k\in\mathbb{Z}^{n},$ (1.39) which are said to be the frequency-uniform decomposition operators. For any $k\in\mathbb{Z}^{n}$, we write $\langle k\rangle=\sqrt{1+|k|^{2}}$. Let $s\in\mathbb{R}$, $0<p,q\leq\infty$. Modulation spaces $M^{s}_{p,q}=M^{s}_{p,q}(\mathbb{R}^{n})$ are defined as: $\displaystyle M^{s}_{p,q}=\left\\{f\in\mathscr{S}^{\prime}(\mathbb{R}^{n}):\;\|f\|_{M^{s}_{p,q}}=\left(\sum_{k\in\mathbb{Z}^{n}}\langle k\rangle^{sq}\|\,\Box_{k}f\|_{p}^{q}\right)^{1/q}<\infty\right\\}.$ (1.40) We will use the function space $\ell^{1,s}_{\Box}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))$ which contains all of the functions $f(t,x)$ so that the following norm is finite: $\displaystyle\|f\|_{\ell^{1,s}_{\Box}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))}:=\sum_{k\in\mathbb{Z}^{n}}\langle k\rangle^{s}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}f\|^{q}_{L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha})}\right)^{1/q}.$ (1.41) Similarly, we can define the space $\ell^{1,s}_{\triangle}\ell^{q}_{\alpha}(L^{p}_{t,x}(I\times Q_{\alpha}))$ with the following norm: $\displaystyle\|f\|_{\ell^{1,s}_{\triangle}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))}:=\sum^{\infty}_{j=0}2^{sj}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\triangle_{j}f\|^{q}_{L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha})}\right)^{1/q}.$ (1.42) A special case is $s=0$, we write $\ell^{1,0}_{\Box}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))=\ell^{1}_{\Box}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))$ and $\ell^{1,0}_{\triangle}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))=\ell^{1}_{\triangle}\ell^{q}_{\alpha}(L^{p}_{t}L^{r}_{x}(I\times Q_{\alpha}))$. The rest of this paper is organized as follows. In Section 2 we give the details of the estimates for the maximal function in certain function spaces. Section 3 is devoted to considering the spatial local versions for the Strichartz estimates and giving some remarks on the estimates of the local smooth effects. In Sections 4–7 we prove our main Theorems 1.1–1.4, respectively. ## 2 Estimates for the maximal function ### 2.1 Time-local version Recall that $S(t)=e^{-{\rm i}t\triangle_{\pm}}=\mathscr{F}^{-1}e^{{\rm i}t|\xi|^{2}_{\pm}}\mathscr{F}$, where $\displaystyle|\xi|^{2}_{\pm}=\sum^{n}_{j=1}\varepsilon_{j}\xi^{2}_{j},\quad\varepsilon_{j}=\pm 1.$ (2.1) Kenig, Ponce and Vega [17] showed the following maximal function estimate: $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|S(t)u_{0}\|^{2}_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}\right)^{1/2}\lesssim C(T)\|u_{0}\|_{H^{s}},$ (2.2) where $s\geq 2+n/2$. If $S(t)=e^{-{\rm i}t\triangle}$, then (2.2) holds for $s>n/2$, $C(T)=(1+T)^{s}$. Using the frequency-uniform decomposition method, we can get the following ###### Proposition 2.1 There exists a constant $C(T)>1$ which depends only on $T$ and $n$ such that $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}S(t)u_{0}\|^{2}_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}\right)^{1/2}\leq C(T)\|u_{0}\|_{M^{1/2}_{2,1}},$ (2.3) In particular, for any $s>(n+1)/2$, $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,S(t)u_{0}\|^{2}_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}\right)^{1/2}\leq C(T)\|u_{0}\|_{H^{s}}.$ (2.4) Proof. By the duality, it suffices to prove that $\displaystyle\int^{T}_{0}(S(t)u_{0},\psi(t))dt\lesssim\|u_{0}\|_{M^{1/2}_{2,1}}\sup_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}\psi(t)\|_{L^{1}_{t,x}([0,T]\times Q_{\alpha})}^{2}\right)^{1/2}.$ (2.5) Since $(M^{1/2}_{2,1})^{*}=M^{-1/2}_{2,\infty}$, we have $\displaystyle\int^{T}_{0}(S(t)u_{0},\psi(t))dt\leq\|u_{0}\|_{M^{1/2}_{2,1}}\left\|\int^{T}_{0}S(-t)\psi(t)dt\right\|_{M^{-1/2}_{2,\infty}}.$ (2.6) Recalling that $\|f\|_{M^{-1/2}_{2,\infty}}=\sup_{k\in\mathbb{Z}^{n}}\langle k\rangle^{-1/2}\|\Box_{k}f\|_{2}$, we need to estimate $\displaystyle\left\|\Box_{k}\int^{T}_{0}S(-t)\psi(t)dt\right\|^{2}_{2}$ $\displaystyle=\int^{T}_{0}\left(\Box_{k}\psi(t),\;\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right)dt$ $\displaystyle\leq\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}\psi\|_{L^{1}_{t,x}([0,T]\times Q_{\alpha})}\left\|\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right\|_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}$ $\displaystyle\leq\|\Box_{k}\psi\|_{\ell^{2}_{\alpha}(L^{1}_{t,x}([0,T]\times Q_{\alpha}))}\left\|\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right\|_{\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}.$ (2.7) If one can show that $\displaystyle\left\|\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right\|_{\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}\lesssim C(T)\langle k\rangle\|\Box_{k}\psi\|_{\ell^{2}_{\alpha}(L^{1}_{t,x}([0,T]\times Q_{\alpha}))},$ (2.8) then from (2.6)–(2.8) we obtain that (2.5) holds. Denote $\displaystyle\Lambda:=\\{\ell\in\mathbb{Z}^{n}:{\rm supp}\,\sigma_{\ell}\cap{\rm supp}\,\sigma_{0}\not=\varnothing\\}.$ (2.9) In the following we show (2.8). In view of Young’s inequality, we have $\displaystyle\left\|\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right\|_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\ell\in\Lambda}\left\|\int^{T}_{0}S(t-\tau)\Box_{k+\ell}\Box_{k}\psi(\tau)d\tau\right\|_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}$ $\displaystyle=\sum_{\ell\in\Lambda}\left\|\int^{T}_{0}[\mathscr{F}^{-1}(e^{{\rm i}(t-\tau)|\xi|^{2}_{\pm}}\sigma_{k+\ell})]*\Box_{k}\psi(\tau)d\tau\right\|_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}$ $\displaystyle\leq\sum_{\ell\in\Lambda}\sum_{\beta\in\mathbb{Z}^{n}}\left\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k+\ell})\right\|_{L^{\infty}_{t,x}([-T,T]\times Q_{\beta})}\|\Box_{k}\psi\|_{L^{1}_{t,x}([0,T]\times(Q_{\alpha}-Q_{\beta}))}.$ (2.10) From (2.10) and Minkowski’s inequality that $\displaystyle\left\|\int^{T}_{0}S(t-\tau)\Box_{k}\psi(\tau)d\tau\right\|_{\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\leq\sum_{\ell\in\Lambda}\sum_{\beta\in\mathbb{Z}^{n}}\left\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k+\ell})\right\|_{L^{\infty}_{t,x}([-T,T]\times Q_{\beta})}\|\Box_{k}\psi\|_{\ell^{2}_{\alpha}(L^{1}_{t,x}([0,T]\times(Q_{\alpha}-Q_{\beta})))}.$ (2.11) It is easy to see that $\displaystyle\|\Box_{k}\psi\|_{\ell^{2}_{\alpha}(L^{1}_{t,x}([0,T]\times(Q_{\alpha}-Q_{\beta})))}\lesssim\|\Box_{k}\psi\|_{\ell^{2}_{\alpha}(L^{1}_{t,x}([0,T]\times Q_{\alpha}))}.$ (2.12) Hence, in order to prove (2.8), it suffices to prove that $\displaystyle\sum_{\beta\in\mathbb{Z}^{n}}\left\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k})\right\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\lesssim C(T)\langle k\rangle.$ (2.13) In fact, observing the following identity, $\displaystyle|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k})|=|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})(\cdot+2tk_{\pm})|,$ (2.14) where $k_{\pm}=(\varepsilon_{1}k_{1},...,\varepsilon_{n}k_{n})$, we have $\displaystyle\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\leq\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q^{*}_{\beta,k})},$ (2.15) where $\displaystyle Q^{*}_{\beta,k}=\\{x:\;x\in 2tk_{\pm}+Q_{\beta}\;\;\mbox{for some }\;t\in[0,T]\\}.$ Denote $\Lambda_{\beta,k}=\\{\beta^{\prime}:\;Q_{\beta^{\prime}}\cap Q^{*}_{\beta,k}\neq\varnothing\\}$. It follows from (2.15) that $\displaystyle\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\leq\sum_{\beta\in\mathbb{Z}^{n}}\sum_{\beta^{\prime}\in\Lambda_{\beta,k}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta^{\prime}})}.$ (2.16) Since each $E_{\beta,k}$ overlaps at most $O(T\langle k\rangle)$ many $Q_{\beta^{\prime}}$, $\beta^{\prime}\in\mathbb{Z}^{n}$, one can easily verify that in the sums of the right hand side of (2.16), each $\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta^{\prime}})}$ repeats at most $O(T\langle k\rangle)$ times. Hence, we have $\displaystyle\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\lesssim\langle k\rangle\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}.$ (2.17) Finally, it suffices to show that $\displaystyle\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\leq C(T).$ (2.18) Denote $\nabla_{t,x}=(\partial_{t},\partial_{x_{1}},...,\partial_{x_{n}})$. By the Sobolev inequality, $\displaystyle\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{\infty}_{t,x}([0,T]\times Q_{\beta})}\lesssim$ $\displaystyle\;\sum_{\beta\in\mathbb{Z}^{n}}\|\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{2n}_{t,x}([0,T]\times Q_{\beta})}$ $\displaystyle+\sum_{\beta\in\mathbb{Z}^{n}}\|\nabla_{t,x}\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\|_{L^{2n}_{t,x}([0,T]\times Q_{\beta})}$ $\displaystyle=$ $\displaystyle I+II.$ (2.19) By Hölder’s inequality, we have $\displaystyle II\lesssim$ $\displaystyle\left(\sum_{\beta\in\mathbb{Z}^{n}}\left\|(1+|x|^{2})^{n}\nabla_{t,x}\mathscr{F}^{-1}(e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{0})\right\|^{2n}_{L^{2n}_{t,x}([0,T]\times Q_{\beta})}\right)^{1/2n}$ $\displaystyle\lesssim$ $\displaystyle\sum^{n}_{i=1}\left\|\mathscr{F}^{-1}(I-\Delta)^{n}(e^{{\rm i}t|\xi|^{2}_{\pm}}\xi_{i}\sigma_{0})\right\|_{L^{2n}_{t,x}([0,T]\times\mathbb{R}^{n})}$ $\displaystyle+\left\|\mathscr{F}^{-1}(I-\Delta)^{n}(e^{{\rm i}t|\xi|^{2}_{\pm}}|\xi|^{2}_{\pm}\sigma_{0})\right\|_{L^{2n}_{t,x}([0,T]\times\mathbb{R}^{n})}$ $\displaystyle\lesssim$ $\displaystyle\;C(T).$ (2.20) One easily sees that $I$ has the same bound as that of $II$. The proof of (2.3) is finished. Noticing that $H^{s}\subset M^{1/2}_{2,1}$ if $s>(n+1)/2$ (cf. [29, 32, 33]), we immediately have (2.4). $\hfill\Box$ ### 2.2 Time-global version Recall that we have the following equivalent norm on Besov spaces ([1, 30]): ###### Lemma 2.2 Let $1\leq p,q\leq\infty$, $\sigma>0$, $\sigma\not\in\mathbb{N}$. Then we have $\displaystyle\|f\|_{B^{\sigma}_{p,q}}\sim\sum_{|\beta|\leq[\sigma]}\|D^{\beta}f\|_{L^{p}(\mathbb{R}^{n})}+\sum_{|\beta|\leq[\sigma]}\left(\int_{\mathbb{R}^{n}}|h|^{-n-q\\{\sigma\\}}\|\vartriangle_{h}D^{\beta}f\|^{q}_{L^{p}(\mathbb{R}^{n})}dh\right)^{1/q},$ (2.21) where $\vartriangle_{h}f=f(\cdot+h)-f(\cdot)$, $[\sigma]$ denotes the minimal integer that is larger than or equals to $\sigma$, $\\{\sigma\\}=\sigma-[\sigma]$. Taking $p=q$ in Lemma 2.2, one has that $\displaystyle\|f\|^{p}_{B^{\sigma}_{p,p}}\sim\sum_{|\beta|\leq[\sigma]}\|D^{\beta}f\|^{p}_{L^{p}(\mathbb{R}^{n})}+\sum_{|\beta|\leq[\sigma]}\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}\frac{|\vartriangle_{h}D^{\beta}f(x)|^{p}}{|h|^{n+p\\{\sigma\\}}}dxdh.$ (2.22) ###### Lemma 2.3 Let $1<p<\infty$, $s>1/p$. Then we have $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|u\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}\lesssim\|(I-\partial^{2}_{t})^{s/2}u\|_{L^{p}(\mathbb{R},B^{ns}_{p,p}(\mathbb{R}^{n}))}.$ (2.23) Proof. We divide the proof into the following two cases. Case 1\. $ns\not\in\mathbb{N}.$ Due to $H^{s}_{p}(\mathbb{R})\subset L^{\infty}(\mathbb{R})$, we have $\displaystyle\|u\|_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|(I-\partial^{2}_{t})^{s/2}u\|_{L^{\infty}_{x}L^{p}_{t}(Q_{\alpha}\times\mathbb{R})}$ $\displaystyle\leq\|(I-\partial^{2}_{t})^{s/2}u\|_{L^{p}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (2.24) Recalling that $\sigma_{\alpha}(x)\gtrsim 1$ for all $x\in Q_{\alpha}$ and $\alpha\in\mathbb{Z}^{n}$, we have from (2.24) that $\displaystyle\|u\|_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\leq\|(I-\partial^{2}_{t})^{s/2}\sigma_{\alpha}u\|_{L^{p}_{t}L^{\infty}_{x}(\mathbb{R}\times\mathbb{R}^{n})}.$ (2.25) Since $B^{ns}_{p,p}(\mathbb{R}^{n})\subset L^{\infty}(\mathbb{R}^{n})$, in view of (2.25), one has that $\displaystyle\|u\|_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\leq\|(I-\partial^{2}_{t})^{s/2}\sigma_{\alpha}u\|_{L^{p}(\mathbb{R},B^{ns}_{p,p}(\mathbb{R}^{n}))}.$ (2.26) For simplicity, we denote $v=(I-\partial^{2}_{t})^{s/2}u$. By (2.22) and (2.26) we have $\displaystyle\sum_{\alpha\in\mathbb{Z}^{n}}\|u\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\lesssim$ $\displaystyle\sum_{|\beta|\leq[ns]}\sum_{\alpha\in\mathbb{Z}^{n}}\int_{\mathbb{R}}\|D^{\beta}(\sigma_{\alpha}v)(t)\|^{p}_{L^{p}(\tilde{Q}_{\alpha})}dt$ $\displaystyle+\sum_{|\beta|\leq[ns]}\sum_{\alpha\in\mathbb{Z}^{n}}\int_{\mathbb{R}}\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}\frac{|\vartriangle_{h}D^{\beta}(\sigma_{\alpha}v)(t,x)|^{p}}{|h|^{n+p\\{ns\\}}}dxdhdt$ $\displaystyle:=$ $\displaystyle I+II.$ (2.27) We now estimate $II$. It is easy to see that $\displaystyle|\vartriangle_{h}\\!D^{\beta}(\sigma_{\alpha}v)|$ $\displaystyle\lesssim\sum_{\beta_{1}+\beta_{2}=\beta}|\vartriangle_{h}\\!(D^{\beta_{1}}\sigma_{\alpha}D^{\beta_{2}}v)|$ $\displaystyle\leqslant\sum_{\beta_{1}+\beta_{2}=\beta}(|D^{\beta_{1}}\sigma_{\alpha}(\cdot+h)\,\vartriangle_{h}\\!D^{\beta_{2}}v|+|(\vartriangle_{h}\\!D^{\beta_{1}}\sigma_{\alpha})D^{\beta_{2}}v|).$ (2.28) Since ${\rm supp}\,\sigma_{\alpha}$ overlaps at most finitely many ${\rm supp}\,\sigma_{\beta}$ and $\sigma_{\beta}=\sigma_{0}(\cdot-\beta)$, $\beta\in\mathbb{Z}^{n}$, it follows from (2.28), $|D^{\beta_{1}}\sigma_{\alpha}|\lesssim 1$ and Hölder’s inequality that $\displaystyle II\lesssim$ $\displaystyle\sum_{|\beta_{1}|,|\beta_{2}|\leq[ns]}\int_{\mathbb{R}}\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}\sum_{\alpha\in\mathbb{Z}^{n}}|D^{\beta_{1}}\sigma_{\alpha}(x+h)|\frac{|\vartriangle_{h}\\!D^{\beta_{2}}v(t,x)|^{p}}{|h|^{n+p\\{ns\\}}}dxdhdt$ $\displaystyle+\sum_{|\beta|\leq[ns]}\sum_{\beta_{1}+\beta_{2}=\beta}\int_{\mathbb{R}^{n}}\frac{\|\vartriangle_{h}\\!D^{\beta_{1}}\sigma_{0}\|^{p}_{L^{\infty}(\mathbb{R}^{n})}}{|h|^{n+p\\{ns\\}}}dh$ $\displaystyle\quad\times\sup_{h}\sum_{\alpha\in\mathbb{Z}^{n}}\int_{\mathbb{R}}\int_{B(0,\sqrt{n})\cup B(-h,\sqrt{n}))}|D^{\beta_{2}}v(t,x+\alpha)|^{p}dxdt$ $\displaystyle\lesssim$ $\displaystyle\sum_{|\beta|\leq[ns]}\int_{\mathbb{R}}\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}\frac{|\vartriangle_{h}\\!D^{\beta}v(t,x)|^{p}}{|h|^{n+p\\{ns\\}}}dxdhdt$ $\displaystyle+\|\sigma_{\alpha}\|^{p}_{B^{ns}_{\infty,p}}\sum_{|\beta|\leq[ns]}\int_{\mathbb{R}}\int_{\mathbb{R}^{n}}|D^{\beta}v(x)|^{p}dxdt$ $\displaystyle\lesssim$ $\displaystyle\;\|v\|^{p}_{L^{p}(\mathbb{R},B^{ns}_{p,p}(\mathbb{R}^{n}))}.$ (2.29) Clearly, one has that $\displaystyle I\lesssim\|v\|^{p}_{L^{p}(\mathbb{R},B^{ns}_{p,p}(\mathbb{R}^{n}))}.$ (2.30) Collecting (2.27), (2.29) and (2.30), we have (2.23). Case 2\. $ns\in\mathbb{N}.$ One can take an $s_{1}<s$ such that $s_{1}>1/p$ and $ns_{1}\not\in\mathbb{N}$. Applying the conclusion as in Case 1, we get the result, as desired. $\hfill\Box$ For the semi-group $S(t)$, we have the following Strichartz estimate (cf. [14]): ###### Proposition 2.4 Let $n\geq 2$. $2\leq p,\rho\leq 2n/(n-2)$ $(2\leq p,\rho<\infty$ if $n=2)$, $2/\gamma(\cdot)=n(1/2-1/\cdot)$. We have $\displaystyle\|S(t)u_{0}\|_{L^{\gamma(p)}(\mathbb{R},L^{p}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\|u_{0}\|_{L^{2}(\mathbb{R}^{n})},$ (2.31) $\displaystyle\|\mathscr{A}F\|_{L^{\gamma(p)}(\mathbb{R},L^{p}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\|F\|_{L^{\gamma(\rho)^{\prime}}(\mathbb{R},L^{\rho^{\prime}}(\mathbb{R}^{n}))}.$ (2.32) If $p$ and $\rho$ equal to $2n/(n-2)$, then (2.31) and (2.32) are said to be the endpoint Strichartz estimates. Using Proposition 2.4, we have ###### Proposition 2.5 Let $p\geq 2+4/n:=2^{*}$. For any $s>n/2$, we have $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,S(t)u_{0}\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}\lesssim\|u_{0}\|_{H^{s}}.$ (2.33) Proof. For short, we write $\langle\partial_{t}\rangle=(I-\partial^{2}_{t})^{1/2}$. By Lemma 2.3, for any $s_{0}>1/2^{*}$, $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|S(t)u_{0}\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}$ $\displaystyle\lesssim\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|S(t)u_{0}\|^{2^{*}}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/{2^{*}}}$ $\displaystyle\lesssim\|\langle\partial_{t}\rangle^{s_{0}}S(t)u_{0}\|_{L^{2^{*}}(\mathbb{R},B^{ns_{0}}_{2^{*},2^{*}}(\mathbb{R}^{n}))}.$ (2.34) We have $\displaystyle\|\langle\partial_{t}\rangle^{s_{0}}S(t)u_{0}\|^{2^{*}}_{L^{2^{*}}(\mathbb{R},B^{ns_{0}}_{2^{*},2^{*}}(\mathbb{R}^{n}))}=\sum_{k=0}^{\infty}2^{ns_{0}k2^{*}}\|\langle\partial_{t}\rangle^{s_{0}}\triangle_{k}S(t)u_{0}\|^{2^{*}}_{L^{2^{*}}_{t,x}(\mathbb{R}^{1+n})}.$ (2.35) Using the dyadic decomposition to the time-frequency, we obtain that $\displaystyle\|\langle\partial_{t}\rangle^{s_{0}}\triangle_{k}S(t)u_{0}\|_{L^{2^{*}}_{t,x}}\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}_{t,x}\langle\tau\rangle^{s_{0}}\varphi_{j}(\tau)\mathscr{F}_{t}e^{{\rm i}t|\xi|^{2}_{\pm}}\varphi_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2^{*}}_{t,x}}.$ (2.36) Noticing the fact that $\displaystyle(\mathscr{F}^{-1}_{t}\langle\tau\rangle^{s_{0}}\varphi_{j}(\tau))\star e^{{\rm i}t|\xi|^{2}_{\pm}}=c\;e^{{\rm i}t|\xi|^{2}_{\pm}}\varphi_{j}(|\xi|^{2}_{\pm})\langle|\xi|^{2}_{\pm}\rangle^{s_{0}},$ (2.37) and using the Strichartz inequality and Plancherel’s identity, one has that $\displaystyle\|\langle\partial_{t}\rangle^{s_{0}}\triangle_{k}S(t)u_{0}\|_{L^{2^{*}}_{t,x}}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\|S(t)\mathscr{F}^{-1}_{x}\langle|\xi|^{2}_{\pm}\rangle^{s_{0}}\varphi_{j}(|\xi|^{2}_{\pm})\varphi_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2^{*}}_{t,x}}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}_{x}\langle|\xi|^{2}_{\pm}\rangle^{s_{0}}\varphi_{j}(|\xi|^{2}_{\pm})\varphi_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}$ $\displaystyle\lesssim 2^{2s_{0}k}\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}_{x}\varphi_{j}(|\xi|^{2}_{\pm})\varphi_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}.$ (2.38) Combining (2.35) and (2.38), together with Minkowski’s inequality, we have $\displaystyle\|\langle\partial_{t}\rangle^{s_{0}}S(t)u_{0}\|_{L^{2^{*}}(\mathbb{R},B^{ns_{0}}_{2^{*},2^{*}}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\left(\sum^{\infty}_{k=0}2^{(n+2)s_{0}k2^{*}}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\varphi_{k}\mathscr{F}u_{0}\|^{2^{*}}_{L^{2}_{x}(\mathbb{R}^{n})}\right)^{1/2^{*}}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|_{B^{(n+2)s_{0}}_{2,2^{*}}}.$ (2.39) In view of $H^{(n+2)s_{0}}\subset B^{(n+2)s_{0}}_{2,2^{*}}$ and Hölder’s inequality, we have for any $\varepsilon>0$, $\displaystyle\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|_{B^{(n+2)s_{0}}_{2,2^{*}}}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|_{H^{(n+2)s_{0}}}$ $\displaystyle\lesssim\left(\sum^{\infty}_{j=0}2^{2j\varepsilon}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|^{2}_{H^{(n+2)s_{0}}}\right)^{1/2}.$ (2.40) By Plancherel’s identity, and ${\rm supp}\varphi_{j}(|\xi|^{2}_{\pm})\subset\\{\xi:\;||\xi|^{2}_{\pm}|\in[2^{j-1},2^{j+1}]\\}$, we easily see that $\displaystyle\left(\sum^{\infty}_{j=0}2^{2j\varepsilon}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|^{2}_{H^{(n+2)s_{0}}}\right)^{1/2}$ $\displaystyle\lesssim\left(\sum^{\infty}_{j=0}\|\langle|\xi|^{2}_{\pm}\rangle^{\varepsilon}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|^{2}_{H^{(n+2)s_{0}}}\right)^{1/2}$ $\displaystyle\lesssim\left(\sum^{\infty}_{j=0}\|\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|^{2}_{H^{(n+2)s_{0}+2\varepsilon}}\right)^{1/2}$ $\displaystyle\lesssim\|u_{0}\|_{H^{(n+2)s_{0}+2\varepsilon}}.$ (2.41) Taking $s_{0}$ such that $(n+2)s_{0}+2\varepsilon<s$, from (2.39)–(2.41) we have the result, as desired. $\hfill\Box$ Next, we consider the estimates for the maximal function based on the frequency-uniform decomposition method. This issue has some relations with the Strichartz estimates in modulation spaces. Recently, the Strichartz estimates have been generalized to various function spaces, for instance, in the Wiener amalgam spaces [10, 11]. Recall that in [32], we obtained the following Strichartz estimate for a class of dispersive semi-groups in modulation spaces: $\displaystyle U(t)=\mathscr{F}^{-1}e^{{\rm i}tP(\xi)}\mathscr{F},$ (2.42) $P(\cdot):\mathbb{R}^{n}\to\mathbb{R}$ is a real valued function, which satisfies the following decay estimate $\displaystyle\|U(t)f\|_{M^{\alpha}_{p,q}}\lesssim(1+|t|)^{-\delta}\|f\|_{M_{p^{\prime},q}},$ (2.43) where $2\leq p<\infty$, $\alpha=\alpha(p)\in\mathbb{R}$, $\delta=\delta(p)>0$, $\alpha,\delta$ are independent of $t\in\mathbb{R}$. ###### Proposition 2.6 Let $U(t)$ satisfy (2.43) and (2.44). We have for any $\gamma\geq 2\vee(2/\delta)$, $\displaystyle\|U(t)f\|_{L^{\gamma}(\mathbb{R},M^{\alpha/2}_{p,1})}\lesssim\|f\|_{M_{2,1}}.$ (2.44) Recall that the hyperbolic Schrödinger semi-group $S(t)=e^{{\rm i}t\Delta_{\pm}}$ has the same decay estimate as that of the elliptic Schrödinger semi-group $e^{{\rm i}t\Delta}$: $\displaystyle\|S(t)u_{0}\|_{L^{\infty}(\mathbb{R}^{n})}\lesssim|t|^{-n/2}\|u_{0}\|_{L^{1}(\mathbb{R}^{n}))}.$ It follows that $\displaystyle\|S(t)u_{0}\|_{M_{\infty,1}}\lesssim|t|^{-n/2}\|u_{0}\|_{M_{1,1}}.$ (2.45) On the other hand, by Hausdorff-Young’s and Hölder’s inequalities we easily calculate that $\displaystyle\|\,\Box_{k}S(t)u_{0}\|_{L^{\infty}(\mathbb{R}^{n})}$ $\displaystyle\lesssim\sum_{\ell\in\Lambda}\|\mathscr{F}^{-1}\sigma_{k+\ell}\mathscr{F}\Box_{k}S(t)u_{0}\|_{L^{\infty}(\mathbb{R}^{n})}$ $\displaystyle\lesssim\sum_{\ell\in\Lambda}\|\sigma_{k+\ell}\mathscr{F}\Box_{k}u_{0}\|_{L^{1}(\mathbb{R}^{n})}\lesssim\|\Box_{k}u_{0}\|_{L^{1}(\mathbb{R}^{n})},$ where $\Lambda$ is as in (2.9). It follows that $\displaystyle\|S(t)u_{0}\|_{M_{\infty,1}}\lesssim\|u_{0}\|_{M_{1,1}}.$ (2.46) Hence, in view of (2.45) and (2.46), we have $\displaystyle\|S(t)u_{0}\|_{M_{\infty,1}}\lesssim(1+|t|)^{-n/2}\|u_{0}\|_{M_{1,1}}.$ (2.47) By Plancherel’s identity, one has that $\displaystyle\|S(t)u_{0}\|_{M_{2,1}}=\|u_{0}\|_{M_{2,1}}.$ (2.48) Hence, an interpolation between (2.47) and (2.48) yields (cf. [33]), $\displaystyle\|S(t)u_{0}\|_{M_{p,1}}\lesssim(1+|t|)^{-n(1/2-1/p)}\|u_{0}\|_{M_{p^{\prime},1}}.$ Applying Proposition 2.6, we immediately obtain that ###### Proposition 2.7 Let $2\leq p<\infty$, $2/\gamma(p)=n(1/2-1/p)$. We have for any $\gamma\geq 2\vee\gamma(p)$, $\displaystyle\|S(t)u_{0}\|_{L^{\gamma}(\mathbb{R},M^{\alpha/2}_{p,1})}\lesssim\|u_{0}\|_{M_{2,1}}.$ (2.49) In particular, if $p\geq 2+4/n:=2^{*}$, then $\displaystyle\|S(t)u_{0}\|_{L^{p}(\mathbb{R},M^{\alpha/2}_{p,1})}\lesssim\|u_{0}\|_{M_{2,1}}.$ (2.50) Let $\Lambda=\\{\ell\in\mathbb{Z}^{n}:\;{\rm supp}\,\sigma_{\ell}\cap{\rm supp}\,\sigma_{0}\not=\varnothing\\}$ be as in (2.9). Using the fact that $\Box_{k}\Box_{k+\ell}=0$ if $\ell\not\in\Lambda$, it is easy to see that (2.50) implies the following frequency-uniform estimates: $\displaystyle\|\,\Box_{k}S(t)u_{0}\|_{L^{p}_{t,x}(\mathbb{R}\times\mathbb{R}^{n})}\lesssim\|\,\Box_{k}u_{0}\|_{2},\quad k\in\mathbb{Z}^{n}.$ (2.51) Applying this estimate, we can get the following ###### Proposition 2.8 Let $p\geq 2+4/n:=2^{*}$ For any $s>(n+2)/p$, we have $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}S(t)u_{0}\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}\lesssim\|u_{0}\|_{M^{s}_{2,1}}.$ (2.52) Proof. Let us follow the proof of Proposition 2.5. Denote $\langle\partial_{t}\rangle=(I-\partial^{2}_{t})^{1/2}$. By Lemma 2.3, for any $s_{0}>1/p$, $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}S(t)u_{0}\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}$ $\displaystyle\lesssim\sum_{k\in\mathbb{Z}^{n}}\|\langle\partial_{t}\rangle^{s_{0}}S(t)\Box_{k}u_{0}\|_{L^{p}(\mathbb{R},B^{ns_{0}}_{p,p}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\sum_{k\in\mathbb{Z}^{n}}\|\langle\partial_{t}\rangle^{s_{0}}S(t)\Box_{k}u_{0}\|_{L^{p}(\mathbb{R},H^{ns_{0}}_{p}(\mathbb{R}^{n}))},$ (2.53) where we have used the fact that $H^{ns_{0}}_{p}(\mathbb{R}^{n})\subset B^{ns_{0}}_{p,p}(\mathbb{R}^{n})$. Since ${\rm supp}\sigma_{k}\subset B(k,\sqrt{n/2})$, applying Bernstein’s multiplier estimate, we get that $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\|\langle\partial_{t}\rangle^{s_{0}}S(t)\Box_{k}u_{0}\|_{L^{p}(\mathbb{R},H^{ns_{0}}_{p}(\mathbb{R}^{n}))}\lesssim\sum_{k\in\mathbb{Z}^{n}}\langle k\rangle^{ns_{0}}\|\langle\partial_{t}\rangle^{s_{0}}S(t)\Box_{k}u_{0}\|_{L^{p}_{t,x}(\mathbb{R}^{1+n})}.$ (2.54) Similarly as in (2.38), using (2.51), we have $\displaystyle\|\langle\partial_{t}\rangle^{s_{0}}S(t)\Box_{k}u_{0}\|_{L^{p}_{t,x}(\mathbb{R}^{1+n})}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}_{x}\langle|\xi|^{2}_{\pm}\rangle^{s_{0}}\varphi_{j}(|\xi|^{2}_{\pm})e^{{\rm i}t|\xi|^{2}_{\pm}}\sigma_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{p}_{t,x}(\mathbb{R}^{1+n})}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\langle k\rangle^{2s_{0}}\|\mathscr{F}^{-1}_{x}\varphi_{j}(|\xi|^{2}_{\pm})\sigma_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}.$ (2.55) In an analogous way as in (2.40) and (2.41), we obtain that $\displaystyle\sum^{\infty}_{j=0}\langle k\rangle^{2s_{0}}\|\mathscr{F}^{-1}_{x}\varphi_{j}(|\xi|^{2}_{\pm})\sigma_{k}(\xi)\mathscr{F}_{x}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}\lesssim\langle k\rangle^{2s_{0}+2\varepsilon}\|\,\Box_{k}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}.$ (2.56) Collecting (2.53)–(2.56), we have $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}S(t)u_{0}\|^{p}_{L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha})}\right)^{1/p}\lesssim\sum_{k\in\mathbb{Z}^{n}}\langle k\rangle^{(n+2)s_{0}+2\varepsilon}\|\,\Box_{k}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}.$ (2.57) Hence, by (2.57) we have (2.52). $\hfill\Box$ Using the ideas as in Lemma 2.3 and Proposition 2.5, we can show the following ###### Proposition 2.9 Let $p\geq 2+4/n:=2^{*}$. Let $2^{*}\leq r,q\leq\infty$, $s_{0}>1/2^{*}-1/q$, $s_{1}>n(1/2^{*}-1/r)$. Then we have $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,S(t)u_{0}\|^{p}_{L^{q}(\mathbb{R},\,L^{r}(Q_{\alpha}))}\right)^{1/p}\lesssim\|u_{0}\|_{H^{s_{1}+2s_{0}}}.$ (2.58) In particular, for any $q,p\geq 2^{*}$, $s>n/2-2/q$, $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,S(t)u_{0}\|^{p}_{L^{q}(\mathbb{R},\,L^{\infty}(Q_{\alpha}))}\right)^{1/p}\lesssim\|u_{0}\|_{H^{s}}.$ (2.59) Sketch of Proof. In view of $\ell^{2^{*}}\subset\ell^{p}$, it suffices to consider the case $p=2^{*}$. Using the inclusions $H^{s_{0}}_{p}(\mathbb{R})\subset L^{q}(\mathbb{R})$ and $B^{s_{1}}_{p,p}(\mathbb{R}^{n})\subset L^{r}(\mathbb{R}^{n})$, we have $\displaystyle\|u\|_{L^{q}(\mathbb{R},\,L^{r}(Q_{\alpha}))}\lesssim\|(I-\partial^{2}_{t})^{s_{0}/2}\sigma_{\alpha}u\|_{L^{p}(\mathbb{R},B^{s_{1}}_{p,p}(\mathbb{R}^{n}))}.$ (2.60) Using the same way as in Lemma 2.3, we can show that $\displaystyle\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|u\|^{p}_{L^{q}(\mathbb{R},\,L^{r}(Q_{\alpha}))}\right)^{1/p}\lesssim\|(I-\partial^{2}_{t})^{s_{0}/2}u\|_{L^{p}(\mathbb{R},B^{s_{1}}_{p,p}(\mathbb{R}^{n}))}.$ (2.61) One can repeat the procedures as in the proof of Lemma 2.3 to conclude that $\displaystyle\sum_{\alpha\in\mathbb{Z}^{n}}\|(I-\partial^{2}_{t})^{s_{0}/2}\sigma_{\alpha}S(t)u_{0}\|^{p}_{L^{p}(\mathbb{R},B^{s_{1}}_{p,p}(\mathbb{R}^{n}))}\lesssim\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|_{H^{s_{1}+2s_{0}}(\mathbb{R}^{n})}.$ (2.62) Applying an analogous way as in the proof of Proposition 2.5, $\displaystyle\sum^{\infty}_{j=0}\|\mathscr{F}^{-1}\varphi_{j}(|\xi|^{2}_{\pm})\mathscr{F}u_{0}\|_{H^{s_{1}+2s_{0}}(\mathbb{R}^{n})}\lesssim\|u_{0}\|_{H^{s_{1}+2s_{0}+2\varepsilon}}.$ (2.63) Collecting (2.61) and (2.63), we immediately get (2.58). $\hfill\Box$ ###### Proposition 2.10 For any $q\geq p\geq 2^{*}$, $s>(n+2)/p-2/q$, $\displaystyle\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}S(t)u_{0}\|^{p}_{L^{q}(\mathbb{R},\,L^{\infty}(Q_{\alpha}))}\right)^{1/p}\lesssim\|u_{0}\|_{M^{s}_{2,1}}.$ (2.64) ## 3 Global-local estimates on time-space ### 3.1 Time-global and space-local Strichartz estimates We need some modifications to the Strichartz estimates, which are global on time variable and local on spatial variable. We always denote by $S(t)$ and $\mathscr{A}$ the generalized Schrödinger semi-group and the integral operator as in (1.13). ###### Proposition 3.1 Let $n\geq 3$. Then we have $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\|S(t)u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{2},$ (3.1) $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|\mathscr{A}F\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|F\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (3.2) $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|\mathscr{A}F\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|F\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}.$ (3.3) Proof. In view of Hölder’s inequality and the endpoint Strichartz estimate, $\displaystyle\|S(t)u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|S(t)u_{0}\|_{L^{2}_{t}L^{2n/(n-2)}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\leq\|S(t)u_{0}\|_{L^{2}_{t}L^{2n/(n-2)}_{x}(\mathbb{R}\times\mathbb{R}^{n})}$ $\displaystyle\lesssim\|u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}.$ (3.4) Using the above ideas and the following Strichartz estimate $\displaystyle\left\|\mathscr{A}F\right\|_{L^{2}_{t}L^{2n/(n-2)}_{x}(\mathbb{R}\times\mathbb{R}^{n})}$ $\displaystyle\lesssim\|F\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times\mathbb{R}^{n})},$ (3.5) $\displaystyle\left\|\mathscr{A}F\right\|_{L^{2}_{t}L^{2n/(n-2)}_{x}(\mathbb{R}\times\mathbb{R}^{n})}$ $\displaystyle\lesssim\|F\|_{L^{2}_{t}L^{2n/(n+2)}_{x}(\mathbb{R}\times\mathbb{R}^{n})},$ (3.6) one can easily get (3.2) and (3.3). $\hfill\Box$ Since the endpoint Strichartz estimates used in the proof of Proposition 3.1 only holds for $n\geq 3$, it is not clear for us if (3.1) still hold for $n=2$. This is why we have an additional condition that $u_{0}\in\dot{H}^{-1/2}$ is small in 2D. However, we have the following (see [17]) ###### Proposition 3.2 Let $n=2$. Then we have for any $1\leq r<4/3$, $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|S(t)u_{0}\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\min\left(\|(-\Delta)^{-1/4}u_{0}\|_{2},\;\|u_{0}\|_{L^{2}\cap L^{r}(\mathbb{R}^{n})}\right).$ (3.7) In the low frequency case, one easily sees that (3.7) is strictly weak than (3.1). Proof. By Lemma 3.4, it suffices to show $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|S(t)u_{0}\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{L^{2}\cap L^{r}(\mathbb{R}^{n})}.$ (3.8) Using the unitary property in $L^{2}$ and the $L^{p}-L^{p^{\prime}}$ decay estimates of $S(t)$, we have $\displaystyle\left\|S(t)u_{0}\right\|_{L^{2}_{x}(Q_{\alpha})}\lesssim(1+|t|)^{1-2/r}\|u_{0}\|_{L^{2}\cap L^{r}(\mathbb{R}^{n})}.$ (3.9) Taking the $L^{2}_{t}$ norm in both sides of (3.9), we immediately get (3.8). Hence, the result follows. $\hfill\Box$ ###### Proposition 3.3 Let $n=2$. Then we have $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|\mathscr{A}F\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{2}}\|F\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (3.10) Proof. We notice that $\displaystyle\|S(t)f\|_{L^{2}_{x}(Q_{\alpha})}$ $\displaystyle\lesssim(1+|t|)^{-1}\|f\|_{L^{1}_{x}\cap L^{2}_{x}(\mathbb{R}^{n})}.$ (3.11) It follows that $\displaystyle\ \left\|\mathscr{A}F\right\|_{L^{2}_{x}(Q_{\alpha})}$ $\displaystyle\lesssim\int_{\mathbb{R}}(1+|t-\tau|)^{-1}\|F(\tau)\|_{L^{1}_{x}\cap L^{2}_{x}(\mathbb{R}^{n})}d\tau.$ (3.12) Using Young’s inequality, one has that $\displaystyle\ \left\|\mathscr{A}F\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|F\|_{L^{1}(\mathbb{R},\,L^{1}_{x}\cap L^{2}_{x}(\mathbb{R}^{n}))}.$ (3.13) In view of Hölder’s inequality, (3.13) yields the result, as desired. $\hfill\Box$ ### 3.2 Note on the time-global and space-local smooth effects Kenig, Ponce and Vega [16, 17] obtained the local smooth effect estimates for the Schrödinger group $e^{{\rm i}t\Delta}$, and their results can also be developed to the non-elliptical Schrödinger group $e^{{\rm i}t\Delta_{\pm}}$ ([18]). On the basis of their results and Proposition 3.1, we can obtain a time-global version of the local smooth effect estimates with the nonhomogeneous derivative $(I-\Delta)^{1/2}$ instead of homogeneous derivative $\nabla$, which is useful to control the low frequency parts of the nonlinearity. ###### Lemma 3.4 ([16]) Let $\Omega$ be an open set in $\mathbb{R}^{n}$, $\phi$ be a $C^{1}(\Omega)$ function such that $\nabla\phi(\xi)\not=0$ for any $\xi\in\Omega$. Assume that there is $N\in\mathbb{N}$ such that for any $\bar{\xi}:=(\xi_{1},...,\xi_{n-1})\in\mathbb{R}^{n-1}$ and $r\in\mathbb{R}$, the equation $\phi(\xi_{1},...,\xi_{k},x,\xi_{k+1},...,\xi_{n-1})=r$ has at most $N$ solutions. For $a(x,s)\in L^{\infty}(\mathbb{R}^{n}\times\mathbb{R})$ and $f\in\mathscr{S}(\mathbb{R}^{n})$, we denote $\displaystyle W(t)f(x)=\int_{\Omega}e^{{\rm i}(t\phi(\xi)+x\xi)}a(x,\phi(\xi))\hat{f}(\xi)d\xi.$ (3.14) Then for $n\geq 2$, we have $\displaystyle\|W(t)f\|_{L^{2}_{t,x}(\mathbb{R}\times B(0,R))}\leq CNR^{1/2}\||\nabla\phi|^{-1/2}\hat{f}\|_{L^{2}(\Omega)}.$ (3.15) ###### Corollary 3.5 Let $n\geq 3$, $S(t)=e^{{\rm i}t\Delta_{\pm}}$. We have $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\|S(t)u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{H^{-1/2}},$ (3.16) $\displaystyle\left\|\mathscr{A}f\right\|_{L^{\infty}(\mathbb{R},H^{1/2})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}.$ (3.17) For $n=2$, (3.17) also holds if one substitutes $H^{1/2}$ by $\dot{H}^{1/2}$. Proof. Let $\Omega=\mathbb{R}^{n}\setminus B(0,1)$, $\phi(\xi)=|\xi|^{2}_{\pm}$ and $\psi$ be as in (1.35), $a(x,s)=1-\psi(s)$ in Lemma 3.4. Taking $W(t):=S(t)\mathscr{F}^{-1}(1-\psi)\mathscr{F}$, from (3.15) we have $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\|S(t)\mathscr{F}^{-1}(1-\psi)\mathscr{F}u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}\lesssim\||\xi|^{-1/2}\hat{u}_{0}\|_{L^{2}_{\xi}(\mathbb{R}^{n}\setminus B(0,1))}.$ (3.18) It follows from Proposition 3.1 that $\displaystyle\|S(t)\mathscr{F}^{-1}\psi\mathscr{F}u_{0}\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|\mathscr{F}^{-1}\psi\mathscr{F}u_{0}\|_{L^{2}_{x}(\mathbb{R}^{n})}$ $\displaystyle\lesssim\|\hat{u}_{0}\|_{L^{2}_{\xi}(B(0,2))}.$ (3.19) From (3.18) and (3.19) we have (3.16), as desired. (3.17) is the dual version of (3.16). $\hfill\Box$ When $n=2$, it is known that for the elliptic case, the endpoint Strichartz estimate holds for the radial function (cf. [28]). So, Corollary 3.5 also holds for the radial function $u_{0}$ in the elliptic case. The following local smooth effect estimates for the nonhomogeneous part of the solutions of the Schrödinger equation is also due to Kenig, Ponce and Vega [17]222In [17], the result was stated for the elliptic case, however, their result is also adapted to the non-elliptic cases.. ###### Proposition 3.6 Let $n\geq 2$, $S(t)=e^{{\rm i}t\Delta_{\pm}}$. We have $\displaystyle\sup_{\alpha\in\mathbb{Z}^{n}}\left\|\nabla\mathscr{A}f\right\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}.$ (3.20) ## 4 Proof of Theorem 1.1 ###### Lemma 4.1 (Sobolev Inequality). Let $\Omega\subset\mathbb{R}^{n}$ be a bounded domain with $\partial\Omega\in C^{m}$, $m,\ell\in\mathbb{N}\cup\\{0\\}$, $1\leq r,p,q\leq\infty$. Assume that $\frac{\ell}{m}\leq\theta\leq 1,\ \ \frac{1}{p}-\frac{\ell}{n}=\theta\left(\frac{1}{r}-\frac{m}{n}\right)+\frac{1-\theta}{q}.$ Then we have $\displaystyle\sum_{|\beta|=\ell}\|D^{\beta}u\|_{L^{p}(\Omega)}\lesssim\|u\|^{1-\theta}_{L^{q}(\Omega)}\|u\|^{\theta}_{W^{m}_{r}(\Omega)},$ (4.1) where $\|u\|_{W^{m}_{r}(\Omega)}=\sum_{|\beta|\leq m}\|u\|_{L^{r}(\Omega)}$. Proof of Theorem 1.1. In order to illustrate our ideas in an exact way, we first consider a simple case $s=[n/2]+5/2$ and there is no difficulty to generalize the proof to the case $s>n/2+3/2$, $s+1/2\in\mathbb{N}$. We assume without loss of generality that $\displaystyle F(u,\,\bar{u},\,\nabla u,\,\nabla\bar{u}):=F(u,\nabla u)=\sum_{\Lambda_{\kappa,\nu}}c_{\kappa\nu_{1}...\nu_{n}}u^{\kappa}u^{\nu_{1}}_{x_{1}}...u^{\nu_{n}}_{x_{n}},$ (4.2) where $\Lambda_{\kappa,\nu}=\\{(\kappa,\nu_{1},...,\nu_{n}):\,m+1\leq\kappa+\nu_{1}+...+\nu_{n}\leq M+1\\}.$ Since we only use the Sobolev norm to control the nonlinear terms, $\bar{u}$ and $u$ have the same norm, whence, the general cases can be handled in the same way. Denote $\displaystyle\lambda_{1}(v):=\|v\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{2}(v):=\|v\|_{\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{3}(v):=\|v\|_{\ell^{2^{*}}_{\alpha}(L^{2m}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ Put $\displaystyle\mathscr{D}_{n}=\left\\{u:\;\sum_{|\beta|\leq[n/2]+3}\lambda_{1}(D^{\beta}u)+\sum_{|\beta|\leq 1}\sum_{i=2,3}\lambda_{i}(D^{\beta}u)\leq\varrho\right\\}.$ (4.3) We consider the mapping $\displaystyle\mathscr{T}:u(t)\to S(t)u_{0}-{\rm i}\mathscr{A}F(u,\,\nabla u),$ (4.4) and we show that $\mathscr{T}:\mathscr{D}_{n}\to\mathscr{D}_{n}$ is a contraction mapping for any $n\geq 2$. Step 1\. For any $u\in\mathscr{D}_{n}$, we estimate $\lambda_{1}(D^{\beta}\mathscr{T}u)$, $|\beta|\leq 3+[n/2]$. We consider the following three cases. Case 1\. $n\geq 3$ and $1\leq|\beta|\leq 3+[n/2]$. In view of Corollary 3.5 and Proposition 3.6, we have for any $\beta$, $1\leq|\beta|\leq 3+[n/2]$, $\displaystyle\lambda_{1}(D^{\beta}\mathscr{T}u)$ $\displaystyle\lesssim\|S(t)D^{\beta}u_{0}\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}+\sum_{\Lambda_{\kappa,\nu}}\|\mathscr{A}D^{\beta}(u^{\kappa}u^{\nu_{1}}_{x_{1}}...u^{\nu_{n}}_{x_{n}})\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{H^{s}}+\sum_{|\beta|\leq 2+[n/2]}\sum_{\Lambda_{\kappa,\nu}}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}(u^{\kappa}u^{\nu_{1}}_{x_{1}}...u^{\nu_{n}}_{x_{n}})\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}.$ (4.5) For simplicity, we can further assume that $u^{\kappa}u^{\nu_{1}}_{x_{1}}...u^{\nu_{n}}_{x_{n}}=u^{\kappa}u^{\nu}_{x_{1}}$ in (4.5) and the general case can be treated in an analogous way333One can see below for a general treating.. So, one can rewrite (4.5) as $\displaystyle\sum_{1\leq|\beta|\leq 3+[n/2]}\lambda_{1}(D^{\beta}\mathscr{T}u)\lesssim\|u_{0}\|_{H^{s}}+\sum_{|\beta|\leq 2+[n/2]}\sum_{\Lambda_{\kappa,\nu}}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}(u^{\kappa}u^{\nu}_{x_{1}})\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}.$ (4.6) It is easy to see that $\displaystyle|D^{\beta}(u^{\kappa}u^{\nu}_{x_{1}})|\lesssim\sum_{\beta_{1}+...+\beta_{\kappa+\nu}=\beta}|D^{\beta_{1}}u....D^{\beta_{\kappa}}uD^{\beta_{\kappa+1}}u_{x_{1}}...D^{\beta_{\kappa+\nu}}u_{x_{1}}|.$ (4.7) By Hölder’s inequality, $\displaystyle\|D^{\beta}(u^{\kappa}u^{\nu}_{x_{1}})\|_{L^{2}_{x}(Q_{\alpha})}\lesssim\sum_{\beta_{1}+...+\beta_{\kappa+\nu}=\beta}\prod^{\kappa}_{i=1}\|D^{\beta_{i}}u\|_{L^{p_{i}}_{x}(Q_{\alpha})}\prod^{\kappa+\nu}_{i=\kappa+1}\|D^{\beta_{i}}u_{x_{1}}\|_{L^{p_{i}}_{x}(Q_{\alpha})},$ (4.8) where $p_{i}=\left\\{\begin{array}[]{ll}2|\beta|/|\beta_{i}|,&|\beta_{i}|\geq 1,\\\ \infty,&|\beta_{i}|=0.\end{array}\right.$ It is easy to see that for $\theta_{i}=|\beta_{i}|/|\beta|$, $\frac{1}{p_{i}}-\frac{|\beta_{i}|}{n}=\theta_{i}\left(\frac{1}{2}-\frac{|\beta|}{n}\right)+\frac{1-\theta_{i}}{\infty}.$ Using Sobolev’s inequality, one has that for $B_{\alpha}:=\\{x:\;|x-\alpha|\leq\sqrt{n}\\}$, $\displaystyle\|D^{\beta_{i}}u\|_{L^{p_{i}}_{x}(Q_{\alpha})}\leq\|D^{\beta_{i}}u\|_{L^{p_{i}}_{x}(B_{\alpha})}\lesssim\|u\|^{1-\theta_{i}}_{L^{\infty}_{x}(B_{\alpha})}\|u\|^{\theta_{i}}_{W^{|\beta|}_{2}(B_{\alpha})},\ \ i=1,...,\kappa;$ (4.9) $\displaystyle\|D^{\beta_{i}}u_{x_{1}}\|_{L^{p_{i}}_{x}(Q_{\alpha})}\lesssim\|u_{x_{1}}\|^{1-\theta_{i}}_{L^{\infty}_{x}(B_{\alpha})}\|u_{x_{1}}\|^{\theta_{i}}_{W^{|\beta|}_{2}(B_{\alpha})},\ \ i=\kappa+1,...,\kappa+\nu.$ (4.10) Since $\sum^{\kappa+\nu}_{i=1}\theta_{i}=1,\quad\sum^{\kappa+\nu}_{i=1}(1-\theta_{i})=\kappa+\nu-1,$ by (4.8)–(4.10) we have $\displaystyle\|D^{\beta}(u^{\kappa}u^{\nu}_{x_{1}})\|_{L^{2}_{x}(Q_{\alpha})}\lesssim$ $\displaystyle\sum_{|\beta|\leq 2+[n/2]}(\|u\|_{W^{|\beta|}_{2}(B_{\alpha})}+\|u_{x_{1}}\|_{W^{|\beta|}_{2}(B_{\alpha})})$ $\displaystyle\times(\|u\|^{\kappa+\nu-1}_{L^{\infty}_{x}(B_{\alpha})}+\|u_{x_{1}}\|^{\kappa+\nu-1}_{L^{\infty}_{x}(B_{\alpha})})$ $\displaystyle\lesssim$ $\displaystyle\sum_{|\gamma|\leq 3+[n/2]}\|D^{\gamma}u\|_{L^{2}_{x}(B_{\alpha})}\sum_{|\beta|\leq 1}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{\infty}_{x}(B_{\alpha})}.$ (4.11) It follows from (4.11) and $\ell^{2^{*}}\subset\ell^{\kappa+\nu-1}$ that $\displaystyle\\!\\!\\!\\!\sum_{|\beta|\leq 2+[n/2]}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}(u^{\kappa}u^{\nu}_{x_{1}})\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\sum_{|\gamma|\leq 3+[n/2]}\|D^{\gamma}u\|_{L^{2}_{t,x}(\mathbb{R}\times B_{\alpha})}\sum_{|\beta|\leq 1}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{\infty}_{t,x}(\mathbb{R}\times B_{\alpha})}$ $\displaystyle\lesssim\sum_{|\gamma|\leq 3+[n/2]}\lambda_{1}(D^{\gamma}u)\sum_{|\beta|\leq 1}\lambda_{2}(D^{\beta}u)^{\kappa+\nu-1}\lesssim\varrho^{\kappa+\nu}.$ (4.12) Hence, in view of (4.6) and (4.12) we have $\displaystyle\sum_{1\leq|\beta|\leq 3+[n/2]}\lambda_{1}(D^{\beta}\mathscr{T}u)\lesssim\|u_{0}\|_{H^{s}}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.13) Case 2\. $n\geq 3$ and $|\beta|=0$. By Corollary 3.5, the local Strichartz estimate (3.2) and Hölder’s inequality, $\displaystyle\lambda_{1}(\mathscr{T}u)$ $\displaystyle\lesssim\|S(t)u_{0}\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}+\|\mathscr{A}F(u,\nabla u)\|_{\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{2}+\sum_{\alpha\in\mathbb{Z}^{n}}\|F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{2}+\sum_{\Lambda_{\kappa,\nu}}\sum_{\alpha\in\mathbb{Z}^{n}}\|u^{\kappa}u^{\nu_{1}}_{x_{1}}...u^{\nu_{n}}_{x_{n}}\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{2}+\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 1}\sup_{\alpha\in\mathbb{Z}^{n}}\|D^{\gamma}u\|_{L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \ \times\sum_{|\beta|\leq 1}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{2(\kappa+\nu-1)}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\|u_{0}\|_{2}+\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 1}\lambda_{1}(D^{\gamma}u)\sum_{i=2,3}\sum_{|\beta|\leq 1}\lambda_{i}(D^{\beta}u)^{\kappa+\nu-1}$ $\displaystyle\lesssim\|u_{0}\|_{2}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.14) Case 3\. $n=2,\;|\beta|=0$. By Propositions 3.2 and 3.3, we have $\displaystyle\lambda_{1}(\mathscr{T}u)$ $\displaystyle\lesssim\|u_{0}\|_{\dot{H}^{-1/2}}+\sum_{\alpha\in\mathbb{Z}^{n}}\|F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (4.15) Using the same way as in Case 2, we have $\displaystyle\lambda_{1}(\mathscr{T}u)$ $\displaystyle\lesssim\|u_{0}\|_{\dot{H}^{-1/2}}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.16) Step 2\. We consider the estimates of $\lambda_{2}(D^{\beta}\mathscr{T}u)$, $|\beta|\leq 1$. Using the estimates of the maximal function as in Proposition 2.5, we have for $|\beta|\leq 1$, $0<\varepsilon\ll 1$, $\displaystyle\lambda_{2}(D^{\beta}\mathscr{T}u)$ $\displaystyle\lesssim\|S(t)D^{\beta}u_{0}\|_{\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}+\|\mathscr{A}D^{\beta}F(u,\nabla u)\|_{\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|D^{\beta}u_{0}\|_{H^{n/2+\varepsilon}}+\sum_{|\beta|\leq 1}\|D^{\beta}F(u,\nabla u)\|_{L^{1}(\mathbb{R},H^{n/2+\varepsilon}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\|u_{0}\|_{H^{n/2+1+\varepsilon}}+\sum_{|\beta|\leq[n/2]+2}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (4.17) Applying the same way as in Step 1, for any $|\beta|\leq[n/2]+2$, $\displaystyle\|D^{\beta}F(u,\nabla u)\|_{L^{2}_{x}(Q_{\alpha})}\lesssim\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\beta|\leq 1}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{\infty}_{x}(B_{\alpha})}\sum_{|\gamma|\leq 3+[n/2]}\|D^{\gamma}u\|_{L^{2}_{x}(B_{\alpha})}.$ (4.18) By Hölder’s inequality, we have from (4.18) that $\displaystyle\|D^{\beta}F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}\lesssim$ $\displaystyle\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 3+[n/2]}\|D^{\gamma}u\|_{L^{2}_{t,x}(\mathbb{R}\times B_{\alpha})}$ $\displaystyle\ \ \ \ \ \times\sum_{|\beta|\leq 1}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{2(\kappa+\nu-1)}_{t}L^{\infty}_{x}(\mathbb{R}\times B_{\alpha})}.$ (4.19) Summarizing (4.19) over all $\alpha\in\mathbb{Z}^{n}$, we have for any $|\beta|\leq 2+[n/2]$, $\displaystyle\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 3+[n/2]}\lambda_{1}(D^{\gamma}u)\sum_{|\beta|\leq 1}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}u\|^{\kappa+\nu-1}_{L^{2(\kappa+\nu-1)}_{t}L^{\infty}_{x}(\mathbb{R}\times B_{\alpha})}$ $\displaystyle\lesssim\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 3+[n/2]}\lambda_{1}(D^{\gamma}u)\sum_{|\beta|\leq 1}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}u\|^{\kappa+\nu-1}_{(L^{2m}_{t}L^{\infty}_{x})\cap L^{\infty}_{t,x}(\mathbb{R}\times B_{\alpha})}$ $\displaystyle\lesssim\sum^{M+1}_{\kappa+\nu=m+1}\sum_{|\gamma|\leq 3+[n/2]}\lambda_{1}(D^{\gamma}u)\sum_{|\beta|\leq 1}(\lambda_{2}(D^{\beta}u)^{\kappa+\nu-1}+\lambda_{3}(D^{\beta}u)^{\kappa+\nu-1})$ $\displaystyle\lesssim\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.20) Combining (4.17) with (4.20), we obtain that $\displaystyle\sum_{|\beta|\leq 1}\lambda_{2}(D^{\beta}\mathscr{T}u)\lesssim\|u_{0}\|_{H^{n/2+1+\varepsilon}}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.21) Step 3\. We estimate $\lambda_{3}(D^{\beta}\mathscr{T}u)$, $|\beta|\leq 1$. In view of Proposition 2.9, one has that $\displaystyle\lambda_{3}(D^{\beta}\mathscr{T}u)$ $\displaystyle\lesssim\|S(t)D^{\beta}u_{0}\|_{\ell^{2^{*}}_{\alpha}(L^{2m}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}+\|\mathscr{A}D^{\beta}F(u,\nabla u)\|_{\ell^{2^{*}}_{\alpha}(L^{2m}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|D^{\beta}u_{0}\|_{H^{n/2-1/m+\varepsilon}}+\sum_{|\beta|\leq 1}\|D^{\beta}F(u,\nabla u)\|_{L^{1}(\mathbb{R},H^{n/2-1/m+\varepsilon}(\mathbb{R}^{n}))}$ $\displaystyle\lesssim\|u_{0}\|_{H^{n/2+1}}+\sum_{|\beta|\leq[n/2+2]}\sum_{\alpha\in\mathbb{Z}^{n}}\|D^{\beta}F(u,\nabla u)\|_{L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha})},$ (4.22) which reduces to the case as in (4.17). Therefore, collecting the estimates as in Steps 1–3, we have for $n\geq 3$, $\displaystyle\sum_{|\beta|\leq 3+[n/2]}\lambda_{1}(D^{\beta}\mathscr{T}u)+\sum_{i=2,3}\sum_{|\beta|\leq 1}\lambda_{i}(D^{\beta}\mathscr{T}u)\lesssim\|u_{0}\|_{H^{s}}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu},$ (4.23) and for $n\geq 2$, $\displaystyle\sum_{|\beta|\leq 4}\lambda_{1}(D^{\beta}\mathscr{T}u)+\sum_{i=2,3}\sum_{|\beta|\leq 1}\lambda_{i}(D^{\beta}\mathscr{T}u)\lesssim\|u_{0}\|_{H^{7/2}\cap\dot{H}^{-1/2}}+\sum^{M+1}_{\kappa+\nu=m+1}\varrho^{\kappa+\nu}.$ (4.24) It follows that for $n\geq 3,$ $\mathscr{T}:\mathscr{D}_{n}\to\mathscr{D}_{n}$ is a contraction mapping if $\varrho$ and $\|u_{0}\|_{H^{s}}$ are small enough (similarly for $n=2$). Before considering the case $s>n/2+3/2$, we first establish a nonlinear mapping estimate: ###### Lemma 4.2 Let $n\geq 2$, $s>0$, $K\in\mathbb{N}$. Let $1\leq p,p_{i},q,q_{i}\leq\infty$ satisfy $1/p=1/p_{1}+(K-1)/p_{2}$ and $1/q=1/q_{1}+(K-1)/q_{2}$. We have $\displaystyle\|v_{1}...v_{K}\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha}))}\lesssim$ $\displaystyle\sum^{K}_{k=1}\|v_{k}\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{1}}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\ \times\prod_{i\not=k,\,i=1,...,K}\|v_{i}\|_{\ell^{1}_{\triangle}\ell^{K-1}_{\alpha}(L^{q_{2}}_{t}L^{p_{2}}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.25) Proof. Denote $S_{r}u=\sum_{j\leq r}\triangle u$. We have $\displaystyle v_{1}...v_{K}=\sum^{\infty}_{r=-1}(S_{r+1}v_{1}...S_{r+1}v_{K}-S_{r}v_{1}...S_{r}v_{K}),$ (4.26) where we assume that $S_{-1}v\equiv 0$. Recall the identity, $\displaystyle\prod^{K}_{k=1}a_{k}-\prod^{K}_{k=1}b_{k}=\sum^{K}_{k=1}(a_{k}-b_{k})\prod_{i\leq k-1}b_{i}\prod_{i\geq k+1}a_{i},$ (4.27) where we assume that $\prod_{i\leq 0}a_{i}=\prod_{i\geq K+1}\equiv 1$. We have $\displaystyle v_{1}...v_{K}=\sum^{\infty}_{r=-1}\sum^{K}_{k=1}\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i}.$ (4.28) Hence, it follows that $\displaystyle\|v_{1}...v_{K}\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle=\sum^{\infty}_{j=0}2^{sj}\sum_{\alpha\in\mathbb{Z}^{n}}\|\triangle_{j}(v_{1}...v_{K})\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum^{K}_{k=1}\sum^{\infty}_{j=0}2^{sj}\sum_{\alpha\in\mathbb{Z}^{n}}\sum^{\infty}_{r=-1}\left\|\triangle_{j}(\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i})\right\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (4.29) Using the support property of $\widehat{\triangle_{r}v}$ and $\widehat{S_{r}v}$, we see that $\displaystyle\triangle_{j}(\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i})\equiv 0,\ \ j>r+C.$ (4.30) Using the fact $\|\int fd\mu\|_{X}\leq\int\|f\|_{X}d\mu$, one has that $\displaystyle\sum_{\alpha\in\mathbb{Z}^{n}}\|\triangle_{j}f\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\leq\sum_{\alpha\in\mathbb{Z}^{n}}\int_{\mathbb{R}^{n}}|\mathscr{F}^{-1}\varphi_{j}(y)|\|f(t,x-y)\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}dy$ $\displaystyle\leq\sup_{y\in\mathbb{R}^{n}}\sum_{\alpha\in\mathbb{Z}^{n}}\|f(t,x-y)\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}\int_{\mathbb{R}^{n}}|\mathscr{F}^{-1}\varphi_{j}(y)|dy$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}.$ (4.31) Collecting (4.29)–(4.31) and using Fubini’s Theorem, we have $\displaystyle\|v_{1}...v_{K}\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\sum^{K}_{k=1}\sum^{\infty}_{r=-1}\sum_{j\leq r+C}2^{sj}\sum_{\alpha\in\mathbb{Z}^{n}}\left\|\triangle_{j}(\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i})\right\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum^{K}_{k=1}\sum^{\infty}_{r=-1}\sum_{j\leq r+C}2^{sj}\sum_{\alpha\in\mathbb{Z}^{n}}\left\|\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i}\right\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum^{K}_{k=1}\sum^{\infty}_{r=-1}2^{sr}\sum_{\alpha\in\mathbb{Z}^{n}}\left\|\triangle_{r+1}v_{k}\prod^{k-1}_{i=1}S_{r}v_{i}\prod^{K}_{i=k+1}S_{r+1}v_{i}\right\|_{L^{q}_{t}L^{p}_{x}(\mathbb{R}\times Q_{\alpha})}$ $\displaystyle\lesssim\sum^{K}_{k=1}\sum^{\infty}_{r=-1}2^{sr}\sum_{\alpha\in\mathbb{Z}^{n}}\|\triangle_{r+1}v_{k}\|_{L^{q_{1}}_{t}L^{p_{1}}_{x}(\mathbb{R}\times Q_{\alpha})}\prod_{i\not=k,\ i=1,...,K}\|v_{i}\|_{\ell^{1}_{\triangle}(L^{q_{2}}_{t}L^{p_{2}}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\sum^{K}_{k=1}\|v_{k}\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{1}}_{x}(\mathbb{R}\times Q_{\alpha}))}\sum_{\alpha\in\mathbb{Z}^{n}}\prod_{i\not=k,\ i=1,...,K}\|v_{i}\|_{\ell^{1}_{\triangle}(L^{q_{2}}_{t}L^{p_{2}}_{x}(\mathbb{R}\times Q_{\alpha}))},$ (4.32) the result follows. $\hfill\Box$ For short, we write $\|\nabla u\|_{X}=\|\partial_{x_{1}}u\|_{X}+...+\|\partial_{x_{n}}u\|_{X}$. ###### Lemma 4.3 Let $n\geq 3$. We have for any $s>0$ $\displaystyle\sum_{k=0,1}\|S(t)\nabla^{k}u_{0}\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{B^{s+1/2}_{2,1}},$ (4.33) $\displaystyle\sum_{k=0,1}\|\mathscr{A}\nabla^{k}F\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|F\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.34) Proof. In view of Corollary 3.5 and Propositions 3.1 and 3.6, we have the results, as desired. $\hfill\Box$ ###### Lemma 4.4 Let $n=2$. We have for any $s>0$ $\displaystyle\sum_{k=0,1}\|S(t)\nabla^{k}u_{0}\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{B^{s+1/2}_{2,1}\cap\dot{H}^{-1/2}},$ (4.35) $\displaystyle\|\mathscr{A}\nabla F\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|F\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ (4.36) $\displaystyle\|\mathscr{A}F\|_{\ell^{1,s}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\|F\|_{\ell^{1,s}_{\triangle}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.37) Proof. By Propositions 3.2, 3.3 and 3.6, we have the results, as desired. $\hfill\Box$ We now continue the proof of Theorem 1.1 and now we consider the general case $s>n/2+3/2$. We write $\displaystyle\lambda_{1}(v):=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1,s-1/2}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{2}(v):=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1,s-1/2}_{\triangle}\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{3}(v):=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1,s-1/2}_{\triangle}\ell^{2^{*}}_{\alpha}(L^{2m}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\mathscr{D}=\\{u:\sum_{i=1,2,3}\lambda_{i}(v)\leq\varrho\\}.$ (4.38) Note $\lambda_{i}$ and $\mathscr{D}$ defined here are different from those in the above. We only give the details of the proof in the case $n\geq 3$ and the case $n=2$ can be shown in a slightly different way. Let $\mathscr{T}$ be defined as in (4.4). Using Lemma 4.3, we have $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{B^{s}_{2,1}}+\|F\|_{\ell^{1,s-1/2}_{\triangle}\ell^{1}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.39) For simplicity, we write $\displaystyle(u)^{\kappa}(\nabla u)^{\nu}=u^{\kappa_{1}}\bar{u}^{\kappa_{2}}u^{\nu_{1}}_{x_{1}}\bar{u}^{\nu_{2}}_{x_{1}}...u^{\nu_{2n-1}}_{x_{n}}\bar{u}^{\nu_{2n}}_{x_{n}},$ (4.40) $|\kappa|=\kappa_{1}+\kappa_{2}$, $|\nu|=\nu_{1}+...+\nu_{2n}$. By Lemma 4.2, we have $\displaystyle\|(u)^{\kappa}(\nabla u)^{\nu}\|_{\ell^{1,s-1/2}_{\triangle}\ell^{1}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\sum_{i=0,1}\|\nabla^{i}u\|_{\ell^{1,s-1/2}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}\sum_{k=0,1}\|\nabla^{k}u\|^{|\kappa|+|\nu|-1}_{\ell^{1}_{\triangle}\ell^{|\kappa|+|\nu|-1}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.41) Hence, if $u\in\mathscr{D}$, in view of (4.39) and (4.41), we have $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{B^{s}_{2,1}}+\sum_{m+1\leq|\kappa|+|\nu|\leq M+1}\varrho^{|\kappa|+|\nu|}.$ (4.42) In view of the estimate for the maximal function as in Proposition 2.5, one has that $\displaystyle\lambda_{2}(S(t)u_{0})\lesssim\|u_{0}\|_{B^{s}_{2,1}}.$ (4.43) and for $i=0,1$, $\displaystyle\|\mathscr{A}\nabla^{i}F\|_{\ell^{1}_{\triangle}\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\leq\sum^{\infty}_{j=0}\int_{\mathbb{R}}\|S(t-\tau)(\triangle_{j}\nabla^{i}F)(\tau)\|_{\ell^{2^{*}}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\sum^{\infty}_{j=0}\int_{\mathbb{R}}\|(\triangle_{j}\nabla^{i}F)(\tau)\|_{H^{s-3/2}(\mathbb{R}^{n})}d\tau$ $\displaystyle\lesssim\sum^{\infty}_{j=0}2^{(s-1/2)j}\int_{\mathbb{R}}\|(\triangle_{j}F)(\tau)\|_{L^{2}(\mathbb{R}^{n})}d\tau.$ (4.44) Hence, by (4.43) and (4.44), $\displaystyle\lambda_{2}(\mathscr{T}u)\lesssim\|u_{0}\|_{B^{s}_{2,1}}+\|F\|_{\ell^{1,s-1/2}_{\triangle}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.45) Similar to (4.45), in view of Proposition 2.9, we have $\displaystyle\lambda_{3}(\mathscr{T}u)\lesssim\|u_{0}\|_{B^{s}_{2,1}}+\|F\|_{\ell^{1,s-1/2}_{\triangle}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.46) In view of Lemma 4.2, we have $\displaystyle\|(u)^{\kappa}(\nabla u)^{\nu}\|_{\ell^{1,s-1/2}_{\triangle}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\lesssim\sum_{k=0,1}\|\nabla^{k}u\|^{|\kappa|+|\nu|-1}_{\ell^{1}_{\triangle}\ell^{(|\kappa|+|\nu|-1)}_{\alpha}(L^{2(|\kappa|+|\nu|-1)}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}$ $\displaystyle\ \ \ \ \times\sum_{i=0,1}\|\nabla^{i}u\|_{\ell^{1,s-1/2}_{\triangle}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (4.47) Hence, if $u\in\mathscr{D}$, we have $\displaystyle\lambda_{2}(\mathscr{T}u)+\lambda_{3}(\mathscr{T}u)\lesssim\|u_{0}\|_{B^{s}_{2,1}}+\sum_{m+1\leq|\kappa|+|\nu|\leq M+1}\varrho^{|\kappa|+|\nu|}.$ (4.48) Repeating the procedures as in the above, we obtain that there exists $u\in\mathscr{D}$ satisfying the integral equation $\mathscr{T}u=u$, which finishes the proof of Theorem 1.1. $\hfill\Box$ ## 5 Proof of Theorem 1.2 We begin with the following ###### Lemma 5.1 Let $\mathscr{A}$ be as in (1.13). There exists a constant $C(T)>1$ which depends only on $T$ and $n$ such that $\displaystyle\sum_{i=0,1}\|\mathscr{A}\nabla^{i}F\|_{\ell^{1}_{\Box}\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}\leq C(T)\|F\|_{\ell^{1,3/2}_{\Box}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}([0,T]\times Q_{\alpha}))}.$ (5.1) Proof. Using Minkowski’s inequality and Proposition 2.1, $\displaystyle\|\mathscr{A}\nabla^{i}F\|_{\ell^{1}_{\Box}\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\leq\sum_{k\in\mathbb{Z}^{n}}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\left(\int^{T}_{0}\|S(t-\tau)\Box_{k}\nabla^{i}F(\tau)\|_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}d\tau\right)^{2}\right)^{1/2}$ $\displaystyle\leq\sum_{k\in\mathbb{Z}^{n}}\int^{T}_{0}\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|S(t-\tau)\Box_{k}\nabla^{i}F(\tau)\|^{2}_{L^{\infty}_{t,x}([0,T]\times Q_{\alpha})}\right)^{1/2}d\tau$ $\displaystyle\leq\sum_{k\in\mathbb{Z}^{n}}\int^{T}_{0}\|\Box_{k}\nabla^{i}F(\tau)\|_{M^{1/2}_{2,1}}d\tau.$ (5.2) It is easy to see that for $i=0,1$, $\displaystyle\|\Box_{k}\nabla^{i}F\|_{M^{1/2}_{2,1}}$ $\displaystyle\lesssim\langle k\rangle^{3/2}\|\Box_{k}F\|_{L^{2}(\mathbb{R}^{n})}\leq\langle k\rangle^{3/2}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}F\|_{L^{2}(Q_{\alpha})}.$ (5.3) By (5.2) and (5.3), we immediately have (5.1). $\hfill\Box$ ###### Lemma 5.2 Let $\mathscr{A}$ be as in (1.13). Let $n\geq 2$, $s>0$. Then we have $\displaystyle\sum_{i=0,1}\|\nabla^{i}\mathscr{A}F\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}\leq\langle T\rangle^{1/2}\|F\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}.$ (5.4) Proof. In view of Proposition 3.6, we have $\displaystyle\|\nabla\mathscr{A}F\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}\lesssim\|F\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}.$ (5.5) By Propositions 3.1 and 3.3, $\displaystyle\|\mathscr{A}F\|_{\ell^{1}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\lesssim\|F\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\leq T^{1/2}\|F\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}.$ (5.6) By (5.5) and (5.6) we immediately have (5.4). $\hfill\Box$ ###### Lemma 5.3 Let $n\geq 2$, $S(t)$ be as in (1.13). Then we have for $i=0,1$, $\displaystyle\|\nabla^{i}S(t)u_{0}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{M^{s+1/2}_{2,1}},\quad n\geq 3,$ (5.7) $\displaystyle\|\nabla^{i}S(t)u_{0}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\lesssim\|u_{0}\|_{M^{s+1/2}_{2,1}\cap\dot{H}^{-1/2}},\quad n=2.$ (5.8) Proof. (5.7) follows from Corollary 3.5. For $n=2$, by Proposition 3.2, we have the result, as desired. $\hfill\Box$ ###### Lemma 5.4 Let $n\geq 2$, $s>0$, $L\in\mathbb{N},L\geq 3$. Let $1\leq p,p_{i},q,q_{i}\leq\infty$ satisfy $1/p=1/p_{1}+(L-1)/p_{2}$ and $1/q=1/q_{1}+(L-1)/q_{2}$. We have $\displaystyle\|v_{1}...v_{L}\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha}))}\lesssim$ $\displaystyle\sum^{L}_{l=1}\|v_{l}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{1}}_{x}(I\times Q_{\alpha}))}$ $\displaystyle\ \times\prod_{i\not=l,\,i=1,...,L}\|v_{i}\|_{\ell^{1}_{\Box}\ell^{L-1}_{\alpha}(L^{q_{2}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}.$ (5.9) Proof. Using the identity $\displaystyle v_{1}...v_{L}=\sum_{k_{1},...,k_{L}\in\mathbb{Z}^{n}}\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L}$ (5.10) and noticing the fact that $\displaystyle\Box_{k}(\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L})=0,\ \ |k-k_{1}-...-k_{L}|\geq C(L,n),$ (5.11) we have $\displaystyle\|v_{1}...v_{L}\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha}))}$ $\displaystyle=\sum_{k\in\mathbb{Z}^{n}}\langle k\rangle^{s}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}(v_{1}...v_{L})\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}$ $\displaystyle\leq\sum_{k_{1},...,k_{n}\in\mathbb{Z}^{n}}\sum_{|k-k_{1}-...-k_{L}|\leq C}\langle k\rangle^{s}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k}(\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L})\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}.$ (5.12) Similar to (4.31) and noticing the fact that $\|\mathscr{F}^{-1}\sigma_{k}\|_{L^{1}{(\mathbb{R}^{n}})}\lesssim 1$, we have $\displaystyle\sum_{\alpha\in\mathbb{Z}^{n}}\|\,\Box_{k}f\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}$ $\displaystyle=\sum_{\alpha\in\mathbb{Z}^{n}}\|(\mathscr{F}^{-1}\sigma_{k})*f\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}$ $\displaystyle\leq\int_{\mathbb{R}^{n}}|(\mathscr{F}^{-1}\sigma_{k})(y)|\left(\sum_{\alpha\in\mathbb{Z}^{n}}\|f(t,x-y)\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}\right)dy$ $\displaystyle\leq\sup_{y\in\mathbb{R}^{n}}\sum_{\alpha\in\mathbb{Z}^{n}}\|f(t,x-y)\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}\|\mathscr{F}^{-1}\sigma_{k}\|_{L^{1}(\mathbb{R}^{n})}$ $\displaystyle\lesssim\sum_{\alpha\in\mathbb{Z}^{n}}\|f\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}.$ (5.13) By (5.12) and (5.13), we have $\displaystyle\|v_{1}...v_{L}\|_{\ell^{1,s}_{\Box}\ell^{1}_{\alpha}(L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha}))}$ $\displaystyle\leq\sum_{k_{1},...,k_{n}\in\mathbb{Z}^{n}}\sum_{|k-k_{1}-...-k_{L}|\leq C}\langle k\rangle^{s}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L}\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}$ $\displaystyle\lesssim\sum_{k_{1},...,k_{n}\in\mathbb{Z}^{n}}(\langle k_{1}\rangle^{s}+...+\langle k_{L}\rangle^{s})\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L}\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}.$ (5.14) By Hölder’s inequality, $\displaystyle\sum_{k_{1},...,k_{n}\in\mathbb{Z}^{n}}\langle k_{1}\rangle^{s}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k_{1}}v_{1}...\Box_{k_{L}}v_{L}\|_{L^{q}_{t}L^{p}_{x}(I\times Q_{\alpha})}$ $\displaystyle\leq\sum_{k_{1},...,k_{n}\in\mathbb{Z}^{n}}\langle k_{1}\rangle^{s}\sum_{\alpha\in\mathbb{Z}^{n}}\|\Box_{k_{1}}v_{1}\|_{L^{q_{1}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha})}\prod^{L}_{i=2}\|\Box_{k_{i}}v_{i}\|_{L^{q_{2}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha})}$ $\displaystyle\leq\|v_{1}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}\sum_{k_{2},...,k_{n}\in\mathbb{Z}^{n}}\sum_{\alpha\in\mathbb{Z}^{n}}\prod^{L}_{i=2}\|\Box_{k_{i}}v_{i}\|_{L^{q_{2}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha})}$ $\displaystyle\leq\|v_{1}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}\sum_{k_{2},...,k_{n}\in\mathbb{Z}^{n}}\prod^{L}_{i=2}\|\Box_{k_{i}}v_{i}\|_{\ell^{L-1}_{\alpha}(L^{q_{2}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}$ $\displaystyle\leq\|v_{1}\|_{\ell^{1,s}_{\Box}\ell^{\infty}_{\alpha}(L^{q_{1}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}\prod^{L}_{i=2}\|v_{i}\|_{\ell^{1}_{\Box}\ell^{L-1}_{\alpha}(L^{q_{2}}_{t}L^{p_{2}}_{x}(I\times Q_{\alpha}))}.$ (5.15) The result follows. $\hfill\Box$ Proof of Theorem 1.2. Denote $\displaystyle\lambda_{1}(v)$ $\displaystyle=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1,3/2}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))},$ $\displaystyle\lambda_{2}(v)$ $\displaystyle=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1}_{\Box}\ell^{2}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}.$ Put $\displaystyle\mathscr{D}=\left\\{u:\;\lambda_{1}(u)+\lambda_{2}(u)\leq\varrho\right\\}.$ (5.16) Let $\mathscr{T}$ be as in (4.4). We will show that $\mathscr{T}:\mathscr{D}\to\mathscr{D}$ is a contraction mapping. First, we consider the case $n\geq 3$. Let $u\in\mathscr{D}$. By Lemmas 5.2 and 5.3, we have $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{2}_{2,1}}+\langle T\rangle^{1/2}\|F\|_{\ell^{1,3/2}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}.$ (5.17) We use the same notation as in (4.40). We have from Lemma 5.4 that $\displaystyle\|(u)^{\kappa}(\nabla u)^{\nu}\|_{\ell^{1,3/2}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}\lesssim$ $\displaystyle\sum_{i=0,1}\|\nabla^{i}u\|_{\ell^{1,3/2}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\ \times\sum_{k=0,1}\|\nabla^{k}u\|^{|\kappa|+|\nu|-1}_{\ell^{1}_{\Box}\ell^{|\kappa|+|\nu|-1}_{\alpha}(L^{\infty}_{t,x}([0,T]\times Q_{\alpha}))}$ $\displaystyle\lesssim\lambda_{1}(u)\lambda_{2}(u)^{|\kappa|+|\nu|-1}\leq\varrho^{|\kappa|+|\nu|}.$ (5.18) Hence, for $n\geq 3$, $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{2}_{2,1}}+\sum^{M}_{|\kappa|+|\nu|=m+1}\varrho^{|\kappa|+|\nu|}.$ (5.19) Next, we consider the estimate of $\lambda_{2}(\mathscr{T}u)$. By Lemma 5.1 and Proposition 2.1, $\displaystyle\lambda_{2}(\mathscr{T}u)$ $\displaystyle\lesssim\|u_{0}\|_{M^{3/2}_{2,1}}+C(T)\|F\|_{\ell^{1,3/2}_{\Box}\ell^{1}_{\alpha}(L^{1}_{t}L^{2}_{x}([0,T]\times Q_{\alpha}))},$ (5.20) which reduces to the estimates of $\lambda_{1}(\cdot)$ as in (5.17). Similarly, for $n=2$, $\displaystyle\lambda_{1}(\mathscr{T}u)+\lambda_{2}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{2}_{2,1}\cap\dot{H}^{-1/2}}+\sum^{M}_{|\kappa|+|\nu|=m+1}\varrho^{|\kappa|+|\nu|}.$ (5.21) Repeating the procedures as in the proof of Theorem 1.1, we can show our results, as desired. $\hfill\Box$ ## 6 Proof of Theorem 1.3 The proof of Theorem 1.3 follows an analogous way as that in Theorems 1.1 and 1.2 and will be sketched. Put $\displaystyle\lambda_{1}(v)=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1,s-1/2}_{\Box}\ell^{\infty}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{2}(v)=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1}_{\Box}\ell^{m}_{\alpha}(L^{\infty}_{t,x}(\mathbb{R}\times Q_{\alpha}))},$ $\displaystyle\lambda_{3}(v)=\sum_{i=0,1}\|\nabla^{i}v\|_{\ell^{1}_{\Box}\ell^{m}_{\alpha}(L^{2m}_{t}L^{\infty}_{x}(\mathbb{R}\times Q_{\alpha}))}.$ Put $\displaystyle\mathscr{D}=\left\\{u:\;\lambda_{1}(u)+\lambda_{2}(u)+\lambda_{3}(u)\leq\varrho\right\\}.$ (6.1) Let $\mathscr{T}$ be as in (4.4). We show that $\mathscr{T}:\mathscr{D}\to\mathscr{D}$. We only consider the case $n\geq 3$. It follows from Lemma 5.3 and 4.3 that $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{s}_{2,1}}+\|F\|_{\ell^{1,s-1/2}_{\Box}\ell^{1}_{\alpha}(L^{2}_{t,x}(\mathbb{R}\times Q_{\alpha}))}.$ (6.2) Using Lemma 5.4 and similar to (5.18), one sees that if $u\in\mathscr{D}$, then $\displaystyle\lambda_{1}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{s}_{2,1}}+\sum_{m+1\leq|\kappa|+|\nu|\leq M+1}\varrho^{|\kappa|+|\nu|}.$ (6.3) Using Proposition 2.10 and combining the proof of (4.44)–(4.46), we see that $\displaystyle\lambda_{2}(\mathscr{T}u)+\lambda_{3}(\mathscr{T}u)\lesssim\|u_{0}\|_{M^{s}_{2,1}}+\sum_{m+1\leq|\kappa|+|\nu|\leq M+1}\varrho^{|\kappa|+|\nu|}.$ (6.4) The left part of the proof is analogous to that of Theorems 1.1 and 1.2 and the details are omitted. $\hfill\Box$ ## 7 Proof of Theorem 1.4 We prove Theorem 1.4 by following some ideas as in Molinet and Ribaud [23] and Wang and Huang [33]. The following is the estimates for the solutions of the linear Schrödinger equation, see [16, 23, 33]. Recall that $\triangle_{j}:=\mathscr{F}^{-1}\delta(2^{-j}\,\cdot)\mathscr{F}$, $j\in\mathbb{Z}$ and $\delta(\cdot)$ is as in Section 1.4. ###### Lemma 7.1 Let $g\in\mathscr{S}(\mathbb{R})$,$f\in\mathscr{S}(\mathbb{R}^{2})$, $4\leq p<\infty$. Then we have $\displaystyle\|\triangle_{j}S(t)g\|_{L_{t}^{\infty}L_{x}^{2}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\lesssim\|\triangle_{j}g\|_{L^{2}},$ (7.1) $\displaystyle\|\triangle_{j}S(t)g\|_{L_{x}^{p}L_{t}^{\infty}}$ $\displaystyle\lesssim 2^{j(\frac{1}{2}-\frac{1}{p})}\|\triangle_{j}g\|_{L^{2}},$ (7.2) $\displaystyle\|\triangle_{j}S(t)g\|_{L_{x}^{\infty}L_{t}^{2}}$ $\displaystyle\lesssim 2^{-j/2}\|\triangle_{j}g\|_{L^{2}},$ (7.3) $\displaystyle\|\triangle_{j}\mathscr{A}f\|_{L_{t}^{\infty}L_{x}^{2}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\lesssim\|\triangle_{j}f\|_{L^{6/5}_{x,t}},$ (7.4) $\displaystyle\|\triangle_{j}\mathscr{A}f\|_{L_{x}^{p}L_{t}^{\infty}}$ $\displaystyle\lesssim 2^{j(\frac{1}{2}-\frac{1}{p})}\|\triangle_{j}f\|_{L^{6/5}_{x,t}},$ (7.5) $\displaystyle\|\triangle_{j}\mathscr{A}f\|_{L_{x}^{\infty}L_{t}^{2}}$ $\displaystyle\lesssim 2^{-j/2}\|\triangle_{j}f\|_{L^{6/5}_{x,t}},$ (7.6) and $\displaystyle\|\triangle_{j}\mathscr{A}(\partial_{x}f)\|_{L_{t}^{\infty}L_{x}^{2}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\lesssim 2^{j/2}\|\triangle_{j}f\|_{L^{1}_{x}L^{2}_{t}},$ (7.7) $\displaystyle\|\triangle_{j}\mathscr{A}(\partial_{x}f)\|_{L_{x}^{p}L_{t}^{\infty}}$ $\displaystyle\lesssim 2^{j/2}2^{j(\frac{1}{2}-\frac{1}{p})}\|\triangle_{j}f\|_{L^{1}_{x}L^{2}_{t}},$ (7.8) $\displaystyle\|\triangle_{j}\mathscr{A}(\partial_{x}f)\|_{L_{x}^{\infty}L_{t}^{2}}$ $\displaystyle\lesssim\|\triangle_{j}f\|_{L_{x}^{1}L_{t}^{2}}.$ (7.9) For convenience, we write for any Banach function space $X$, $\|f\|_{\ell_{\triangle}^{1,s}(X)}=\sum_{j\in\mathbb{Z}}2^{js}\|\triangle_{j}f\|_{X},\quad\|f\|_{\ell_{\triangle}^{1}(X)}:=\|f\|_{\ell_{\triangle}^{1,0}(X)}.$ ###### Lemma 7.2 Let $s>0$, $1\leq p,p_{i},\gamma,\gamma_{i}\leq\infty$ satisfy $\displaystyle\frac{1}{p}=\frac{1}{p_{1}}+...+\frac{1}{p_{N}},\quad\frac{1}{\gamma}=\frac{1}{\gamma_{1}}+...+\frac{1}{\gamma_{N}}.$ (7.10) Then $\displaystyle\left\|u_{1}...u_{N}\right\|_{\ell_{\triangle}^{1,s}(L^{p}_{x}L^{\gamma}_{t})}$ $\displaystyle\lesssim\|u_{1}\|_{\ell_{\triangle}^{1,s}(L^{p_{1}}_{x}L^{\gamma_{1}}_{t})}\prod^{N}_{i=2}\|u_{i}\|_{\ell_{\triangle}^{1}(L^{p_{i}}_{x}L^{\gamma_{i}}_{t})}$ $\displaystyle\quad+\|u_{2}\|_{\ell_{\triangle}^{1,s}(L^{p_{2}}_{x}L^{\gamma_{2}}_{t})}\prod_{i\not=2,\,i=1,...,N}\|u_{i}\|_{\ell_{\triangle}^{1}(L^{p_{i}}_{x}L^{\gamma_{i}}_{t})}$ $\displaystyle\quad+...+\prod^{N-1}_{i=1}\|u_{i}\|_{\ell_{\triangle}^{1}(L^{p_{i}}_{x}L^{\gamma_{i}}_{t})}\|u_{N}\|_{\ell_{\triangle}^{1,s}(L^{p_{N}}_{x}L^{\gamma_{N}}_{t})},$ (7.11) and in particular, if $u_{1}=...=u_{N}=u$, then $\displaystyle\left\|u^{N}\right\|_{\ell_{\triangle}^{1,s}(L^{p}_{x}L^{\gamma}_{t})}$ $\displaystyle\lesssim\|u\|_{\ell_{\triangle}^{1,s}(L^{p_{1}}_{x}L^{\gamma_{1}}_{t})}\prod^{N}_{i=2}\|u\|_{\ell_{\triangle}^{1}(L^{p_{i}}_{x}L^{\gamma_{i}}_{t})}.$ (7.12) Substituting the spaces $L^{p}_{x}L^{\gamma}_{t}$ and $L^{p_{i}}_{x}L^{\gamma_{i}}_{t}$ by $L^{\gamma}_{t}L^{p}_{x}$ and $L^{\gamma_{i}}_{t}L^{p_{i}}_{x}$, respectively, (7.11) and (7.12) also holds. Proof. We only consider the case $N=2$ and the case $N>2$ can be handled in a similar way. We have $\displaystyle u_{1}u_{2}$ $\displaystyle=\sum^{\infty}_{r=-\infty}[(S_{r+1}u_{1})(S_{r+1}u_{2})-(S_{r}u_{1})(S_{r}u_{2})]$ $\displaystyle=\sum^{\infty}_{r=-\infty}[(\triangle_{r+1}u_{1})(S_{r+1}u_{2})+(S_{r}u_{1})(\triangle_{r+1}u_{2})],$ (7.13) and $\displaystyle\triangle_{j}(u_{1}u_{2})=\triangle_{j}\Big{(}\sum_{r\geq j-10}[(\triangle_{r+1}u_{1})(S_{r+1}u_{2})+(S_{r}u_{1})(\triangle_{r+1}u_{2})]\Big{)}.$ (7.14) We may assume, without loss of generality that there is only the first term in the right hand side of (7.14) and the second term can be handled in the same way. It follows from Bernstein’s estimate, Hölder’s and Young’s inequalities that $\displaystyle\sum_{j\in\mathbb{Z}}2^{sj}\|\triangle_{j}(u_{1}u_{2})\|_{L^{p}_{x}L^{\gamma}_{t}}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}2^{sj}\sum_{r\geq j-10}\|(\triangle_{r+1}u_{1})(S_{r+1}u_{2})\|_{L^{p}_{x}L^{\gamma}_{t}}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}2^{sj}\sum_{r\geq j-10}\|\triangle_{r+1}u_{1}\|_{L^{p_{1}}_{x}L^{\gamma_{1}}_{t}}\|S_{r+1}u_{2}\|_{L^{p_{2}}_{x}L^{\gamma_{2}}_{t}}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}2^{s(j-r)}\sum_{r\geq j-10}2^{rs}\|\triangle_{r+1}u_{1}\|_{L^{p_{1}}_{x}L^{\gamma_{1}}_{t}}\|S_{r+1}u_{2}\|_{L^{p_{2}}_{x}L^{\gamma_{2}}_{t}}$ $\displaystyle\lesssim\|u_{1}\|_{\ell_{\triangle}^{1,s}(L^{p_{1}}_{x}L^{\gamma_{1}}_{t})}\|u_{2}\|_{\ell_{\triangle}^{1}(L^{p_{2}}_{x}L^{\gamma_{2}}_{t})},$ (7.15) which implies the result, as desired. $\quad\quad\Box$ ###### Remark 7.3 One easily sees that (7.12) can be slightly improved by $\displaystyle\left\|u^{N}\right\|_{\ell^{1,s}(L^{p}_{x}L^{\gamma}_{t})}$ $\displaystyle\lesssim\|u\|_{\ell^{1,s}(L^{p_{1}}_{x}L^{\gamma_{1}}_{t})}\prod^{N}_{i=2}\|u\|_{L^{p_{i}}_{x}L^{\gamma_{i}}_{t}}.$ (7.16) In fact, from Minkowski’s inequality it follows that $\displaystyle\|S_{r}u\|_{L^{p}_{x}L^{\gamma}_{t}}\lesssim\|u\|_{L^{p}_{x}L^{\gamma}_{t}}.$ (7.17) From (7.15) and (7.17) we get (7.16). Proof of Theorem 1.4. We can assume, without loss of generality that $\displaystyle F(u,\bar{u},u_{x},\bar{u}_{x})=\sum_{m+1\leq\kappa+\nu\leq M+1}\lambda_{\kappa\nu}u^{\kappa}u^{\nu}_{x}$ (7.18) and the general case can be handled in the same way. Step 1\. We consider the case $m>4.$ Recall that $\displaystyle\|u\|_{X}$ $\displaystyle=\sup_{s_{m}\leq s\leq\tilde{s}_{M}}\sum_{i=0,1}\sum_{j\in\mathbb{Z}}|\\!|\\!|\partial^{i}_{x}\triangle_{j}u|\\!|\\!|_{s},$ (7.19) $\displaystyle|\\!|\\!|\triangle_{j}v|\\!|\\!|_{s}:=$ $\displaystyle 2^{sj}(\|\triangle_{j}v\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}+2^{j/2}\|\triangle_{j}v\|_{L^{\infty}_{x}L^{2}_{t}})$ $\displaystyle+2^{(s-\tilde{s}_{m})j}\|\triangle_{j}v\|_{L_{x}^{m}L_{t}^{\infty}}+2^{(s-\tilde{s}_{M})j}\|\triangle_{j}v\|_{L_{x}^{M}L_{t}^{\infty}}.$ (7.20) Considering the mapping $\displaystyle\mathscr{T}:u(t)\to S(t)u_{0}-{\rm i}\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x}),$ (7.21) we will show that $\mathscr{T}:X\to X$ is a contraction mapping. We have $\displaystyle\|\mathscr{T}u(t)\|_{X}\lesssim\|S(t)u_{0}\|_{X}+\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{X}.$ (7.22) In view of (7.1), (7.2) and (7.3) we have, $\displaystyle|\\!|\\!|\partial^{i}_{x}\triangle_{j}S(t)u_{0}|\\!|\\!|_{s}\lesssim 2^{sj}\|\partial^{i}_{x}\triangle_{j}u_{0}\|_{2}.$ (7.23) It follows that $\displaystyle\|S(t)u_{0}\|_{X}\lesssim\sup_{s_{m}\leq s\leq\tilde{s}_{M}}\sum_{i=0,1}\sum_{j\in\mathbb{Z}}2^{sj}\|\partial^{i}_{x}\triangle_{j}u_{0}\|_{2}\lesssim\|u_{0}\|_{\dot{B}^{s_{m}}_{2,1}\cap\dot{B}^{1+\tilde{s}_{M}}_{2,1}}.$ (7.24) We now estimate $\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{X}.$ We have from (7.4), (7.5) and (7.6) that $\displaystyle|\\!|\\!|\triangle_{j}(\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x}))|\\!|\\!|_{s}\lesssim 2^{sj}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{6/5}_{x,t}}.$ (7.25) From (7.7), (7.8) and (7.9) it follows that $\displaystyle|\\!|\\!|\triangle_{j}(\mathscr{A}\partial_{x}F(u,\bar{u},u_{x},\bar{u}_{x}))|\\!|\\!|_{s}\lesssim 2^{sj}2^{j/2}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{1}_{x}L^{2}_{t}}.$ (7.26) Hence, from (7.19), (7.25) and (7.26) we have $\displaystyle\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{X}\lesssim$ $\displaystyle\sum_{j\in\mathbb{Z}}2^{sj}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{6/5}_{x,t}}$ $\displaystyle+\sum_{j\in\mathbb{Z}}2^{sj}2^{j/2}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{1}_{x}L^{2}_{t}}=I+II.$ (7.27) Now we perform the nonlinear estimates. By Lemma 7.2, $\displaystyle I$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\Big{(}\|u\|_{\ell_{\triangle}^{1,s}(L^{6}_{x,t})}\|u\|^{\kappa-1}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}\|u_{x}\|^{\nu}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}$ $\displaystyle\quad\quad\quad\quad\quad\quad\quad+\|u_{x}\|_{\ell_{\triangle}^{1,s}(L^{6}_{x,t})}\|u_{x}\|^{\nu-1}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}\|u\|^{\kappa}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}\Big{)}$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\big{(}\sum_{i=0,1}\|\partial_{x}^{i}u\|_{\ell_{\triangle}^{1,s}(L^{6}_{x,t})}\big{)}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|^{\kappa+\nu-1}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}\big{)}.$ (7.28) For any $m\leq\lambda\leq M$, we let $\frac{1}{\rho}=\frac{1}{2}-\frac{4}{3\lambda}$. It is easy to see that the following inclusions hold: $\displaystyle L^{\infty}_{t}(\mathbb{R},\dot{H}^{s_{\lambda}})\cap L^{6}_{t}(\mathbb{R},\dot{H}^{s_{\lambda}}_{6})\subset L^{3\lambda/2}_{t}(\mathbb{R},\dot{H}^{s_{\lambda}}_{\rho})\subset L^{3\lambda/2}_{x,t}.$ (7.29) More precisely, we have $\displaystyle\sum_{j\in\mathbb{Z}}\|\triangle_{j}u\|_{L^{3\lambda/2}_{x,t}}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}\|\triangle_{j}u\|_{L^{3\lambda/2}_{t}(\mathbb{R},\,\dot{H}^{s_{\lambda}}_{\rho})}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}\|\triangle_{j}u\|^{4/\lambda}_{L^{6}_{t}(\mathbb{R},\,\dot{H}^{s_{\lambda}}_{6})}\|\triangle_{j}u\|^{1-4/\lambda}_{L^{\infty}_{t}(\mathbb{R},\,\dot{H}^{s_{\lambda}})}$ $\displaystyle\lesssim\|u\|^{4/\lambda}_{\ell^{1,s_{\lambda}}(L^{6}_{x,t})}\|u\|^{1-4/\lambda}_{\ell^{1,s_{\lambda}}(L^{\infty}_{t}L^{2}_{x})}.$ (7.30) Using (7.30) and noticing that $s_{m}\leq s_{\kappa+\nu-1}\leq s_{M}<\tilde{s}_{M}$, we have $\displaystyle\|\partial^{i}_{x}u\|^{\kappa+\nu-1}_{\ell_{\triangle}^{\,1}(L^{3(\kappa+\nu-1)/2}_{x,t})}$ $\displaystyle\lesssim\|\partial^{i}_{x}u\|^{4}_{\ell_{\triangle}^{1,\,s_{\kappa+\nu-1}}(L^{6}_{x,t})}\|\partial^{i}_{x}u\|^{\kappa+\nu-5}_{\ell_{\triangle}^{1,\,s_{\kappa+\nu-1}}(L^{\infty}_{t}L^{2}_{x})}$ $\displaystyle\lesssim\|u\|^{\kappa+\nu-1}_{X}.$ (7.31) Combining (7.28) with (7.31), we have $\displaystyle I\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X}.$ (7.32) Now we estimate $II$. By Lemma 7.2, $\displaystyle II$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\Big{(}\|u\|_{\ell_{\triangle}^{1,\,s+1/2}(L^{\infty}_{x}L^{2}_{t})}\|u\|^{\kappa-1}_{\ell_{\triangle}^{\,1}(L^{\kappa+\nu-1}_{x}L^{\infty}_{t})}\|u_{x}\|^{\nu}_{\ell_{\triangle}^{\,1}(L^{\kappa+\nu-1}_{x}L^{\infty}_{t})}$ $\displaystyle\quad\quad\quad\quad\quad\quad\quad+\|u_{x}\|_{\ell_{\triangle}^{1,\,s+1/2}(L^{\infty}_{x}L^{2}_{t})}\|u_{x}\|^{\nu-1}_{L^{\kappa+\nu-1}_{x}L^{\infty}_{t}}\|u\|^{\kappa}_{L^{\kappa+\nu-1}_{x}L^{\infty}_{t}}\Big{)}$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\big{(}\sum_{i=0,1}\|\partial_{x}^{i}u\|_{\ell_{\triangle}^{1,\,s+1/2}(L^{\infty}_{x}L^{2}_{t})}\big{)}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|^{\kappa+\nu-1}_{L^{\kappa+\nu-1}_{x}L^{\infty}_{t}}\big{)}$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\|u\|_{X}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|^{\kappa+\nu-1}_{L^{m}_{x}L^{\infty}_{t}\,\cap\,L^{M}_{x}L^{\infty}_{t}}\big{)}$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X}.$ (7.33) Collecting (7.27), (7.28), (7.32) and (7.33), we have $\displaystyle\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{X}$ $\displaystyle\lesssim\sum_{m+1\leq\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X}.$ (7.34) By (7.22), (7.24) and (7.34) $\displaystyle\|\mathscr{T}u(t)\|_{X}\lesssim\|u_{0}\|_{\dot{B}^{s_{m}}_{2,1}\cap\dot{B}^{1+\tilde{s}_{M}}_{2,1}}+\sum_{m+1\leq\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X}.$ (7.35) Step 2\. We consider the case $m=4$. Recall that $\displaystyle\|u\|_{X}$ $\displaystyle=\sum_{i=0,1}\big{(}\|\partial^{i}_{x}u\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}+\sup_{s_{5}\leq s\leq\tilde{s}_{M}}\sum_{j\in\mathbb{Z}}|\\!|\\!|\partial^{i}_{x}\triangle_{j}u|\\!|\\!|_{s}\big{)}.$ By (7.1), (7.2) and (7.3), $\displaystyle\|S(t)u_{0}\|_{X}\lesssim\|u_{0}\|_{2}+\sup_{s_{5}\leq s\leq\tilde{s}_{M}}\sum_{i=0,1}\sum_{j\in\mathbb{Z}}2^{sj}\|\partial^{i}_{x}\triangle_{j}u_{0}\|_{2}\lesssim\|u_{0}\|_{B^{1+\tilde{s}_{M}}_{2,1}}.$ (7.36) We now estimate $\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{X}.$ By Strichartz’ and Hölder’s inequality, we have $\displaystyle\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\|(|u|+|u_{x}|)^{\kappa+\nu}\|_{L^{6/5}_{x,t}}$ $\displaystyle\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\|(|u|+|u_{x}|)\|_{L^{6}_{x,t}}\|(|u|+|u_{x}|)\|^{\kappa+\nu-1}_{L^{3(\kappa+\nu-1)/2}_{x,t}}$ $\displaystyle\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\|(|u|+|u_{x}|)\|_{L^{6}_{x,t}}\|(|u|+|u_{x}|)\|^{\kappa+\nu-1}_{L^{6}_{x,t}\cap L^{3M/2}_{x,t}}$ $\displaystyle\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|_{L^{6}_{x,t}}\big{)}^{\kappa+\nu}$ $\displaystyle\ \ \ +\sum_{5\leq\kappa+\nu\leq M+1}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|_{L^{6}_{x,t}}\big{)}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u\|_{L^{3M/2}_{x,t}}\big{)}^{\kappa+\nu-1}.$ (7.37) Applying (7.30), we see that (7.37) implies that $\displaystyle\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\|u\|_{X}^{\kappa+\nu}.$ (7.38) From Bernstein’s estimate and (7.7) it follows that $\displaystyle\|\partial_{x}\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\leq\|P_{\leq 1}(\mathscr{A}\partial_{x}F(u,\bar{u},u_{x},\bar{u}_{x}))\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\ \ \ +\|P_{>1}(\mathscr{A}\partial_{x}F(u,\bar{u},u_{x},\bar{u}_{x}))\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\lesssim\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\ \ \ +\sum_{j\gtrsim 1}2^{j/2}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{1}_{x}L^{2}_{t}}$ $\displaystyle\lesssim\|\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}$ $\displaystyle\ \ \ +\sum_{j\in\mathbb{Z}}2^{\tilde{s}_{M}}2^{j/2}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{1}_{x}L^{2}_{t}}=III+IV.$ (7.39) The estimates of $III$ and $IV$ have been given in (7.38) and (7.33), respectively. We have $\displaystyle\|\partial_{x}\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{\infty}_{t}L^{2}_{x}\,\cap\,L^{6}_{x,t}}\lesssim\sum_{5\leq\kappa+\nu\leq M+1}\|u\|_{X}^{\kappa+\nu}.$ (7.40) We have from (7.4)–(7.6), (7.7)–(7.9) that $\displaystyle\sum_{j\in\mathbb{Z}}|\\!|\\!|\triangle_{j}(\mathscr{A}F(u,\bar{u},u_{x},\bar{u}_{x}))|\\!|\\!|_{s}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}2^{sj}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{6/5}_{x,t}},$ (7.41) $\displaystyle\sum_{j\in\mathbb{Z}}|\\!|\\!|\triangle_{j}(\mathscr{A}\partial_{x}F(u,\bar{u},u_{x},\bar{u}_{x}))|\\!|\\!|_{s}$ $\displaystyle\lesssim\sum_{j\in\mathbb{Z}}2^{sj}2^{j/2}\|\triangle_{j}F(u,\bar{u},u_{x},\bar{u}_{x})\|_{L^{1}_{x}L^{2}_{t}}$ (7.42) hold for all $s>0$. The right hand side in (7.42) has been estimated by (7.33). So, it suffices to consider the estimate of the right hand side in (7.41). Let us observe the equality $\displaystyle F(u,\bar{u},u_{x},\bar{u}_{x})=\sum_{\kappa+\nu=5}\lambda_{\kappa\nu}u^{\kappa}u_{x}^{\nu}+\sum_{5<\kappa+\nu\leq M+1}\lambda_{\kappa\nu}u^{\kappa}u_{x}^{\nu}:=V+VI.$ (7.43) For any $s_{5}\leq s\leq\tilde{s}_{M}$, $VI$ has been handled in (7.28)–(7.32): $\displaystyle\sum_{5<\kappa+\nu\leq M+1}\sum_{j\in\mathbb{Z}}2^{sj}\|\triangle_{j}(u^{\kappa}u_{x}^{\nu})\|_{L^{6/5}_{x,t}}$ $\displaystyle\lesssim\sum_{5<\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X}.$ (7.44) For the estimate of $V$, we use Remark 7.3, for any $s_{5}\leq s\leq\tilde{s}_{M}$, $\displaystyle\sum_{\kappa+\nu=5}\sum_{j\in\mathbb{Z}}2^{sj}\|\triangle_{j}(u^{\kappa}u_{x}^{\nu})\|_{L^{6/5}_{x,t}}$ $\displaystyle\lesssim\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u_{x}\|^{4}_{L^{6}_{x,t}}\big{)}\big{(}\sum_{i=0,1}\|\partial^{i}_{x}u_{x}\|_{\ell_{\triangle}^{1,s}(L^{6}_{x,t})}\big{)}$ $\displaystyle\lesssim\|u\|^{5}_{X}.$ (7.45) Summarizing the estimate above, $\displaystyle\|\mathscr{T}u(t)\|_{X}\lesssim\|u_{0}\|_{B^{1+\tilde{s}_{M}}_{2,1}}+\sum_{5\leq\kappa+\nu\leq M+1}\|u\|^{\kappa+\nu}_{X},$ (7.46) whence, we have the results, as desired. $\quad\quad\Box$ Acknowledgment. This work is supported in part by the National Science Foundation of China, grants 10571004 and 10621061; and the 973 Project Foundation of China, grant 2006CB805902. ## References * [1] J. Bergh and J. Löfström, Interpolation Spaces, Springer–Verlag, 1976. * [2] I. Bejenaru and D. Tataru, Large data local solutions for the derivative NLS equation, arXiv:math.AP/0610092 v1. * [3] J. Bourgain, Fourier transform restriction phenomena for certain lattice subsets and applications to nonlinear evolution equations, GAFA, 3 (1993), 107 - 156 and 209 - 262. * [4] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, and T. Tao, A refined global well-posedness result for the Schrödinger equation with derivative, SIAM J. Math. Anal., 34 (2002), 64–86. * [5] P. Constantin and J. C. Saut, Local smoothing properties of dispersive equations, J. Amer. Math. Soc., 1 (1988), 413–446. * [6] H. Chihara, Global existence of small solutions to semilinear Schrödinger equations with guage invariance, Publ. RIMS, 31 (1995), 731–753. * [7] H. Chihara, The initial value problem for cubic semilinear Schrödinger equations with guage invariance, Publ. RIMS, 32 (1996), 445–471. * [8] H. Chihara, Gain of regularity for semilinear Schrdinger equations, Math. Ann. 315 (1999), 529-567. * [9] M. Christ, Illposedness of a Schrödinger equation with derivative regularity, Preprint. * [10] E. Cordero and F. Nicola. Strichartz estimates inWiener amalgam spaces for the Schrödinger equation. Math. Nachr., 281 (2008), 25–41. * [11] E Cordero, F Nicola, Metaplectic representation on Wiener amalgam spaces and applications to the Schrödinger equation, J. Funct. Anal., 254 (2008), 506-534. * [12] H. G. Feichtinger, Modulation spaces on locally compact Abelian group, Technical Report, University of Vienna, 1983. Published in: “Proc. Internat. Conf. on Wavelet and Applications”, 99–140. New Delhi Allied Publishers, India, 2003. http://www.unive.ac.at/nuhag-php/bibtex/ open_files/fe03-1_modspa03.pdf. * [13] K. Gröchenig, Foundations of Time-Frequency Analysis, Birkhäuser, Boston, MA, 2001. * [14] M. Keel and T. Tao, Endpoint Strichartz estimates, Amer. J. Math., 120 (1998), 955-980. * [15] A. Grünrock On the Cauchy- and periodic boundary value problem for a certain class of derivative nonlinear Schrödinger equations, arXiv:math/0006195v1. * [16] C. E. Kenig, G. Ponce and L. Vega, Oscillatory integrals and regularity of dispersive equations, Indiana Univ. Math. J., 40 (1991), 253–288. * [17] C. E. Kenig, G. Ponce, L. Vega, Small solutions to nonlinear Schrodinger equation, Ann. Inst. Henri Poincaré, Sect C, 10 (1993), 255-288. * [18] C. E. Kenig, G. Ponce and L. Vega, Smoothing effects and local existence theory for the generalized nonlinear Schrödinger equations, Invent. Math., 134 (1998), 489–545. * [19] C. E. Kenig, G. Ponce, L. Vega, The Cauchy problem for quasi-linear Schrodinger equations, Invent. Math. 158 (2004), 343–388. * [20] C. E. Kenig, G. Ponce, C. Rolvent, L. Vega, The genreal quasilinear untrahyperbolic Schrodinger equation, Advances in Mathematics 206 (2006), 402–433. * [21] S. Klainerman, Long-time behavior of solutions to nonlinear evolution equations, Arch. Rational Mech. Anal., 78 (1982), 73–98. * [22] S. Klainerman, G. Ponce, Global small amplitude solutions to nonlinear evolution equations, Commun. Pure Appl. Math., 36 (1983), 133–141. * [23] L. Molinet and F. Ribaud, Well posedness results for the generalized Benjamin-Ono equation with small initial data, J. Math. Pures Appl., 83 (2004), 277-311. * [24] L. Molinet, J.D.Saut and N.Tzvetkov, Ill-posedness issues for the Benjamin-Ono equation and related equations, SIAM J.Math. Anslysis, 33 (2001), 982-988. * [25] T. Ozawa and J. Zhang, Global existence of small classical solutions to nonlinear Schrödinger equations, Ann. I. H. Poincaré, AN, to appear. * [26] P. Sjölin, Regularity of solutions to the Schrödinger equations, Duke Math. J., 55 (1987), 699–715. * [27] M. Sugimoto amd N. Tomita, The dilation property of modulation spaces and their inclusion relation with Besov spaces, Preprint. * [28] T. Tao, Spherically averaged endpoint Strichartz estimates for the two-dimensional Schrödinger equation, Commun. PDE, 25 (2000), 1471–1485. * [29] J. Toft, Continuity properties for modulation spaces, with applications to pseudo-differential calculus, I. J. Funct. Anal., 207 (2004), 399–429. * [30] H. Triebel, Theory of Function Spaces, Birkhäuser–Verlag, 1983. * [31] Baoxiang Wang, Lifeng Zhao and Boling Guo, Isometric decomposition operators, function spaces $E^{\lambda}_{p,q}$ and applications to nonlinear evolution equations, J. Funct. Anal., 233 (2006), 1–39. * [32] Baoxiang Wang and Henryk Hudzik, The global Cauchy problem for the NLS and NLKG with small rough data, J. Differential Equations, 231 (2007), 36–73. * [33] Baoxiang Wang and Chunyan Huang, Frequency-uniform decomposition method for the generalized BO, KdV and NLS equations, J. Differential Equations, 239 (2007), 213–250. * [34] L. Vega, The Schrödinger equation: pointwise convergence to the initial data, Proc. Amer. Math. Soc., 102 (1988), 874–878.

arxiv-papers

2024-09-04T02:48:54.381703

0803.2679

11institutetext: Department of Mathematics, University of Milan, via Saldini 50, 10133 Milan Italy giacomo.aletti@mat.unimi.it bongio@mat.unimi.it vincenzo.capasso@mat.unimi.it # Statistical aspects of birth–and–growth stochastic processes Giacomo Aletti Enea G. Bongiorno Vincenzo Capasso ###### Abstract The paper considers a particular family of set–valued stochastic processes modeling birth–and–growth processes. The proposed setting allows us to investigate the nucleation and the growth processes. A decomposition theorem is established to characterize the nucleation and the growth. As a consequence, different consistent set–valued estimators are studied for growth process. Moreover, the nucleation process is studied via the hitting function, and a consistent estimator of the nucleation hitting function is derived. ## Introduction Nucleation and growth processes arise in several natural and technological applications (cf. [6, 5] and the references therein) such as, for example, solidification and phase–transition of materials, semiconductor crystal growth, biomineralization, and DNA replication (cf., e.g., [17]). During the years, several authors studied stochastic spatial processes (cf. [10, 31, 23] and references therein) nevertheless they essentially consider static approaches modeling real phenomenons. For what concerns the dynamical point of view, a parametric _birth–and–growth process_ was studied in [25, 26]. A birth–and–growth process is a RaCS family given by $\Theta_{t}=\bigcup_{n:T_{n}\leq t}\Theta_{T_{n}}^{t}(X_{n})$, for $t\in\mathbb{R}_{+}$, where $\Theta^{t}_{T_{n}}\left(X_{n}\right)$ is the RaCS obtained as the evolution up to time $t>T_{n}$ of the germ born at (random) time $T_{n}$ in (random) location $X_{n}$, according to some growth model. An analytical approach is often used to model birth–and–growth process, in particular it is assumed that the growth of a spherical nucleus of infinitesimal radius is driven according to a non–negative normal velocity, i.e. for every instant $t$, a border point of the crystal $x\in\partial\Theta_{t}$ “grows” along the outwards normal unit (e.g. [16, 4, 3, 8]). In view of the chosen framework, different parametric and non–parametric estimations are proposed over the years (cf. [27, 24, 12, 5, 7, 2, 9] and references therein). Note that the existence of the outwards normal vector imposes a regularity condition on $\partial\Theta_{t}$ (and also on the nucleation process: it cannot be a point process). On the other hand, it is well known that random sets are particular cases of fuzzy sets. Now, in the class of all convex fuzzy sets having compact support, Doob–type decomposition for sub- and super–martingales was studied (e.g. [13, 15, 14, 32]). Nevertheless, a more general case (than the convex one) has not yet been considered; surely, in order to do this, the first easiest step is to consider decomposition for random set–valued processes. After which, the step forward, to be considered in a following paper, can be to generalize results of this paper to birth–and–growth fuzzy set–value processes. This paper is an attempt to offer an original approach based on a purely geometric stochastic point of view in order to avoid regularity assumptions describing birth–and–growth processes. The pioneer work [21] studies a growth model for a single convex crystal based on Minkowski sum, whilst in [1], the authors derive a computationally tractable mathematical model of such processes that emphasizes the geometric growth of objects without regularity assumptions on the boundary of crystals. Here, in view of this approach, we introduce different set–valued parametric estimators of the rate of growth of the process. They arise naturally from a decomposition via Minkowski sum and they are consistent as the observation window expands to the whole space. On the other hand, keeping in mind that distributions of random closed sets are determined by Choquet capacity functionals and that the nucleation process cannot be observed directly, the paper provides an estimation procedures of the hitting function of the nucleation process. The article is organized as follows. Section 0.1 contains preliminary properties. Section 0.2 introduces a birth–and–growth model for random closed sets as the combination of two set–valued processes (nucleation and growth respectively). Further, a decomposition theorem is established to characterize the nucleation and the growth. Section 0.3 studies different estimators of the growth process and correspondent consistent properties are proved. In Section 0.4, the nucleation process is studied via the hitting function, and a consistent estimator of the nucleation hitting function is derived. ## 0.1 Preliminary results Let $\mathbb{N}$, $\mathbb{Z}$, $\mathbb{R}$, $\mathbb{R}_{+}$ be the sets of all non–negative integer, integer, real and non–negative real numbers respectively, and let $\mathfrak{X}=\mathbb{R}^{d}$. Let $\mathbb{F}$ be the family of all closed subsets of $\mathfrak{X}$ and $\mathbb{F}^{\prime}=\mathbb{F}\setminus\\{\emptyset\\}$. The suffixes $b$, $k$ and $c$ denote boundedness, compactness and convexity properties respectively (e.g. $\mathbb{F}_{kc}$ denotes the family of all compact convex subsets of $\mathfrak{X}$). For all $A,B\subseteq\mathfrak{X}$ and $\alpha\in\mathbb{R}_{+}$, let us define $\begin{array}[]{rll}A+B=&\left\\{a+b:a\in A,\ b\in B\right\\}=\bigcup_{b\in B}b+A,&\textrm{(Minkowski Sum)},\\\ \alpha\cdot A=&\alpha A=\left\\{\alpha a:a\in A\right\\},&\textrm{(Scalar Product)},\\\ A\ominus B=&\left(A^{C}+B\right)^{C}=\bigcap_{b\in B}b+A,&\textrm{(Minkowski Subtraction)},\\\ \check{A}=&\left\\{-a:a\in A\right\\},&\textrm{(Symmetric Set)},\end{array}$ where $A^{C}=\left\\{x\in\mathfrak{X}:x\not\in A\right\\}$ is the complementary set of $A$, $x+A$ means $\\{x\\}+A$ (i.e. $A$ translate by vector $x$), and, by definition, $\emptyset+A=\emptyset=\alpha\emptyset$. It is well known that $+$ is a commutative and associative operation with a neutral element but, in general, $A\subseteq\mathfrak{X}$ does not admit opposite (cf. [29, 18]) and $\ominus$ is not the inverse operation of $+$. The following relations are useful in the sequel (see [30]): for every $A,B,C\subseteq\mathfrak{X}$ $\begin{array}[]{c}(A\cup B)+C=(A+C)\cup(B+C),\\\ \textrm{if }B\subseteq C,\quad A+B\subseteq A+C,\\\ (A\ominus B)+\check{B}\subseteq A\quad\textrm{and}\quad(A+B)\ominus\check{B}\supseteq A,\\\ (A\cup B)\ominus C\supseteq(A\ominus C)\cup(B\ominus C).\end{array}$ In the following, we shall work with closed sets. In general, if $A,B\in\mathbb{F}$ then $A+B$ does not belong to $\mathbb{F}$ (e.g., in $\mathfrak{X}=\mathbb{R}$ let $A=\left\\{n+1/n:n>1\right\\}$ and $B=\mathbb{Z}$, then $\left\\{1/n=\left(n+1/n\right)+(-n)\right\\}\subset A+B$ and $1/n\downarrow 0$, but $0\not\in A+B$). In view of this fact, we define $A\oplus B=\overline{A+B}$ where $\overline{(\cdot)}$ denotes the closure in $\mathfrak{X}$. It can be proved that, if $A\in\mathbb{F}$ and $B\in\mathbb{F}_{k}$ then $A+B\in\mathbb{F}$ (see [30]). For any $A,B\in\mathbb{F}^{\prime}$ the _Hausdorff distance_ (or _metric_) is defined by $\delta_{H}(A,B)=\max\left\\{\sup_{a\in A}\inf_{b\in B}\left\|a-b\right\|_{\mathfrak{X}},\sup_{b\in B}\inf_{a\in A}\left\|a-b\right\|_{\mathfrak{X}}\right\\}.$ A random closed set (RaCS) is a map $X$ defined on a probability space $(\Omega,\mathfrak{F},\mathbb{P})$ with values in $\mathbb{F}$ such that $\left\\{\omega\in\Omega:X(\omega)\cap K\neq\emptyset\right\\}$ is measurable for each compact set $K$ in $\mathfrak{X}$. It can be proved (see [19]) that, if $X,X_{1},X_{2}$ are RaCS and if $\xi$ is a measurable real–valued function, then $X_{1}\oplus X_{2}$, $X_{1}\ominus X_{2}$, $\xi X$ and $(\textrm{Int}\ X)^{C}$ are RaCS. Moreover, if $\left\\{X_{n}\right\\}_{n\in\mathbb{N}}$ is a sequence of RaCS then $X=\overline{\bigcup_{n\in\mathbb{N}}X_{n}}$ is so. Let $X$ be a RaCS, then $T_{X}(K)=\mathbb{P}(X\cap K\neq\emptyset)$, for all $K\in\mathbb{F}_{k}$, is its _hitting function_ (or _Choquet capacity functional_). The well known Matheron Theorem states that, the probability law $\mathbb{P}_{X}$ of any RaCS $X$ is uniquely determined by its hitting function (see [20]) and hence by $Q_{X}(K)=1-T_{X}(K)$. ###### Remark 0.1 _(See[22].) _ If both $X$ and $Y$ are RaCS, then, for every $K\in\mathbb{F}_{k}$, $T_{X\oplus Y}(K)=\mathbb{E}\left[\mathbb{E}\left[\left.{T_{X}\left(K\oplus\check{Y}\right)}\right|{Y}\right]\right].$ Moreover, if $X,Y$ are independent, then, for every $K\in\mathbb{F}_{k}$, $T_{X\cup Y}\left(K\right)=T_{X}\left(K\right)+T_{Y}\left(K\right)-T_{X}\left(K\right)T_{Y}\left(K\right).$ A RaCS $X$ is _stationary_ if the probability laws of $X$ and $X+v$ coincide for every $v\in\mathfrak{X}$. Thus, the hitting function of a stationary RaCS clearly is invariant up to translation $T_{X}(K)=T_{X}(K+{v})$ for each $K\in\mathbb{F}_{k}$ and any $v\in\mathfrak{X}$. A stationary RaCS $X$ is _ergodic_ , if, for all $K_{1},K_{2}\in\mathbb{F}$, $\frac{1}{\left|W_{n}\right|}\int_{W_{n}}Q_{X}((K_{1}+v)\cup K_{2})dv\rightarrow Q_{X}(K_{1})Q_{X}(K_{2}),\qquad\textrm{as}\qquad n\rightarrow\infty;$ where $\left\\{W_{n}\right\\}_{n\in\mathbb{N}}$ is a _convex averaging sequence of sets_ in $\mathfrak{X}$ (see [11]), i.e. each $\left\\{W_{n}\right\\}$ is convex and compact, $W_{n}\subset W_{n+1}$ for all $n\in\mathbb{N}$ and $\sup\left\\{r\geq 0:B(x,r)\subset W_{n}\textrm{ for some }x\in W_{n}\right\\}\uparrow\infty,\qquad\textrm{as}\qquad n\rightarrow\infty.$ ###### Proposition 0.2 __Let $X,Y$ be RaCS with $Y\in\mathbb{F}^{\prime}_{k}$ a.s. and $X$ stationary, then $X+Y$ is a stationary RaCS. Moreover, if $X$ is ergodic, then $X+Y$ is so. Proof. Let $Z=X+Y$, it is a RaCS. Note that $\displaystyle T_{Z}(K)$ $\displaystyle=$ $\displaystyle\mathbb{E}\left[\mathbb{E}\left[\left.{T_{X}\left(K+\check{Y}\right)}\right|{Y}\right]\right]=\mathbb{E}\left[\mathbb{E}\left[\left.{T_{X}\left(K+\check{Y}+v\right)}\right|{Y}\right]\right]=T_{Z}(K+v),$ for every $K\in\mathbb{F}_{k}$ and $v\in\mathfrak{X}$, then $Z=X+Y$ is stationary. Further, let us suppose that $X$ is ergodic, then, by Tonelli’s Theorem and by dominated convergence theorem, we obtain $\displaystyle\int_{W_{n}}\frac{Q_{Z}((K_{1}+v)\cup K_{2})}{\left|W_{n}\right|}dv$ $\displaystyle=$ $\displaystyle\mathbb{E}\left[\mathbb{E}\left[\left.{\frac{1}{\left|W_{n}\right|}\int_{W_{n}}Q_{X}(((K_{1}+v)\cup K_{2})+\check{Y})dv}\right|{Y}\right]\right]$ $\displaystyle\rightarrow$ $\displaystyle\mathbb{E}\left[\mathbb{E}\left[\left.{Q_{X}(K_{1}+\check{Y})Q_{X}(K_{2}+\check{Y})}\right|{Y}\right]\right]$ $\displaystyle=$ $\displaystyle Q_{Z}(K_{1})Q_{Z}(K_{2}),$ for every $K_{1},K_{2}\in\mathbb{F}_{k}$. Hence $X+Y$ is ergodic. $\blacksquare$ ## 0.2 A Birth–and–Growth process Let $(\Omega,\mathfrak{F},\left\\{\mathfrak{F}_{n}\right\\}_{n\in\mathbb{N}},\mathbb{P})$ be a filtered probability space with the usual properties. Let $\left\\{B_{n}:n\geq 0\right\\}$ and $\left\\{G_{n}:n\geq 1\right\\}$ be two families of RaCS such that $B_{n}$ is $\mathfrak{F}_{n}$–measurable and $G_{n}$ is $\mathfrak{F}_{n-1}$–measurable. These processes represent the _birth_ (or _nucleation_) _process_ and the _growth process_ respectively. Thus, let us define recursively a birth–and–growth process $\Theta=\left\\{\Theta_{n}:n\geq 0\right\\}$ by $\Theta_{n}=\left\\{\begin{array}[]{ll}(\Theta_{n-1}\oplus G_{n})\cup B_{n},&n\geq 1,\\\ B_{0},&n=0.\end{array}\right.$ (1) Roughly speaking, Equation (1) means that $\Theta_{n}$ is the enlargement of $\Theta_{n-1}$ due to a Minkowski _growth_ $G_{n}$ while _nucleation_ $B_{n}$ occurs. Without loss of generality let us consider the following assumption. (A-1) For every $n\geq 1$, $0\in G_{n}$. Note that, Assumption 0.2 is equivalent to $\Theta_{n-1}\subseteq\Theta_{n}$. In [1], the authors derive (1) from a continuous time birth–and–growth process; here, in order to make inference, the discrete time case is sufficient. Indeed, a sample of a birth–and–growth process is usually a time sequence of pictures that represent process $\Theta$ at different temporal step; namely $\Theta_{n-1}$, $\Theta_{n}$. Thus, in view of (1), it is interesting to investigate $\\{G_{n}\\}$ and $\left\\{B_{n}\right\\}$; in particular, we shall estimate the maximal growth $G_{n}$ and the capacity functional of $B_{n}$. For the sake of simplicity, $Y$, $X$, $G$ and $B$ will denote RaCS $\Theta_{n}$, $\Theta_{n-1}$, $G_{n}$ and $B_{n}$ respectively (then $X\subseteq Y$). Thus, let us consider the following general definition. ###### Definition 0.3 __Let $Y$, $X$ be RaCS with $X\subseteq Y$. A _$X$ –decomposition of $Y$_ is a couple of RaCS $(G,B)$ for which ††margin: ††margin: $Y=(X\oplus G)\cup B.$ (2) Note that, since we can consider $\left(G,B\right)=\left(\left\\{0\right\\},Y\right)$, there always exists a $X$–decomposition of $Y$. It can happen that $G$ and $B$ in (2) are not unique. As example, let $Y=[0,1]$ and $X=\\{0\\}$, then both $(G_{1},B_{1})=(Y,Y)$ and $(G_{2},B_{2})=(X,Y)$ satisfy (2). As a consequence, since we can not distinguish between two different decompositions, we shall choose a maximal one according to the following proposition. ###### Proposition 0.4 _(See[30]) _ Let $Y$, $X$ be RaCS with $X\subseteq Y$. Then $G=Y\ominus\check{X}=\left\\{g\in\mathfrak{X}:g+X\subseteq Y\right\\}.$ (3) is the greatest RaCS, with respect to set inclusion, such that $(X\oplus G)\subseteq Y$. ###### Corollary 0.5 __The couple $(G=Y\ominus\check{X},B=Y\cap\overline{(X\oplus G)^{C}})$ is the max-min $X$–decomposition of $Y$. As a consequence, $(G,B)$ is a $X$–decomposition of $Y$ and for any other $X$–decomposition of $Y$, say $(G^{\prime},B^{\prime})$, then $G^{\prime}\subseteq G$ and $B^{\prime}\supseteq B$. In other words, if $X,G^{\prime},B^{\prime}$ are RaCS and $Y=(X\oplus G^{\prime})\cup B^{\prime}$, then $G=Y\ominus\check{X}\supseteq G^{\prime}$ and $Y=(X\oplus G)\cup B^{\prime}$. Let $\Theta$ be as in (1). From now on, $G_{n}$ denotes $\Theta_{n}\ominus\check{\Theta}_{n-1}$ that, as a consequence of Assumption 0.2, contains the origin. Moreover, we shall suppose (A-2) There exists $K\in\mathbb{F}^{\prime}_{b}$ such that $G_{n}=\Theta_{n}\ominus\check{\Theta}_{n-1}\subseteq K$ for every $n\in\mathbb{N}$. (A-3) For every $n\geq 1$, $\left(B_{n}\ominus\check{\Theta}_{n-1}\right)=\emptyset$ almost surely. Roughly speaking, Assumption 0.5 means that process $\Theta$ does not grow too “fast”, whilst Assumption 0.5 means that it cannot born something that, up to a translation, is larger (or equal) than what there already exists. ††margin: ††margin: Let us remark that Assumption 0.5 implies $\left\\{G_{n}\right\\}\subset\mathbb{F}^{\prime}_{k}$ and $X\oplus G_{n}=X+G_{n}$, for any RaCS $X$. ## 0.3 Estimators of $G$ On the one hand Proposition 0.4 gives a theoretical formula for $G$, but, on the other hand, in practical cases, data are bounded by some observation window and edge effects may cause problems. Hence, as the standard statistical scheme for spatial processes (e.g. [23]) suggests, we wonder if there exists a consistent estimator of $G$ as the observation window expands to the whole space $\mathfrak{X}$. ###### Proposition 0.6 __If $\left\\{W_{i}\right\\}_{i\in\mathbb{N}}\subset\mathbb{F}^{\prime}_{ck}$ is a convex averaging sequence of sets, then, for any $K\in\mathbb{F}^{\prime}_{k}$, $\mathfrak{X}=\bigcup_{i\in\mathbb{N}}W_{i}\ominus\check{K}$. In this case, we say that $\left\\{W_{i}\right\\}_{i\in\mathbb{N}}$ expands to $\mathfrak{X}$ and we shall write $W_{i}\uparrow\mathfrak{X}$. Proof. At first note that $\mathfrak{X}=\bigcup_{i\in\mathbb{N}}\textrm{Int}\ W_{i}$ and for any $i\in\mathbb{N}$, $W_{i}\subseteq W_{i+1}$. Let $x\in\mathfrak{X}$ and $K\in\mathbb{F}^{\prime}_{k}$. Note that, $x+K\in\mathbb{F}^{\prime}_{k}$ is a compact set. Then there exists a finite family of indices $I\subset\mathbb{N}$ such that, if $N=\max I$, then $x+K\subseteq\bigcup_{j\in I}\textrm{Int}\ W_{j}=\textrm{Int}\ W_{N}.$ Hence, we have that $x\in\textrm{Int}\ W_{N}\ominus\check{K}\subseteq W_{N}\ominus\check{K}$, i.e., for any $x\in\mathfrak{X}$, there exists $n_{0}\in\mathbb{N}$ such that $x\in W_{n_{0}}\ominus\check{K}$. $\blacksquare$ Let $W\in\left\\{W_{i}\right\\}_{i\in\mathbb{N}}$ be an observation window and let us denote by ${Y}_{W}$ and ${X}_{W}$, the (random) observation of $Y$ and $X$ through $W$, i.e. $Y\cap W$ and $X\cap W$ respectively. Let us consider the estimator of $G$ given by the maximal ${X}_{W}$–decomposition of ${Y}_{W}$: $\widehat{G}_{W}=\left({Y}_{W}\ominus{\check{X}}_{W}\right)$ (4) so that ${X}_{W}\oplus\widehat{G}_{W}\subseteq{Y}_{W}\subseteq W$. Notice that, whenever $Y$ and $X$ are bounded, then there exists $W_{j}\in\left\\{W_{i}\right\\}_{i\in\mathbb{N}}$ such that $Y\subseteq W_{j}$ and $\check{X}\subseteq W_{j}$, hence $\widehat{G}_{W_{j}}=Y\ominus\check{X}=G$. In other words, on the set $\left\\{\omega\in\Omega:X(\omega),Y(\omega)\textrm{ bounded}\right\\}$, the estimator (4) is consistent $\widehat{G}_{W_{i}}(Y,X|Y,X\textrm{ bounded})\rightarrow G,\quad\qquad\textrm{as }W_{i}\uparrow\mathfrak{X};$ otherwise, as we already said, if $Y$ and $X$ are unbounded, edge effects may cause problems and the estimator (4) is, in general, not consistent as we discussed in the following example. ###### Example 0.7 __Let $\mathfrak{X}=\mathbb{R}^{2}$, let us consider $X=\left(\\{x=0\\}\cup\\{y=0\\}\right)$ and $Y=X+B(0,1)$ where $B(0,1)$ is the closed unit ball centered in the origin. Surely $X\subset Y$, and they are unbounded. Note that $Y=(X+G)$ for any $G$ such that $\left(\\{0\\}\times[-1,1]\cup[-1,1]\times\\{0\\}\right)\subseteq G\subseteq B(0,1)$. On the other hand, by Proposition 0.4, there exists a unique $G$ that is the greatest set, with respect to set inclusion; in this case $G=[-1,1]\times[-1,1]$. Let us suppose $0\in W_{0}$ and let $W\in\left\\{W_{i}\right\\}_{i\in\mathbb{N}}$, then, by Equation (4), the estimator of $G$ is $\widehat{G}_{W}=\\{0\\}\neq G$. This is an edge effect due to the fact that, for every $G^{\prime}$ with $\left\\{0\right\\}\subset G^{\prime}\subseteq G$, it holds $\left({X}_{W}+G^{\prime}\right)\cap W^{C}\neq\emptyset$ and then ${X}_{W}+G^{\prime}\not\subseteq{Y}_{W}$ that does not agree with Proposition 0.4. Edge effects can be reduced by considering the following estimators of $G$††margin: ††margin: $\displaystyle\widehat{G}^{1}_{W}$ $\displaystyle=$ $\displaystyle\left({Y}_{W}\ominus{\check{X}}_{W\ominus\check{K}}\right)\cap K,$ (5) $\displaystyle\widehat{G}^{2}_{W}$ $\displaystyle=$ $\displaystyle\left(\left[{Y}_{W}\cup\left(\partial^{+K}_{W}{X}_{W}\right)\right]\ominus{\check{X}}_{W}\right)\cap K;$ (6) where $K$ is given in Assumption 0.5 and where $\left(\partial^{+K}_{W}{X}_{W}\right)=\overline{({X}_{W}+K)\setminus W}$. The role of $K$ will be clarified in Proposition 0.8 where it guarantees the monotonicity of $\widehat{G}^{1}_{W}$. Note that, estimators (5) (6) are bounded (i.e. compact) RaCS, moreover, if $Y$ and $X$ are bounded, then $\widehat{G}^{1}_{W_{j}}$, $\widehat{G}^{2}_{W_{j}}$ eventually coincide with the estimator (4); i.e. there exists $n_{0}$ such that for all $j\geq n_{0}$, $\widehat{G}_{W_{j}}=\widehat{G}^{1}_{W_{j}}=\widehat{G}^{2}_{W_{j}}=G$. Let us explain how $\widehat{G}^{1}_{W}$ and $\widehat{G}^{2}_{W}$ work. Estimator $\widehat{G}^{1}_{W}$ is obtained by reducing the information given by $X$ to the smaller window $W\ominus\check{K}$, whilst $Y$ is observed in $W$. Then $\widehat{G}^{1}_{W}$ is the greatest subset of $K$, with respect to set inclusion, such that ${X}_{W\ominus\check{K}}+\widehat{G}^{1}_{W}\subseteq{Y}_{W}$ (see Proposition 0.4). Estimator $\widehat{G}^{2}_{W}$ is obtained by observing $X$ in $W$ (and not $W\ominus\check{K}$), whilst $Y$ is increased (at least) by $\left(\partial^{+K}_{W}{X}_{W}\right)$, that is the greatest possible set of growth for $X$ outside of the observed window $W$. Then $\widehat{G}^{2}_{W}$ is the greatest subset of $K$, with respect to set inclusion, such that $({X}_{W}+\widehat{G}^{2}_{W})\cap W\subseteq{Y}_{W}$, or, alternatively, ${X}_{W}+\widehat{G}^{2}_{W}\subseteq{Y}_{W^{\prime}}$, where ${Y}_{W^{\prime}}={Y}_{W}\cup\left(\partial^{+K}_{W}{X}_{W}\right)$ (see Proposition 0.4). Note that by definition of Minkowski Subtraction $\displaystyle\widehat{G}^{1}_{W}=$ $\displaystyle\bigcap_{x\in{X}_{W\ominus\check{K}}}$ $\displaystyle x+\left((-x+K)\cap{Y}_{W}\right),$ $\displaystyle\widehat{G}^{2}_{W}=$ $\displaystyle\bigcap_{x\in{X}_{W}}$ $\displaystyle x+\left((-x+K)\cap Y_{W^{\prime}}\right);$ i.e. every $x\in{X}_{W\ominus\check{K}}$ (resp. $x\in{X}_{W}$) “grows” at most as $(-x+K)\cap{Y}_{W}$ (resp. $(-x+K)\cap{Y}_{W^{\prime}}$). Now, we are ready to show the consistency property of $\widehat{G}^{1}_{W_{i}}$ and $\widehat{G}^{2}_{W}$. In particular, Proposition 0.8 proves that $\widehat{G}^{1}_{W_{i}}$ decreases, with respect to set inclusion, to the theoretical $G$, whenever $W_{i}$ expands to the whole space ($W_{i}\uparrow\mathfrak{X}$). Proposition 0.9 proves that, for every $W\in\mathbb{F}^{\prime}$, $\widehat{G}^{2}_{W}$ is a better estimator than $\widehat{G}^{1}_{W}$ and hence it is a consistent estimator of $G$. ###### Proposition 0.8 __Let $Y$, $X$ be RaCS, let $0\in G=Y\ominus\check{X}\subseteq K$. The following statements hold for $\widehat{G}^{1}_{W}$: (1) $G\subseteq\widehat{G}^{1}_{W}$ for every $W$; (2) $\widehat{G}^{1}_{W_{2}}\subseteq\widehat{G}^{1}_{W_{1}}$ if $W_{2}\supseteq W_{1}$; (3) If $W_{i}\uparrow\mathfrak{X}$, then $\bigcap_{i\in\mathbb{N}}\widehat{G}^{1}_{W_{i}}=G$. Moreover, $\lim_{i\to\infty}\delta_{H}(\widehat{G}^{1}_{W_{i}},G)=0.$ (7) Proof. (1) Since $0\in K$, $\bigcap_{k\in K}-k+W=W\ominus\check{K}\subseteq W$ and then ${X}_{W\ominus\check{K}}\subseteq W$. Let $g\in G$, then $g+X\subseteq Y$. Since $g\in K$, last inclusion still holds when $X$ and $Y$ are substituted by ${X}_{W\ominus\check{K}}$ and ${Y}_{W}$ respectively: $g+{X}_{W\ominus\check{K}}\subseteq{Y}_{W}$. Thus $g\in\widehat{G}^{1}_{W}$ follows by Definition (5) and Proposition 0.4. (2) In order to obtain $\widehat{G}^{1}_{W_{2}}\subseteq\widehat{G}^{1}_{W_{1}}$, it is sufficient to prove that ${X}_{W_{1}\ominus\check{K}}+\widehat{G}^{1}_{W_{2}}\subseteq{Y}_{W_{1}}$ (8) since $\widehat{G}^{1}_{W_{1}}$ is the greatest set, with respect to set inclusion, for which the inclusion (8) holds. In fact, $W_{1}\ominus\check{K}\subseteq\left(W_{1}\ominus\check{K}\right)+K\subseteq W_{1}\subseteq W_{2}$, then ${X}_{W_{1}\ominus\check{K}}\subseteq{X}_{W_{2}}$. Let $x\in{X}_{W_{1}\ominus\check{K}}=X\cap\left(W_{1}\ominus\check{K}\right)$, then $x\in{X}_{W_{2}}$. By definition of $\widehat{G}^{1}_{W_{2}}$, we have $x+\widehat{G}^{1}_{W_{2}}\subseteq{Y}_{W_{2}}\subseteq Y.$ On the other hand, since $x\in W_{1}\ominus\check{K}$ and $\widehat{G}^{1}_{W_{2}}\subseteq K$, we have $x+\widehat{G}^{1}_{W_{2}}\subseteq\left(W_{1}\ominus\check{K}\right)+K\subseteq W_{1};$ i.e. $x+\widehat{G}^{1}_{W_{2}}$ is included both in $Y$ and in $W_{1}$. (3) Since $G\subseteq\bigcap_{i\in\mathbb{N}}\widehat{G}^{1}_{W_{i}}$, it remains to prove that $\bigcap_{i\in\mathbb{N}}\widehat{G}^{1}_{W_{i}}\subseteq G;$ i.e. if $g\in\widehat{G}^{1}_{W_{i}}$ for each $i\in\mathbb{N}$, then $g\in G$. Take $g\in\bigcap_{i\in\mathbb{N}}\widehat{G}^{1}_{W_{i}}$. By definition of $\widehat{G}^{1}_{W_{1}}$, we have $g+x\in Y\textrm{ for all }x\in{X}_{W_{i}\ominus\check{K}}\textrm{ and }\forall i\in\mathbb{N}.$ (9) By contradiction, assume $g\not\in G$. Then $g+X\not\subseteq Y$, i.e. there exists $\overline{x}\in X$ such that $\left(g+\overline{x}\right)\not\in Y$. On the one hand, Proposition 0.6 implies that there exists $j\in\mathbb{N}$ such that $\overline{x}\in W_{j}\ominus\check{K}$. On the other hand, Equation (9) implies $g+\overline{x}\in Y$ which is a contradiction. Thus Theorem 1.1.18 in [19] implies (7). $\blacksquare$ ###### Proposition 0.9 __For every $W\in\mathbb{F}^{\prime}$, $G\subseteq\widehat{G}^{2}_{W}\subseteq\widehat{G}^{1}_{W}$. Proof. Let us divide the proof in two parts; in the first one we prove that $\widehat{G}^{2}_{W}\subseteq\widehat{G}^{1}_{W}$, in the second one that $G\subseteq\widehat{G}^{2}_{W}$. Let $g\in\widehat{G}^{2}_{W}$ and $x\in X_{W\ominus\check{K}}$. Since $\widehat{G}^{2}_{W}\subseteq K$, we have $x+g\in\left(W\ominus\check{K}\right)+\widehat{G}^{2}_{W}\subseteq\left(W\ominus\check{K}\right)+K\subseteq W;$ (10) where we use properties of monotonicity of the Minkwoski Subtraction and Sum. Moreover, by definition of $\widehat{G}^{2}_{W}$, $x+g\in{Y}_{W},\qquad\textrm{or}\qquad x+g\in\left(\partial^{+K}_{W}{X}_{W}\right)\subseteq W^{C}.$ By (10), $x+g\in{Y}_{W}$. The arbitrary choice of $x\in{X}_{W\ominus\check{K}}$ completes the first part of the proof. For the second part, let $g\in G$ and $x\in{X}_{W}$. By definition of $G$, $x+g\in Y$. We have two cases: \- $x+g\in W$, and therefore $x+g\in{Y}_{W}$, \- $x+g\not\in W$. Since $x\in{X}_{W}$, $x+g\in\left({X}_{W}+G\right)\setminus W\subseteq\left(\partial^{+K}_{W}{X}_{W}\right).$ $\blacksquare$ ###### Corollary 0.10 __ $\widehat{G}^{2}_{W}$ is consistent (i.e. $\widehat{G}^{2}_{W}\downarrow G$ whenever $W\uparrow\mathfrak{X}$). Figure 1: We consider two pictures of a simulated birth–and–growth process, at two different time instants, that in our notations are $X$ and $Y$. Emphasizing the differences, we report here the magnified pictures of the true growth used for the simulation, the computed $\widehat{G}^{2}_{W}$, $\widehat{G}^{1}_{W}$ and $\widehat{G}^{1}_{W\ominus\check{K}}$. Note that they agree with Proposition 0.8 and Proposition 0.9 since $\widehat{G}^{1}_{W\ominus\check{K}}\supseteq\widehat{G}^{1}_{W}\supseteq\widehat{G}^{2}_{W}$. ### A General Definition of $\widehat{G}^{2}_{W}$. The following proposition shows that the estimator in (6) can be defined in an equivalent way by $\widehat{G}^{2}_{W}(Z)=\left\\{\left[{Y}_{W}\cup\left(\partial^{+K}_{W}Z\right)\right]\ominus{\check{X}}_{W}\right\\}_{K};$ where $\left(\partial^{+K}_{W}X\right)$ in (6) is substituted by $\left(\partial^{+K}_{W}Z\right)$ with ${X}_{W\setminus\left(W\ominus\check{K}\right)}\subseteq Z\subseteq W.$ (11) In other words, we are saying that, under condition (11), $\widehat{G}^{2}_{W}(Z)$ does not depend on $Z$. From a computational point of view, this means that $Z$ can be chosen in a way that reduces the computational costs. On the one hand, the best choice of $Z$ seems to be the smallest possible set, i.e. $Z={X}_{W\setminus\left(W\ominus\check{K}\right)}$. On the other hand, in order to get ${X}_{W\setminus\left(W\ominus\check{K}\right)}$, we have to compute $\left(W\ominus\check{K}\right)$ that may be costly if at least one between $W$ and $K$ has a “bad shape” (for instance it is not a rectangular one). ###### Proposition 0.11 __If $Z_{1},Z_{2}\in\mathfrak{P}^{\prime}$ both satisfy condition (11), then $\widehat{G}^{2}_{W}(Z_{1})=\widehat{G}^{2}_{W}(Z_{2})$. Proof. It is sufficient to prove: (1) $Z_{1}\subseteq Z_{2}$ implies $\widehat{G}^{2}_{W}(Z_{1})\subseteq\widehat{G}^{2}_{W}(Z_{2})$; (2) $\widehat{G}^{2}_{W}(W)\subseteq\widehat{G}^{2}_{W}\left({X}_{W\setminus\left(W\ominus\check{K}\right)}\right)$. In fact, (1) and (2) imply that $\widehat{G}^{2}_{W}(W)=\widehat{G}^{2}_{W}\left({X}_{W\setminus\left(W\ominus\check{K}\right)}\right)$. At the same time they imply $\widehat{G}^{2}_{W}(Z)=\widehat{G}^{2}_{W}\left({X}_{W\setminus\left(W\ominus\check{K}\right)}\right)$ holds for every $Z$ that satisfies (11); that is the thesis. _STEP (1)_ is a consequence of the following implications $\displaystyle Z_{1}\subseteq Z_{2}$ $\displaystyle\Rightarrow$ $\displaystyle Z_{1}+K\subseteq Z_{2}+K,$ $\displaystyle\Rightarrow$ $\displaystyle{Y}_{W}\cup\left[\left(Z_{1}+K\right)\setminus W\right]\subseteq{Y}_{W}\cup\left[\left(Z_{2}+K\right)\setminus W\right],$ $\displaystyle\Rightarrow$ $\displaystyle\widehat{G}^{2}_{W}(Z_{1})\subseteq\widehat{G}^{2}_{W}(Z_{2});$ where the last one holds since $X_{1}\ominus Y\subseteq X_{2}\ominus Y$ if $X_{1}\subseteq X_{2}$ (see [30]). Before proving the second step, we show that $\widehat{G}^{2}_{W}\left(Z\right)=\widehat{G}^{2}_{W}\left({Z}_{W\setminus\left(W\ominus\check{K}\right)}\right)$ for all $Z$ that satisfies (11). This statement is true if $\left({Z}_{W\setminus\left(W\ominus\check{K}\right)}+K\right)\setminus W$ and $\left(Z+K\right)\setminus W$ are the same set. Since Minkowski sum is distributive with respect to union, we get $\displaystyle\left(Z+K\right)\setminus W$ $\displaystyle=$ $\displaystyle\left[\left({Z}_{W\setminus\left(W\ominus\check{K}\right)}\cup{Z}_{W\ominus\check{K}}\right)+K\right]\setminus W$ $\displaystyle=$ $\displaystyle\left[\left({Z}_{W\setminus\left(W\ominus\check{K}\right)}+K\right)\setminus W\right]\cup\left[\left({Z}_{W\ominus\check{K}}+K\right)\setminus W\right].$ Then we have to prove that $\left[\left({Z}_{W\ominus\check{K}}+K\right)\setminus W\right]=\emptyset$ : $\displaystyle\left({Z}_{W\ominus\check{K}}+K\right)\setminus W$ $\displaystyle=$ $\displaystyle\left\\{\left[Z\cap\left(W\ominus\check{K}\right)\right]+K\right\\}\setminus W$ $\displaystyle\subseteq$ $\displaystyle\left\\{\left(Z+K\right)\cap\left[\left(W\ominus\check{K}\right)+K\right]\right\\}\setminus W$ $\displaystyle\subseteq$ $\displaystyle\left[\left(Z+K\right)\cap W\right]\setminus W=\emptyset.$ _STEP (2)_. Since $\widehat{G}^{2}_{W}\left({X}_{W}\right)=\widehat{G}^{2}_{W}\left({X}_{W\setminus\left(W\ominus\check{K}\right)}\right)$, thesis becomes $\widehat{G}^{2}_{W}(W)\subseteq\widehat{G}^{2}_{W}({X}_{W})$. Let $g\in\widehat{G}^{2}_{W}(W)$. We must prove $g\in\widehat{G}^{2}_{W}({X}_{W})$, i.e. for every $x\in{X}_{W}$ $g+x\in{Y}_{W},\qquad\textrm{ or }\qquad g+x\in\left({X}_{W}+K\right)\setminus W.$ Since $g\in\widehat{G}^{2}_{W}(W)$, for any $x\in{X}_{W}$ we can have two possibilities (a) $g+x\in{Y}_{W}$, (b) $g+x\in\left(W+K\right)\setminus W$. It remains to prove that (b) implies $g+x\in\left({X}_{W}+K\right)\setminus W$. In particular, (b) implies $g+x\in W^{C}$. At the same time $g+x\in{X}_{W}+K$, i.e. $g+x\in\left({X}_{W}+K\right)\setminus W$. $\blacksquare$ ## 0.4 Hitting Function Associated to $B$ In many practical cases, an observer, through a window $W$ and at two different instants, observes the nucleation and growth processes namely $X$ and $Y$. According to Section 0.3 we can estimate $G$ via the consistent estimator $\widehat{G}^{2}_{W}$ or $\widehat{G}^{1}_{W}$ (in the following we shall write $\widehat{G}_{W}$ meaning one of them). From the birth–and–growth process point of view, it is also interesting to test whenever the nucleation process $B=\left\\{B_{n}\right\\}_{n\in\mathbb{N}}$ is a specific RaCS (for example a Boolean model or a point process). In general, we cannot directly observe the $n$–th nucleation $B_{n}$ since it can be overlapped by other nuclei or by their evolutions. Nevertheless, we shall infer on the hitting function associated to the nucleation process $T_{B_{n}}(\cdot)$. Let us consider the decomposition given by (2) $Y=(X+G)\cup B$ then the following proposition is a consequence of Remark 0.1. ###### Proposition 0.12 __If $(G,B)$ is a $X$–decomposition of $Y$ such that $B$ is independent on $X$ and on $G$, then, for each $K\in\mathbb{F}_{k}$, $T_{Y}\left(K\right)=T_{X+G}\left(K\right)+T_{B}\left(K\right)-T_{X+G}\left(K\right)T_{B}\left(K\right),$ that, in terms of $Q_{\cdot}(K)=\left(1-T_{\cdot}\left(K\right)\right)$, is equivalent to $Q_{Y}(K)=Q_{B}(K)Q_{X+G}(K).$ In other words, the probability for the exploring set $K$ to miss $Y$ is the probability for $K$ to miss $B$ multiplied by the probability for $K$ to miss $X+G$. ###### Remark 0.13 __Working with data we shall consider two estimators of the hitting function (we refer to[23, p. 57–63] and references therein). In particular, if $X$ is a stationary ergodic RaCS, then $T_{X}(\cdot)$ can be estimated by a single realization of $X$ and two empirical estimators are given by $\widehat{T}_{X,W}(K)=\frac{\mu_{\lambda}\left(\left(X+\check{K}\right)\cap\left(W\ominus K_{0}\right)\right)}{\mu_{\lambda}\left(W\ominus K_{0}\right)},\quad K\in\mathbb{F}_{k};$ where $\mu_{\lambda}$ is the Lebesgue measure on $\mathfrak{X}=\mathbb{R}^{d}$ and $K_{0}$ is a compact set such that $K\subset K_{0}$ for all $K\in\mathbb{F}_{k}$ of interest. A _regular closed_ set in $\mathfrak{X}$ is a closed set $G\in\mathbb{F}^{\prime}$ for which $G=\overline{\textrm{Int}\ G}$; i.e. $G$ is the closure (in $\mathfrak{X}$) of its interior. ###### Proposition 0.14 __Let $G\in\mathbb{F}^{\prime}_{k}$ be a regular closed subset in $\mathfrak{X}$. Then, for every $X\in\mathbb{F}^{\prime}$, $X+G$ is a regular closed set. Proof. Since $X+G$ is a closed set, $\overline{\textrm{Int}\ (X+G)}\subseteq X+G$. It remains to prove that $X+G\subseteq\overline{\textrm{Int}\ (X+G)}$. Let $y\in X+G$, then there exists $x\in X$ and $g\in G$ such that $y=x+g$. If $g\in\textrm{Int}\ G$, then there exists an open neighborhood of $g$ for which $U(g)\subseteq\textrm{Int}\ G$ and $x+U(g)$ is an open neighborhood of $x+g$ included in $X+G$; i.e. $x+g\in\textrm{Int}\ (X+G)$. On the other hand, let $g\in\partial G=G\setminus\textrm{Int}\ G$, then there exists $\left\\{g_{n}\right\\}_{n\in\mathbb{N}}\subset G$ such that $g_{n}\rightarrow g$ and $g_{n}\in\textrm{Int}\ G$, for all $n\in\mathbb{N}$. Thus, for every $n\in\mathbb{N}$, $x+g_{n}$ is an interior point of $X+G$ and $x+g_{n}\rightarrow x+g\in\overline{\textrm{Int}\ (X+G)}$. $\blacksquare$ ###### Proposition 0.15 _(See[23, Theorem 4.5 p. 61] and references therein) _ Let $X$ be an ergodic stationary random closed set. If the random set $X$ is almost surely regular closed $\sup_{\textrm{\tiny$\begin{array}[]{c}K\in\mathbb{F}_{k}\\\ K\subseteq K_{0}\end{array}$}}\left|\widehat{T}_{X,W}(K)-T_{X}(K)\right|\rightarrow 0,\quad\textrm{a.s.}$ (12) as $W\uparrow\mathfrak{X}$ and for every $K_{0}\in\mathbb{F}^{\prime}$. ###### Remark 0.16 __Proposition 0.14, together to Equation (1) means that, if $\left\\{G_{n}\right\\}_{n\in\mathbb{N}}$ is a sequence of almost surely regular closed sets, then $\left\\{\Theta_{n}\right\\}_{n\in\mathbb{N}}$ is so. The following Theorem shows that the hitting functional ${Q}_{B}$ of the hidden nucleation process can be exstimated by the observable quantity $\widetilde{Q}_{B,W}$, where for every $K\in\mathbb{F}_{k}$, $\widetilde{Q}_{B,W}(K):=\frac{\widehat{Q}_{Y,W}(K)}{\widehat{Q}_{X+\widehat{G}_{\textrm{\tiny$W$}},W}(K)},$ (13) and $\widehat{G}_{W}$ is given by (5) or (6). ###### Theorem 0.17 __Let $X,Y$ be two RaCS a.s. regular closed. Let $(G,B)$ be a $X$–decomposition of $Y$ with $B$ a stationary ergodic RaCS independent on $G$ and $X$. Assume that $G$ is an a.s. regular closed set and $\widetilde{Q}_{B,W}$ defined in Equation (13). Then, for any $K\in\mathbb{F}_{k}$, $\left|\widetilde{Q}_{B,W}(K)-Q_{B}(K)\right|\mathop{\longrightarrow}_{W\uparrow\mathfrak{X}}0,\quad\textrm{a.s.}$ Proof. Let $K\in\mathbb{F}_{k}$ be fixed. For the sake of simplicity, $Q_{\cdot}$, $\widetilde{Q}_{\cdot}$ and $\widehat{Q}_{\cdot}$ denote $Q_{\cdot}(K)$, $\widetilde{Q}_{\cdot,W}(K)$ and $\widehat{Q}_{\cdot,W}(K)$ respectively. Thus, $\left|\widetilde{Q}_{B}-Q_{B}\right|=\left|\frac{\widehat{Q}_{Y}}{\widehat{Q}_{X+\widehat{G}_{W}}}-\frac{Q_{Y}}{Q_{X+G}}\right|=\left|\frac{\widehat{Q}_{Y}Q_{X+G}-Q_{Y}\widehat{Q}_{X+\widehat{G}_{W}}}{\widehat{Q}_{X+\widehat{G}_{W}}Q_{X+G}}\right|.$ Since $Y\supseteq X+\widehat{G}_{W}$, $\widehat{Q}_{X+\widehat{G}_{W}}>\widehat{Q}_{Y}$. Accordingly to (12), $\widehat{Q}_{Y}$ converges to $Q_{Y}$ that is a positive quantity. Thus, thesis is equivalent to prove that $\left|\widehat{Q}_{Y}Q_{X+G}-Q_{Y}\widehat{Q}_{X+\widehat{G}_{W}}\right|\rightarrow 0,\quad\textrm{a.s.}$ as $W\uparrow\mathfrak{X}$. The following inequalities hold $\displaystyle\left|\widehat{Q}_{Y}Q_{X+G}-Q_{Y}\widehat{Q}_{X+\widehat{G}_{W}}\right|$ $\displaystyle\leq$ $\displaystyle Q_{X+G}\left|\widehat{Q}_{Y}-Q_{Y}\right|+Q_{Y}\left|Q_{X+G}-\widehat{Q}_{X+\widehat{G}_{W}}\right|$ $\displaystyle\leq$ $\displaystyle Q_{X+G}\left|\widehat{Q}_{Y}-Q_{Y}\right|+$ $\displaystyle Q_{Y}\left|Q_{X+G}-Q_{X+\widehat{G}_{W}}\right|+Q_{Y}\left|Q_{X+\widehat{G}_{W}}-\widehat{Q}_{X+\widehat{G}_{W}}\right|.$ Proposition 0.2 and Proposition 0.14 guarantee that $X+G$ is a stationary ergodic RaCS and a.s. regular closed, then we can apply (12) to the first and the third addends. It remains to prove that $\left|Q_{X+G}-Q_{X+\widehat{G}_{W}}\right|\rightarrow 0\quad\textrm{as }W\uparrow\mathfrak{X}.$ (14) Since Minkowski sum is a continuous map from $\mathbb{F}\times\mathbb{F}_{k}$ to $\mathbb{F}$ (see [30]), $\widehat{G}_{W}\downarrow G$ a.s. implies $X+\widehat{G}_{W}\downarrow X+G$ a.s. As a consequence, we get that $X+\widehat{G}_{W}\downarrow X+G$ in distribution [28, p. $182$], which is Equation (14). $\blacksquare$ ## References * [1] G. Aletti, E. G. Bongiorno, and V. Capasso, A set–valued framework for birth–and–growth processes. (submitted). * [2] G. Aletti and D. Saada, Survival analysis in Johnson–Mehl tessellation, Stat. Inference Stoch. Process., 11 (2008) 55–76. * [3] M. Burger, V. Capasso, and A. Micheletti, An extension of the Kolmogorov–Avrami formula to inhomogeneous birth–and–growth processes, in: G. Aletti et al., (Eds.), Math Everywhere, Springer, Berlin, 2007 63–76. * [4] M. Burger, V. Capasso, and L. Pizzocchero, Mesoscale averaging of nucleation and growth models, Multiscale Model. Simul., 5 (2006) 564–592. * [5] V. Capasso, (Ed.), Mathematical Modelling for Polymer Processing. Polymerization, Crystallization, Manufacturing, Mathematics in Industry, 2, Springer–Verlag, Berlin, 2003. * [6] V. Capasso, On the stochastic geometry of growth, in: Sekimura, T. et al., (Eds.), Morphogenesis and Pattern Formation in Biological Systems, Springer, Tokyo, 2003 45–58. * [7] V. Capasso and E. Villa, Survival functions and contact distribution functions for inhomogeneous, stochastic geometric marked point processes, Stoch. Anal. Appl., 23 (2005) 79–96. * [8] S. N. Chiu, Johnson–Mehl tessellations: asymptotics and inferences, in: Probability, finance and insurance, World Sci. Publ., River Edge, NJ, 2004 136–149. * [9] S. N. Chiu, I. S. Molchanov, and M. P. Quine, Maximum likelihood estimation for germination–growth processes with application to neurotransmitters data, J. Stat. Comput. Simul., 73 (2003) 725–732. * [10] N. Cressie, Modeling growth with random sets, in: Spatial statistics and imaging (Brunswick, ME, 1988), IMS Lecture Notes Monogr. Ser., Vol. 20, Inst. Math. Statist., Hayward, CA, 1991 31–45. * [11] D. J. Daley and D. Vere-Jones, An Introduction to the Theory of Point Processes. Vol. I, Probability and its Applications, Springer–Verlag, New York, second edition, 2003. * [12] T. Erhardsson, Refined distributional approximations for the uncovered set in the Johnson–Mehl model, Stochastic Process. Appl., 96 (2001) 243–259. * [13] W. Fei and R. Wu, Doob’s decomposition theorem for fuzzy (super) submartingales, Stochastic Anal. Appl., 22 (2004) 627–645. * [14] Y. Feng, Decomposition theorems for fuzzy supermartingales and submartingales, Fuzzy Sets and Systems, 116 (2000) 225–235. * [15] Y. Feng and X. Zhu, Semi-order fuzzy supermartingales and submartingales with continuous time, Fuzzy Sets and Systems, 130 (2002) 75–86. * [16] H. J. Frost and C. V. Thompson, The effect of nucleation conditions on the topology and geometry of two–dimensional grain structures, Acta Metallurgica, 35 (1987) 529–540. * [17] J. Herrick, S. Jun, J. Bechhoefer, and A. Bensimon, Kinetic model of DNA replication in eukaryotic organisms, J.Mol.Biol., 320 (2002) 741–750. * [18] K. Keimel and W. Roth, Ordered Cones and Approximation, Lecture Notes in Mathematics, Vol. 1517, Springer–Verlag, Berlin, 1992\. * [19] S. Li, Y. Ogura, and V. Kreinovich, Limit Theorems and Applications of Set–Valued and Fuzzy Set–Valued Random Variables, Kluwer Academic Publishers Group, Dordrecht, 2002. * [20] G. Matheron, Random Sets and Integral Geometry, John Wiley & Sons, New York-London-Sydney, 1975. * [21] A. Micheletti, S. Patti, and E. Villa, Crystal growth simulations: a new mathematical model based on the Minkowski sum of sets, in: D.Aquilano et al., (Eds.), Industry Days 2003-2004, volume 2 of The MIRIAM Project, Esculapio, Bologna, 2005 130–140. * [22] I. S. Molchanov, Limit Theorems for Unions of Random Closed Sets, Lecture Notes in Mathematics, Vol. 1561, Springer–Verlag, Berlin, 1993\. * [23] I. S. Molchanov, Statistics of the Boolean Model for Practitioners and Mathematicians, Wiley, Chichester, 1997. * [24] I. S. Molchanov and S. N. Chiu, Smoothing techniques and estimation methods for nonstationary Boolean models with applications to coverage processes, Biometrika, 87 (2000) 265–283. * [25] J. Møller, Random Johnson–Mehl tessellations, Adv. in Appl. Probab., 24 (1992) 814–844. * [26] J. Møller, Generation of Johnson–Mehl crystals and comparative analysis of models for random nucleation, Adv. in Appl. Probab., 27 (1995) 367–383. * [27] J. Møller and M. Sørensen, Statistical analysis of a spatial birth–and–death process model with a view to modelling linear dune fields, Scand. J. Statist., 21 (1994) 1–19. * [28] H. T. Nguyen, An Introduction to Random Sets, Chapman & Hall/CRC, Boca Raton, FL, 2006. * [29] H. Rådström, An embedding theorem for spaces of convex sets, Proc. Amer. Math. Soc., 3 (1952) 165–169. * [30] J. Serra, Image Analysis and Mathematical Morphology, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], London, 1984\. * [31] D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic Geometry and its Applications, John Wiley & Sons Ltd., Chichester, second edition, 1995. * [32] P. Terán, Cones and decomposition of sub- and supermartingales. Fuzzy Sets and Systems, 147 (2004) 465–474.

arxiv-papers

2024-09-04T02:48:54.390696

0803.2713

# Layered Structures Favor Superconductivity in Compressed Solid SiH4 X. J. Chen1,2,3, J. L. Wang1, V. V. Struzhkin2, H. K. Mao2, R. J. Hemley,2 and H. Q. Lin1 1Department of Physics and Institute of Theoretical Physics, Chinese University of Hong Kong, Hong Kong, China 2Geophysical Laboratory, Carnegie Institution of Washington, Washington, DC 20015, USA 3School of Physics, South China University of Technology, Guangzhou 510640, China ###### Abstract The electronic and lattice dynamical properties of compressed solid SiH4 have been calculated in the pressure range up to 300 GPa with density functional theory. We find that structures having a layered network with eight-fold SiH8 coordination favor metallization and superconductivity. SiH4 in these layered structures is predicted to have superconducting transition temperatures ranging from 20 to 80 K, thus presenting new possibilities for exploring high temperature superconductivity in this hydrogen-rich system. ###### pacs: 74.10.+v, 74.70.Ad, 74.62.Fj SiH4 offers a source for high purity silicon in epitaxial and thin film deposition which is at the base of electronics and microdevices. There is ongoing interest in this material as well due to the suggestion of Ashcroft ashc that SiH4 would eventually undergo a transition to metallic and then a superconducting state at pressures considerably lower than may be necessary for solid hydrogen. Exploring the possibility of metallic hydrogen has long been a major driving force in high-pressure condensed matter science and remains an important challenge in modern physics and astrophysics. Recent experimental work on SiH4, using diamond-anvil cell techniques, has revealed an enhanced reflectivity with increasing pressure sun ; chen . It was found chen that solid SiH4 becomes opaque at 27-30 GPa and exhibits Drude-like behavior at around 60 GPa, signalling the onset of pressure-induced metallization. Structural information is the primary step toward understanding these observed electronic properties. SiH4 has a rich phase diagram with at least seven known phases chen ; clus ; wild . Only one solid phase has been reported in the pressure range between 10 and 25 GPa and at room temperature, with a monoclinic structure (Space group $P2_{1}/c$) degt . The very low hydrogen scattering cross section of hydrogen- containing materials in all diffraction methods makes structural determination very difficult, specifically in determining the H positions. Although the neutron diffraction is powerful in detecting the H-bonding structures, the current accessible pressure range of this technique is limited to 30 GPa ding . Therefore it is still impossible to use neutron diffraction to obtain the interesting structural information of SiH4 in the metallic state. The challenge of experimentally or theoretically determining the high-pressure structures of SiH4 is still enormous. The sequence of SiH4 structures provides the basis for understanding whether the material is a favorable candidate of a high temperature superconductor. For a metallic $Pman$ SiH4 phase, Feng et al. feng obtained a superconducting transition temperature $T_{c}$ of 166 K at 202 GPa by using the electron- phonon coupling strength for lead under ambient pressure as both materials have the same characteristic density of states per volume at relevant pressures. Pickard and Needs pick also studied the structural properties of SiH4 and mentioned the possibility of superconductivity in a $C2/c$ phase. All previous work degt ; feng ; pick was done without including the calculations of the phonon spectra and electron-phonon coupling parameters. Later phonon calculations by Yao $et$ $al.$ yao showed that the $Pman$ structure is in reality not stable and that a new $C2/c$ structure is dynamically stable from 65 to 150 GPa. This $C2/c$ SiH4 phase was predicted to exhibit superconductivity close to 50 K at 125 GPa yao . It is not known whether there exists a common structural feature that favors superconductivity in metallic SiH4, and no study on superconductivity with other stable structures has been attempted. In this Letter we report a theoretical study of superconductivity in compressed solid SiH4 including structural, electronic, and vibrational calculations. We find six energetically favorable structures in which the $P\overline{1}$, $Cmca$, and $C2/c$ structures have layered networks with eight-fold SiH8 coordination. This layered feature favors metallization and superconductivity. The layered phases are predicted to have $T_{c}$’s in the range of 20 and 80 K, suggesting that SiH4 is indeed a good candidate for high-temperature superconductivity. To study the structural and electronic behavior of SiH4 over a wide range of pressure, we used the Perdew-Burke-Ernzerhof generalized gradient approximation (GGA) density functional and projector augmented wave method as implemented in the Vienna $ab$ $initio$ simulation package (VASP) vasp . An energy cutoff of 450 eV is used for the plane wave basis sets, and 16$\times$16$\times$16 and 8$\times$8$\times$8 Monkhorst-Pack $k$-point grids are used for Brillouin zone sampling of two SiH4 molecular cells and four SiH4 molecular cells, respectively. The lattice dynamical and superconducting properties were calculated by the Quantum-Espresso package pwscf using Vanderbilt-type ultrasoft potentials with a cut-off energy of 25 Ry and 200 Ry for the wave functions and the charge density, respectively. 12$\times$12$\times$12 Monkhorst-Pack $k$-point grids with Gaussian smearing of 0.02 Ry were used for the phonon calculations at 4$\times$4$\times$4 $q$-point mesh and double $k$-point grids were used for calculation of the electron-phonon interaction matrix element. Figure 1: (color online) The enthalpy versus pressure for competitive structures of SiH4. The enthalpy of the $I4_{1}/a$ phase is taken as the reference point. We performed a systematic study of the phase stability of SiH4 based on ab initio first-principles calculations. Out of more than one hundred structures we studied, six new polymorphs of SiH4 with low enthalpies are found in the pressure range from 0 to 300 GPa. In Fig. 1 we plot the pressure dependence of their enthalpies along with the results for the $C2/c$ and $I4_{1}/a$ structure reported previously pick . A monoclinic structure with $P2_{1}/c$ symmetry has the lowest enthalpy below 27 GPa, in good agreement with the recent experiments degt . The $P2_{1}/c$ structure consists of four isolated covalently bonded SiH4 tetrahedra with the H atom of one molecule pointing away from the H atoms of a neighboring molecule. We find that a face-centered orthorhombic structure with $Fdd2$ symmetry appears stable from 27 to 60 GPa. These two phases are insulating. Figure 2: (color online) The energetically most favorable structures computed for SiH4 polymorphs at various pressures. Near 50 GPa, there are three other competitive, low-enthalpy structures with the $C2/c$, $I4_{1}/amd$, and $I4_{1}/a$ symmetry. Our $C2/c$ structure was based on the UI4 arrangement, which as exhibits a layered structure. However, the $C2/c$ structure predicted by Pickard and Needs pick forms three- dimensional networks at high pressures. For clarification, we name their structure as $C2/c$(3D) and our layered structure as $C2/c$(2D). There are subtle differences in band structures between the $C2/c$ predicted by Yao et al. yao and our $C2/c$(2D). Our $C2/c$(2D) structure is composed of six-fold coordinated SiH6 octahedra inside a layer, but Si-H bonding between different layers is absent. It is still dynamically stable up to 250 GPa. However, the stability of the $C2/c$ phase of Yao et al. yao is only stable up to 150 GPa. On compression, the six-fold coordinated SiH6 octahedra are transformed into eight-fold coordinated SiH8 dodecahedra with a S2H2 bridge-like bonding arrangement within the layer. Above 110 GPa, the $C2/c$(2D) structure becomes metallic. The $I4_{1}/amd$ SiH4 phase is composed of eight-fold coordinated SiH8 dodecahedra and every Si atom shares two H atoms with other Si atoms, as in the $I4_{1}/a$ structure. The $I4_{1}/amd$-type SiH4 is a semimetal between 40 and 70 GPa. Figure 3: (color online) (a) Electronic band structure for the $Cmca$ SiH4 phase at 250 GPa. The horizontal line shows the location of the Fermi level. (b) The phonon dispersion and phonon density of states (PDOS) projected on Si and H atoms for $Cmca$ SiH4 at 250 GPa. The $I4_{1}/a$ structure was found to have the lowest enthalpy over a wide pressure range (60 to 220 GPa), consistent with previous calculations pick . In an early powder x-ray diffraction study sear , $I4_{1}/a$ was considered as one of most plausible structures for the low-temperature phase II. Raman measurements chen indicate that this structure may not exist for SiH4 under high pressure and at room temperature. SiH4 in the $I4_{1}/a$ structure is believed to be stable only at low temperature. Between 220-270 GPa, a metallic $Cmca$ structure has the lowest enthalpy. Upon further compression, the $C2/c$(3D) phase possesses the lowest enthalpy between 270-300 GPa, which confirms the previous calculations pick . The $Cmca$ phase is also a layered structure that consists of eight-fold coordinated SiH8 dodecahedra. In each layer, the Si-H bonding arrangement is the same as that in the $I4_{1}/a$ structure. Thus, the $Cmca$ phase can be viewed as the two-dimensional analogue of the $I4_{1}/a$ phase. In the 60 to 100 GPa range, we found a metallic triclinic structure with $P\overline{1}$ space group with eight-fold SiH8 coordination. It has almost the same enthalpy as the $Cmca$ structure between 60 and 100 GPa. Both of their enthalpies are within 0.25 eV of the insulating $I4_{1}/a$ structure. As pressure is increased, the $P\overline{1}$ structure transforms gradually into the $Cmca$ structure. It is instructive to note that the $P\overline{1}$ and $Cmca$ phases are metallic over the pressure regime between 60 GPa and 270 GPa, which is in a good agreement with recent experiments chen . Although the insulating $I4_{1}/a$ phase has the lowest enthalpy between 60 and 220 GPa, both $P\overline{1}$ and $Cmca$ are good candidates for metallic phases in this regime. We thus obtain six energetically favorable structures for SiH4 at high pressures. The atomic arrangements for each of these structures is shown in Fig. 2. The $P\overline{1}$, $Cmca$, and $C2/c$(2D) phases have layered structures and are metallic. Figure 4: (color online) Electron-phonon spectral function $\alpha^{2}F(\omega)$ vs frequency $\omega$ of metallic SiH4 with the $Cmca$, $C2/c$(2D) and $C2/c$(3D) structures at various pressures. Figure 3(a) shows the calculated band structure for $Cmca$ SiH4 at 250 GPa. It can be seen that the $Cmca$ structure is metallic. The valence bands cross the Fermi level $E_{F}$ along the Y$\Gamma$ direction, while the conduction bands cross $E_{F}$ near the $\Gamma$ point. Upon compression, the conduction band crossing oss $E_{F}$ at the $\Gamma$ point shifts lower in energy, while the valence band across $E_{F}$ along the Y$\Gamma$ direction only moves up slightly in energy. The net effect of the pressure-induced band shifts is to increase the volume of the Fermi surface and the phase space for the electron- phonon interaction. The structural stability of each SiH4 phase has been examined through lattice dynamics calculations. The typical results of the phonon dispersion and projected phonon density of states for the $Cmca$ SiH4 at 250 GPa are displayed in Fig. 3(b). The $Cmca$ stability is confirmed by the absence of imaginary frequency modes. There are weak interactions between the Si framework and H atoms over the whole frequency range. The heavy Si atoms dominate the low-frequency vibrations, and the light H atoms contribute significantly to the high-frequency modes. Three separate regions of bands can be recognized. The modes for the frequencies below 750 cm-1 are mainly due to the motions of Si. The bands around 200 cm-1 are caused by acoustic phonons. The Si-H-Si bending vibrations dominate the intermediate-frequency region between 750 and 1200 cm-1. At high frequencies above 1200 cm-1, the phonon spectrum belongs to the Si-H bond stretching vibrations. Figure 5: (color online) Calculated (a) logarithmic average phonon frequency $\omega_{\rm log}$, (b) electron-phonon coupling parameter $\lambda$, and (c) superconducting transition temperature $T_{c}$ of SiH4 with the $Cmca$, $C2/c$(2D), and $C2/c$(3D) structure as a function of pressure up to 300 GPa. The electron-phonon spectral function $\alpha^{2}F(\omega)$ is essential in determining the electron-phonon coupling strength $\lambda$ and logarithmic average phonon frequency $\omega_{\rm log}$. We have calculated $\alpha^{2}F(\omega)$ for the metallic SiH4 phases in the pressure range of interest. Figure 4 shows the results for the $Cmca$, $C2/c$(2D) and $C2/c$(3D) structures over the pressure range from 70 to 300 GPa. Below 600 cm-1 the major contributions to $\alpha^{2}F(\omega)$ come from the phonon modes involving Si-Si vibrations, and the remaining part of the electron-phonon coupling is mainly due to the phonon modes involving H-H vibrations. The $\alpha^{2}F(\omega)$ on the wide high-energy side is significantly higher than that on the narrow low-energy side. Thus the high-energy H-H vibrations dominate the total $\lambda$ value. Among these metallic structures, the $C2/c$(2D) phase at 250 GPa has a relatively large $\alpha^{2}F(\omega)$ over the entire frequency range studied, resulting in a large $\lambda$. We now examine whether superconductivity in metallic SiH4 is possible, using the $T_{c}$ equation derived by Allen-Dynes alle . In the calculations, we took the Coulomb pseudopotential $\mu^{*}$ to be 0.1 which was found to reproduce $T_{c}$ in MgB2 sing . Figure 5 shows the pressure dependence of $\omega_{\rm log}$, $\lambda$, and $T_{c}$ for SiH4 with the metallic $Cmca$, $C2/c$(2D) and $C2/c$(3D) structures. The calculated $\omega_{\rm log}$ decreases [increases] with pressure for the $C2/c$(2D) [$C2/c$(3D)] structure. However, $\lambda$ changes with pressure in an opposite sense to $\omega_{\rm log}$ for both structures. For the $Cmca$ structure, both $\omega_{\rm log}$ and $\lambda$ do not show a monotonic pressure dependence. In these metallic phases studied, the variation of $T_{c}$ with pressure is found to resemble the $\lambda$ behavior. It is therefore indicated that the pressure effect on $T_{c}$ in SiH4 is primarily controlled by $\lambda$. The $Cmca$ phase has a $T_{c}$ of 75 K even at 70 GPa. For the $C2/c$(2D) phase, we calculate a relatively large $T_{c}$ as high as 80 K at 250 GPa. A decreasing $T_{c}$ from 47.7 K at 200 GPa to 26.1 K at 300 GPa is obtained for the $C2/c$(3D) phase. The $P\overline{1}$ structure is also estimated to have a $T_{c}$ of 46.6 K at 70 GPa. The current results thus suggest new possibilities for exploring high temperature superconductivity in this hydrogen-rich system. In summary, we have investigated the structural stability of silane under pressure up to 300 GPa. The $P2_{1}/c$ phase is confirmed to be a good candidate for the low-pressure insulating phase. Between 27 and 60 GPa, $Fdd2$ is predicted to be the structure of another insulating phase. At higher pressures, silane enters the metallic state having a structure with $P\overline{1}$ symmetry. As pressure is further increased, the $P\overline{1}$ structure transforms gradually into the $Cmca$ structure. The three-dimensional $C2/c$ structure is most stable only after 270 GPa. The layered feature of this material favors metallization and superconductivity. The relatively high transition temperatures in metallic silane are mainly attributed to the strong electron-phonon coupling due to the phonon modes involving H-H vibrations. We are grateful to J. S. Tse, R. E. Cohen, and S. A. Gramsch for discussions and comments. This work was supported by the HKRGC (402205); the U.S. DOE-BES (DEFG02-02ER34P5), DOE-NNSA (DEFC03-03NA00144), and NSF (DMR-0205899). X.J.C. wishes to thank CUHK for kind hospitality during the course of this work. When revising this manuscript, we were aware of the discovery of superconductivity in SiH4 [M. I. Eremets et al., Science 319, 1506 (2008)]. ## References * (1) N. W. Ashcroft, Phys. Rev. Lett. 92, 187002 (2004). * (2) L. L. Sun, A. L. Ruoff, C. S. Zha, and G. Stupian, J. Phys.: Condens. Matter 18, 8573 (2006); 19, 479001 (2007). * (3) X. J. Chen, V. V. Struzhkin, Y. Song, A. F. Goncharov, M. Ahart, Z. X. Liu, H. K. Mao, and R. J. Hemley, Proc. Natl. Acad. Sci. U.S.A. 105, 20 (2008). * (4) K. Clusius, Z. Phys. Chem. B 23, 213 (1933). * (5) R. E. Wilde and T. K. K. Srinivasan, J. Phys. Chem. Solids 36, 119 (1975). * (6) O. Degtyareva, M. Martinez-Canales, A. Bergara, X. J. Chen, Y. Song, V. V. Struzhkin, H. K. Mao, and R. J. Hemley, Phys. Rev. B 76, 064123 (2007). * (7) Y. Ding, J. Xu, C. T. Prewitt, R. J. Hemley, H. K. Mao, J. A. Cowan, J. Z. Zhang, J. Qian, S. C. Vogel, K. Lokshin, and Y. S. Zhao, Appl. Phys. Lett. 86, 052505 (2005). * (8) J. Feng, W. Grochala, T. Jaroń, R. Hoffmann, A. Bergara, and N. W. Ashcroft, Phys. Rev. Lett. 96, 017006 (2006). * (9) C. J. Pickard and R. J. Needs, Phys. Rev. Lett. 97, 045504 (2006). * (10) Y. Yao, J. S. Tse, Y. Ma, and K. Tanaka, Europhys. Lett. 78, 37003 (2007). * (11) V. L. Ginzburg, Rev. Mod. Phys. 76, 981 (2004). * (12) G. Kresse and J. Furthmuller, Comput. Mater. Sci. 6, 15 (1996). * (13) See http://www.pwscf.org, also S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Rev. Mod. Phys. 73, 515 (2001). * (14) W. M. Sears and J. A. Morrison, J. Chem. Phys. 62, 2736 (1975). * (15) P. B. Allen and R. C. Dynes, Phys. Rev. B 12, 905 (1975). * (16) P. P. Singh, Phys. Rev. Lett. 97, 247002 (2006).

arxiv-papers

2024-09-04T02:48:54.396512

0803.2724

# Combined experimental and theoretical investigation of the premartensitic transition in Ni2MnGa C.P. Opeil Physics Department Boston College, 140 Commonwealth Avenue, Chestnut Hill, MA 02467 B. Mihaila Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 R.K. Schulze Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 L. Mañosa Departament d´Estructura i Constituents de la Matèria, Facultat de Física, Universitat de Barcelona, Diagonal 647, 08028 Barcelona, Catalonia, Spain A. Planes Departament d´Estructura i Constituents de la Matèria, Facultat de Física, Universitat de Barcelona, Diagonal 647, 08028 Barcelona, Catalonia, Spain W.L. Hults Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 R.A. Fisher Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 P.S. Riseborough Physics Department, Temple University, Philadelphia, PA 19122 P.B. Littlewood Cavendish Laboratory, Madingley Road, Cambridge CB3 0HE, United Kingdom J.L. Smith Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 J.C. Lashley Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM 87545 ###### Abstract Ultraviolet-photoemission (UPS) measurements and supporting specific-heat, thermal-expansion, resistivity and magnetic-moment measurements are reported for the magnetic shape-memory alloy Ni2MnGa over the temperature range $100~{}K<T<250~{}K$. All measurements detect clear signatures of the premartensitic transition ($T_{\mathrm{PM}}\sim 247~{}K$) and the martensitic transition ($T_{\mathrm{M}}\sim 196~{}K$). Temperature-dependent UPS shows a dramatic depletion of states (pseudogap) at $T_{\mathrm{PM}}$ located 0.3 eV below the Fermi energy. First-principles electronic structure calculations show that the peak observed at 0.3 eV in the UPS spectra for $T>T_{\mathrm{PM}}$ is due to the Ni-d minority-spin electrons. Below $T_{\mathrm{M}}$ this peak disappears, resulting in an enhanced density of states at energies around 0.8 eV. This enhancement reflects Ni-d and Mn-d electronic contributions to the majority-spin density of states and is accompanied by significant reconstruction of the Fermi surface. ###### pacs: 81.30.Kf, 71.20.Be, 79.60.-i ††preprint: LA-UR-08-1278 Ni-Mn-Ga alloys with near-stoichiometric compositions, Ni2MnGa, are important functional materials Magnetism05 owing to their magnetic shape-memory Kakeshita02 , magnetocaloric Marcos03 and magnetoresistive Biswas2005 properties. This ferromagnetic fcc $L2_{1}$ Heusler ($a_{\mathrm{fcc}}=5.81$ Å) was first identified by Webster _et al._ Webster1 as a system undergoing a martensitic transition (MT) in its ferromagnetic phase ($T_{\mathrm{C}}\sim 380~{}K$) with little magnetic hysteresis. In the last decade research on these alloys has focused on the structural and magnetic characterization and on their shape-memory applications Soderberg06 . First-principles calculations Rabe ; Lee02 and measurements on shape-memory alloys Trevor ; Ross ; Dugdale ; Lashley indicate the driving role of the electronic structure and its relation to the lattice dynamics. Figure 1: (Color online) Temperature dependence of the heat capacity (a), thermal expansion (b), temperature-derivative of the resistivity (c), and magnetic moment (d) in the region of the PMT in Ni2MnGa. The lattice dynamics of Ni2MnGa has been investigated from ultrasonic measurements Worgull96 and neutron diffraction experiments Zheludev95 ; manosa01 ; Steve07 . It was found that the transverse TA2 phonon branch exhibits pronounced softening at $1/3$ of the zone boundary on decreasing the temperature, and this softening was described as a Bain distortion in the context of the Wechler, Lieberman, and Read theory of martensite formation Lieberman . In similar structural shape-memory alloys InTl Trevor , AuZn Ross and NiAl Dugdale , this softening is associated with nesting features of the Fermi surface Lee02 . Below a certain temperature, there is a freezing of the displacements associated with this soft phonon so that a micro-modulated phase forms, which is described as a periodic distortion of the parent cubic phase Zheludev95 . In Ni2MnGa, the premartensitic phase develops with little or no thermal hysteresis and is driven by a magnetoelastic coupling Planes97 . On further cooling, Ni2MnGa transforms to an approximately five-layered quasi- tetragonal martensitic structure. The low-temperature phase is _incommensurate_ Zheludev95 with a period (0.43,0.43,0) and exhibits well- defined phasons best characterized as charge-density wave (CDW) excitations Steve07 . In this paper we study the role of conduction electrons in the two-step MT in Ni2MnGa using photoemission spectroscopy and thermodynamic measurements. LEED and X-ray diffraction Laue measurements show the quality of our sample is appropriate for high-resolution photoemission spectroscopy. Ultraviolet photoemission (UPS) measurements show the opening of a pseudogap at 0.3 eV below the Fermi energy at the MT and provide further evidence that the Fermi surface is strongly nested at the MT and is only partially nested at the premartensitic transition (PMT). Single crystals were grown by a Bridgman technique. All samples were spark cut from a large Ni2MnGa single crystal and used for thermal expansion, specific heat, resistivity, magnetic moment, X-ray diffraction and UPS experiments. We note that in the low-temperature phase, the crystal structure of our sample is found to be a five-layered martensite (termed $5R$ or $10M$): the unit cell is monoclinic with the parameters $a$=4.2Å, $b$=5.5Å, $c$=21Å, $\alpha$=$\gamma$=90∘ and $\beta$=91∘ manosa01 . The linear coefficient of thermal expansion was measured in a three-terminal capacitive dilatometer George over the range of 100 K $<$ T $<$ 250 K. The specific heat was measured using a thermal-relaxation method Lashley1 . The magnetic moment was measured using a vibrating-sample magnetometer on a Quantum Design Physical Properties Measurement System (PPMS). Resistivity measurements were made on the PPMS using an ac technique. Figure 1 shows clear signatures of a PMT and MT in the specific heat, thermal expansion, temperature-derivative of the resistivity and the magnetic moment. The specific-heat data show a broad peak centered at $T=225~{}K$ in agreement with previous measurements in a sample of the same composition hc . This feature does not have any measurable thermal hysteresis associated with it. Conversely, the MT at $T_{\mathrm{M}}=196~{}K$ is a sharp, first-order transition associated with 8 K of thermal hysteresis. Low-temperature specific-heat measurements (not shown) give an electronic specific heat coefficient, $\gamma=10.6$ mJ K-2mol-1, and Debye temperature, $\Theta_{D}=205.6$ K. The high-temperature effects are mirrored in the thermal expansion where the sharp, discontinuous MT is preceded by a broad feature with an onset temperature of $T_{\mathrm{PM}}=247~{}K$. The temperature- derivative of the resistivity shows a break in the resistivity slope at $T=214~{}K$ followed by a discontinuity at $T_{\mathrm{M}}=196~{}K$, as does the magnetic moment. The lack of thermal hysteresis at the PMT and the behavior of the thermal expansion, specific heat and electrical resistivity is consistent with CDW formation: a continuous transition associated with CDW onset (PMT), and discontinuous behavior at lock-in (MT)cravenmeyer77 . Figure 2: (Color online) Energy dependence of the normal photoemission intensity for selected temperatures in the region of the PMT in Ni2MnGa. The inset shows _all_ intensity plots at 1 K temperature intervals. Further insight into the nature of the PMT can be obtained using temperature- dependent photoemission spectroscopy photo-exp . Figure 2 illustrates the normal-emission UPS spectra (He I, $h\nu$=21.2 eV) for several temperatures of interest. In addition, the inset in Fig. 2 shows the UPS spectra in the temperature range of the PMT at 1K temperature intervals. In the inset, two sudden redistributions of the UPS intensity are observed corresponding to the onset of the PMT and the MT, respectively. Above $T_{\mathrm{PM}}$ and below $T_{\mathrm{M}}$ the UPS spectra remain unchanged and are not shown. For $T>T_{\mathrm{PM}}$, the UPS spectrum exhibits a prominent peak located at a binding energy of 1.3 eV, together with two smaller features at 0.8 eV and 0.3 eV. As the temperature is lowered below $T_{\mathrm{PM}}$, we note a rapid drop in the intensity of the 0.3 eV peak, followed by a slow, but continuous, decrease in the intensity of this peak with decreasing temperature. A second sudden drop in the intensity of this peak occurs just above $T_{\mathrm{M}}$. Throughout the process of depletion of the allowed electronic states close to the Fermi energy, we notice an enhancement in the number of available states above 0.8 eV. The temperature dependence of the UPS spectra reported here indicates the formation of a shallow pseudogap in the allowed density of states, in the PMT region, for energies between 0.3 eV and the Fermi energy. Similar effects have been reported in the past for high-$T_{\mathrm{C}}$ cuprates cuprates and purple bronze purple . We note that the UPS spectra depicted in the inset of Fig. 2 show a much sharper transition at $T_{\mathrm{M}}$ then $T_{\mathrm{PM}}$, consistent with the data illustrated in Fig. 1. The thermal expansion, resistivity and magnetic moment data support the notion of a first-order transition at the MT, $T_{\mathrm{M}}$. The smooth nature of the PMT makes difficult the identification of the onset temperature for the PMT, $T_{\mathrm{PM}}$. This is perhaps best illustrated by the thermal expansion data. Nevertheless, from the UPS and thermal expansion data, we estimate $T_{\mathrm{M}}-T_{\mathrm{PM}}\approx$ 50 K. Figure 3: (Color online) The majority- and minority-spin DOS for the spin- polarized martensitic phase ($T<T_{\mathrm{M}}$) and the spin-polarized fcc ($T_{\mathrm{PM}}<T<T_{\mathrm{C}}$) of Ni2MnGa. The top panel depicts the total DOS, whereas the middle and bottom panels depict the contributions due to the d electrons of the Ni and Mn atoms, respectively. The contributions due to the Ga electronic degrees of freedom are smaller by an order of magnitude and have been disregarded for the purpose of this comparison. Our UPS data was compared with results of first-principles band-structure calculations using the generalized gradient approximation approach GGA in the full-potential linearized-augmented-plane-wave method Blaha . We use the experimental lattice constants and calculations are performed on grids of 286 and 726 $k$ points in the irreducible Brillouin zone for the austenitic and martensitic phases, respectively. Figure 3 depicts the majority-spin (spin-up) and minority-spin (spin-down) densities of states (DOS) for the spin-polarized martensitic phase ($T<T_{\mathrm{M}}$) and the spin-polarized fcc phase ($T_{\mathrm{PM}}<T<T_{\mathrm{C}}$) of bulk Ni2MnGa. The main contributions to the total density of states in either phase or spin state are due to the Ni-d and Mn-d electrons. We note that our electronic structure results also show a redistribution of the density of states away from the Fermi surface in the martensitic phase as compared to the fcc phase. Our calculations indicate that the missing spectral weight, observed experimentally near the Fermi energy in the martensitic phase, is due to the disappearance of the Ni-d peak located at 0.3 eV in the spin-down DOS corresponding to the fcc phase. In the martensitic phase, the spectral weight shifts to slightly lower binding energies and results in the enhancement of the peak observed at 0.8 eV in the spin-up DOS. The latter peak has a combined Ni-d and Mn-d character, as reflected by the peaks highlighted in the plots of the Ni-d and Mn-d spin-up partial DOS. Although we predict that near the Fermi surface there is a spectral-weight transfer from minority- to spin-up electrons (which could potentially be resolved by spin-polarised photoemission), the net magnetization shown in Fig. 1d in fact drops at the transition. The calculated net magnetization change (integrated over all energies) is found to be close to zero, since the low energy redistribution of spin density is approximately canceled by higher energy states. The disagreement with experiment indicates that correlation effects may be important for the deeper Mn d-levels - note that the magnetization is mostly dominated by Mn whereas the Fermi surface is preponderantly from Ni states. Figure 4: Fermi surfaces in the Brillouin zones corresponding to a) the non- magnetic fcc phase ($T>T_{\mathrm{C}}$), b) the spin-polarized fcc phase ($T_{\mathrm{PM}}<T<T_{\mathrm{C}}$), and c) the martensitic ($T<T_{\mathrm{PM}}$) of Ni2MnGa. Both the merged and the individual band contributions to the Fermi surface are depicted. Plots performed with XCrySDen xcrysden using the structural data from Ref. struct_data . The Fermi surfaces (FS) in the Brillouin zones corresponding to the various phases of Ni2MnGa are displayed in Fig. 4. Two electronic bands contribute to the FS in the non-magnetic phase ($T>T_{\mathrm{C}}$), whereas three and two electronic bands contribute to the spin-up and spin-down FS below $T_{\mathrm{C}}$. The character of the spin-down Fermi surfaces for $T_{\mathrm{PM}}<T<T_{\mathrm{C}}$ is dominated by the Ni-d electrons, whereas the character of the spin-up Fermi surfaces is shared by both Ni-d and Mn-d electrons. Below $T_{\mathrm{M}}$ the spin-up contribution to the FS look similar to the spin-up contribution above $T_{\mathrm{M}}$, but the FS is nested along the $k_{z}$ direction in the martensitic phase. The change in the FS of the spin-down electrons above $T_{\mathrm{PM}}$ and below $T_{\mathrm{M}}$ is more dramatic and reflects the changes noted in the calculated DOS and the measured UPS spectra. We note that the PMT is not accompanied by a change in the magnetic moment, whereas $T_{M}$ marks a substantial rearrangement of the spin density between bands, as seen in both experiment and theory. Thus we speculate that the transition at $T_{PM}$ corresponds to a nesting feature of a single band. As the amplitude of this instability grows on lowering the temperature, mixing with other bands becomes inevitable - which apparently triggers a second and stronger instability in the martensitic phase involving a redistribution between the spin directions and the development of quasi-1D-like reconstructed Fermi surface for the two spin-down bands. To summarize, in this paper we report the presence of a pseudogap in photoemission spectra, 0.3 eV below the Fermi energy, at the PMT temperature, $T_{\mathrm{PM}}$. Based on first-principles electronic structure calculations, we conclude that the changes in the experimental photoemission spectra are due to a redistribution of the electronic density of states associated with the Mn-d and Ni-d electronic degrees of freedom. We show that the peak observed at 0.3 eV in the UPS spectra above $T_{\mathrm{PM}}$ is due to the Ni-d spin-down electrons. Below $T_{\mathrm{M}}$ this peak disappears, resulting in an enhanced DOS at energies around 0.8 eV. This enhancement reflects Ni-d and Mn-d electronic contributions to the spin-up DOS. The calculated innermost Fermi surface of spin-up electronic states exhibits weak momentum dispersion, which, together with the changes in the material properties observed experimentally in the premartensitic temperature region (see Fig. 1), suggests a CDW pseudogap-opening mechanism from Fermi surface nesting in Ni2MnGa. ###### Acknowledgements. This work was supported in part by the Los Alamos National Laboratory under the auspices of the U.S. Department of Energy, and the Trustees of Boston College. LM and AP acknowledge partial support from projects MAT2007-61200 (CICyT, Spain) and 2005SGR00969 (DURSI, Catalonia). PSR acknowledges support from the U.S. Department of Energy through award DEFG02-01ER45872. The authors would like to acknowledge useful conversations with Y. Lee and B. Harmon. ## References * (1) Magnetism and Structure in functional materials ed. A. Planes, L. Mañosa, and A. Saxena, Materials Science Series, Vol. 79 (Springer Verlag, Berlin 2005). * (2) T. Kakeshita and K. Ullakko, MRS Bull. 27, 105 (2002). * (3) J. Marcos _et al._ , Phys. Rev. B 68, 094401 (2003). * (4) C. Biswas, R. Rawat and S.R. Barman, Appl. Phys. Lett. 86, 202508 (2005). * (5) P.J. Webster _et al._ , Philos. Mag. B 49, 295 (1984). * (6) See O. Soderberg _et al._ , Giant magnetostrictive materials in Handbook of Magnetic Materials, Vol. 79, ed. K.J.H. Buschow (Elsevier, Amsterdam, 2006) and references therein. * (7) C. Bungaro, K. M. Rabe, and A. Dai Corso, Phys. Rev. B 68, 134104 (2003); A.T. Zayak _et al._ , Phys. Rev. B 68, 132402 (2003); G.I. Velikohkhatnyi and I.I. Naumov, Phys. Sol. State 41, 617 (1999). * (8) Y. Lee, J.Y. Rhee and B.N. Harmon, Phys. Rev. B 66, 054424 (2002); P. Entel _et al._ , J. Phys. D: Appl. Phys. 39, 865 (2006). * (9) M. Liu, T. R. Finlayson, and T. F. Smith, Phys. Rev. B 48, 3009 (1993). * (10) P.A. Goddard _et al._ , Phys. Rev. Lett. 94, 116401 (2005). * (11) S.B. Dugdale _et al._ , Phys. Rev. Lett. 96, 046406 (2006). * (12) J.C. Lashley _et al._ , Phys. Rev. B 75, 205119 (2007); G. Jakob, T. Eichhorn, M. Kallmayer, and H. J. Elmers, Phys. Rev. B 76, 174407 (2007). * (13) J. Worgull, E. Petti, and J. Trivisonno, Phys. Rev. B 54, 15695 (1996); M. Stipcich _et al._ , Phys. Rev. B 70, 054115 (2004). * (14) A. Zheludev _et al._ , Phys. Rev. B 51, 11310 (1995); ibid. Phys. Rev. B 54, 15045 (1996) * (15) L. Mañosa _et al._ , Phys. Rev. B 64, 024305 (2001). * (16) S.M. Shapiro _et al._ , Euro. Phys. Lett. 77, 56004 (2007). * (17) M.S. Wechlsher, D.S. Lieberman, and T.A. Read, J. Metals 5, 1503 (1953). * (18) A. Planes, E. Obradó, A. Gonzàlez-Comas and L. Mañosa, Phys. Rev. Lett. 79, 3926 (1997). * (19) G.M. Schmeideshoff _et al._ , Rev. Sci. Instrum. 77, 123907 (2006). * (20) J.C. Lashley _et al._ , Cryogenics 43, 369 (2003). * (21) F.J. Pérez-Reche, E. Vives, L. Mañosa, and A. Planes, Mater. Sci. Eng. A 378, 353 (2004). * (22) R.A. Craven and S.F. Meyer, Phys. Rev. B 16, 4583 (1977). * (23) The crystal surface preparation and details of the experimental setup have been discussed in C.P. Opeil _et al._ , Phys. Rev. B 73, 165109 (2006); _ibid._ 75, 045120 (2006). The sample surface quality and alignment were verified using LEED measurements. * (24) M.R. Norman, D. Pines, and C. Kallin, Adv. Phys. 54, 715 (2005). * (25) M.A. Valbuena _et al._ , J. Phys. Chem. Solids 67, 213 (2006). * (26) J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). * (27) P. Blaha _et al._ , WIEN2k, An Augmented Plane Wave Plus Local Orbitals Program for Calculating Crystal Properties (Karlheinz Schwarz, Technische Universit t Wien, Austria, 2001). * (28) A. Kokalj, Comp. Mater. Sci. 28, 155 (2003). Code available from http://www.xcrysden.org/. * (29) D.Y. Cong, P. Zetterström, and Y.D. Wang, Appl. Phys. Lett. 87, 111906 (2005).

arxiv-papers

2024-09-04T02:48:54.401008

0803.2740

# Unconventional sign-reversing superconductivity in LaFeAsO1-xFx I.I. Mazin Code 6393, Naval Research Laboratory, Washington, D.C. 20375 D.J. Singh Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831-6114 M.D. Johannes Code 6393, Naval Research Laboratory, Washington, D.C. 20375 M.H. Du Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831-6114 (Printed ) ###### Abstract We argue that the newly discovered superconductivity in a nearly magnetic, Fe- based layered compound is unconventional and mediated by antiferromagnetic spin fluctuations, though different from the usual superexchange and specific to this compound. This resulting state is an example of extended $s$-wave pairing with a sign reversal of the order parameter between different Fermi surface sheets. The main role of doping in this scenario is to lower the density of states and suppress the pair-breaking ferromagnetic fluctuations. The discovery of superconductivity with $T_{c}\gtrsim$ 26 K JACS in a compound that contains doped Fe2+ square lattice sheets raises immediate questions about the nature of the superconducting state and the pairing mechanism. A number of highly unusual properties suggest, even at this early stage, that the superconductivity is not conventional. We argue that not only is it unconventional, but that doped LaFeAsO represents the first example of multigap superconductivity with a discontinuous change of the order parameter (OP) phase between bands, a state discussed previously ($e.g.,$ Refs.MazinGolubov ; GorkovAgterberg ), but not yet observed in nature. We suggest that superconductivity here is mediated by spin fluctuations (SF), as many believe is the case in cuprates, heavy fermion materials, or ruthenates. SF can only induce a triplet superconducting OP, or a singlet one that changes sign over the Fermi surface (FS). The latter condition often, but not always, dictates strong angular anisotropy of the OP (cf. d-wave). In our scenario it is satisfied despite full angular isotropy, since the sign changes between the two sets of FSs. Our model is also principally different from the convetional s-wave superconductivity discovered in MgB2: there the pairing interaction is attractive, in our case it is repulsive (but pairing thanks to the sign reversal). Finally, similar to d- or p-wave superconductivity our OP has a nearest-neighbor structure in the real space, thus reducing the Coulomb repulsion within the pair. We begin by arguing against conventional superconductivity. The pure compound, LaFeAsO, is on the verge of a magnetic instability: it has a very high magnetic susceptibilityJACS and is strongly renormalized compared to density functional (DFT) calculations singh . This renormalization is higher than in any known conventional superconductor, including MgCNi3, where superconductivity is believed to be strongly depressed by spin fluctuations. The susceptibility in the pure compound is large and upon doping with F grows even larger and becomes Curie-Weiss like. This suggests nearness to a critical point in the pure compound and non-trivial competition between different spin fluctuations (SF). Very strong electron phonon interactions would be required to overcome the destructive effects of such SF. We have calculated ab initio the electron-phonon spectral function, $\alpha^{2}F(\omega)$, and coupling, $\lambda$, for the stoichiometric compound espresso . Some moderate coupling exists, mostly to As modes, but the total $\lambda$ appears to be $\sim 0.2$, with $\omega_{log}\sim 250$K ($\omega_{log}$ is the logarithmically averaged boson frequency of Eliashberg theory), which can in no way explain $T_{c}\gtrsim 26$ K. The calculated DFT Fermi surfaces singh for undoped LaFeAsO consist of two small electron cylinders around the tetragonal $M$ point, and two hole cylinders, plus a heavy 3D hole pocket around $\Gamma$. To study doping effects, we performed full-potential calculations using the WIEN package in the virtual crystal approximation and the PBE GGA functional. The lattice parameters we took from experiment JACS and optimized the internal positions positions . In Fig. 1 we show the bands near the Fermi level for $x=0.1$ and the corresponding Fermi surface. As expected, the 3D pocket fills with electron doping (at $x=0.04$-0.05) and the fermiology radically simplifies, leaving a highly 2D electronic structure with two heavy hole cylinders and two lighter (and larger) electron cylinders. Figure 1: color online(a) Calculated band structure at $x=0.1$ near the Fermi Level. (b) Calculated Fermi surface at 10% F doping. Note that the only 3D parts are the far ends of the electron cylinders around M. The fully three- dimensional surface present in the undoped compound is suppressed beneath EF by increased electron count. Figure 2: color online Fermi surface formation upon backfolding of the large BZ corresponding to a simple Fe square lattice. (a) Real space: the four small unit cells of the Fe sublattice only (dashed) with the larger solid diamond of actual two-Fe unit cell. Dark and light circles indicate superexchange (checkerboard) ordering. The inset shows the spin density wave corresponding to X̃ point SF. (b) Reciprocal space: the black square is the unfolded BZ, the blue diamond is the downfolded zone, the blue ellipse is the electron pocket from the Ỹ point downfolded onto the X̃ point (which is M in the small BZ). This fermiology imposes strong constraints on the superconductivity. In particular, with the exception of phonons, it is hard to identify pairing interactions with a strong $k_{z}$ dependence. Thus, states with strong variations of the OP along $k_{z}$ are unlikely. An angular variation of the OP in the $xy$ plane is possible, but would require an unrealistically strong $\mathbf{q}$-dependence of the pairing interaction on the scale of the small Fermi surface size, and would also be extremely sensitive to impurities. On the other hand, the small Fermi surfaces are readily compatible with a pairing state with weak variations of the OP within the sheets, but a $\pi$ phase shift between electron and hole cylinders. Here we show that a SF pairing interaction favoring exactly such a state is present in this material and we discuss the expected consequent physical properties. The SF spectrum is unusually rich for this compound and comes from three separate sources. First, the system is relatively close to a Stoner ferromagnetic instability. Second, there is a nearest-neighbor antiferromagnetic (AFM) superexchangeadded (d). Third, there are nesting- related AFM spin-density-wave type SF near wave vectors connecting the electron and hole pockets. The latter appear to be the strongest ones. The corresponding interaction connects the well-separated FS pockets located around $\Gamma$, and around $M$. Though repulsive in the singlet channel, these would nevertheless be strongly pairing provided that the OPs on the two sets of the FSs have opposite signs. The main message of our paper is that this “$s_{\pm}$” superconducting state is both consistent with experimental observations and most favored by SF in this system. As opposed to the undoped materialsingh , we do not find any FM solution for the doped compound, even with the more magnetic GGA functional. This suggests that the main function of doping is to move the system away from a ferromagnetic instability (see, however, Ref. added ). That the system becomes less magnetic is due to the fact that upon doping the heavy 3D hole pocket near $\Gamma$ rapidly fills and the total DOS drops (by a factor of two). At a doping level $x=0.1$, we obtain in the GGA an unrenormalized Pauli susceptibility $\operatorname{Re}\chi_{0}(\mathbf{q}=0,\omega=0)\approx 4\times 10^{-5}$ emu/mole ($N(0)=0.64$ states/eV/spin/Fe). Using 1.1 eV for the Stoner $I$ on Fe, we obtain a renormalized $\chi(0)$ of 0.14$\times 10^{-3}$ emu/mole, a large renormalization, but much smaller than what is needed to explain the experimental valueJACS . Note that in the undoped system the calculated susceptibility is larger, and the experimental one smaller, than in the doped one. This suggests that besides FM SF there are other, more important spin excitations in the system. The first candidate for these is a superexchange corresponding to the standard simple nearest-neighbor checkerboard antiferromagnetism. Importantly, there is also substantial direct Fe-Fe overlapsingh , which leads to an additional AFM exchange of comparable strength and with the same checkerboard geometry. Further discussion requires an understanding of the fermiology in clearer terms. The band structure may at first appear intractably complex, but it is in fact relatively simple, involving only three Fe orbitals near the Fermi level. The hole pockets around $\Gamma$ originate from Fe $d_{xz}$ and $d_{yz}$ bands that are degenerate at $\Gamma,$ and form two nearly perfect concentric cylinders. The electron surfaces are better understood if we recall that the underlying Fe layer forms a square lattice with the period $\tilde{a}=a/\sqrt{2}$ (Fig. 2). The corresponding 2D Brillouin zone (BZ) is twice larger and rotated by 45${}^{\circ}.$ If we could “unfold” the Fermi surface of Fig. 1, we would find the same two hole pockets at $\tilde{\Gamma}$ and $one$ electron pocket around the point X̃ in the large BZ. The latter is formed by the $d_{yz}$ band (or $d_{xz}$ near Ỹ) that starts 180 meV below the Fermi energy, disperses up along X̃ M̃ and is practically flat along X̃ $\tilde{\Gamma}$. This band is hybridized with the $d_{xy}$ band (or $d_{x^{2}-y^{2}}$ in the two-Fe cell), which starts from X̃ at an energy of -520 meV, and is instead dispersive along X̃ $\tilde{\Gamma}$ . Upon hybridization, these two bands yield an oval cylindrical electron pocket, elongated along $\tilde{\Gamma}$X̃ notelee . Electronic transitions from the hole pockets to the electron pocket should lead to a broad (because the electron pocket is oval rather than circular) peak in the noninteracting susceptibility $\chi_{0}(q,\omega\rightarrow 0)$, at $q=(\pi/\tilde{a},0)$, while the superexchange interaction $J(q)$ on the square lattice should be peaked at $q=(\pi/\tilde{a},\pi/\tilde{a})$ ($\Gamma$ in the downfolded BZ). The renormalized susceptibility, $\chi(q)=\chi_{0}(q)/[1-J(q)\chi_{0}(q)]$ then has a rich structure with maxima at $\tilde{\Gamma},$ X̃ and M̃. For the true unit cell with two Fe, both $\tilde{\Gamma}$ and M̃ fold down into the $\Gamma$ point, while X̃ folds down into the M point. The corresponding folding of the Fermi surfaces makes the electron pockets overlap, forming two intersecting surfaces (Fig. 1). It is important to appreciate from this gedanken unfolding that already on the level of the noninteracting susceptibility there is a tendency for antiferromagnetic correlations with a wave vector different from the superexchange one. In Fig. 3 we present the calculated chi $\chi_{0}(q,\omega)$ = $\frac{f(\epsilon_{k})-f(\epsilon_{k+q})}{\epsilon_{k}-\epsilon_{k+q}-\omega-i\delta}$ at $\omega\rightarrow 0$ and $q_{z}=\pi/c$ ($\chi$ is practically independent of $q_{z}$), The peak at M, derived from interband transitions, is very broad, as expected, with some structure around the M point, due to the particular orientation of the two oval pockets at M and the size difference between the hole and electron cylinders for finite doping. With minor modification, this structure is present also in the undoped compound. Figure 3: color online The imaginary (a) and the real (b) parts of the non- interacting susceptibility $\chi_{0}(q,\omega\rightarrow 0)$, in arbitrary units. Within common approximations Im$\chi$=Im$[\frac{\chi_{0}}{(1-J(\mathbf{q})\chi_{0}(q,w)}]$ is measurable by neutron scattering and Re$\chi$ controls the pairing interaction, in the singlet channel proportional to $1/[1-J(\mathbf{q})\chi_{0}(q,0)]$ (see Ref. MU for a review.) Note that the RPA enhancement will strengthen both peaks. We have also performed magnetic calculations (discussed in a separate publication) in a supercell corresponding to the superexchange, $\mathbf{q}=$ M̃ (=$\Gamma$ in the downfolded Brillouin zone) and nesting-induced $\mathbf{q}=$ X̃ (=M) spin density waves. We find, indeed, that the tendency to ferromagnetic ordering is suppressed in the doped compound, while the tendency to nesting-based antiferromagnetism is present, even leading to an actual instability at the mean field level. It is worth noting that while this instability does not appear in actual superconducting materials, it has been now found experimentally in the low doping regime added . The strong AFM SF around M favor our proposed $s_{\pm}$ state. Cases where SF- induced interactions connect two pockets of the Fermi surfaces, including SF originating from electronic transitions between the very same pockets have been considered in the past. odd However, they involved FS pockets related by symmetry, which strongly restricts the phase relations between them. In our case the two sets of pockets are not symmetry related, and nothing prevents them from assuming arbitrary phases. For the singlet case, the coupling matrix between the hole ($h)$ and the electron pockets ($e)$ is negative, $\lambda_{eh}<0.$ The diagonal components emerge from competition between the attractive phonon-mediated and repulsive SF-mediated interactions and are, presumably, weak. If $\lambda_{hh}^{ph}$ is the average of the phonon-mediated interaction over the wave vectors $q<0<2k_{F}^{h}$, and $\lambda_{hh}^{sf}$ the same for the SF, then $\lambda_{hh}=\lambda_{hh}^{ph}-\lambda_{hh}^{sf}.$ Our calculated electron- phonon coupling comes mostly from small wave vectors, that is, $\lambda_{hh}^{ph}+\lambda_{ee}^{ph}\approx 0.2$, and $\lambda_{eh}^{ph}\approx 0$. Phonons, therefore, though weak, promote the $s_{\pm}$ state. On the other hand, $\lambda_{hh}^{sf}$ and $\lambda_{ee}^{sf}$, come from FM (small-q) SF and are pair breaking. Finally, the superexchange interaction does not affect $\lambda_{hh},$ but the corresponding wave vector $\mathbf{q}=\tilde{\Gamma}$M̃ connects the $e$ pockets near X̃ and Ỹ and are also pair breakingadded (c). The $T_{c}$, as usual for a two-gap superconductorMazinAntropovReview , is defined by the maximal eigenvalue of the $\lambda$ matrix, $\lambda_{eff}=[\lambda_{hh}+\lambda_{ee}+\sqrt{4\lambda_{eh}\lambda_{he}+(\lambda_{ee}-\lambda_{hh})^{2}}]/2,$ and the OPs $\Delta_{e,h}$ are defined by the corresponding eigenvector. In our case, the signs of $\Delta_{h}$ and $\Delta_{e}$ will be opposite, and their absolute values (despite the different densities of states) will be similar since presumably $(\lambda_{ee}-\lambda_{hh})^{2}\ll\lambda_{eh}\lambda_{he}.$ This “$s_{\pm}$” state is analogous to the states proposed previously for semimetalsAronovSonin and bilayer cupratesLichtMazin . From the point of view of neutron scattering, the structure of the peak in $\chi$ near M is important, but for SF induced superconductivity it does not matter at all. A well defined spin-excitation requires a sharp peak, but the pairing interaction is integrated over all possible $q$ vectors spanning the two sets of Fermi surfaces so that only the total weight of the peak is important. Dimensionality is however very important, as 3D coupling supports a long range magnetic ordering (competing with superconductivity). Indeed, the experiment shows an abrupt appearence of superconductivity at $x\approx 0.03,$ roughly where the 3D pockets around $\Gamma$ disappear. One might envision a triplet state similar to the described singlet one, fully gapped, as expected for unitary 2D $p$-wave states, and with different amplitudes (or signs) on the two cylinders. The similarity, however, is misleading. In the triplet channel, SF induce attraction, but given the relatively large width of the AF peak (Fig.3), a large part of the pairing will be lost as only SF with a wave vector exactly equal to ($\pi/a,\pi/a)$ will be fully pairing, and some others will even be pair-breaking. Therefore, in this scenario where the antiferromagnetic SF around the $M$ point provide the primary pairing interaction, we expect the lowest energy superconducting state to be $s_{\pm}$. The structure of the OP in real space can be evaluated using the lowest Fourier component, compatible with the proposed $s_{\pm}$ state, namely $\cos k_{x}+\cos k_{y}$ (equivalently, $\cos\widetilde{k_{x}}\cos\widetilde{k_{y}}).$ This corresponds to pairing of electrons that reside on the nearest neighbors, just as in the d-wave case, so that the onsite Coulomb repulsion is not particularly destructive for this state. Finally, we discuss the experimental ramifications of the $s_{\pm}$ state. In many aspects they are similar to that of the $s_{\pm}$ state proposed in Ref. LichtMazin, and discussed in some detail in Refs.LichtMazin ; MazGZ ; MazY . The thermodynamic and tunneling characteristics are the same as for a conventional two-gap superconductorMazinAntropovReview . The two-gap character, however, may be difficult to resolve, given the dominance of the interband interactions which will render the two gap magnitudes similar. Nonmagnetic small-$q$ intraband ($e-e$ or $h-h)$ scattering, as well as the interband spin-flip pairing will not be pair breaking, but paramagnetic interband scattering will, resulting in finite DOS below the gapMazinGolubov , consistent with specific heat measurements heat . The most interesting feature of the $s_{\pm}$ state MazY is that the coherence factors for exciting Bogolyubov quasiparticles on FS sheets with opposite signs of the OP are reversed compared to conventional coherence factors. We outline a few important consequences of the relatively straigtforward application of this concept to experimental probes. First, one expects a qualitative difference between experiments that probe vertical transitions (q=0) and those that probe transitions with q close to $\pi/a,\pi/a.$ For instance, the spin susceptibility at q=0 will behave conventionally, i.e. exponentially decay below $T_{c}$ without any coherence peak, while the susceptibility for q$\approx\pi/a,\pi/a$ will have a coherence peak that should be detectable by neutron scattering as an enhancement below $T_{c}$ MazY . AFM SF near $\pi/a,\pi/a$ dominate in the doped material, and so the usual coherence peak in the NMR relaxation rate, which averages equally over all wave vectors is expected to disappear or be strongly reduced. It was shown in Ref. MazGZ that Josephson currents from FSs with different signs of the OP interfere destructively, and the net phase corresponds to the sign of the FS with the higher normal conductance. In the constant relaxation time approximation, both in-plane and out of plane conductivities are dominated by the electron pockets. This is unfortunate, since there would otherwise be a $\pi$ phase shift between the $ab$ and $c$ tunneling and corner-junction experiments could be used, as in high-Tc cuprates. To summarize, we argue that the fermiology found in doped LaAsFeO gives rise to strong, but broad antiferromagnetic spin fluctuations near the M point in the Brillouin zone, while the tendency to magnetism existing at zero doping is suppressed. These fluctuations, while too broad to induce a magnetic instability, are instrumental in creating a superconducting state with OPs of opposite signs on the electron and hole pockets. We would like to acknowledge helpful discussions with D. Scalapino. Work at NRL is supported by ONR. Work at ORNL was supported by DOE, Division of Materials Science and Engineering. ## References * (1) Y. Kamihara, $et$ $al$, J. Am. Chem. Soc. 108, (in press) doi:10.1021/ja800073m * (2) A. A. Golubov and I. I. Mazin, Phys. Rev, B55, 15146 (1997) * (3) D. F. Agterberg, V. Barzykin, and L. P. Gor’kov, Phys. Rev. B60, 14868 (1999) * (4) G.F. Chen, Z. Li, G. Li, J. Zhou, D. Wu, J. Dong, W. Z. Hu, P. Zheng, Z.J. Chen, J.L. Luo, and N.L. Wang, cond-mat:0803.0128. * (5) G. Mu, X. Zhu, L. Fang, L. Shan, C. Ren, and H.H. Wen, cond-mat:0803.0928 * (6) A.S. Sefat, M.A. McGuire, B.C. Sales, R. Jin, J.Y. Howe, and D.G. Mandrus, cont-mat:0803.2439 * (7) H. Yang, X. Zhu, L. Fang, G. Mu, and H.H. Wen, cond-mat:0803:0623 * (8) D.J. Singh, M.H. Du, cond-mat:0803.0429 * (9) We used linear response theory with ultrasoft pseudopotentials as implemented within the Quantum Espresso package by S. Baroni et al, http://www.pwscf.org/. We used a zone sampling of 1377 inequivalent $\mathbf{k}$-points, a 4x4x2 phonon mesh and a planewave cut-off of 50 Ry. * (10) The optimized positions actually depend slightly on doping. For $x=0.1$, in GGA, they are: $z_{La}=0.148$, $z_{As}=0.638$. * (11) We thank P.A. Lee for pointing out an erroneous transmutation of X̃ and Ỹ in our original description. * (12) We used a three-dimensional grid of $\approx$ 75,000 $k$ and $q$ points with a temperature smearing of 1 mRy, and the constant matrix element approximation * (13) T. Moria and K. Ueda, Rep. Prog. Phys. 66, 1299 (2003) * (14) M.D. Johannes, I.I. Mazin, D.J. Singh, and D.A. Papaconstantopoulos, Phys. Rev. Lett. 93, 097005 (2004); K. Voelker and M. Sigrist, cond-mat/0208367 * (15) I.I. Mazin and V.P. Antropov, Physica C 385, 49 (2003). * (16) A.G. Aronov and E.B. Sonin, Zh. Eksp. Teor. Fiz. 63, 1059 (1972) [Sov. Phys. -JETP 36, 556 (1973)] * (17) A.I.Liechtenstein, I.I. Mazin, and O.K. Andersen, Phys. Rev. Lett. 74, 2303 (1995). * (18) I.I.Mazin, A.A. Golubov, and A.D. Zaikin, Phys. Rev. Lett. 75, 2574 (1995). * (19) I.I. Mazin and V. Yakovenko, Phys. Rev. Lett.,75, 4134 (1995). * (20) After this manuscript was submitted for publication, several dozen relevant preprints have been posted, in particular (a) Experimental verification of the SDW state predicted here (inset of Fig.2), at $x\lesssim 0.05$ (arXiv:0804.0796), (b) While we calculated the SDW stripe ordered phase at $x=0.1,$ Dong et al (arXiv:0803.2883) have subsequently shown that, contrary to our initial conjecture, at zero doping this instability is even stronger, (c) Kuroki et al (arXiv:0803.3325) (cond-mat) have studied the SF induced superconductivity in LaFeAsO adding the RPA renormalization to our model and found that the $s_{\pm}$ state has the lowest energy as long as the hole pockets are present (otherwise, a nodeless d-wave state is the most stable one) and (d) Yildirim (arXiv:0804.2252)has pointed out that the next n.n. superexchange in this system is of the same order as the n.n. superxchange, adding a fourth nontrivial nagnetic interaction to those we considered. These and other recent results support our main proposition that the role of doping is to drive the system away from a magnetic instability, though we misidentified the offending instability as ferromagnetic.

arxiv-papers

2024-09-04T02:48:54.405144

0803.2741

# Spin-Orbit Coupling and Ion Displacements in Multiferroic TbMnO3 H. J. Xiang National Renewable Energy Laboratory, Golden, Colorado 80401, USA Su-Huai Wei National Renewable Energy Laboratory, Golden, Colorado 80401, USA M.-H. Whangbo Department of Chemistry, North Carolina State University, Raleigh, North Carolina 27695-8204, USA Juarez L. F. Da Silva National Renewable Energy Laboratory, Golden, Colorado 80401, USA ###### Abstract The magnetic and ferroelectric (FE) properties of TbMnO3 were investigated on the basis of relativistic density functional theory (DFT) calculations. We show that, due to spin-orbit coupling, the spin-spiral plane of TbMnO3 can be either the $bc$\- or $ab$-plane, but not the $ac$-plane. As for the mechanism of FE polarization, our work reveals that the “pure electronic” model by Katsura, Nagaosa and Balatsky is inadequate in predicting the absolute direction of FE polarization. Our work indicates that to determine the magnitude and the absolute direction of FE polarization in spin-spiral states, it is crucial to consider the displacements of the ions from their centrosymmetric positions. ###### pacs: 75.80.+q,77.80.-e,75.30.Gw,71.20.-b Recent studies on magnetic ferroelectric (FE) materials have shown that electric polarization can be significantly modified by the application of a magnetic field Cheong2007 ; Kimura2003 ; Hur2004 ; Lawes2005 ; Katsura2005 ; Sergienko2006 ; Mostovoy2006 . Perovskite TbMnO3 with a spin-spiral magnetic order is a prototypical multiferroic compound with a gigantic magnetoelectric effect Kimura2003 . Currently, there are two important issues concerning the FE polarization of TbMnO3. One concerns the origin of FE polarization. Model Hamiltonians studies of spin-spiral multiferroic compounds have provided two different pictures. In the Katsura-Nagaosa-Balatsky (KNB) model Katsura2005 , the hybridization of electronic states induced by spin-orbit coupling (SOC) leads to a FE polarization of the charge density distribution even if the ions are not displaced from their centrosymmetric positions. In contrast, the model study by Sergienko and Dagotto Sergienko2006 concluded that oxygen ion displacements from their centrosymmetric positions are essential for the FE polarization in multiferroic compounds Picozzi2007 . When carried out with the ions kept at their centrosymmetric positions, density functional theory (DFT) calculations Xiang2007 for the spin-spiral states of LiCuVO4 predict FE polarizations that agree reasonably well in magnitude with experiment Schrettle , which is in apparent support of the KNB model Katsura2005 . It is, therefore, important to check which model, the KNB or the “ion displacement” model, is relevant for the FE polarization in TbMnO3. The other issue concerns the spin-spiral plane of TbMnO3. Under a magnetic field, the spin-spiral plane of TbMnO3 can be either the $bc$-plane or the $ab$-plane, but not the $ac$-plane. To explain this observation, it is necessary to probe the magnetic anisotropy of the Mn3+ ion. The magnetic anisotropy of the Tb3+ ion might be also relevant for the magnetoelectric effect, as suggested by Prokhnenko et al. Prokhnenko2007 . In this Letter, we investigated these issues on the basis of DFT calculations and found that the consideration of the ion displacements is essential for the FE polarizations in the spin-spiral state of TbMnO3, and the KNB model can be erroneous even for predicting the absolute direction of FE polarization. The absence of the $ac$-plane spin-spiral in TbMnO3 is explained by the magnetic anisotropy of the Mn3+ ion. Our calculations were based on DFT plus the on-site repulsion U method Liechtenstein1995 within the generalized gradient approximation Perdew1996 . We used $U_{eff}=6.0$ eV on Tb 4f states. With other $U_{eff}$ values for Tb, similar results were obtained. For Mn 3d states, we employed $U_{eff}=2.0$ eV, which leads to the spin exchange interactions between the Mn3+ ions that are consistent with the observed magnetic structure of TbMnO3 (see below). For the calculation of FE polarization, the Berry phase method King-Smith1993 encoded in the Vienna ab initio simulation package (VASP) was employed VASP ; PAW , in which the Tb 4f electrons were treated as core electrons. For the study of the SOC effect associated with the Tb 4f electrons, we used the full-potential augmented plane wave plus local orbital method as implemented in the WIEN2k code wien2k . Due to the small value of the spin anisotropy, we employed the convergence threshold of $10^{-7}$ for electron density. As shown in Fig. 1, the experimental crystal structure Alonso2000 of TbMnO3 has a distorted GdFeO3-type orthorhombic perovskite structure with space group Pbnm. In our calculations, the experimental structure was used unless otherwise stated. Figure 1: (Color online) Perspective view of the orthorhombic structure of TbMnO3. The large, medium, and small spheres represent the Tb, Mn, and O ions, respectively. The Mn-Mn spin exchange paths $J_{ab}$, $J_{aa}$, $J_{bb}$, and $J_{cc}$ are also indicated. The solid vectors denote the easy axes for the Tb3+ and Mn3+ spins. Figure 2: (Color online) Electronic structure obtained for the FM state of TbMnO3 from the WIEN2k calculations. The total DOS, the Tb 4f partial DOS, and the Mn 3d partial DOS are shown. The position of the $f_{z(x^{2}-y^{2})}$ state is also indicated. The basic electronic structure of TbMnO3 is shown in Fig. 2 in terms of the density of states (DOS) of the FM state calculated by WIEN2k with the Tb 4f states treated as valence states. The system is an insulator with energy gap of about 0.5 eV between the spin-up Mn 3d $e_{g}$ states. The spin-up Tb 4f states are all occupied while the spin-down Tb 4f states occur at around $-2.8$ eV with the remaining down-spin Tb 4f states above the Fermi level. This feature is consistent with the f8 configuration of the Tb3+ ion. Partial DOS analysis indicates that the occupied down-spin states have mainly the $f_{z(x^{2}-y^{2})}$ orbital character with a slight contribution from the $f_{xyz}$ orbital. For the spin exchange between Mn3+ ions, there are four spin exchange paths to consider as shown in Fig. 1: $J_{ab}$ is the intralayer ($ab$ plane) nearest neighbor (NN) exchange interaction, $J_{aa}$ and $J_{bb}$ are the intralayer next-nearest neighbor (NNN) exchange interactions along the $a$ and $b$ directions, respectively, and $J_{cc}$ is the interlayer NN exchange interaction along the $c$ direction. The high-spin Mn3+ ions have the $t_{2g}^{3}e_{g}^{1}$ configuration with the $e_{g}$ states forming a staggered orbital ordering of the $d_{3x^{2}-r^{2}}$ and $d_{3y^{2}-r^{2}}$ states. Such an orbital ordering induces an NN intralayer ferromagnetic (FM) exchange, and an NN interlayer antiferromagnetic (AFM) exchange. The cooperative Jahn-Teller distortion associated with the orbital ordering leads to a large NNN intralayer super-superexchange along the $b$ direction Kimura2003B ; Picozzi2006 . Our VASP calculations lead to $J_{ab}=-1.52$ meV, $J_{aa}=0.57$ meV, $J_{bb}=0.85$ meV, and $J_{cc}=0.50$ meV. The calculated $J_{ab}$, $J_{bb}$ and $J_{cc}$ values are consistent with those expected for TbMnO3 Kimura2003B . $J_{aa}$ was predicted to be FM by Kimura et al. Kimura2003B , but our calculations show it to be AFM. The classical spin analysis based on the Freiser method Freiser1961 predicts that the spin ground state is a spin spiral with the modulation vector $\mathbf{q}=(0,q_{y},0)$ with $q_{y}=\mathrm{acos}(-\frac{J_{ab}}{2J_{bb}})/\pi=0.15$. The prediction of a spin-spiral ground state is in agreement with the experimental observation Kenzelmann2005 , which shows an incommensurate spiral with $\mathbf{q}=(0,0.27,0)$ below 28 K. It is worthwhile to point out that if a smaller $U_{eff}$ is used $J_{bb}$ will be stronger, and $J_{ab}$ be weaker, hence leading to a larger $q_{y}$. To examine the magnetic anisotropy of the high-spin Mn3+ ion in TbMnO3, it is necessary to consider DFT+U calculations with SOC included. TbMnO3 has four Mn3+ ions per unit cell. The easy axes of these ions are not collinear but related by symmetry. Since the effect of SOC is largely local in nature, we consider for simplicity only one Mn ion (i.e., Mn1 in Fig. 1) per unit cell by replacing the remaining three Mn3+ ions with Sc3+ ions that have no magnetic moment. To remove any possible coupling with the Tb 4f moment, we replace the Tb3+ ions of the unit cell with La3+ ions that have no magnetic moment. The energy dependence upon the Mn spin direction obtained from DFT+U+SOC calculations is shown in Fig. 3(a) for the cases when the Mn spin lies in the $ab$ and $bc$ planes. The energy minimum occurs at 60∘ and 70∘ for the case when the spin lies in the $ab$ and $bc$ planes, respectively, and the energy minimum for the $ab$-plane is lower than that for $bc$-plane by $0.12$ meV/Mn. The above results can be readily understood by analyzing the effect of the SOC Hamiltonian Xiang2007B ; Xiang2008 . As shown in Fig. 3(b), each MnO6 octahedron of TbMnO3 is axially elongated. With one of the two longest Mn-O bonds taken as the local $z$ axis and neglecting the slight difference between the other four short Mn-O bonds, the d-block levels of the Mn3+ (d4) ion are described as shown on the right hand side of Fig. 3(b), which leads to the electron configuration $(xz)^{1}(yz)^{1}(xy)^{1}(z^{2})^{1}(x^{2}-y^{2})^{0}$. The SOC Hamiltonian $\lambda\hat{\mathbf{L}}\cdot\hat{\mathbf{S}}$ leads to an interaction of the empty $d_{x^{2}-y^{2}}$ ($e_{g}$) state with the other d-states. The strongest SOC occurs between two $e_{g}$ states ($d_{xy}$ and $d_{x^{2}-y^{2}}$) with the maximum energy lowering when the spin lies in the local $z$ direction Xiang2008 . For the Mn1 ion, the $\theta$ and $\phi$ angles of the local $z$ direction in the global coordinate system are $\theta=80.2^{\circ}$ and $\phi=60.5^{\circ}$, respectively. In good agreement with these values, our SOC analysis based on tight-binding calculations Harrison shows that the actual easy axis of Mn1 has $\theta=84^{\circ}$ and $\phi=60^{\circ}$. Thus the easy axis is close to the $ab$ plane, and is far from the $c$ axis. This explains why the energy minimum for the Mn spin lying in the $ab$-plane has a lower energy than that lying in the $bc$-plane. In addition, the easy axis is closer to the $b$ axis than to the $a$ axis. Our result is consistent with the experimental observation Quezel1977 that the sine-wave modulation of the Mn magnetic moment below 40 K has the direction parallel to the $b$-axis. Since the easy axis of the Mn spin is far from the $c$ direction, it is expected that the spin-spiral plane of TbMnO3 is either the $ab$-plane or the $bc$-plane, but not the $ac$-plane test . Experimentally Kenzelmann2005 , it was found that the Mn moments of TbMnO3 form a $bc$-plane elliptical spiral with $m_{b}$(Mn) $=3.9$ $\mu_{B}$ $>$ $m_{c}$(Mn) $=2.8$ $\mu_{B}$. This can be readily explained by the fact that the easy axis of the Mn spins is close to the $b$ axis, but far from the $c$ direction. Figure 3: (Color online) (a) The dependence of energy on the direction of the Mn spin in the $ab$-plane or the $bc$-plane. (b) The local structure of the distorted MnO6 octahedron. The numbers give the Mn-O bond lengths in Å. The local coordinate system is also indicated. The right-hand-side diagram illustrates the electron configuration of the Mn3+ ($d^{4}$) ion. The label “SOC” denotes that the largest SOC mixing occurs between the $d_{xy}$ and $d_{x^{2}-y^{2}}$ states. (c) The dependence of energy on the direction of Tb spins in the $ab$-plane. Figure 4: (Color online) The atomic displacements (indicated by solid arrows) of the spin-spiral states after geometry optimization for (a) the $bc$-plane spiral and (b) the $ab$-plane spiral. The dashed arrows indicate the directions of the Mn spins in the spiral states. The directions of the electric polarizations are also shown. The magnetic anisotropy of the Tb3+ ion in TbMnO3 was also calculated in a similar manner; three of the four Tb3+ ions in a unit cell were replaced by La3+ ions with all the Mn3+ ions of a unit cell replaced by Sc3+ ions. Our DFT+U+SOC calculations show that the Tb spin prefers to lie in the $ab$ plane since the state for the spin parallel to the $c$-axis is higher than that for the spin in the $ab$ plane by at least 0.83 meV/Tb. The energy calculated for Tb1 (as labeled in Fig. 1 as a function of the angle $\phi$ with $\theta=90^{\circ}$ is presented in Fig. 3(c), which reveals that the energy miminum occurs at around $\phi=145^{\circ}$. The large anisotropy energy is consistent with the Ising behavior of Tb moment Prokhnenko2007 . The easy axes of the Tb3+ ions presented in Fig. 1 show that they lie symmetrically around the $b$-axis with the angle $55^{\circ}$. This is in excellent agreement with the observed angle of $57^{\circ}$ Quezel1977 . The electric polarization of TbMnO3 in the spin-spiral state was calculated using the VASP. To reduce the computational task, we considered the $\mathbf{q}=(0,1/3,0)$ state, which we simulated by using a $1\times 3\times 1$ supercell. Using the experimental centrosymmetric structure, the electric polarization from the pure electronic effect was calculated to be $\mathbf{P}=(0,0,0.5)$ $\mu C/m^{2}$ polarization ; Seki2008 ; Malashevich2008 for the $bc$-plane spiral shown in Fig. 4(a). Experimentally, the magnitude of the electric polarization for the $bc$-plane spiral is about 600 $\mu C/m^{2}$ Prokhnenko2007B , which is three orders of magnitude larger than the value calculated with no geometry relaxation. For the $ab$-plane spiral shown in Fig. 4(b), the electric polarization is calculated to be $\mathbf{P}=(-331.0,0,0)$ $\mu C/m^{2}$ in the absence of geometry relaxation. The absolute directions of the FE polarizations obtained for the $bc$\- and $ab$-plane spin-spiral states from DFT calculations without geometry relaxation are opposite to those predicted by the KNB model (Eq. 6 of ref. Katsura2005 ). To examine the effect of ion displacements in the spin-spiral states on the FE polarization, we optimized the atom positions of TbMnO3 in the $bc$\- and $ab$-plane spiral states by performing DFT+U+SOC calculations and calculated the electric polarizations of TbMnO3 using the relaxed structures. These calculations lead to $P_{c}$ = $-424.0$ $\mu C/m^{2}$ for the $bc$-plane spiral, and to $P_{a}$ = $131.2$ $\mu C/m^{2}$ for the $ab$-plane spiral. The absolute directions of the FE polarizations are switched for both spin-spiral states under geometry relaxation. The calculated FE polarization for the $bc$-plane spiral using the relaxed structure is now much closer to the experimental value. Furthermore, the direction of the FE polarization is in agreement with experiment Yamasaki2007B ; polarization ; Seki2008 ; Malashevich2008 as well as the KNB model. However, this agreement between the KNB model and experimental values is fortuitous; for LiCuVO4 and LiCu2O2, which consists of CuO2 ribbon chains, the KNB model predicts the wrong absolute direction of the FE polarization for the $ab$-plane spiral, but the correct absolute direction of the FE polarization for the $bc$-plane spiral, regardless of whether the unrelaxed or the relaxed crystal structures are employed Xiang2007 . The failure of the KNB model could be due to the fact that it was derived for the $t_{2g}$ systems while the $e_{g}$ states are important in Mn3+ and Cu2+ systems. Furthermore, we notice that the FE polarization of the $ab$-plane spiral calculated by using the relaxed structure, now along the $a$ direction, is smaller in magnitude than that calculated by using the experimental structure. In the case of LiCuVO4 and LiCu2O2, the geometry relaxation in the spin-spiral states was found to enhance the magnitudes of the FE polarizations without changing their absolute directions Xiang2007 . To show how the ion displacements break the centrosymmetry, we present the pattern of the atom displacements with respect to the centrosymmetric structure in Fig. 4. For the $bc$-plane spiral, the O atoms at the Wyckoff 4c position (i.e., O1) lying in the $ab$ plane of the Tb atoms almost do not move. The movement of all the O atoms at the Wyckoff 8d position (i.e., O2) close to the $ab$ plane of the Mn atoms have a component along the $c$ direction. All the Mn and Tb atoms have a displacement along the $-c$ direction. The largest displacements along the $c$ direction (about $2.4\times 10^{-4}$ Å) occur at the Mn sites rather than at the O sites. This finding is in contradiction to the assumption introduced in the “ion-displacement” model Sergienko2006 . Considering that the Born effective charge is positive for the Tb3+ and Mn3+ cations, and negative for the O2- anion, the total electric polarization is expected to be along the $-c$ direction, in agreement with the DFT calculation. For the $ab$-plane spiral, some O atoms have displacements along the $-a$ direction, but other O atoms have displacements along the $a$ direction. The sum of the O ion displacements is along the $-a$ direction. The occurrence of alternating O displacements is consistent with the prediction made by Sergienko and Dagotto Sergienko2006 , and is responsible for the smaller electric polarization when compared with the case of the $bc$-plane spiral. All the Mn atoms have a displacement along the $a$ direction. Another unexpected finding is that some Tb atoms have the largest displacements with a large component along the $b$ or $-b$ direction and a small component along the $a$ direction. In summary, the absence of the $ac$-plane spin-spiral in TbMnO3 is explained by the magnetic anisotropy of the Mn3+ ion. The calculated easy axis for Tb is in excellent agreement with the experimental result. The consideration of the ion displacements in the spin-spiral states of TbMnO3 is essential in determining the magnitude and the absolute direction of the FE polarizations, which is in support of the “ion-displacement” model. Surprisingly, the displacements of the Mn3+ and Tb3+ ions are generally greater than those of the O2- ions. The KNB model, however, can fail to describe both the magnitude and the absolute direction of FE polarization. Work at NREL was supported by the U.S. Department of Energy, under Contract No. DE-AC36-99GO10337. The research at NCSU was supported by the Office of Basic Energy Sciences, Division of Materials Sciences, U.S. Department of Energy, under Grant No. DE-FG02-86ER45259. We thank Prof. D. Vanderbilt for the discussion about the definition of positive electric polarization Yamasaki2007B ; Seki2008 ; Malashevich2008 . ## References * (1) S.-W. Cheong and M. Mostvoy, Nature Mater. 6, 13 (2007). * (2) T. Kimura et al., Nature (London) 426, 55 (2003). * (3) G. Lawes et al., Phys. Rev. Lett. 95, 087205 (2005). * (4) N. Hur et al., Nature (London) 429, 392 (2004). * (5) H. Katsura et al., Phys. Rev. Lett. 95, 057205 (2005). * (6) M. Mostovoy, Phys. Rev. Lett. 96, 067601 (2006). * (7) I. A. Sergienko and E. Dagotto, Phys. Rev. B 73, 094434 (2006). * (8) S. Picozzi et al., Phys. Rev. Lett. 99, 227201 (2007). * (9) H. J. Xiang and M.-H. Whangbo, Phys. Rev. Lett. 99, 257203 (2007). The definition of the positive direction of electric polarization adopted in this work is opposite to the conventional one polarization ; Seki2008 ; Malashevich2008 . * (10) F. Schrettle et al., arXiv:0712.3583v1. * (11) O. Prokhnenko et al., Phys. Rev. Lett. 99, 177206 (2007). * (12) A. I. Liechtenstein et al., Phys. Rev. B 52, R5467 (1995). * (13) J. P. Perdew et al., Phys. Rev. Lett. 77, 3865 (1996). * (14) R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993); R. Resta, Rev. Mod. Phys. 66, 899 (1994). * (15) P. E. Blöchl, Phys. Rev. B 50, 17953 (1994); G. Kresse and D. Joubert, ibid 59, 1758 (1999). * (16) G. Kresse and J. Furthmüller, Comput. Mater. Sci. 6, 15 (1996); Phys. Rev. B 54, 11169 (1996); * (17) P. Blaha et al., in WIEN2K, An Augmented Plane Wave Plus Local Orbitals Program for Calculating Crystal Properties, edited by K. Schwarz (Techn. Universität Wien, Austria, 2001). * (18) J. A. Alonso et al., Inorg. Chem. 39, 917 (2000). * (19) T. Kimura et al., Phys. Rev. B 68, 060403(R) (2003). * (20) S. Picozzi et al., Phys. Rev. B 74, 094402 (2006). * (21) M. J. Freiser, Phys. Rev. 123, 2003 (1961). * (22) M. Kenzelmann et al., Phys. Rev. Lett. 95, 087206 (2005). * (23) H. J. Xiang and M. -H. Whangbo, Phys. Rev. B 75, 052407 (2007). * (24) H. J. Xiang et al., Phys. Rev. Lett. 100, 167207 (2008). * (25) W. A. Harrison, Electronic Structure and Properties of Solids (Freeman, San Francisco, 1980). * (26) S. Quezel et al., Physica B 86, 916 (1977). * (27) VASP calculations show that the $ac$-plane spiral has higher energy by 0.23 meV/Mn than the $ab$-plane or $bc$-plane spiral. The energy difference between the $ab$-plane spiral and $bc$-plane spiral is within 0.02 meV/Mn. This difference is much smaller than that (0.12 meV/Mn) between the minima on the $bc$ and $ab$ plane since in a spiral not all spins lie along the minimum direction. * (28) In this work we adopt the conventional definition of the positive direction of electric polarization (i.e., from the negative to the positive charge site). * (29) S. Seki et al., Phys. Rev. Lett. 100, 127201 (2008). * (30) A. Malashevich and D. Vanderbilt, Phys. Rev. Lett. 101, 037210 (2008). * (31) O. Prokhnenko et al., Phys. Rev. Lett. 98, 057206 (2007). * (32) Y. Yamasaki et al., Phys. Rev. Lett. 98, 147204 (2007). The polarization direction of this report is in error. It should have been along the $-c$ direction polarization ; Seki2008 ; Malashevich2008 .

arxiv-papers

2024-09-04T02:48:54.409841

0803.2747

# Irreducible multi-particle correlations in states without maximal rank D.L. Zhou Beijing National Laboratory for Condensed Matter Physics Institute of Physics, Chinese Academy of Sciences, Beijing 100080, China ###### Abstract In a system of $n$ quantum particles, the correlations are classified into a series of irreducible $k$-particle correlations ($2\leq k\leq n$), where the irreducible $k$-particle correlation is the correlation appearing in the states of $k$ particles but not existing in the states of $k-1$ particles. A measure of the degree of irreducible $k$-particle correlation is defined based on the maximal entropy construction. By adopting a continuity approach, we overcome the difficulties in calculating the degrees of irreducible multi- particle correlations for the multi-particle states without maximal rank. In particular, we obtain the degrees of irreducible multi-particle correlations in the $n$-qubit stabilizer states and the $n$-qubit generalized GHZ states, which reveals the distribution of multi-particle correlations in these states. ###### pacs: 03.65.Ud, 03.67.Mn, 89.70.Cf Introduction. — How to classify and quantify different types of correlations in a multi-particle quantum state is fundamental in many-particle physics and quantum information. The traditional method to characterize the multi-particle correlation in many-particle physics is to use the multi-particle correlation functions, which are directly associated with experimental observables. Another method is originated from Shannon’s deep insight of information Shannon , where the entropy is used as a measure of information. Since different types of correlations in a multi-particle state can be regarded as different types of nonlocal information, a natural idea is to build a relation between correlation and entropy. Along this direction, the concept of the irreducible $n$-particle correlation in an $n$-particle quantum state was first proposed in Ref. Linden by Linden et al.. The counterpart for a probability distribution of $n$ classical variables was given in Ref. Schneidman . Note that, the concept in Ref. Linden was naturally generalized in Ref. Schneidman : a series of the connected information of order $k$ are defined, which corresponds to the irreducible $k$-particle correlation in an $n$-particle quantum state. The degree of irreducible $2$-particle correlation in a $2$-particle quantum state is equal to the $2$-particle mutual entropy Linden , which is also obtained in different contexts Groisman ; Schumacher . The irreducible $n$-particle correlation in an $n$-particle state has been shown to be zero for most $n$-particle pure states Linden2 , e.g., among $n$-qubit pure states, the irreducible $n$-particle correlation is not zero only for the GHZ type pure states Walck . In this letter, based on the maximal entropy construction, we give the definition of a measure of the degree of irreducible $k$-particle correlation in an $n$-particle state. This definition can be regarded not only as a direct generalization of the concept in Ref. Linden , but also as the quantum version of connected correlation of order $k$ in Ref. Schneidman . It is worthy to note that the correlations in the $n$-particle system can be classified into the irreducible $k$-particle correlations. In another word, the degrees of irreducible multi-particle correlations tell us how the correlations are distributed into the system. Because the measure of the degree of irreducible $k$-particle correlation is defined by the constrained optimization over the $n$-particle quantum states, its explicit calculation in a general $n$-particle state ($n>2$) is challenging, even for a $3$-qubit state. To our best knowledge, no explicit calculations of irreducible multi-particle correlations exist in the available literature. The main purpose of this letter is to provide a continuity approach to calculate the degrees of irreducible multi-particle correlations for the multi-particle states without maximal rank, which are interested or useful in many particle physics or quantum information. In particular, we obtain the analytic results for the degrees of irreducible multi-particle correlations for the stabilizer states Gottesman ; Raussendorf ; Hein and the generalized GHZ states Walck . Notations and definitions. — For simplicity, we introduce the following notations. The set $\mathbf{e}(n)=\\{1,2,\cdots,n\\}$. An $m$-element subset of the set $\mathbf{e}(n)$ is denoted as $\mathbf{a}(m)=\\{a_{1},a_{2},\cdots,a_{m}\\}$, and the complementary set of the set $\mathbf{a}(m)$ relative to the set $\mathbf{e}(n)$ is denoted as $\mathbf{\bar{a}}(n-m)=\\{\bar{a}_{1},\bar{a}_{2},\cdots,\bar{a}_{(n-m)}\\}$. In a system of $n$ quantum particles, the complete knowledge of its state is specified by the $n$-particle density matrix $\rho^{\mathbf{e}(n)}$. The irreducible $k$-particle correlation $(2\leq k\leq n)$ in the state is defined as the correlation appearing in the $k$-particle reduced density matrixes $\rho^{\mathbf{a}(k)}$, but not existing in the $(k-1)$-particle states $\rho^{\mathbf{a}(k-1)}$. To define a measure of the degree of irreducible multi-particle correlations in the state $\rho^{\mathbf{e}(n)}$, similar to the method we adopted in Ref. Zhou1 , we introduce an $n$-particle density matrix $\tilde{\rho}_{l}^{\mathbf{e}(n)}$ for each $l\in\mathbf{e}(n)$ that satisfies: $\tilde{\rho}_{l}^{\mathbf{e}(n)}\in\\{\rho_{l}^{\mathbf{e}(n)}|\mathrm{max}S(\rho_{l}^{\mathbf{e}(n)})\\},$ (1) where the $l$-particle reduced density matrix $\rho_{l}^{\mathbf{a}(l)}=\rho^{\mathbf{a}(l)}$ (2) for arbitrary subset $\mathbf{a}(l)$, and the function $S$ is the von Neumann entropy defined as $S(x)=-\mathrm{Tr}(x\log_{2}x)$. In another word, the $n$-particle density matrix $\tilde{\rho}_{l}^{\mathbf{e}(n)}$ has the same $l$-particle reduced density matrixes as those of $\rho^{\mathbf{e}(n)}$, but it is maximally noncommittal to the other missing information contained in the state $\rho^{\mathbf{e}(n)}$ Jaynes . A measure of the degree of irreducible $k$-particle correlation in the state $\rho^{\mathbf{e}(n)}$ is then defined as $C^{(k)}(\rho^{\mathbf{e}(n)})=S(\tilde{\rho}_{k-1}^{\mathbf{e}(n)})-S(\tilde{\rho}_{k}^{\mathbf{e}(n)}).$ (3) The total correlation in the state $\rho^{\mathbf{e}(n)}$ is referred to the nonlocal information appearing in $\rho^{\mathbf{e}(n)}$, but not existing in the single particle states $\rho^{\mathbf{a}(1)}$. A measure of the degree of the total correlation in the state $\rho^{\mathbf{e}(n)}$ is then defined as $C^{T}(\rho^{\mathbf{e}(n)})=S(\tilde{\rho}_{1}^{\mathbf{e}(n)})-S(\tilde{\rho}_{n}^{\mathbf{e}(n)}).$ (4) Substituting Eqs. (3) into Eq. (4), we find that $C^{T}(\rho^{\mathbf{e}(n)})=\sum_{k=2}^{n}C^{(k)}(\rho^{\mathbf{e}(n)}).$ (5) Eq. (5) not only justifies that Eq. (4) is a legitimate measure of the total correlation, but also implies that all the irreducible multi-particle correlations construct a classification of the total correlation. Note that the above definitions of the degrees of different types of correlations, Eqs. (3) and Eq. (4), are intimately related with the von Neaumann entropy. The underlying reason is as follows. On the one hand, the von Neaumann entropy of a quantum state is a measure of the degree of uncertainties of the state. On the other hand, the existence of correlation in the multi-particle quantum state decreases the uncertainties of the state. Therefore the decreasing of uncertainties, i.e, the entropy difference, is reasonable to be used as a measure of the degree of correlation. Standard exponential form. — Note that the $n$-particle density matrixes $\tilde{\rho}_{l}^{\mathbf{e}(n)}$ are essential elements in the definitions in Eqs. (3) and Eq. (4). The following important theorem gives the standard exponential form of the state $\tilde{\rho}_{l}^{\mathbf{e}(n)}$. Theorem $1$: For an $n$-particle quantum state $\rho^{\mathbf{e}(n)}$ with maximal rank, a state $\tilde{\rho}_{l}^{\mathbf{e}(n)}$ ($1\leq l\leq n$) which satisfies Eqs. (1) and (2) can be written in the following exponential form: $\tilde{\rho}_{l}^{(\mathbf{e}(n))}=\exp\Big{(}\sum_{\mathbf{a}(l)}\Lambda^{\mathbf{a}(l)}\otimes 1^{\mathbf{\bar{a}}(n-l)}\Big{)},$ (6) where $1^{\mathbf{\bar{a}}(n-l)}$ is the identity operators on the Hilbert space of particles $\mathbf{\bar{a}}(n-l)$, and the operators $\Lambda^{\mathbf{a}(l)}$ are to be determined by Eqs. (2). Proof: The Lagrange multipliers $\Lambda^{\mathbf{a}(l)}$ are introduced to transform the constrained maximization defined by Eqs. (1) and (2) to the following unconstrained minimization: $\displaystyle-S(\rho_{m}^{\mathbf{e}(n)})-\sum_{\mathbf{a}(l)}\mathrm{Tr}\big{(}\Lambda^{\mathbf{a}(l)}(\rho_{m}^{\mathbf{a}(l)}-\rho^{\mathbf{a}(l)})\big{)}\geq\mathrm{Tr}\big{(}\Lambda^{\mathbf{a}(l)}\rho^{\mathbf{a}(l)}\big{)},$ where the Klein inequality Wehrl is used. The equality is satisfied if and only if Eq. (6) is satisfied. The Lagrange multiplies $\Lambda^{\mathbf{a}(l)}$ are to be determined by Eqs. (2). Because the Klein inequality Wehrl involves only the positive operators, we need to limit ourselves to the states with maximal rank. A direct result of theorem $1$ is the following corollary. Corollary $1$: For $m=1$ case, we have $\tilde{\rho}_{1}^{\mathbf{e}(n)}=\exp\Big{(}\sum_{\mathbf{a}(1)}\Lambda^{\mathbf{a}(1)}\otimes 1^{\mathbf{\bar{a}}(n-1)}\Big{)}=\prod_{i=1}^{n}\otimes\rho^{(i)}.$ Therefore the degree of the total correlation (5) in the state $\rho^{\mathbf{e}(n)}$ is give by $C^{T}(\rho^{\mathbf{e}(n)})=\sum_{i=1}^{n}S(\rho^{(i)})-S(\rho^{\mathbf{e}(n)}),$ (7) where we used $\tilde{\rho}_{n}^{\mathbf{e}(n)}=\rho^{\mathbf{e}(n)}$. Although the degree of the total correlation has an analytical expression (7), we have not been able to give similar analytical results for the degrees of irreducible multi-particle correlations $C^{(k)}(\rho^{\mathbf{e}(n)})$. Note that theorem $1$ is a direct generalization of the corresponding result in Ref. Linden . It shows that the feature of multi-particle correlation in the state $\tilde{\rho}_{l}^{\mathbf{e}(n)}$ is directly embodied in the exponential form of the state. As emphasized in Ref. Linden , theorem $1$ are not available for the multi-particle states without maximal rank. However, most multi-particle states interested in many particle physics or quantum information have not maximal rank, e.g., the $n$-qubit stabilizer states and the generalized GHZ states discussed below. To overcome the difficulties in using theorem $1$ to treat with the states without maximal rank, we adopt the following approach. A multi-particle state without maximal rank can always be regarded as the limit case of a series of multi-particle states with maximal rank. If the degrees of irreducible multi- particle correlations for the series of states with maximal rank can be obtained, then we take the limit to get the degrees of irreducible multi- particle correlations for the state without maximal rank. We called the above method as the continuity approach. The proofs of theorem $2$ and $3$ below are typical applications of this approach. Correlations in stabilizer states. — An $n$-qubit stabilizer state state $\rho_{S}^{\mathbf{e}(n)}$ is defined as $\rho_{s}^{\mathbf{e}(n)}=\frac{1}{2^{n}}\sum_{\\{\alpha_{i}\in\mathbf{z}\\}}\prod_{i=1}^{m}g_{i}^{\alpha_{i}},$ (8) where the set $\mathbf{z}=\\{0,1\\}$, $\\{g_{i}\\}$ are $m$ ($m\leq n$) independent commute $n$-qubit Pauli group elements. The set $\mathfrak{g}(\rho_{s}^{\mathbf{e}(n)})=\\{g_{i}\\}$ is called the stabilizer generator of the state $\rho_{s}^{\mathbf{e}(n)}$, and the group generated by the generator $\mathfrak{g}$, denoted as $\mathcal{G}(\rho_{s}^{\mathbf{e}(n)})=\\{\prod_{i}g_{i}^{\alpha_{i}},\alpha_{i}\in\mathbf{z}\\}$, is called the stabilizer of the state. To make the state (8) to be a legitimate state, the minus identity operator is required not to be an element of the stabilizer $\mathcal{G}$. Note that our definition of the $n$-qubit stabilizer state is an extension of the usual definition Hein , which corresponds to the case when $m=n$. When $m<n$, the stabilizer states defined by Eq. (8) are no longer pure states. According to the definition of the $n$-qubit Pauli group, an element $h\in\mathcal{G}(\rho_{s}^{\mathbf{e}(n)})$ can be written as $h=\pm\prod_{i=1}^{n}O^{(i)}$ for $O\in\\{I,X,Y,Z\\}$, where $I$ is the $2\times 2$ identity operator, and $X$, $Y$, $Z$ are three Pauli matrixes. The number of identity operator $I$ in the element $h$ is $N_{I}(h)=\sum_{i}\mathrm{Tr}O^{(i)}/2$. The stabilizer $\mathcal{G}(\rho_{s}^{\mathbf{e}(n)})$ can be classified into a series of sets $\mathcal{G}_{k}(\rho_{s}^{\mathbf{e}(n)})=\\{h|h\in\mathcal{G}(\rho_{s}^{\mathbf{e}(n)}),N_{I}(h)\geq n-k\\}$ for $k\in\mathbf{e}(n)$. Although in general $\mathcal{G}_{k}(\rho_{s}^{\mathbf{e}(n)})$ is not a group, we can still define the generator $\mathfrak{g}_{k}(\rho_{s}^{\mathbf{e}(n)})$ for the set $\mathcal{G}_{k}(\rho_{s}^{\mathbf{e}(n)})$ as a set of elements in $\mathcal{G}_{k}(\rho_{s}^{\mathbf{e}(n)})$ such that every element in $\mathcal{G}_{k}(\rho_{s}^{\mathbf{e}(n)})$ can be written as a unique product of elements in the set. As in group theory, the choice of the generator $\mathfrak{g}_{k}(\rho_{s}^{\mathbf{e}(n)})$ is not unique in general. The irreducible $k$-particle correlation in an $n$-qubit stabilizer state is given by the following theorem. Theorem $2$: The irreducible $k$-particle irreducible correlation in an $n$-qubit stabilizer state $\rho_{s}^{\mathbf{e}(n)}$ is $C^{(k)}(\rho_{s}^{\mathbf{e}(n)})=|\mathfrak{g}_{k}(\rho_{s}^{\mathbf{e}(n)})|-|\mathfrak{g}_{k-1}(\rho_{s}^{\mathbf{e}(n)})|,$ (9) where $|\cdot|$ denotes the size of a set. Proof: Note that we can always take $\mathfrak{g}_{1}(\rho_{s}^{\mathbf{e}(n)})\subseteq\mathfrak{g}_{2}(\rho_{s}^{\mathbf{e}(n)})\subseteq\cdots\subseteq\mathfrak{g}_{n}(\rho_{s}^{\mathbf{e}(n)})$. Then the elements contained in $\mathfrak{g}_{k}(\rho_{s}^{\mathbf{e}(n)})$ but not in $\mathfrak{g}_{k-1}(\rho_{s}^{\mathbf{e}(n)})$ are reexpressed as $g_{ki}$ for $i\in\mathbf{e}(|\mathfrak{g}_{k}|-|\mathfrak{g}_{k-1}|)$. Thus we can construct an $n$-qubit state with a real parameter $\lambda$ as $\rho^{\mathbf{e}(n)}_{m}(\lambda)=\exp\Big{(}\eta+\lambda\sum_{k=1}^{m}\sum_{i=1}^{|\mathfrak{g}_{k}|-|\mathfrak{g}_{k-1}|}g_{ki}\Big{)},$ (10) where $\eta=-\ln(2^{n}\cosh^{|\mathfrak{g}_{m}|}\lambda)$, which is determined by the normalization condition $\mathrm{Tr}(\rho_{m}^{\mathbf{e}(n)})=1$. Then the above state is expanded as $\rho^{\mathbf{e}(n)}_{m}(\lambda)=\frac{1}{2^{n}}\Big{(}1+\sum_{d=1}^{|\mathfrak{g}_{m}|}\tanh^{d}\lambda\sum_{\sum\alpha_{ki}=d}\prod_{k\leq m}g_{ki}^{\alpha_{ki}}\Big{)}.$ (11) Note that if $\exists$ $\alpha_{ki}=1$ for $k>m$, then $\forall\mathbf{a}(m)$, we have $\mathrm{Tr}_{\mathbf{\bar{a}}(n-m)}\Big{(}\prod_{k\leq n}g_{(ki)}^{\alpha_{ki}}\Big{)}=0$. Thus the $m$-particle reduced density matrix $\rho_{m}^{\mathbf{a}(m)}(\lambda)=\rho_{n}^{\mathbf{a}(m)}(\lambda)$. According to theorem $1$, the degree of irreducible $k$-particle correlation in the $n$-qubit state $\rho_{n}^{\mathbf{e}(n)}(\lambda)$ is $C^{(k)}(\rho_{n}^{\mathbf{e}(n)}(\lambda))=S(\rho_{k-1}^{\mathbf{e}(n)}(\lambda))-S(\rho_{k}^{\mathbf{e}(n)}(\lambda)).$ (12) From Eq. (11), we observe that when the parameter $\lambda$ takes the limit of positive infinity, the states $\rho^{\mathbf{e}(n)}_{m}(\lambda)$ are stabilizer states. In particular, $\lim_{\lambda\rightarrow+\infty}\rho^{\mathbf{e}(n)}_{n}(\lambda)=\rho_{s}^{\mathbf{e}(n)}.$ (13) It is easy to prove that $S(\rho_{m}^{\mathbf{e}(n)}(+\infty))=n-|l_{m}|$. Then the degree of irreducible $k$-particle correlation in the stabilizer state $\rho_{s}^{\mathbf{e}(n)}$ is $C^{(k)}(\rho_{s}^{\mathbf{e}(n)})=|\mathfrak{g}_{k}(\rho_{s}^{\mathbf{e}(n)})|-|\mathfrak{g}_{k-1}(\rho_{s}^{\mathbf{e}(n)})|.$ (14) Note that the result of theorem $2$ has been given in Ref. Zhou , which is based on some reasonable arguments in the context of multi-party threshold secret sharing. Theorem $2$ can be used to analyze the multi-particle correlation distributions in all the stabilizer states. Let us illustrate its power with analysis of the correlations in the two stabilizer states: $\sigma_{1}^{\mathbf{e}(3)}=1/2(|000\rangle\langle 000|+|111\rangle\langle 111|)$ and $\sigma_{2}^{\mathbf{e}(3)}=|\textrm{GHZ}\rangle\langle\textrm{GHZ}|$ with $|\textrm{GHZ}\rangle=1/\sqrt{2}(|000\rangle+|111\rangle)$. A simple calculation yields the following results. For the state $\sigma_{1}^{\mathbf{e}(3)}$, there are $2$ bits of correlations altogether, and these $2$ bits of correlations are irreducible $2$-particle correlations. For the other state $\sigma_{2}^{\mathbf{e}(3)}$, the total correlations become $3$ bits, and these $3$ bits of correlations are classified into $2$ bits of irreducible $2$-particle correlation and $1$ bit of irreducible $3$-particle correlation. Correlations in generalized GHZ states. — A generalized $n$-qubit GHZ states is defined as $|G_{n}\rangle=\alpha|00\cdots 0\rangle+\beta|11\cdots 1\rangle,$ (15) where the parameters $\alpha$ and $\beta$ satisfy $|\alpha|^{2}+|\beta|^{2}=1$ and $\alpha\beta\neq 0$. The degrees of irreducible multi-particle correlations for the state $\rho_{G}^{\mathbf{e}(n)}$ ($\equiv|G_{n}\rangle\langle G_{n}|$) are given by the following theorem. Theorem $3$: The degrees of irreducible multi-particle correlation in the generalized GHZ state (15) are given by $\displaystyle C^{(2)}(\rho_{G}^{\mathbf{e}(n)})$ $\displaystyle=$ $\displaystyle(n-1)E(|\alpha|^{2}),$ (16) $\displaystyle C^{(n)}(\rho_{G}^{\mathbf{e}(n)})$ $\displaystyle=$ $\displaystyle E(|\alpha|^{2}),$ (17) where $E(x)=-x\log_{2}x-(1-x)\log_{2}(1-x)$. The degrees of the other types of irreducible multi-particle correlation are zero identically, i.e., $C^{(3)}(\rho_{G}^{\mathbf{e}(n)})=C^{(4)}(\rho_{G}^{\mathbf{e}(n)})=\cdots=C^{(n-1)}(\rho_{G}^{\mathbf{e}(n)})=0$. Proof: Let us construct an $n$-qubit state $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})=\exp\Big{(}\eta+\gamma\sum_{i=2}^{n}Z^{(1)}Z^{(i)}+\vec{\lambda}\cdot\vec{\Sigma}\Big{)},$ (18) where the vector $\vec{\lambda}=\lambda_{x}\hat{x}+\lambda_{y}\hat{y}+\lambda_{z}\hat{z}=\lambda\hat{\lambda}$ , the operator vector $\vec{\Sigma}=\hat{x}X^{(1)}\prod_{i=2}^{n}X^{(i)}+\hat{y}Y^{(1)}\prod_{i=2}^{n}X^{(i)}+\hat{z}Z^{(1)}$, the parameter $\eta$ is determined by the normalization condition $\mathrm{Tr}\big{(}\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})\big{)}=1$, and the notation $\hat{v}$ represents the unit vector along the direction of the vector $\vec{v}$. The $\hat{\lambda}$ component of the operator vector $\vec{\Sigma}$ is denoted as $\Sigma_{\lambda}=\hat{\lambda}\cdot\vec{\Sigma}$. Note that $\Sigma_{\lambda}^{\dagger}=\Sigma_{\lambda}$, $\Sigma_{\lambda}^{2}=1$, and $[\Sigma_{\lambda},Z^{(1)}Z^{(i)}]=0$ for $i\in\mathbf{e}(n)$. The state (18) can thus be written as $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})=\frac{1}{2^{n}}\sum_{i=2}^{n}[1+\tanh(\gamma)Z^{(1)}Z^{(i)}][1+\tanh(\lambda)\Sigma_{\lambda}].$ Note that in the above equation, only the term $Z_{1}\lambda_{z}/\lambda$ in $\Sigma_{\lambda}$ contributes to the reduced $(n-1)$-particle reduced density matrixes. Therefore the state $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda}^{\prime})$ has the same $(n-1)$-particle reduced density matrixes as the state $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})$ if the following condition is satisfied: $\displaystyle\lambda^{\prime}_{x}=\lambda^{\prime}_{y}=0,\;\;\tanh\lambda^{\prime}_{z}=\frac{\lambda_{z}}{\lambda}\tanh\lambda.$ (19) According to theorem $1$, we find $\tilde{\rho}_{m}^{\mathbf{e}(n)}(\gamma,\vec{\lambda})=\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda^{\prime}})$ (20) for $m=2,3,\cdots,n-1$. Therefore the degrees of irreducible multi-particle correlations for the state $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})$ can be obtained via Eqs. (3). Without loss of generality, we assume that in Eq. (15) $\alpha=\cos(\theta/2)$ and $\beta=\sin(\theta/2)e^{i\phi}$. Then we define the Bloch vector $\hat{u}=(\sin\theta\cos\phi,\sin\theta\sin\phi,\cos\theta)$. Let us take $\hat{\lambda}=\hat{u}$, then $\Sigma_{u}|G_{n}\rangle=|G_{n}\rangle$. The operators $\\{Z^{(1)}X^{(i)}\\}$ and $\Sigma_{u}$ can be regarded as the stabilizer generator of the state $\rho_{G}^{\mathbf{e}(n)}$. According to theorem $2$, the relation between the generalized GHZ state $\rho_{G}^{\mathbf{e}(n)}$ and $\rho^{\mathbf{e}(n)}(\gamma,\vec{\lambda})$ is $\rho_{G}^{\mathbf{e}(n)}=\lim_{\lambda\rightarrow+\infty}\rho^{\mathbf{e}(n)}(\lambda,\lambda\hat{u}).$ (21) In this case, we find that, for $m=2,3,\cdots,n-1$, $\displaystyle\lim_{\lambda\rightarrow+\infty}\tilde{\rho}_{m}^{\mathbf{e}(n)}(\lambda,\lambda\hat{u})$ (22) $\displaystyle=$ $\displaystyle|\alpha|^{2}|00\cdots 0\rangle\langle 00\cdots 0|+|\beta|^{2}|11\cdots 1\rangle\langle 11\cdots 1|.$ A direct calculation will yield the results of theorem $3$. Theorem $3$ shows that in the generalized $n$-qubit GHZ state (15), only the irreducible $2$-particle correlation and the irreducible $n$-particle correlation exist, and the previous one is $(n-1)$ times of the later one. Summary. — In summary, the definitions of the degrees of irreducible $k$-particle correlations in an $n$-particle state are given as a natural generalization of those defined in Linden ; Schneidman . The significance of exponential form of a multi-particle state in characterizing irreducible multi-particle correlation is emphasized by theorem $1$. Adopting the continuity approach, we are capable of applying theorem $1$ to deal with the irreducible multi-particle correlations in the multi-particle states without maximal rank. Particularly, we successfully obtained the degrees of irreducible $k$-particle correlations in the $n$-qubit stabilizer states and the $n$-qubit generalized GHZ states. The multi-particle correlation structures in these states are revealed by our results. We hope that the concepts of irreducible multi-particle correlations will contribute to the characterizations of multi-particle correlations in condensed matter system, e.g., topological orders Kitaev ; Wen ; Yang in degenerate ground states. The author thanks for helpful discussions with L. You, Z.D. Wang, and C.P. Sun. This work is supported by NSF of China under Grant 10775176, and NKBRSF of China under Grant 2006CB921206 and 2006AA06Z104. ## References * (1) C.E. Shannon, Bell Syst. Tech. J. 27, 379 (1948). * (2) N. Linden, S. Popescu, and W.K. Wootters, Phys. Rev. Lett. 89, 207901 (2002). * (3) E. Schneidman, S. Still, M.J. Berry II, and W. Bialek, Phys. Rev. Lett. 91, 238701 (2003). * (4) B. Groisman, S. Popescu, and A. Winter, Phys. Rev. A 72, 032317 (2005). * (5) B. Schumacher and M.D. Westmoreland, Phys. Rev. A 74, 042305 (2006). * (6) N. Linden and W.K. Wootters, Phys. Rev. Lett. 89, 277906 (2002). * (7) S.N. Walck and D.W. Lyons, Phys. Rev. Lett. 100, 050501 (2008). * (8) D. Gottesman, PhD thesis, Caltech (1997), arXiv:quant-ph/9705052. * (9) R. Raussendorf and H.J. Briegel, Phys. Rev. Lett. 86, 5188 (2001). * (10) M. Hein, W. Dür, J. Eisert, R. Raussendor, M. Van den Nest, and H.J. Briegel, in Proceedings of the International School of Physics“Enrico Fermi”, Course 162, Varenna, edited by G. Casati, et al. (IOS Press, Amsterdam, 2006), arXiv:quant-ph/0602096. * (11) D.L. Zhou, B. Zeng, Z. Xu, and L. You, Phys. Rev. A 74, 052110 (2006). * (12) E.T. Jaynes, Phys. Rev. 106, 620 (1957). * (13) A. Wehrl, Rev. Mod. Phys. 50, 221 (1978). The Klein inequality: for positive operators $A$ and $B$, $\mathrm{Tr}(A(\ln A-\ln B))\geq 0$, the equality is satisfied if and only if $A=B$. * (14) D.L. Zhou and L. You, arXiv: quant-ph/0701029. * (15) A. Kitaev and J. Preskill, Phys. Rev. Lett. 96, 110404 (2006). * (16) M. Levin and X.-G. Wen, Phys. Rev. Lett. 96, 110405 (2006). * (17) S. Yang, D.L. Zhou, and C.P. Sun, Phys. Rev. B 76, 180404(R) (2007).

arxiv-papers

2024-09-04T02:48:54.413873

0803.2753

# Resolving the chemistry in the disk of TW Hydrae I. Deuterated species Chunhua Qi11affiliation: Harvard–Smithsonian Center for Astrophysics, 60 Garden Street, MS 42, Cambridge, MA 02138, USA; cqi@cfa.harvard.edu, dwilner@cfa.harvard.edu. , David J. Wilner11affiliation: Harvard–Smithsonian Center for Astrophysics, 60 Garden Street, MS 42, Cambridge, MA 02138, USA; cqi@cfa.harvard.edu, dwilner@cfa.harvard.edu. , Yuri Aikawa22affiliation: Department of Earth and Planetary Sciences, Kobe University, Kobe 657-8501, Japan; aikawa@kobe-u.ac.jp. , Geoffrey A. Blake33affiliation: Divisions of Geological & Planetary Sciences and Chemistry & Chemical Engineering, California Institute of Technology, MS 150–21, Pasadena, CA 91125, USA; gab@gps.caltech.edu. , Michiel R. Hogerheijde44affiliation: Leiden Observatory, Leiden University, PO Box 9513, 2300 RA, Leiden, The Netherlands; michiel@strw.leidenuniv.nl. ###### Abstract We present Submillimeter Array (SMA) observations of several deuterated species in the disk around the classical T Tauri star TW Hydrae at arcsecond scales, including detections of the DCN J=3–2 and DCO+ J=3–2 lines, and upper limits to the HDO 31,2–22,1, ortho-H2D+ 11,0–11,1 and para-D2H+ 11,0–10,1 transitions. We also present observations of the HCN J=3–2, HCO+ J=3–2 and H13CO+ J=4–3 lines for comparison with their deuterated isotopologues. We constrain the radial and vertical distributions of various species in the disk by fitting the data using a model where the molecular emission from an irradiated accretion disk is sampled with a 2D Monte Carlo radiative transfer code. We find that the distribution of DCO+ differs markedly from that of HCO+. The D/H ratios inferred change by at least one order of magnitude (0.01 to 0.1) for radii $<$30 AU to $\geq$70 AU and there is a rapid falloff of the abundance of DCO+ at radii larger than 90 AU. Using a simple analytical chemical model, we constrain the degree of ionization, x(e-)=n(e-)/n(H2), to be $\sim 10^{-7}$ in the disk layer(s) where these molecules are present. Provided the distribution of DCN follows that of HCN, the ratio of DCN to HCN is determined to be 1.7$\pm$0.5 $\times$ 10-2; however, this ratio is very sensitive to the poorly constrained vertical distribution of HCN. The resolved radial distribution of DCO+ indicates that in situ deuterium fractionation remains active within the TW Hydrae disk and must be considered in the molecular evolution of circumstellar accretion disks. ###### Subject headings: circumstellar matter —comets: general —ISM: molecules —planetary systems: protoplanetary disks —stars: individual (TW Hydrae) —stars: pre-main-sequence ††slugcomment: ApJ accepted: March 18, 2008 ## 1\. Introduction Millimeter-wave interferometers have imaged the gas and dust surrounding over a dozen T Tauri and Herbig Ae stars (see Dutrey et al., 2007 for a review). These studies demonstrate the potential to dramatically improve our understanding of disk physical and chemical structure, providing insights that will ultimately enable a more comprehensive understanding of star and planet formation. As analogs to the Solar Nebula, these circumstellar disks offer a unique opportunity to study the conditions during the planet formation process, especially the complex chemical evolution that must occur. In the outer parts of the disk directly accessible to millimeter-wave interferometry, observations of deuterated species are particularly important because they can constrain the origin of primitive solar system bodies such as comets and other icy planetesimals. Deuterated molecule chemistry is sensitive to the temperature history of interstellar and circumstellar gas, as well as to density-sensitive processes such as molecular depletion in the cold ($<$ 20 K) and dense ($>$ 106 cm-3) disk midplane. The similarity of molecular D/H ratios between comets and high mass hot cores has been used to argue for an “interstellar” origin of cometary matter, but ambiguity remains and this argument is not secure (Bergin et al., 2007). Aikawa & Herbst (1999) investigate the chemistry of deuterium-bearing molecules in outer regions of protoplanetary disks and find that molecules formed in the disk have similar D/H ratios as those in comets. To date, the determination of D/H ratios in disks has been limited to the measurements of DCO+/HCO+ in two classical T Tauri Stars (cTTs) with single dish telescopes (TW Hydrae, van Dishoeck et al. 2003; DM Tauri, Guilloteau et al. 2006), but DCO+ has not been observed in comets. Although H2D+ and HDO lines have been detected in disks (Ceccarelli et al. 2004, 2005), there are no corresponding H${}_{3}^{+}$ and H2O observations that allow derivations of the D/H ratio. Therefore, direct measurements in disks of species observed in comets, such as DCN and HCN, will be important to help relate deuterium fractionation in disks to that in comets. Deuterated molecules in cold pre-stellar and protostellar cores are found to be enhanced by orders of magnitude over the elemental D/H abundance ratio of 1.5$\times$10-5 through fractionation in the gas phase at low temperatures, an effect driven primarily by deuterated H${}_{3}^{+}$, which is formed by exchange reaction between H${}_{3}^{+}$ and HD, then transfers its deuteron to neutral species. Theoretical models of disks (Aikawa et al., 2002; Willacy, 2007) with realistic temperature and density structure show that the DCO+/HCO+ ratio increases with radius due to the decreasing temperature moving out from the star. Spatially resolved data of deuterated molecules will help test disk physical models through these chemical consequences. The current capabilities of millimeter-wave observatories are limited both by sensitivity and by the small angular size of circumstellar disks, which makes spatially resolving the deuterium fractionation difficult. In this work we take advantage of the proximity of the most nearby classical T Tauri star, TW Hydrae, which is surrounded by a disk of radius $3\farcs 5$, or 200 AU at a distance of 56 pc (Qi et al., 2004), and a possible orbiting planet of mass (9.8$\pm$3.3) MJupiter at 0.04 AU (Setiawan et al., 2008), to study the physical and chemical structure of a protoplanetary environment at high spatial and spectral resolution using interferometry. The TW Hydrae disk is viewed nearly face-on, and so is well posed to investigate the radial distribution of molecular emission. Although the current angular resolution of $\sim$1–2′′ places only a few independent beams ( or ‘pixels’) across the disk, effectively improved resolution of the disk chemistry can be achieved thanks to the fact that the gas kinematics are essentially Keplerian. Beckwith & Sargent (1993) show that molecular line emission from Keplerian disks displays a “dipole” pattern near the systematic velocity due to the shear created by the orbital motion. The emission near the systemic velocity is dominated by the outer disk regions, and the separation of the emission peaks depends sensitively on the abundance gradient of the emitting species with radius. This important feature, commonly seen in velocity channel maps of disks with a range of inclination angles (even as small as 6–7 degrees in the case of TW Hydrae, Qi et al., 2004), may be used to study the radial distribution of molecular emission far more precisely than is afforded by the limited number of “pixels” provided by the available resolution. Here we report on Submillimeter Array (SMA)111The Submillimeter Array is a joint project between the Smithsonian Astrophysical Observatory and the Academia Sinica Institute of Astronomy and Astrophysics, and is funded by the Smithsonian Institution and the Academia Sinica. (Ho et al., 2004) observations of deuterated species in the disk around TW Hydrae, including the first detection and images of DCN and DCO+. In §2 we describe the observations, while in §3 we introduce the analysis method used and the molecular distribution parameters derived. We describe the model fitting results and discuss the implications in §4, and we present a summary and conclusions in §5. ## 2\. Observations All of the observations of TW Hydrae (R.A.: 11h01m51.s875; Dec: -34∘42′17.′′155; J2000.0) were made between 2005 February and 2006 December using the SMA 8 antenna interferometer located atop Mauna Kea, Hawaii. Table 1 and Table 2 summarize the observational parameters for the detected and undetected species, respectively. The SMA receivers operate in a double- sideband (DSB) mode with an intermediate frequency band of 4–6 GHz which is sent over fiber optic transmission lines to 24 “overlapping” digital correlator “chunks” covering a 2 GHz spectral window. Two settings were used for observing the HCN/DCN and HCO+/DCO+ lines: at 265.7–267.7 GHz (lower sideband) the tuning was centered on the HCN J=3–2 line at 265.8862 GHz in chunk S22 while the HCO+ J=3–2 line at 267.5576 GHz was simultaneously observed in chunk S02; at 215.4–217.4 GHz (lower sideband) the tuning was centered on the DCO+ J=3–2 line at 216.1126 GHz in chunk S16 while the DCN J=3–2 line at 217.2386 GHz was simultaneously observed in chunk S02 (the HDO 31,2–22,1 line at 225.90 GHz was also covered in the USB in chunk S06). Combinations of two array configurations (compact and extended) were used to obtain projected baselines ranging from 6 to 180 meters. Cryogenic SIS receivers on each antenna produced system temperatures (DSB) of 200-1400 K at 260 GHz and 100–200 K at 210 GHz. Observations of ortho-H2D+ and H13CO+ were simultaneously carried out (in the dual receiver observational mode) on 28 December 2006 using only the compact array configuration. Three out of eight antennas were equipped with 400 GHz receivers (DSB system temperature between 400 and 800 K) and were available for observations of the 372.4213 GHz ortho-H2D+ 11,0–11,1 line, and all eight antennas were operating with 345 GHz receivers (DSB system temperature between 150 and 300 K) for observations of the 346.9985 GHz H13CO+ J=4–3 and 345.796 GHz CO J=3–2 lines. The correlator was configured with a narrow band of 512 channels over the 104 MHz chunk, which provided 0.2 MHz frequency resolution, or 0.28 km s-1 at 217 GHz, 0.23 km s-1 at 267 GHz and 0.16 km s-1 at 372 GHz. Observations of the 691.6604 GHz para-D2H+ 11,0–10,1 line were shared with observations of the CO J=6–5 line made on 17 February 2005 (details provided in Qi et al., 2006). Calibrations of the visibility phases and amplitudes were achieved with interleaved observations of the quasars J1037-295 and J1147-382, typically at intervals of 20-30 minutes. Observations of Callisto provided the absolute scale for the calibration of flux densities. The uncertainties in the flux scale are estimated to be 15%. All of the calibrations were done using the MIR software package 222http://www.cfa.harvard.edu/$\sim$cqi/mircook.html, while continuum and spectral line images were generated and CLEANed using MIRIAD. Table 1Observational Parameters for SMA TW Hydrae (detected species) | HCN 3–2 | HCO+ 3–2 | H13CO+ 4- -3 | DCN 3–2c | DCO+ 3–2 ---|---|---|---|---|--- Rest Frequency (GHz): | 265.886 | 267.558 | 346.999 | 217.239 | 216.113 Observations | 2005 Mar 04 | 2005 Mar 04 | 2006 Dec 28 | 2006 Apr 28 | 2006 Apr 28 | 2005 Apr 21 | 2005 Apr 21 | | | 2006 Feb 03 | 2005 April 26 | 2005 Apr 26 | | | Antennas used | 7 | 7 | 8 | 8 | 8 Synthesized beam: | $1\farcs 6\times 1\farcs 1$ PA -0.5∘ | $1\farcs 6\times 1\farcs 1$ PA -6.3∘ | $4\farcs 1\times 1\farcs 8$ PA 3.3∘ | $5\farcs 9\times 3\farcs 2$ PA -1.5∘ | $2\farcs 6\times 1\farcs 6$ PA 2.8∘ Channel spacing (km s-1): | 0.23 | 0.23 km | 0.70 | 0.56 | 0.28 R.M.S.a (Jy beam-1): | 0.35 | 0.29 | 0.16 | 0.10 | 0.10 Peak intensityb (Jy) | 4.7 | 2.0 | 0.70 | 0.31 | 0.56 aafootnotetext: SNR limited by the dynamic range. bbfootnotetext: Intensity averaged over the corresponding beam. ccfootnotetext: Only compact configuration data used for DCN observation. Table 2Observational Parameters for SMA TW Hydrae (undetected species) | HDO 31,2–22,1 | o-H2D+ 11,0–11,1 | p-D2H+ 11,0–10,1 ---|---|---|--- Rest Frequency (GHz): | 225.897 | 372.421 | 346.999 Observations: | 2006 April 28 | 2006 Dec 28 | 2005 Feb 17 Antenna used: | 8 | 3 | 4 Synthesized beam: | $5\farcs 7\times 3\farcs 1$ PA -0.6∘ | $4\farcs 7\times 3\farcs 8$ PA 14.5∘ | $3\farcs 3\times 1\farcs 3$ PA 7.5∘ Channel spacing (km s-1): | 0.54 | 0.65 | 0.35 R.M.S. (Jy beam-1): | 0.11 | 1.17 | 7.39 Nmolecule (1 $\sigma$) (cm-2): | $<$2.0 $\times$ 1014 | $<$1.7 $\times$ 1012 | $<$9.0 $\times$ 1014 ## 3\. Spectral Line Modeling Several model approaches have been developed to interpret interferometric molecular line and continuum observations from disks (Dutrey et al., 2007). In brief, the approach of Qi et al. (2004, 2006) works as follows: the kinetic temperature and density structure of the disk is determined by modeling the spectral energy distribution (SED) assuming well mixed gas and dust, with the results confirmed by the resolved (sub)mm continuum images. Then, a grid of models with a range of disk parameters including the outer radius Rout, the disk inclination $i$, position angle P.A. and the turbulent line width vturb are produced and a 2D accelerated Monte Carlo model (Hogerheijde & van der Tak, 2000) is used to calculate the radiative transfer and molecular excitation. The collisional rates are taken from the Leiden Atomic and Molecular Database 333http://www.strw.leidenuniv.nl/$\sim$moldata (Schoier et al. 2005) for non-LTE line radiative transfer calculations. Specifically in the models used in this paper, the rate coefficients for HCO+ and DCO+ in collisions with H2 have been calculated by Flower (1999); the rate coefficients for HCN and DCN in collisions with H2 have been scaled from the rate of HCN-He calculated by Green & Thaddeus (1974). The model parameters are fitted using a $\chi^{2}$ analysis in the $(u,v)$ plane. In the studies by Qi et al. (2004, 2006), the distribution of CO molecules was assumed to follow the H nuclei as derived from the dust density structure and a gas of solar composition. However, molecules in disks do not necessarily share the same distribution as molecular hydrogen. Theoretical models (e.g. Aikawa et al., 1996; Aikawa & Nomura, 2006) predict so-called three-layered structure; most molecules are photo-dissociated in the surface layer of the disk and frozen out in the mid-plane where most of hydrogen resides, with an abundance that peaks in the warm molecular layer at intermediate scale heights. To approximate this more complex behavior, we introduce new molecular distribution parameters for use in spectral line modeling. For first-order analysis in the radial molecular distributions, the column densities are assumed to vary as a power law as a function of radius: $\displaystyle\Sigma_{i}(r)=\Sigma_{i}(10{\rm AU})(\frac{r}{10{\rm AU}})^{p_{i}}$ Here $\Sigma_{i}$ and $p_{i}$ describe column density distribution of a specific species ($i$) rather than disk (hydrogen) column density. In most disk models to date, a single radial power-law is adopted for fitting the hydrogen or dust surface density. By assuming CO follows the distribution of hydrogen column density, Qi et al. (2004, 2006) also find satisfactory power- law fits for multiple CO transtions. When a single power law is insufficient for fitting the radial distribution, then a broken power law with two different indices will be used. Figure 1.— This schematic diagram shows an arbitrary molecular vertical distribution as a function of $\Sigma_{21}$ measured from the disk surface at a certain radius. $\sigma_{m}$ and $\sigma_{s}$ are the midplane and surface boundary parameters for model fitting (see §3 of the text). Figure 2.— Top: The HCO+, DCO+, HCN and DCN J=3–2 spectra at the peak continuum (stellar) position. The fluxes of HCO+ and DCO+ are averaged over the beam of DCO+ J=3–2 ($2\farcs 6\times 1\farcs 6$ PA 2.8∘). The fluxes of HCN and DCN are averaged over the beam of DCN J=3–2 ($5\farcs 9\times 3\farcs 2$ PA -1.5∘). The vertical dotted lines indicate the positions of the fitted VLSR for each molecular transition except for DCN 3-2 where we adopt the VLSR from that of HCN J=3–2. Bottom: Velocity channel maps of the HCO+, DCO+, HCN and DCN J=3–2 emissions toward TW Hydrae. The angular resolutions are $1\farcs 6\times 1\farcs 1$ at PA -6.3∘ for HCO+ J=3–2 and $1\farcs 6\times 1\farcs 1$ at PA -0.5∘ for HCN J=3–2. The cross indicates the continuum (stellar) position. The axes are offsets from the pointing center in arcseconds. The 1$\sigma$ contour steps are 0.4, 0.12, 0.35 and 0.09 Jy beam-1 for HCO+, DCO+, HCN and DCN J=3–2 respectively and the contours start at 2$\sigma$. To calculate the surface density at different radii, the vertical molecular distribution is needed, but this distribution may well vary with distance from the star. However, theoretical models (Aikawa & Nomura, 2006) indicate that the vertical distribution of molecules at different radii is similar as a function of the hydrogen column density measured from the disk surface, $\Sigma_{21}\equiv\Sigma_{\rm H}/(1.59\times 10^{21}~{}{\rm cm}^{-2})$, where the denominator is the conversion factor of the hydrogen column density to $A_{v}$ for the case of interstellar dust. As indicated by Figure 5 of Aikawa & Nomura (2006), the vertical distribution of molecular abundances shows a good correlation with $\Sigma_{21}$. Figure 1 shows a schematic diagram of an arbitrary distribution of a molecule in disk where the x-axis shows $\Sigma_{21}$ and the y-axis shows the normalized molecular fraction. Smaller $\Sigma_{21}$ values (to the left of the plot) approach the surface of the disk, while larger $\Sigma_{21}$ values (to the right) approach the midplane of the disk. For modeling, we make the further simplifying assumption that gaseous molecules exist with a constant abundance in layers between $\sigma_{s}$ and $\sigma_{m}$, the surface and midplane boundaries of $\Sigma_{21}$, respectively, as indicated in Figure 1 by the shaded area. While the vertical distribution of molecules in disks may have a more complex distribution than assumed here, the adopted parameters provide a gross approximation to the vertical location where the species is most abundant. This is adequate for a first description, given the quality of the data available. For example, the model of Aikawa & Nomura (2006) shows two vertical peaks of HCO+, but the secondary peak is an order of magnitude smaller, and provides effectively negligible emission. Also, at least for the nearly face- on disk of TW Hydrae, the uncertainties in vertical distributions do not affect the derived radial distribution. This simple model captures the basic characteristics of three-layered structure predicted by theoretical models. Using the new distribution parameters: $\Sigma_{i}(10{\rm AU})$ and $p_{i}$ (radial), $\sigma_{s}$ and $\sigma_{m}$ (vertical), the radial and vertical distributions of the molecules in disks can be constrained by observation. The best fit model is obtained by minimizing the $\chi^{2}$, the weighted difference between the real and imaginary part of the complex visibility measured at the selected points of the ($u,v$) plane. The $\chi^{2}$ values are computed by the simultaneous fitting of channels covering LSR velocities from 1 to 4 km s-1. Rout and $p_{i}$ are calculated on grids in steps of 5 (AU) and 0.2, respectively. $\log\left(\sigma_{s}\right)$ and $\log\left(\sigma_{m}\right)$ are calculated on grids in steps of 0.2 within the range from $-$2 to 2. $\Sigma_{i}(10{\rm AU})$ and $i$ are found with each pair of radial and vertical distribution parameters by minimizing $\chi^{2}$. The systemic velocity VLSR is fit separately since it is not correlated with those distribution parameters. For each fit parameter, the 1$\sigma$ uncertainties are estimated as $\chi^{2}_{1\sigma}=\chi^{2}_{m}+\sqrt{2n}$, where $n$ is the number of degrees of freedom and $\chi^{2}_{m}$ is the $\chi^{2}$ value of best-fit model, as in Isella et al. (2007). We tested values for the turbulent velocity in the range from 0.0 to 0.15 km s-1 and found that exact value does not have a significant impact, in part because of the coarse spectral resolution of the data. Therefore, we fixed the turbulent velocity at an intermediate value, 0.08 km s-1. ## 4\. Results and Discussion Figure 3.— The solid and dotted contours show the temperature and density profiles from the TW Hya model of Calvet et al. (2002). The blue and red lines confine the locations of HCO+ and HCN in the model (see text). Figure 4.— Radial distribution of molecular column densities and DCO+/HCO+ ratio for the best-fit models for TW Hydrae. Solid lines depict single power-law fits, while the dashed lines are for DCO+ Model 3 where two power-law indices are used. Figure 2 shows the spectra of HCO+, DCO+, HCN and DCN at the stellar (peak continuum) position and their velocity channel maps. Table 3 summarizes the power-law fitting results on HCO+, DCO+ and HCN. The DCN and H13CO+ lines are too weak for a $\chi^{2}$ analysis of their distribution parameters, and so only the ratios of DCN/HCN and HCO+/H13CO+ are fit. Figure 3 shows the density and temperature contours of the disk model (adopted from Calvet et al., 2002) and the locations of HCO+ and HCN derived from the model fitting procedure. Figure 4 shows the radial distribution of the molecular column densities of the best fit models for HCO+, DCO+ and HCN. Not surprisingly, we found that the radial distributions are better constrained than the vertical ones: the local minimum of $\chi^{2}$ for different grids of vertical parameters are within the noise limit, i.e. the vertical parameters are the least well determined, due in part to the face-on nature of the TW Hydrae disk. We therefore treat the best-fit vertical results as fixed, and investigate the uncertainty of other parameters. Table 3Fitting results Parameters | HCO+ | DCO+a | DCO+b | HCN ---|---|---|---|--- Stellar Mass: M∗(M⊙) | 0.6 | 0.6 | 0.6 | 0.6 Inclination: $i$(deg) | 6.8$\pm$0.3 | 7.4$\pm$0.6 | 7.4 | 6.6$\pm$0.8 Systemic velocity: VLSR(km s-1) | 2.88$\pm$0.05 | 2.94$\pm$0.06 | 2.9 4 | 2.73$\pm$0.06 Position angle: P.A.(deg) | -27.4 | -27.4 | -27.4 | -27.4 Turbulent line width: vturb(km s-1) | 0.08 | 0.08 | 0.08 | 0.08 Outer radius: Rout(AU) | 200 $\pm$ 10 | 90 $\pm$ 5 | 160 | 100 $\pm$ 10 Column Density at 10AU: $\Sigma(10{\rm AU})$ (cm-2) | 3.8 $\pm$0.5 $\times$ 1015 | 1.9$\pm$0.2 $\times$ 1010 | 4.8 $\times$ 106 | 2.4$\pm$0.4 $\times$ 1014 Radial power index: p | -2.9 $\pm$ 0.3 | 2.4 $\pm$ 0.8 | 7,-6 c cfootnotemark: | -1.0 $\pm$ 1.2 Vertical Parameters: $\sigma_{s}$,$\sigma_{m}$ | 0.1, 10 | 0.1, 10 | 0.1,10 | 0.3, 30 Minimum $\chi^{2}$: | 329588 | 575347 | 575347 | 297198 Reduced $\chi^{2}$: $\chi^{2}_{r}$ | 1.21 | 1.95 | 1.95 | 1.35 aafootnotetext: DCO+ Model 2. bbfootnotetext: DCO+ Model 3, no error estimation. ccfootnotetext: DCO+ turning point at radius 70 AU. ### 4.1. HCO+ and DCO+ The first detection of DCO+ in the disk around TW Hydrae was obtained by van Dishoeck et al. (2003) with the James Clerk Maxwell Telescope (JCMT), who reported a beam-averaged DCO+/HCO+ abundance ratio of 0.035. Our spatially resolved observations of the DCO+ and HCO+ J=3–2 emission from the disk suggest a more complex chemical situation. Figure 5.— Top: Velocity channel maps of the HCO+ J=3–2 emission toward TW Hydrae. The angular resolution is $1\farcs 6\times 1\farcs 1$ at PA -6.3∘. The cross indicates the continuum (stellar) position. The axes are offsets from the pointing center in arcseconds. The 1$\sigma$ contour step is 0.4 Jy beam-1 and the contours start at 2$\sigma$. Middle: channel map of the best-fit model with the same contour levels. Bottom: difference between the best-fit model and data on the same contour scale. Figure 5 shows the channel maps of HCO+ J=3–2, together with the best-fit model and the data-model residuals. Table 3 lists the best-fit model parameters. The $\chi^{2}$ surface for the radial power index p${}_{\rm HCO^{+}}$ and the outer radius Rout is shown in the top panel of Figure 6. The 1$\sigma$ contour confines Rout to 200$\pm$10 AU and p${}_{\rm HCO^{+}}$ to $-$2.9$\pm$0.3. The H13CO+ J=4–3 line was also detected, but the emission is not strong enough to constrain its distribution. Assuming that H13CO+ has the same distribution as HCO+, then fitting the ratio of HCO+/H13CO+ to match the intensity of the H13CO+ emission indicates an HCO+/H13CO+ ratio of 100$\pm$25\. This value is consistent with the nominal solar system value of 89. Figure 7 presents the best-fit model spectra of H13CO+ J=4–3 overlayed on the SMA data. Figure 6.— Iso-$\chi^{2}$ surfaces of (Rout,pi) for HCO+, DCO+ (as in Model 2) and HCN. Contours correspond to the 1 to 6 $\sigma$ errors. For DCO+, the index values larger than 3 at around 90 AU indicate the ratios of DCO+/HCO+ larger than 1, so the $\chi^{2}$ surface is not calculated beyond that. Figure 7.— The beam-averaged H13CO+ J=4–3 spectra at the continuum (ste llar) position. The SMA data are presented by the solid histogram, and the simulated model by the dashed histogram. The vertical dotted line indicates the position of the fitted VLSR for HCO+ J=3–2. The vertical distribution of DCO+ is even less well constrained than that of HCO+, and we choose to adopt the same values of $\sigma_{s}$ and $\sigma_{m}$ for both HCO+ and DCO+ (as indicated in Table 3). Because of the nearly face- on viewing geometry for the disk, the fitting of radial distribution power-law index pi is not affected significantly by the uncertainties in vertical structure, but only the value of $\Sigma_{i}$(10AU) needs to be adjusted. Figure 8.— Models of DCO+ with different distributions of radial column densities. Examination of the channel maps shows upon inspection that the radial distributions of DCO+ and HCO+ are different. To demonstrate how differences in the radial distribution affect the resulting images, Figure 8 presents three models of DCO+ radial distribution and Figure 9 shows the corresponding simulated channel maps of the DCO+ J=3–2 line. Figure 9.— DCO+ J=3–2 channel maps toward TW Hydrae and the simulated model distributions depicted in Figure 8. Model 1 assumes that the DCO+ distribution follows the best-fit model of HCO+. The minimum $\chi^{2}$ is determined to be 575396 and the corresponding DCO+/HCO+ is found to be 4.7 $\times$ 10-2. Comparing the simulated maps from Model 1 with the data in Figure 9 shows a distinct difference in that the double peaked nature of the central channel in the data is more obvious than in this model. The contrast of the contour levels between the central channel and the adjacent channels are also smaller in the data than in this model. Because the emission of the central channel mostly originates at large disk radii, these differences suggest that the DCO+ emission arising from the outer regions of the disk is stronger than predicted by this model. The slope of radial DCO+ distribution does not decrease as steeply as does HCO+. In other words, the D/H ratio must increase with increasing radius. Model 2 shows the best-fit result for the radial distribution of DCO+ assuming a single power index. As shown in Figure 4, the radial distribution of DCO+ is strikingly different from that of HCO+, with a positive power index of 2.4 and a smaller but better constrained Rout 90$\pm$5 AU. The contours of the iso-$\chi^{2}$ surface for DCO+ in the middle panel of Figure 6 indicate that the uncertainty of the radial power index is very large but that the index is still larger than 1.6 within the 1$\sigma$ error, much larger than $-$2.9 found for HCO+. The simulated DCO+ channel maps for Model 2 shown in Figure 9 are an improvement over Model 1 in matching the data. However in this model, Rout ($\sim$ 90 AU) is much smaller than the disk radius observed with HCO+ and CO ($\sim$ 200 AU) and DCO+ increases with radius but then disappears sharply at Rout$\sim$90 AU, as a step function, which is hard to understand. Comparison with the data shows that there are fewer complete contours around the double peaks in the central channel, which suggests that the DCO+ emission is maximum at an intermediate radius rather than at the edge. For Model 3, instead of using a single power index (p) to fit for the radial distribution of DCO+ column density N(DCO+), the fit uses two power indices (p1 and p2) and a turning point (Rt) where the power law index changes from p1 to p2, i.e. the location of the peak of N(DCO+). The parameters p1, p2 and Rt are searched within limited grids to minimize $\chi^{2}$. In this model, N(DCO+) is found to increase with radius out to $\sim 70$ AU, and then to decrease. Table 3 presents the best-fit parameters; no error estimation is provided due to the computation difficulties. Figure 9 shows the best-fit model images. Comparison of Model 3 with the data shows a visual improvement over Model 2 in the central channel, although the $\chi^{2}$ value of the two models are not distinguishable. We thus cannot clearly discriminate with the $\chi^{2}$ statistic if there is indeed a peak with radius for N(DCO+) as in Model 3, or if DCO+ increases with radius and disappears suddenly around 90 AU as in Model 2. Even with this ambiguity, however, both models imply that the D/H ratios change by at least an order of magnitude (0.01 to 0.1) from radii $<$30 AU to $>$70 AU and that there is a rapid falloff of N(DCO+) at radii larger than 90 AU. Because the emission in the central channel comes from the outer part of the disk (Keplerian rotation) projected along the line-of-sight with a very small inclination of around 7 degrees for TW Hydrae, the central velocity channel is most important for constraining the radial distribution. Based on the difference in the central velocity channel map between the models, we believe Model 3 is the more plausible description of the radial distribution of DCO+. Figure 10 shows the channel maps of the DCO+ J=3–2 emission, together with those of Model 3 and the data-model residuals. Figure 10.— Top: Velocity channel map of the DCO+ J=3–2 toward TW Hydrae. The angular resolution is $2\farcs 6\times 1\farcs 6$ at PA 2.8∘. The cross indicates the continuum (stellar) position. The axes are offsets from the pointing center in arcseconds. The 1$\sigma$ contour step is 0.12 Jy beam-1 and the contours start at 2$\sigma$. Middle: channel map of Model 3 with the same contour levels. Bottom: difference between Model 3 and data on the same contour scale. Observations of deuterated molecular ions and the level of deuterium fractionation have been used to estimate the ionization degree in molecular clouds, and a similar analysis can be applied to circumstellar disks. If we consider only the ionization balance determined by HCO+, H${}_{3}^{+}$, DCO+ and electrons in steady state as shown in Equation 14 of Caselli (2002), the electron fractional abundance can be derived to be around 10-7. Of course, this value is only valid in the intermediate layer where HCO+ is abundant and multiply deuterated H${}_{3}^{+}$ is less abundant than HCO+. Several important complications are also neglected in this analysis, including the presence of other atomic and molecular ions, neutral species besides CO which destroy H2D+, and negatively charged dust grains and refractory metals. Still, accurate measurements of DCO+ and HCO+ are the first steps toward an understanding of the ionization fraction in the disk. The increase of D/H ratios from inner to outer disk is generally consistent with the current theoretical models of the gas-deuterium fractionation processes that consider the effect of cold temperature. But the quick disappearance of DCO+ at radii beyond 90 AU (comparing with Rout around 200 AU for CO and HCO+) is puzzling, since DCO+ is expected to be abundant and observable in the cold outer region of the disk where HCO+ is still available. More theoretical work is needed to explain the disappearance of DCO+ in the outer part of the disk. ### 4.2. HCN and DCN Figure 11 shows the HCN J=3–2 channel maps, together with the best fit model and residuals. Table 3 lists the best-fit model parameters, and Figure 4 shows the radial distribution of column density derived. The $\chi^{2}$ surface shown in the bottom panel of Figure 6 indicates Rout 100$\pm$10 AU and pHCN $-$1.0$\pm$1.2. The radial power index is poorly constrained probably due to more complex distributions for HCN. A detailed comparison of the molecular distributions of HCN and CN will be presented elsewhere. Figure 11.— Top: Velocity channel map of the HCN J=3–2 toward TW Hydrae. The angular resolution is $1\farcs 6\times 1\farcs 1$ at PA -0.5∘. The cross indicates the continuum (stellar) position. The axes are offsets from the pointing center in arcseconds. The 1$\sigma$ contour step is 0.35 Jy beam-1 and the contours start at 2$\sigma$. Middle: channel map of the best-fit model with the same contour levels. Bottom: difference between the best-fit model and data on the same contour scale. Although the best fit vertical parameters seem to indicate that HCN ($\sigma_{m,\rm HCN}=30$) is found much deeper toward the midplane than is HCO+ ($\sigma_{m,\rm HCO^{+}}=10$), we emphasize that we are not able to accurately constrain the vertical distributions from the present data. This ambiguity strongly affects the column density of HCN (i.e. $\Sigma_{\rm HCN}$(10AU)) needed to fit the data (not the power index pHCN of radial distribution). A worse fit to the data ($\chi^{2}$ larger by 3$\sigma$ over the best-fit model) can be obtained by assuming that the vertical distributions of HCN and HCO+ are the same, but the HCN column density is 1.5 times larger than that needed for the best fit model due to higher density near the midplane. The DCN J=3–2 transition is detected at a signal-to-noise ratio of 3 near the fitted HCN VLSR of 2.73 km s-1 (Figure 2). While this signal-to-noise ratio is not high, the significance of the detection is further supported by the channel maps (Figure 12 upper panel) where the velocity gradient along the disk position angle of $\sim$ $-$30∘ is consistent with that seen in CO J=2–1 and J=3–2 (Qi et al., 2004) and the other molecular lines presented in this paper. Since the DCN 3–2 emission is weak, we are not able to fit for the molecular distribution and so make the simplifying assumption that the distribution of DCN follows that of HCN. As with H13CO+, we fit the DCN/HCN ratio over the whole disk and determine the DCN/HCN ratio to be 1.7$\pm$0.5 $\times$ 10-2. To again emphasize the impact of the assumed vertical distribution on the derived column densities, the DCN/HCN ratio could be as high as 5 $\times$ 10-2 if HCN and DCN are distributed vertically over the same region as is HCO+. Figure 12.— Top left: DCN J=3–2 velocity channel maps (red: 2.84 km s-1, blue: 2.28 km s-1) from TW Hydrae, overlaid on the 217 GHz dust continuum map (gray scale). The cross indicates the position of the continuum peak. Top right: the simulated model for DCN J=3–2. The 1$\sigma$ contour step is 0.09 Jy beam-1 and the contours start with 2$\sigma$. Bottom: the beam-averaged DCN 3–2 spectra at the continuum (stellar) position. The SMA data are presented by the solid histogram, and the simulated model by the dashed histogram. The vertical dotted line indicates the position of the fitted VLSR for HCN J=3–2. Highly fractionated DCN/HCN ratios have been measured in comets. In the coma of comet Hale-Bopp, for example, Meier et al. (1998) reported the ratio to be around 2.3 $\times$ 10-3, but higher DCN/HCN ratios – (D/H)HCN,jet $\approx$0.025 are detected from the pristine material sublimed from icy grains ejected in jets from the nucleus which may present a more representative sampling of cometary ices that have not experienced significant thermal processing (Blake et al., 1999). Such ratios are consistent with those found here in the TW Hydrae disk, indicating that high D/H ratios in comets could originate from material in the outer regions of disks where in situ deuterium fractionation is ongoing, rather than requiring an inheritance from interstellar material. ### 4.3. Upper limits for H2D+, D2H+ and HDO In the disk midplane, H2 is expected to be gaseous and the molecular ion formed by the cosmic ray ionization of H2, H${}_{3}^{+}$, is expected to be the most abundant ion. Unfortunately, H${}_{3}^{+}$ is only detectable in cold gas via infrared absorption. In its deuterated forms, however, H2D+ and even D2H+ are expected to be abundant in the cold, dense gas (Ceccarelli & Dominik, 2005). The ground-state transition of ortho-H2D+ was first detected in a young stellar object (NGC 1333 IRAS 4A) by Stark et al. (1999) and in a prestellar core (L1544) by Caselli et al. (2003). Both H2D+ and D2H+ have been detected toward another pre-stellar core 16293E via their ground-state submm rotational lines (Vastel et al., 2004). The inclusion of multiply deuterated H${}_{3}^{+}$ in chemical models leads to predictions of higher values of the D/H ratio in cold, high-density regions of the interstellar medium. Similarly, in the dense, cold disk midplane, CO is depleted, and high abundances of H2D+ and D2H+ are expected. For this reason, Ceccarelli et al. (2004) searched for the ground-state transition of ortho-H2D+ with the Caltech Submillimeter Observatory (CSO), and reported 3.2$\sigma$ and 4.7$\sigma$ detections toward TW Hydrae and DM Tauri, respectively. With the 400 GHz and 690 GHz receiver- equipped SMA antennas, we searched for the 372 GHz ortho-H2D+ 11,1–11,0 and 690 GHz para-D2H+ 110–101 lines toward TW Hydrae. No significant emission signals were detected. Here we discuss the upper limits and their implications. Our 3-antenna SMA observations give a 1$\sigma$ upper limit for the ortho-H2D+ 11,1–11,0 line emission of 1.2 Jy beam-1 km s-1 with a $4\farcs 7\times 3\farcs 8$ synthesized beam. To compare the result with single dish data, the extent of the source emission must be known. Since H2D+ was observed along with the H13CO+ 4–3 line in a dual-receiver observation on 28 December 2006 and H13CO+ 4–3 has been clearly detected at JCMT (van Dishoeck et al., 2003), we can compare the intensities of these two lines between the SMA data and the single dish observations to constrain the emitting areas. The H13CO+ 4–3 line was detected at JCMT with an integrated intensity of 0.07 K km s-1 in a 13′′ beam (van Dishoeck et al., 2003); while for the SMA the integrated intensity of this line is determined to be 0.61 Jy beam-1 km s-1 with a beam of $4\farcs 1\times 1\farcs 8$. If the extent of the emission of H13CO+ is similar to that of H2D+, our interferometric upper limit of Jy beam-1 km s-1 for H2D+ (considering the small change of the beam sizes) becomes 0.09 K km s-1 (1$\sigma$) or 0.27 K km s-1 (3 $\sigma$) upper limits, which is consistent with 2$\sigma$ upper limits of about 0.2 K km s-1 by the JCMT (Thi et al., 2004) but slightly lower than the 3.2$\sigma$ detection of 0.39 K km s-1 by Ceccarelli et al. (2004). We estimate the 1$\sigma$ upper limit of the ortho-H2D+ column density to be 1.7 $\times$ 1012 cm-2 according to the Equation 4 of Vastel et al. (2004), which is slightly less than the 2$\sigma$ upper limit estimate of 4.4 $\times$ 1012 cm-2 by Thi et al. (2004) since the SMA data had a smaller noise level. Of course this analysis assumes the extent of H2D+ is similar to that of H13CO+. With the deployment of more 400 GHz receivers on the SMA and further H2D+ observations, it should be possible to provide rather better constraints on the H2D+ abundance in the disk. We estimate the 1$\sigma$ upper limit for para-D2H+ 110–101 to be 5.35 Jy beam-1 km s-1 with a beam of $3\farcs 3\times 1\farcs 3$. The 1$\sigma$ upper limit to the para-D2H+ column density is estimated to be 9.0 $\times$ 1014 cm-2. This is less constrained than H2D+ due to the relatively poor system sensitivity at 690 GHz. For HDO 31,2–22,1, the 1$\sigma$ upper limit is 0.10 Jy beam-1 km s-1. Assuming an excitation temperature of 30 K, the upper limit for the HDO column density is 2.0 $\times$ 1014 cm-2. Since the lower state energy level of this line is nearly 160 K, it must originate from warm regions of the disk which are quite distinct from the cold layers where HDO ground state transition absorption, as found in DM Tauri (Ceccarelli et al., 2005), must arise – although we note that the detection of the HDO absorption line in DM Tauri has been disputed by Guilloteau et al. (2006). ## 5\. Summary and Conclusions Observations of deuterated species in circumstellar disks are important to understand the origin of primitive solar system bodies in that they can directly constrain the deuterium fractionation in the outer regions where cometary ices are likely formed. Spatially resolved observations of the D/H ratios in disks enable the comparison of the fractionation measured in comets such as Hale-Bopp (Blake et al., 1999) with the specific conditions at each disk radius. We have presented the first images of DCO+ and DCN emission from the disk around a classical T Tauri star, TW Hydrae, along with images of the HCN and HCO+ J=3–2 lines. The observations of deuterium fractionation serve as a clear measure of the importance of low-temperature gas-phase deuterium fractionation processes. These observations strongly support the proposed link among high gas densities, cold temperature and enhanced deuterium fractionation. Detailed chemical models are still needed to explain how DCO+ disappears from the outer part of the disk. The similarity of the D/H ratios in cold clouds, disks and pristine cometary material has been used to argue that the gas spends most of its lifetime at low temperatures and is incorporated into the disks before the envelope is heated, i.e. before the Class I stage. By combining self-consistent physical models and 2D radiative transfer codes to interpret high spatial resolution millimeter-wave molecular images, we are only now beginning to investigate the radial and vertical distributions of molecules in disks. The radial distribution of DCO+ in the disk of TW Hydrae indicates that in situ deuterium fractionation is ongoing. The molecular evolution within disks must therefore be considered in the investigation of the origin of primitive solar system bodies. We have obtained less stringent constraints on the vertical distributions of molecules in the disk of TW Hydrae. To address the ambiguity present in the analysis of single objects, data from a robust sample of disks is needed, in particular one that covers a range of disk inclinations. More sensitive observations are also needed for the rare isotopologues H13CN, H13CO+ and, of course, DCN, to understand the radial and vertical gradient of deuterated species in these disks. In the future, observations of DCN and other species with the Atacama Large Millimeter Array will provide further insight into the chemical state of protoplanetary disks. Partial support for this work comes from NASA Origins of Solar Systems Grant NNG05GI81G. M.R.H is supported by a VIDI grant from the Netherlands Organization for Scientific Research. C.Q. acknowledges Paola Caselli for her help and useful suggestions. We thank the referee for very useful comments. ## References * Aikawa & Herbst (1999) Aikawa, Y. & Herbst, E. 1999, ApJ, 526, 314 * Aikawa et al. (1996) Aikawa, Y., Miyama, S. M., Nakano, T., & Umebayashi, T. 1996, ApJ, 467, 684 * Aikawa & Nomura (2006) Aikawa, Y. & Nomura, H. 2006, ApJ, 642, 1152 * Aikawa et al. (2002) Aikawa, Y., van Zadelhoff, G. J., van Dishoeck, E. F., & Herbst, E. 2002, A&A, 386, 622 * Beckwith & Sargent (1993) Beckwith, S. V. W. & Sargent, A. I. 1993, ApJ, 402, 280 * Bergin et al. (2007) Bergin, E. A., Aikawa, Y., Blake, G. A., & van Dishoeck, E. F. 2007, in Protostars and Planets V, ed. B. Reipurth, D. Jewitt, & K. Keil, 751–766 * Blake et al. (1999) Blake, G. A., Qi, C., Hogerheijde, M. R., Gurwell, M. A., & Muhleman, D. O. 1999, Nature, 398, 213 * Calvet et al. (2002) Calvet, N., D’Alessio, P., Hartmann, L., Wilner, D., Walsh, A., & Sitko, M. 2002, ApJ, 568, 1008 * Caselli (2002) Caselli, P. 2002, Planet. Space Sci., 50, 1133 * Caselli et al. (2003) Caselli, P., van der Tak, F. F. S., Ceccarelli, C., & Bacmann, A. 2003, A&A, 403, L37 * Ceccarelli & Dominik (2005) Ceccarelli, C. & Dominik, C. 2005, A&A, 440, 583 * Ceccarelli et al. (2005) Ceccarelli, C., Dominik, C., Caux, E., Lefloch, B., & Caselli, P. 2005, ApJ, 631, L81 * Ceccarelli et al. (2004) Ceccarelli, C., Dominik, C., Lefloch, B., Caselli, P., & Caux, E. 2004, ApJ, 607, L51 * Dutrey et al. (2007) Dutrey, A., Guilloteau, S., & Ho, P. 2007, in Protostars and Planets V, ed. B. Reipurth, D. Jewitt, & K. Keil, 495–506 * Flower (1999) Flower, D. R. 1999, MNRAS, 305, 651 * Green & Thaddeus (1974) Green, S., & Thaddeus, P. 1974, ApJ, 191, 653 * Guilloteau et al. (2006) Guilloteau, S., Piétu, V., Dutrey, A., & Guélin, M. 2006, A&A, 448, L5 * Ho et al. (2004) Ho, P. T. P., Moran, J. M., & Lo, K. Y. 2004, ApJ, 616, L1 * Hogerheijde & van der Tak (2000) Hogerheijde, M. R. & van der Tak, F. F. S. 2000, A&A, 362, 697 * Isella et al. (2007) Isella, A., Testi, L., Natta, A., Neri, R., Wilner, D., & Qi, C. 2007, A&A, 469, 213 * Meier et al. (1998) Meier, R., Owen, T. C., Jewitt, D. C., Matthews, H. E., Senay, M., Biver, N., Bockelee-Morvan, D., Crovisier, J., & Gautier, D. 1998, Science, 279, 1707 * Qi et al. (2004) Qi, C., Ho, P. T. P., Wilner, D. J., Takakuwa, S., Hirano, N., Ohashi, N., Bourke, T. L., Zhang, Q., Blake, G. A., Hogerheijde, M., Saito, M., Choi, M., & Yang, J. 2004, ApJ, 616, L11 * Qi et al. (2006) Qi, C., Wilner, D. J., Calvet, N., Bourke, T. L., Blake, G. A., Hogerheijde, M. R., Ho, P. T. P., & Bergin, E. 2006, ApJ, 636, L157 * Schöier et al. (2005) Schöier, F. L., van der Tak, F. F. S., van Dishoeck, E. F., & Black, J. H. 2005, A&A, 432, 369 * Setiawan et al. (2008) Setiawan, J., Henning, T., Launhardt, R., Müller, A., Weise, P., & M., K. 2008, Nature, 451, 38 * Stark et al. (1999) Stark, R., van der Tak, F. F. S., & van Dishoeck, E. F. 1999, ApJ, 521, L67 * Thi et al. (2004) Thi, W.-F., van Zadelhoff, G.-J., & van Dishoeck, E. F. 2004, A&A, 425, 955 * van Dishoeck et al. (2003) van Dishoeck, E. F., Thi, W.-F., & van Zadelhoff, G.-J. 2003, A&A, 400, L1 * Vastel et al. (2004) Vastel, C., Phillips, T. G., & Yoshida, H. 2004, ApJ, 606, L127 * Willacy (2007) Willacy, K. 2007, ApJ, 660, 441

arxiv-papers

2024-09-04T02:48:54.418329

0803.2762

On the string solution in the SUSY - Skyrme model Pham Thuc Tuyen 111Email: tuyenpt@coltech.vnu.vn Department of Theoretical Physics, Hanoi University of Science, 334 Nguyen Trai, Thanh Xuan, Ha Noi, Viet Nam. Do Quoc Tuan 222Email: do.tocxoan@gmail.com. Associate address after February 2008: Department of Computing Physics, Hanoi University of Science, 334 Nguyen Trai, Thanh Xuan, Ha Noi Department of Theoretical Physics, Hanoi University of Science, 334 Nguyen Trai, Thanh Xuan, Ha Noi, Viet Nam. Abstract: In this paper, we have found the string solution in the SUSY Skyrme model. Moreover, the mechanics of decay of SUSY - string was discussed. Keywords : String, SUSY, Skyrme model. ## 1 Introduction String - like solution firstly obtained [2] from equaton of motion of the Skyrme model with a pion’s mass term by A. Jackson. This string-like solution may be closely related to QCD string [2]. This string solution is unstable and it may decay by emitting pions [2]. During the decay a baryon current flows along the string, producing half a baryon and half an antibaryon. Long strings can decay via many different decay models, some producing baryon-antibaryon pairs [2]. Recently, M. Nitta and M. Skiiki constructed non-topological string solutions with $U(1)$ Noether charge in the Skyrme model with a pion mass term. And they also showed that this string were not stabilized by $U(1)$ rotation and decay into baryon-antibaryon pairs or mesons in the same way as the string without the charge. They showed that a rotating confugutation would reduce its rotational energy by emmiting pions. When this string becomes longer than $\pi/{\hat{m}_{\pi}}$, it will decay by emitting pions [3]. In this paper we want to connect idea of the string [2] to the supersymmetric skyrme model proposed by E. A. Bershoeff at el. [4]. This way may give us the SUSY string solution (superstring) and a mechanics of decay squark-antisquark, baryon-antibaryons, etc. ## 2 The SUSY string solution Let us consider the lagrangian with a mass of pion [8] ${\cal L}=-\frac{{F_{\pi}^{2}}}{{16}}Tr\left({\partial_{\mu}U^{\dagger}\partial^{\mu}U}\right)+\frac{1}{{32e^{2}}}Tr\left({\left[{U^{\dagger}\partial_{\mu}U,U^{\dagger}\partial_{\nu}U}\right]^{2}}\right)+\frac{1}{8}m_{\pi}^{2}F_{\pi}^{2}Tr\left[{1-U}\right].$ (1) In terms of complex scalars $A_{i}$ [4], this equation can be rewritten as ${\cal L}={\cal L}_{susy}+\frac{1}{8}m_{\pi}^{2}F_{\pi}^{2}\left[{\bar{A}_{1}+A_{1}-2}\right].$ (2) Now, let us consider the rotating soliton developed by M. Nitta at el. [3] from the original soliton constructed by A. Jackson [2] $U=\left[{\begin{array}[]{*{20}c}{\cos f\left(r\right)}&{i\sin f\left(r\right)e^{-i\left({\theta+\alpha\left(t\right)}\right)}}\\\ {i\sin f\left(r\right)e^{i\left({\theta+\alpha\left(t\right)}\right)}}&{\cos f\left(r\right)}\\\ \end{array}}\right]$ (3) $\to A_{1}=\cos f\left(r\right);A_{2}=i\sin f\left(r\right)e^{i\left({\theta+\alpha\left(t\right)}\right)},$ (4) in the cylindrical coordinate system with the metric $ds^{2}=-dt^{2}+dz^{2}+dr^{2}+r^{2}d\theta^{2}.$ (5) Substituting this solution into Lagrangian (2), we obtain the string tension (see more in the appendix) ${\cal E}=\int{4\pi r^{2}}dr\left\\{{\frac{{-f_{\pi}^{2}}}{8}\left[{\left({-\dot{\alpha}^{2}+\frac{1}{{r^{2}}}}\right)\sin^{2}f+f^{\prime 2}}\right]+}\right.$ $+\frac{1}{{8e^{2}}}\left[{\left({a-b}\right)\left({\left({\dot{\alpha}^{4}-2\frac{{\dot{\alpha}^{2}}}{{r^{2}}}+\frac{1}{{r^{4}}}}\right)\sin^{4}f+2f^{\prime 2}\sin^{2}f\left({\dot{\alpha}^{2}+\frac{1}{{r^{2}}}}\right)+f^{\prime 4}}\right)+}\right.$ $\left.{\left.{+b\left({\left({\dot{\alpha}^{4}+\ddot{\alpha}^{2}+1}\right)\sin^{2}f+f^{\prime\prime 2}+f^{\prime 4}}\right)}\right]+\frac{1}{4}m_{\pi}^{2}f_{\pi}^{2}\left({1-\cos f}\right)}\right\\}.$ (6) Setting the dimensionless variable $\rho=f_{\pi}er\equiv\gamma r$, and taking the variations of $f\left(\rho\right)$ in the string tension $\delta_{f}{\cal E}=0$, we have the Euler-Lagrange equation $\frac{\partial}{{\partial\rho}}\frac{{\delta{\cal E}}}{{\delta f^{\prime}}}-\frac{{\delta{\cal E}}}{{\delta f}}=0$ $\to f^{\prime\prime}\left[{-2\rho^{2}+4\rho^{2}(a-b)\sin^{2}f\left({\frac{{\dot{\alpha}^{2}}}{{\gamma^{2}}}+\frac{1}{{\rho^{2}}}}\right)+12\rho^{2}af^{\prime 2}}\right]+$ $+f^{\prime 3}\left[{8\rho a-2\sin 2f\left({\frac{{\dot{\alpha}^{2}}}{{\gamma^{2}}}+\frac{1}{{\rho^{2}}}}\right)}\right]+4f^{\prime 2}\rho^{2}(a-b)\sin 2f\left({\frac{{\dot{\alpha}^{2}}}{{\gamma^{2}}}+\frac{1}{{\rho^{2}}}}\right)+$ $+f^{\prime}\left[{-4\rho+8\rho\left({a-b}\right)\sin^{2}f\left({\frac{{\dot{\alpha}^{2}}}{{\gamma^{2}}}+\frac{1}{{\rho^{2}}}}\right)-2\sqrt{\rho}\left({a-b}\right)\sin^{2}f-}\right.$ $\left.{-2\rho^{2}\left({a-b}\right)\left({\frac{{\dot{\alpha}^{4}}}{{\gamma^{4}}}-\frac{{2\dot{\alpha}^{2}}}{{\gamma^{2}\rho^{2}}}+\frac{1}{{\rho^{4}}}}\right)\sin 2f\sin^{2}f}\right]+$ $+\rho^{2}\left[\frac{-b}{{\gamma^{4}}}\left({\ddot{\alpha}^{2}+\dot{\alpha}^{4}+1}\right)+{\frac{{\dot{\alpha}^{2}}}{{\gamma^{2}}}+\frac{1}{{\rho^{2}}}}\right]\sin 2f-2\frac{{m_{\pi}^{2}}}{{\gamma^{2}}}\rho^{2}\sin f=0.$ (7) For the finiteness and regularity conditions of the string tension, one requires $f\left(0\right)=n\pi;f\left(\infty\right)=0,$ (8) where $n$ is a positive integer. This field equation can be solved numerically with two boundary conditions (8). ### 2.1 the case of $b=0$ In the configuration of rotating soliton $\alpha\left(t\right)$ is angular rotation of soliton in time [2, 3] thus the quantity $\hat{\omega}$ defined as $\dot{\alpha}=\frac{\omega}{\gamma}\equiv\hat{\omega}$ will be understood as an angular velocity of rotating soliton. To obtain the asymptotic form of the profile $f\left(\rho\right)$ as $\rho\to\infty$ we need linearize the field equation (7). Setting $f=\delta f$, we get $\rho^{2}\delta f^{\prime\prime}+2\rho\delta f^{\prime}+\left[{1-\rho^{2}m^{2}}\right]\delta f-=0,$ (9) where $m^{2}=\left({\frac{{m_{\pi}^{2}}}{{\gamma^{2}}}-\hat{\omega}^{2}}\right)$ . This is the Bessel equation, to obtain solutions of Bessel function, we require $0<\hat{\omega}<\hat{m}_{\pi},$ (10) where $\hat{m}_{\pi}=\frac{{m_{\pi}}}{\gamma}$ . Let us mention in ref. [3], the condition for angular velocity was obtained as $0<\hat{\omega}<\frac{{\hat{m}_{\pi}}}{{\sqrt{2}}}$ and they shown the mechanics of emitting pions of string when $\hat{\omega}$ increases over the critical value $\hat{\omega}_{+}=\hat{m}_{\pi}/\sqrt{2}$. However, in SUSY case, our critical value is $\hat{\omega}_{+}^{susy}=\hat{m}_{\pi}=\sqrt{2}\hat{\omega}_{+}$. In cases of critical values are larger than this critical value, the SUSY - String solution can be emitted into baryons - sbaryons pairs and pions - spions. ### 2.2 the case of $b\neq 0$ For this case, we have the equation $\delta f^{\prime\prime}+\frac{2}{\rho}\delta f^{\prime}+\frac{1}{{\rho^{2}}}\delta f+\left[{-\frac{b}{{\gamma^{2}}}\left({\frac{{d\hat{\omega}}}{{dt}}}\right)^{2}-b\hat{\omega}^{4}+\hat{\omega}^{2}-\frac{b}{{\gamma^{4}}}-\hat{m}_{\pi}^{2}}\right]\delta f=0.$ (11) We now consider an angular velocity of soliton is constant, means $\frac{{d\omega}}{{dt}}=0$. Similarly to a case of $b=0$, we require $\left[-b\hat{\omega}^{4}+\hat{\omega}^{2}-c\right]>0,$ (12) where $c=\frac{b}{{\gamma^{4}}}+\hat{m}_{\pi}^{2}$. Following conditions for $b$ and $\omega$ are given as $-\gamma^{2}m_{\pi}^{2}<b<0\to 0<\omega<\sqrt{x_{2}},$ (13) $0<b<\frac{{\gamma^{2}\left({-1+\sqrt{1+m_{\pi}^{2}}}\right)}}{2}\to\sqrt{x_{1}}<\omega<\sqrt{x_{2}},$ (14) where $x_{1}=\frac{{-1-\sqrt{1-4bc}}}{{-2b}},x_{2}=\frac{{-1+\sqrt{1-4bc}}}{{-2b}}$. Eqs (13), (14) show us areas of $\omega$ in which SUSY string may be decay. This is a different point between non-SUSY string [2, 3] and SUSY string. ## 3 Appendix ### 3.1 Recall the SUSY Skyrme model Let us consider the SUSY Lagrangian Skyrme [4, 20] ${\cal L}=-\frac{{f_{\pi}^{2}}}{{16}}Tr\left({\partial_{\mu}U^{\dagger}\partial^{\mu}U}\right)+\frac{1}{{32e^{2}}}Tr\left({\left[{U^{\dagger}\partial_{\mu}U,U^{\dagger}\partial_{\nu}U}\right]^{2}}\right),$ (15) where $U$ is can $SU(2)$ matrix, $f_{\pi}$ is the pion decay constant, $e$ is a free parameter. The ordinary derivatives in the Lagrangian density (15) is replaced by the covariant derivatives $\partial_{\mu}U\to D_{\mu}U=\partial_{\mu}U-iV_{\mu}U\tau_{3},$ (16) Eq (15) becomes as ${\cal L}=-\frac{{f_{\pi}^{2}}}{{16}}Tr\left({D_{\mu}U^{\dagger}D^{\mu}U}\right)+\frac{1}{{32e^{2}}}Tr\left({\left[{U^{\dagger}D_{\mu}U,U^{\dagger}D_{\nu}U}\right]^{2}}\right).$ (17) Eq (17) is invariant under the local $U\left(1\right)_{R}$ and the global $SU\left(2\right)_{L}$ transformations $U\left(r\right)\to AU\left(r\right)e^{i\lambda\left(r\right)\tau_{3}},A\in SU\left(2\right)_{L}$ , $V_{\mu}\left(r\right)\to V_{\mu}\left(r\right)+\partial_{\mu}\lambda\left(r\right).$ (18) where the gauge field $V_{\mu}\left(r\right)$ is defined as $V_{\mu}=-\frac{i}{2}Tr\left({U^{\dagger}\partial_{\mu}U\tau_{3}}\right).$ (19) One parametrizes the $SU(2)$ matrix $U$ in terms of the complex scalars $A_{i}$ $U\left(r\right)=\left({\begin{array}[]{*{20}c}{A_{1}}&{-A_{2}^{*}}\\\ {A_{2}}&{A_{1}^{*}}\\\ \end{array}}\right),$ (20) where $\bar{A}^{i}A_{i}=A_{1}^{*}A_{1}+A_{2}^{*}A_{2}=1$ . Eq (16) can be rewritten as $D_{\mu}A_{i}=\left({\partial_{\mu}-iV_{\mu}}\right)A_{i}$ , $D_{\mu}\bar{A}_{i}=\left({\partial_{\mu}+iV_{\mu}}\right)\bar{A}_{i}$ (21) and the new form of gauge field is $V_{\mu}\left(r\right)=-\frac{i}{2}\bar{A}^{i}\mathord{\buildrel{\lower 3.0pt\hbox{$\scriptscriptstyle\leftrightarrow$}}\over{\partial}}A_{i}.$ (22) Finally, we obtain Lagrangian in terms of complex scalars $A_{i}$ ${\cal L}=-\frac{{f_{\pi}^{2}}}{8}\bar{D}_{\mu}\bar{A}D^{\mu}A-\frac{1}{{16e^{2}}}F_{\mu\nu}^{2},$ (23) where $F_{\mu\nu}\left(V\right)=\partial_{\mu}V_{\nu}-\partial_{\nu}V_{\mu}$ One supersymmetrised Skyrme model by extending $A_{i}$ to chiral scalar multiplets $\left({A_{i},\psi_{\alpha i},F_{i}}\right)$ $\left({i,\alpha=1,2}\right)$ and the vector $V_{\mu}\left(x\right)$ to real vector multiplets $\left({V_{\mu},\lambda_{\alpha},D}\right)$. Here, the fields $F_{i}$ are complex scalars, $D$ is real scalar, $\psi_{\alpha i}$, $\lambda_{\alpha}$ are Majorana two-component spinors. $\psi_{\alpha i}$ corresponds to a left-handed chiral spinor, $\bar{\psi}^{\alpha i}=\left({\psi_{i}^{\alpha}}\right)^{*}$ corresponds to a right-handed one. The SUSY Lagrangian density is given by ${\cal L}_{susy}=\frac{{f_{\pi}^{2}}}{8}\left[{-D^{\mu}\bar{A}^{i}D_{\mu}A_{i}-\frac{1}{2}i\bar{\psi}^{\dot{\alpha}i}\left({\sigma_{\mu}}\right)_{\alpha\dot{\alpha}}\mathord{\buildrel{\lower 3.0pt\hbox{$\scriptscriptstyle\leftrightarrow$}}\over{D}}^{\mu}\psi_{i}^{\alpha}+\bar{F}^{i}F_{i}-}\right.$ $\left.{-i\bar{A}^{i}\lambda^{\alpha}\psi_{\alpha i}+iA_{i}\bar{\lambda}^{\dot{\alpha}}\bar{\psi}_{\dot{\alpha}}^{i}+D\left({\bar{A}^{i}A_{i}-1}\right)}\right]+$ $+\frac{1}{{8e^{2}}}\left[{-\frac{1}{2}F_{\mu\nu}^{2}-i\bar{\lambda}^{\dot{\alpha}}\left({\sigma^{\mu}}\right)_{\dot{\alpha}}^{\alpha}\partial_{\mu}\lambda_{\alpha}+D^{2}}\right].$ (24) This Lagrangian is invariant under the following set of supersymmetric transformations $\delta A_{i}=-\varepsilon^{\alpha}\psi_{\alpha i},$ (25) $\delta\psi_{\alpha i}=-i\bar{\varepsilon}^{\dot{\alpha}}\left({\sigma^{\mu}}\right)_{\alpha\dot{\alpha}}D_{\mu}A_{i}+\varepsilon_{\alpha}F_{i},$ (26) $\delta F_{i}=-i\bar{\varepsilon}^{\dot{\alpha}}\left({\sigma^{\mu}}\right)_{\dot{\alpha}}^{\alpha}D_{\mu}\psi_{\alpha i}-i\bar{\varepsilon}^{\dot{\alpha}}A_{i}\bar{\lambda}_{\dot{\alpha}},$ (27) $\delta V_{\mu}=-\frac{1}{2}i\left({\sigma_{\mu}}\right)^{\alpha\dot{\alpha}}\left({\bar{\varepsilon}_{\dot{\alpha}}\lambda_{\alpha}+\varepsilon_{\alpha}\bar{\lambda}_{\dot{\alpha}}}\right),$ (28) $\delta\lambda_{\alpha}=\varepsilon^{\beta}\left({\sigma^{\mu\nu}}\right)_{\beta\alpha}F_{\mu\nu}+i\varepsilon_{\alpha}D,$ (29) $\delta D=\frac{1}{2}\left({\sigma^{\mu}}\right)_{\alpha\dot{\alpha}}\partial_{\mu}\left({\bar{\varepsilon}^{\dot{\alpha}}\lambda^{\alpha}-\varepsilon^{\alpha}\bar{\lambda}^{\dot{\alpha}}}\right).$ (30) The field equation and their supersymmetric transformations lead to the following constraints $\bar{A}^{i}A_{i}=0,$ (31) $\bar{A}^{i}\psi_{\alpha i}=0,$ (32) $\bar{A}^{i}F_{i}=0,$ (33) and following algebraic expressions for $V_{\mu}=-\frac{1}{2}\left\\{{i\bar{A}^{i}\mathord{\buildrel{\lower 3.0pt\hbox{$\scriptscriptstyle\leftrightarrow$}}\over{\partial}}_{\mu}A_{i}+\left({\sigma_{\mu}}\right)^{\alpha\dot{\alpha}}\bar{\psi}_{\dot{\alpha}}^{i}\psi_{\alpha i}}\right\\},$ (34) $\lambda_{\alpha}=-i\left\\{{\bar{F}^{i}\psi_{\alpha i}+i\left({\sigma^{\mu}}\right)_{\alpha\dot{\alpha}}\left({D_{\mu}A_{i}}\right)\bar{\psi}^{\dot{\alpha}i}}\right\\},$ (35) $D=D^{\mu}\bar{A}^{i}D_{\mu}A_{i}+\frac{1}{2}i\bar{\psi}^{\dot{\alpha}i}\left({\sigma^{\mu}}\right)_{\alpha\dot{\alpha}}\left({\mathord{\buildrel{\lower 3.0pt\hbox{$\scriptscriptstyle\leftrightarrow$}}\over{D}}_{\mu}\psi_{i}^{\alpha}}\right)-\bar{F}^{i}F_{i}.$ (36) To obtain the minimum of SUSY extension, one set $\psi_{\alpha i}=F_{i}=0$. Therefore, Eq (24) becomes as ${\cal L}_{susy}=-\frac{{f_{\pi}^{2}}}{8}\bar{D}^{\mu}\bar{A}D_{\mu}A+\frac{1}{{8e^{2}}}\left[{-\frac{1}{2}F_{\mu\nu}^{2}+\left({\bar{D}^{\mu}\bar{A}D_{\mu}A}\right)^{2}}\right].$ (37) However, there is another four-derivatives term of $A_{i}$, one gave the general form of SUSY Lagrangian $\mathcal{L}_{susy}=-\frac{{f_{\pi}^{2}}}{8}\bar{D}^{\mu}\bar{A}D_{\mu}A+\frac{1}{{8e^{2}}}\left\\{{a\left[{-\frac{1}{2}F_{\mu\nu}^{2}+\left({\bar{D}^{\mu}\bar{A}D_{\mu}A}\right)^{2}}\right]+b\left\\{{\square\bar{A}\square A-\left({\bar{D}^{\mu}\bar{A}D_{\mu}A}\right)^{2}}\right\\}}\right\\},$ (38) where $a$, $b$ are constants, $\square=D^{\mu}D_{\mu}$ is the gauge covariant $D^{\prime}alembertian$. ### 3.2 Some main results for Eq (6) * The first term We have $D_{\mu}A_{i}=\left({\partial_{\mu}-iV_{\mu}}\right)A_{i}=\left[{\partial_{\mu}+\frac{1}{2}\left({\partial_{\mu}\bar{A}^{i}}\right)A_{i}+\frac{1}{2}\bar{A}^{i}\left({\partial_{\mu}A_{i}}\right)}\right]A_{i},$ (39) $\bar{D}^{\mu}\bar{A}^{i}=\left({\partial^{\mu}+iV^{\mu}}\right)\bar{A}^{i}=\left[{\partial^{\mu}-\frac{1}{2}\left({\partial^{\mu}\bar{A}^{i}}\right)A_{i}-\frac{1}{2}\bar{A}^{i}\left({\partial^{\mu}A_{i}}\right)}\right]\bar{A}^{i}.$ (40) Inserting forms of $A_{i}$ (4), let us final results $D_{0}A_{i}=-\dot{\alpha}\sin fe^{i\left({\theta+\alpha}\right)}$ ; $D_{1}A_{i}=0$ ; $D_{2}A_{i}=i\cos ff^{\prime}e^{i\left({\theta+\alpha}\right)}$; $D_{3}A_{i}=-\sin fe^{i\left({\theta+\alpha}\right)}$ , $\bar{D}^{0}\bar{A}_{i}=\dot{\alpha}\sin fe^{-i\left({\theta+\alpha}\right)}$; $\bar{D}^{1}\bar{A}_{i}=0$; $\bar{D}^{2}\bar{A}_{i}=-i\cos ff^{\prime}e^{-i\left({\theta+\alpha}\right)}$; $\bar{D}^{3}\bar{A}_{i}=-\sin fe^{-i\left({\theta+\alpha}\right)}$ . $\Rightarrow\bar{D}^{\mu}\bar{A}_{i}D_{\mu}A_{i}=\left({-\dot{\alpha}^{2}+1}\right)\sin^{2}f+f^{\prime 2}.$ (41) * The second term We have $\left({F_{\mu\nu}}\right)^{2}=\left({\partial_{1}V_{2}-\partial_{2}V_{1}}\right)\left({\partial^{1}V^{2}-\partial^{2}V^{2}}\right),$ (42) in terms of forms of $A_{i}$ (4), we obtain final results $F_{12}^{2}=F_{21}^{2}=0$; $F_{13}^{2}=F_{31}^{2}=0$; $F_{23}^{2}=F_{32}^{2}=0$ . $\Rightarrow\left({F_{\mu\nu}}\right)^{2}=0$ (43) * The third term We have $\square=D^{\mu}D_{\mu}=\left({\partial^{\mu}-iV^{\mu}}\right)\left({\partial_{\mu}-iV_{\mu}}\right)$ $=\partial^{\mu}\partial_{\mu}-i\partial^{\mu}V_{\mu}-iV^{\mu}\partial_{\mu}-V^{\mu}V_{\mu}=\partial^{\mu}\partial_{\mu}$ (44) Or $\square\bar{A}\square A=\partial_{0}^{2}\bar{A}^{i}\partial_{0}^{2}A_{i}+\partial_{1}^{2}\bar{A}^{i}\partial_{1}^{2}A_{i}+\partial_{2}^{2}\bar{A}^{i}\partial_{2}^{2}A_{i}+\partial_{3}^{2}\bar{A}^{i}\partial_{3}^{2}A_{i}.$ (45) Final results are $\partial_{0}^{2}\bar{A}^{i}\partial_{0}^{2}A_{i}=\cos^{2}f\left({f^{\prime 4}}\right)+2\sin 2f\left({f^{\prime 2}}\right)f^{\prime\prime}+\sin^{2}f\left({f^{\prime\prime 2}}\right)+\sin^{2}f\left({\ddot{\alpha}^{2}+\dot{\alpha}^{4}}\right),$ (46) $\partial_{1}^{2}\bar{A}^{i}\partial_{1}^{2}A_{i}=0,$ (47) $\partial_{2}^{2}\bar{A}^{i}\partial_{2}^{2}A_{i}=\sin^{2}f\left({f^{\prime 4}}\right)-2\sin 2f\left({f^{\prime 2}}\right)f^{\prime\prime}+\cos^{2}f\left({f^{\prime\prime 2}}\right),$ (48) $\partial_{3}^{2}\bar{A}^{i}\partial_{3}^{2}A_{i}=\sin^{2}f.$ (49) $\Rightarrow\square\bar{A}\square A=\sin^{2}f\left[{\dot{\alpha}^{4}+\ddot{\alpha}^{2}+1}\right]+f^{\prime\prime 2}+f^{\prime 4}$ (50) ## 4 Conclusion In this paper, we have performed analytic calculations, new results were found. In near future, they will be computed clearly by numerical methods, this way will give us the brilliant picture of mechanics of SUSY string’s decay. ACKNOWLEDGMENT We would like to thank Department of Theoretical Physics because of helps for us. One of us (DQT) want to thank Department of Computing Physics, HUS for giving me good conditions of working. ## REFERENCES * [1] A. Jackson, Nucl. Phys. A 493, (1989) 365. * [2] A. Jackson, Nucl. Phys. A 496, (1989) 667. * [3] M. Nitta, N. Skiiki, $\it arXiv$: 0706.0316 v2 (hep-th) * [4] E. A. Bergshoeff, R. I. Nepomachie, H. J. Schnitzer Nucl. Phys. B 249 (1985) 93. * [5] T. H. R. Skyrme, Pro. Roy. Soc. Lon, Vol 260, No 1300 127\. * [6] G. S. Adkins, C. R. Nappi, E. Witten, Nucl. Phys. B 228 (1983) 552. * [7] G. S. Adkins, Nucl. Phys. B 249 (1985) 507. * [8] G. S. Adkins, C. R. Nappi, Nucl. Phys. B 233 (1984) 109. * [9] J. Wess, B. Zumino, Phys. Lett. B 37 (1971) 95. * [10] A. P. Balachandra, S. Digal, Phys. Rev. D 66 (2002) 034018. * [11] A. P. Balachandra, S. Digal, Int. J. Mod. Phys. A 17 (2002) 1149. * [12] R. Rajaraman, H. M. Sommermann, J. Wambach, H. W. Wyld, Phys. Rev. D 33 (1986) 287. * [13] I. Zahed, G. E. Brown, Phys. Rept 142, No 1,2 (1986) 1. * [14] P. Fayet, S. Ferrara, Phys. Rept 32, No 5 (1977) 249. * [15] H. B. Nielsen, P. Olesen, Nucl. Phys. B 61 (1973) 45. * [16] A. D. Jackson, M. Rho, Phys. Rev. Lett 51 (1983) 751. * [17] N. A. Viet, P. T. Tuyen, J. Phys. G 15 (1989) 937. * [18] H. Y. Cheung, F. Gursey, Mod. Phys. Lett. A, Vol 5, No 21 (1990) 1685. * [19] P. T. Tuyen, D. Q. Tuan, $\it arXiv$: 0710.0971 v1 (nucl-th) * [20] N. Shiiki, N. Sawado, S. Oryu, hep-th/0603069

arxiv-papers

2024-09-04T02:48:54.424085

0803.2954

# Entanglement Monogamy of Tripartite Quantum States Chang-shui Yu He-shan Song hssong@dlut.edu.cn School of Physics and Optoelectronic Technology, Dalian University of Technology, Dalian 116024, P. R. China ###### Abstract An interesting monogamy equation with the form of Pythagorean theorem is found for $2\otimes 2\otimes n$-dimensional pure states, which reveals the relation among bipartite concurrence, concurrence of assistance, and genuine tripartite entanglement. At the same time, a genuine tripartite entanglement monotone as a generalization of 3-tangle is naturally obtained for $(2\otimes 2\otimes n)$\- dimensional pure states in terms of a distinct idea. For mixed states, the monogamy equation is reduced to a monogamy inequality. Both results for tripartite quantum states can be employed to multipartite quantum states. ###### pacs: 03.67.Mn, 03.65.Ta, 03.65.Ud ## I I. Introduction Entanglement is an essential feature of quantum mechanics, which distinguishes quantum from classical world. A key property of entanglement as well as one of the fundamental differences between quantum entanglement and classical correlations is the degree of sharing among many parties —–Unlike classical correlations, quantum entanglement is monogamous [1-3], i.e., the degree to which either of two parties can be entangled with anything else seems to be constrained by the entanglement that may exist between the two quantum parties. For the systems of three qubits, a kind of monogamy of bipartite quantum entanglement measured by concurrence [4] was described by Coffman- Kundu-Wootters (CKW) inequality [1]. The generalization to the case of multiple qubits was conjectured by CKW and has been proven recently by Osborne et al [5]. The monogamy inequality dual to CKW inequality based on concurrence of assistance (CoA) [6] was presented for tripartite systems of qubits by Gour et al [7] and the generalized one for multiple qubits was proven in Ref. [8]. In this paper, we find a new and very interesting monogamy equation for $\left(2\otimes 2\otimes n\right)$-dimensional (or multiple qubits) quantum pure states which relates the bipartite concurrence, CoA and genuine tripartite entanglement. In fact, CKW inequality and the dual one correspond to a residual quantity, respectively. It is only for tripartite pure states of qubits that so far the two residual quantities have been shown to be the same and have clear physical meanings. From Ref. [1] and [6], one can learn that it just corresponds to 3-tangle [1]. One of the distinguished advantages of our monogamy equation will be found that the residual quantity has clear physical meanings not only for $\left(2\otimes 2\otimes n\right)$-dimensional quantum pure state but also for a general multipartite pure state including a pair of qubits. Recently it has been realized that entanglement is a useful physical resource for various kinds of quantum information processing [9-12]. Based on the different physics of implementation, there are usually three alternative ways [7] to producing entanglement. The specially important way for quantum communication is the reduction of a multipartite entangled state to an entangled state with fewer parties, which is called ”assisted entanglement” quantified by entanglement of assistance (EoA) [13]. An important application of EoA is for tripartite quantum entangled state to maximize the entanglement of two parties (qubits) denoted by Alice and Bob with the assistance of the third party (qudit) named Charlie who is only allowed to do local operations. However, because EoA is not an entanglement monotone [14], one would prefer to the remarkable entanglement monotone——concurrence of assistance (CoA) where concurrence is employed to quantify the entanglement between Alice and Bob. In this process of entanglement preparation, Charlie only makes local operations and classical communications in order to increase the entanglement shared by Alice and Bob, therefore it is impossible to produce new entanglement. There must exist some trade-off between the increment of entanglement shared by Alice and Bob induced by Charlie and quantum correlations with other forms. Then what are those? The question is answered in this paper by our interesting monogamy equation. From the equation, one can find that the increment of entanglement shared by Alice and Bob just corresponds to the degree of genuine tripartite entanglement (3-way entanglement) of $\left(2\otimes 2\otimes n\right)$\- dimensional quantum pure state and is analytically calculable. Hence, the increment naturally characterizes the genuine tripartite entanglement, which is shown to be an entanglement monotone and can be considered as an interesting generalization of 3-tangle in terms of a new idea. In addition, the monogamy equation is reduced to a monogamy inequality for mixed states. The results are also suitable for multipartite quantum states. This paper is organized as follows. We first introduce our interesting monogamy equation for pure states; Then for mixed states, we reduce this monogamy equation for pure state to a monogamy inequality; Next we point out these results are suitable for multipartite quantum states; The conclusion is drawn finally. ## II II. Monogamy equation for pure states Given a tripartite $\left(2\otimes 2\otimes n\right)$\- dimensional quantum pure state $\left|\Psi\right\rangle_{ABC}$ shared by three parties Alice, Bob and Charlie, where Charlie’s aim is to maximize the entanglement shared by Alice and Bob by local measurements on Charlie’s particle C, the reduced density matrix by tracing over party C can be given by $\rho_{AB}=Tr_{C}$ $\left(\left|\Psi\right\rangle_{ABC}\left\langle\Psi\right|\right)$. Let $\mathcal{E}=\\{p_{i},\left|\varphi_{i}^{AB}\right\rangle\\}$ is any a decomposition of $\rho_{AB}$ such that $\rho_{AB}=\sum\limits_{i}p_{i}\left|\varphi_{i}^{AB}\right\rangle\left\langle\varphi_{i}^{AB}\right|,\sum\limits_{i}p_{i}=1,$ (1) then CoA is defined [5,6] by $\displaystyle C_{a}\left(\left|\Psi\right\rangle_{ABC}\right)$ $\displaystyle=$ $\displaystyle\max_{\mathcal{E}}\sum\limits_{i}p_{i}C\left(\left|\varphi_{i}^{AB}\right\rangle\right)$ (2) $\displaystyle=C_{a}\left(\rho_{AB}\right)$ $\displaystyle=$ $\displaystyle tr\sqrt{\sqrt{\rho_{AB}}\tilde{\rho}_{AB}\sqrt{\rho_{AB}}}$ (3) $\displaystyle=$ $\displaystyle\sum\limits_{i=1}^{4}\lambda_{i},$ (4) where $\tilde{\rho}_{AB}=\left(\sigma_{y}\otimes\sigma_{y}\right)\rho_{AB}^{\ast}\left(\sigma_{y}\otimes\sigma_{y}\right)$, $\sigma_{y}$ is Pauli matrix and $C\left(\rho_{AB}\right)=\max\\{0,\lambda_{1}-\sum\limits_{i>1}\lambda_{i}\\}$ (5) is the concurrence of the reduced density matrix $\rho_{AB}$ with $\lambda_{i}$ being the square roots of the eigenvalues of $\rho_{AB}\tilde{\rho}_{AB}$ in decreasing order. With the definitions of CoA and concurrence, we can obtain the following theorem. Theorem 1: _For a_ $\left(2\otimes 2\otimes n\right)$_\- dimensional quantum pure state_ $\left|\Psi\right\rangle_{ABC}$_,_ $C_{a}^{2}\left(\rho_{AB}\right)=C^{2}\left(\rho_{AB}\right)+\tau^{2}\left(\rho_{AB}\right),$ (6) _ where _$\tau\left(\rho_{AB}\right)=\tau\left(\left|\Psi\right\rangle_{ABC}\right)$_is the genuine tripartite entanglement measure for_ $\left|\Psi\right\rangle_{ABC}$_._ It is very interesting that eq. (6) has an elegant form that is analogous to Pythagorean theorem if one considers CoA as the length of the hypotenuse of a right-angled triangle and considers bipartite concurrence and genuine tripartite entanglement as the lengths of the other two sides of the triangle. Note that the lengths of all the sides are allowed to be zero. The illustration of the relation is shown in Fig. 1 (See the left triangle). Proof. According to the definition of CoA and concurrence, it is obvious that $C_{a}^{2}\left(\rho_{AB}\right)-C^{2}\left(\rho_{AB}\right)\geq 0.$ (7) Then the remaining is to prove that $\tau\left(\rho_{AB}\right)$ is an entanglement monotone and characterizes the genuine tripartite entanglement of $\left|\Psi\right\rangle_{ABC}$. Next, we first prove that $\tau\left(\rho_{AB}\right)$ does not increase under a general tripartite local operation and classical communication (LOCC) denoted by $\mathcal{M}_{k}$ where subscript $k$ labels different outcomes. We first assume that Alice and Bob perform quantum operations $M_{Akj}$, and $M_{Bkj}$ on their qubits respectively, where $\sum\limits_{k,j}$ $M_{Akj}^{\dagger}M_{Akj}\leq I_{A}$ and $\sum\limits_{k,j}$ $M_{Bkj}^{\dagger}M_{Bkj}\leq I_{B}$ are the most general local operations given in terms of the Kraus operator [15] with $I_{A}$ and $I_{B}$ being the identity operators in Alice’s and Bob’s systems. After local operations, the average CoA can be written as $\displaystyle\sum\limits_{kk^{\prime}}P_{kk^{\prime}}\tau\left(\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)\right)$ (8) $\displaystyle=$ $\displaystyle\sum\limits_{kk^{\prime}}P_{kk^{\prime}}\sqrt{C_{a}^{2}\left(\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)\right)-C^{2}\left(\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)\right)}$ $\displaystyle\leq$ $\displaystyle\left\\{\left[\sum\limits_{kk^{\prime}}P_{kk^{\prime}}C_{a}\left(\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)\right)\right]^{2}\right.$ $\displaystyle\left.-\left[\sum\limits_{kk^{\prime}}P_{kk^{\prime}}C\left(\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)\right)\right]^{2}\right\\}^{1/2}$ $\displaystyle=$ $\displaystyle\sum\limits_{kk^{\prime}jj^{\prime}}\left|\det\left(M_{Akj}\right)\det\left(M_{Bk^{\prime}j^{\prime}}\right)\right|\sqrt{C_{a}^{2}\left(\rho_{AB}\right)-C^{2}\left(\rho_{AB}\right)}$ $\displaystyle\leq$ $\displaystyle\sqrt{C_{a}^{2}\left(\rho_{AB}\right)-C^{2}\left(\rho_{AB}\right)}=\tau\left(\rho_{AB}\right),$ where $\displaystyle\mathcal{M}_{kk^{\prime}}\left(\rho_{AB}\right)=\sum\limits_{jj^{\prime}}\left(M_{Akj}\otimes M_{Bk^{\prime}j^{\prime}}\otimes I_{C}\right)$ (9) $\displaystyle\times$ $\displaystyle\left|\Psi\right\rangle_{ABC}\left\langle\Psi\right|\left(M_{Akj}^{\dagger}\otimes M_{Bk^{\prime}j^{\prime}}^{\dagger}\otimes I_{C}\right)/P_{kk^{\prime}},$ and $P_{kk^{\prime}}=tr\mathcal{M}_{kk^{\prime}}\left(\left|\Psi\right\rangle_{ABC}\left\langle\Psi\right|\right)$. Here the first inequality follows from Cauchy-Schwarz inequality: $\sum\limits_{i}x_{i}y_{i}\leq\left(\sum\limits_{i}x_{i}^{2}\right)^{1/2}\left(\sum\limits_{j}y_{j}^{2}\right)^{1/2},$ (10) the second inequality follows from the geometric-arithmetic inequality $\sum\limits_{kj}\left|\det\left(M_{xkj}\right)\right|\leq\frac{1}{2}\sum\limits_{kj}trM_{xkj}^{\dagger}M_{xkj}\leq 1,x=A,B,$ (11) and the second equation is derived from the fact [4,16] that $C_{a}\left(M_{Akj}\rho_{AB}M_{Akj}^{\dagger}\right)=\left|\det\left(M_{Akj}\right)\right|C_{a}\left(\rho_{AB}\right),$ (12) $C\left(M_{Akj}\rho_{AB}M_{Akj}^{\dagger}\right)=\left|\det\left(M_{Akj}\right)\right|C\left(\rho_{AB}\right)$ (13) and the analogous relations for $M_{Bk^{\prime}j^{\prime}}$. Eq. (8) shows that $\tau\left(\rho_{AB}\right)$ does not increase under Alice’s and Bob’s local operations. Figure 1: The illustration of the relation among CoA, bipartite concurrence and genuine tripartite entanglement. The left right-angled triangle corresponds to Theorem 1 (for pure states) and the right obtuse-angled triangle corresponds to Theorem 2 (for mixed states). All the quantities given in the figures are defined the same as the corresponding theorems. Next we prove that $\tau\left(\rho_{AB}\right)$ does not increase under Charlie’s local operations either. Suppose $\rho_{AB}=\lambda\rho_{1}^{AB}+(1-\lambda)\rho_{2}^{AB}$, $\lambda\in[0,1]$, then $\displaystyle\lambda\tau\left(\rho_{1}^{AB}\right)+(1-\lambda)\tau\left(\rho_{2}^{AB}\right)$ (14) $\displaystyle=$ $\displaystyle\lambda\sqrt{C_{a}^{2}\left(\rho_{1}^{AB}\right)-C^{2}(\rho_{1}^{AB})}$ $\displaystyle+(1-\lambda)\sqrt{C_{a}^{2}\left(\rho_{2}^{AB}\right)-C^{2}(\rho_{2}^{AB})}$ $\displaystyle\leq$ $\displaystyle\left\\{\left[\lambda C_{a}\left(\rho_{1}^{AB}\right)+(1-\lambda)C_{a}\left(\rho_{1}^{AB}\right)\right]^{2}\right.$ $\displaystyle\left.-\left[\lambda C\left(\rho_{2}^{AB}\right)+(1-\lambda)C\left(\rho_{2}^{AB}\right)\right]^{2}\right\\}^{1/2}$ $\displaystyle\leq$ $\displaystyle\sqrt{C_{a}^{2}\left(\rho_{AB}\right)-C^{2}\left(\rho_{AB}\right)}=\tau\left(\rho_{AB}\right),$ where the first inequality follows from the Cauchy-Schwarz inequality (10) and the second inequality follows from the definitions of $C_{a}\left(\rho_{AB}\right)$ and $C\left(\rho_{AB}\right)$. Eq. (14) shows that $\tau\left(\rho_{AB}\right)$ is a concave function of $\rho_{AB}$, which proves that $\tau\left(\rho_{AB}\right)$ does not increase under Charlie’s local operations following the same procedure (or Theorem 3) in Ref. [17]. All above show that $\tau\left(\rho_{AB}\right)$ is an entanglement monotone. Now we prove that $\tau\left(\rho_{AB}\right)$ characterizes genuine tripartite entanglement. Based on eq. (4) and eq. (5), it is obvious that $\tau\left(\rho_{AB}\right)=\left\\{\begin{array}[]{cc}\sum\limits_{i=1}^{4}\lambda_{i},&\lambda_{1}\leq\sum\limits_{i=2}^{4}\lambda_{i},\\\ 2\sqrt{\lambda_{1}\sum\limits_{i=2}^{4}\lambda_{i}},&\lambda_{1}>\sum\limits_{i=2}^{4}\lambda_{i},\end{array}\right.$ (15) which is an explicit formulation. Ref. [18] has given a special quantity named ”entanglement semi-monotone” that characterizes the genuine tripartite entanglement. One can find that it requires the same conditions as the quantity introduced in Ref. [18] for $\tau\left(\rho_{AB}\right)$ to reach _zero_ , which shows that $\tau\left(\rho_{AB}\right)$ characterizes the genuine tripartite entanglement. The proof is completed.$\hfill\Box$ In general, multipartite entanglement is quantified in terms of different classifications [19-21]. However, $\tau\left(\rho_{AB}\right)$ quantifies genuine tripartite entanglement in a new way, i.e., we consider the entanglement of GHZ-state class as the minimal unit [22] in terms of tensor treatment [23] and summarize all the genuine tripartite inseparability without further classifications. It is an interesting generalization of 3-tangle. Theorem 1 shows a very clear physical meaning, i.e. the increment of entanglement between Alice and Bob induced by Charlie is just the genuine tripartite entanglement among them. The meaning can especially easily be understood for tripartite quantum state of qubits. In this case, $\tau\left(\rho_{AB}\right)=2\sqrt{\lambda_{1}\lambda_{2}}$. Two most obvious examples are GHZ state and W state. The entanglement of reduced density matrix of GHZ state is zero, hence Theorem 1 shows that the CoA of GHZ state all comes from the three-way entanglement and equals to 1 (the value of 3-tangle). On the contrary, the W state has no three-way entanglement (only two-way entanglement) [24], hence its CoA is only equal to the concurrence ($\frac{2}{3}$) of two parties. That is to say, for W state, Charlie can not provide any help to increase the entanglement between Alice and Bob. ## III III. Monogamy inequality for mixed states For a given mixed state $\rho_{ABC}$, CoA can be extended to mixed states in terms of convex roof construction [15], i.e., $C_{a}(\rho_{ABC})=\min\sum\limits_{i}p_{i}C_{a}(\left|\psi^{i}\right\rangle_{ABC}),$ (16) where the minimum is taken over all decompositions $\\{p_{i},\left|\psi\right\rangle_{ABC}\\}$ of $\rho_{ABC}$. Thus we have the following theorem. Theorem 2.-_For a_ $\left(2\otimes 2\otimes n\right)$_\- dimensional mixed state_ $\rho_{ABC}$_,_ $C_{a}^{2}\left(\rho_{ABC}\right)\geqslant C^{2}\left(\rho_{AB}\right)+\tau^{2}\left(\rho_{ABC}\right),$ (17) _where_ $\tau\left(\rho_{ABC}\right)$_is the genuine tripartite entanglement measure for mixed states by extending_ $\tau\left(\cdot\right)$_of pure states in terms of convex roof construction and_ $\rho_{AB}=tr_{C}\rho_{ABC}$_._ Analogous to Theorem 1, one can easily find that the relation of Theorem 2 corresponds to an obtuse-angled triangle after a simple algebra, where CoA corresponds to the length of the side opposite to the obtuse angle. See the right triangle in Fig.1 for the illustration. Proof. Suppose $\\{p_{k},\left|\psi^{k}\right\rangle_{ABC}\\}$ is the optimal decomposition in the sense of $\tau\left(\rho_{ABC}\right)=\sum\limits_{k}p_{k}\tau\left(\left|\psi^{k}\right\rangle_{ABC}\right)=\sum\limits_{k}p_{k}\tau\left(\sigma_{AB}^{k}\right),$ (18) where $\sigma_{AB}^{k}=tr_{C}\left[\left|\psi^{k}\right\rangle_{ABC}\left\langle\psi^{k}\right|\right]$. According to Theorem 1, we have $\displaystyle\tau\left(\rho_{ABC}\right)$ $\displaystyle=$ $\displaystyle\sum\limits_{k}p_{k}\sqrt{C_{a}^{2}\left(\sigma_{AB}^{k}\right)-C^{2}\left(\sigma_{AB}^{k}\right)}$ (19) $\displaystyle\leq$ $\displaystyle\sqrt{\left[\sum\limits_{k}p_{k}C_{a}\left(\sigma_{AB}^{k}\right)\right]^{2}-\left[\sum\limits_{k}p_{k}C\left(\sigma_{AB}^{k}\right)\right]^{2}}$ $\displaystyle\leq$ $\displaystyle\sqrt{C_{a}^{2}\left(\rho_{ABC}\right)-C^{2}\left(\rho_{AB}\right)},$ where the first inequality follows from Cauchy-Schwarz inequality (10) and the second inequality holds based on the definitions of $C_{a}\left(\rho_{AB}\right)$ and $C\left(\rho_{AB}\right)$. Eq. (19) finishes the proof. $\hfill\Box$ ## IV IV. Monogamy for multipartite quantum states Any a given $N$-partite quantum state can always be considered as a $\left(2\otimes 2\otimes X\right)$\- dimensional tripartite quantum states with $X$ denoting the total dimension of $N-2$ subsystems so long as the state includes at least two qubits, hence both the two theorems hold in these cases. However, it is especially worthy of being noted that the two qubits must be owned by Alice and Bob respectively and the other $N-2$ subsystems should be at Charlie’s hand and be considered as a whole. Charlie is allowed to perform any nonlocal operation on the $N-2$ subsystems. In addition, there may be different groupings [25] of a multipartite quantum state especially for multipartite quantum states of qubits, hence there exist many analogous monogamy equations (for pure states) or monogamy inequalities (for mixed states) for the same quantum state. For pure states, every monogamy equation will lead to a genuine $\left(2\otimes 2\otimes X\right)$\- dimensional tripartite entanglement monotone that quantifies the genuine tripartite entanglement of the tripartite state generated by the corresponding grouping. ## V V. Conclusion and discussion We have presented an interesting monogamy equation with elegant form for $(2\otimes 2\otimes n)$\- dimensional quantum pure states, which, for the first time, reveals the relation among bipartite concurrence, CoA and genuine tripartite entanglement. The equation naturally leads to a genuine tripartite entanglement measure for $(2\otimes 2\otimes n)$-dimensional tripartite quantum pure states, which quantifies tripartite entanglement in terms of a new idea. The monogamy equation can be reduced to a monogamy inequality for mixed states. Both the results for tripartite quantum states are also suitable for multipartite quantum states. We hope that the current results can shed new light on not only the monogamy of entanglement but also the quantification of multipartite entanglement. ## VI Acknowledgement This work was supported by the National Natural Science Foundation of China, under Grant No. 10747112 and No. 10575017. ## References * (1) V. Coffman, J. Kundu and W. K. Wootters, Phys. Rev. A 61, 052306 (2000). * (2) B. M. Terhal, IBM J. Res. Dev. 48, 71 (2004). * (3) M. Koashi and A. Winter, Phys. Rev. A 69, 022309 (2004) and the references therein. * (4) W. K. Wootters, Phys. Rev. Lett. 80, 2245 (1998). * (5) T. J. Osborne and F. Verstraete, Phys. Rev. Lett. 96, 220503 (2006). * (6) T. Laustsen, F. Verstraete and S. J. Van Enk, Quantum Information and Computation 4, 64 (2003). * (7) G. Gour, D. A. Meyer and B. C. Sanders, Phys. Rev. A 72, 042329 (2005). * (8) S. Bandyopadhyay, G. Gour and B. C. Sanders, J. Math. Phys. 48, 012108 (2007). * (9) M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, Cambridge, 2000). * (10) M. Zukowski, A. Zeilinger, M. A. Horne, and A. K. Ekert, Phys. Rev. Lett. 71, 4287 (1993). * (11) C. H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres and W. K. Wootters, Phys. Rev. Lett. 70, 1895 (1993). * (12) C. H. Bennett and S. J. Wiesner, Phys. Rev. Lett. 69, 2881 (1992). * (13) D. P. Divincenzo, C. A. Fuchs, H. Mabuchi, J. A. Smolin, A. Thapliyal and A. Uhlmann, quant-ph/9803033. * (14) G. Gour and R. W. Spekkens, Phys. Rev. A 73, 062331 (2006). * (15) G. Gour, Phys. Rev. A 72, 042318 (2005). * (16) G. Gour, Phys. Rev. A 71, 012318 (2005). * (17) G. Vidal, J. Mod. Opt 47, 355 (2000). * (18) Chang-shui Yu, He-shan Song and Ya-hong Wang, Quantum Information and Computation 7, 584 (2007). * (19) W. Dür, G. Vidal and J. I. Cirac, Phys. Rev. A 62, 062314 (2000). * (20) A. Miyake, Phys. Rev. A 67, 012108 (2003). * (21) A. Miyake and F. Verstraete, Phys. Rev. A 69, 012101 (2004). * (22) Ref. [19,20] have introduced onionlike classification of tripartite entanglement for $2\otimes 2\otimes n$\- dimensional quantum pure states. It has been shown that the quantum states in the outer class can always irreversely converted into the states in the inner class. The GHZ-state class belongs to the most inner class that characterizes genuine tripartite entanglement (three-way entanglement). Thus entanglement of GHZ-state class can be understood as the minimal unit that quantifies genuine tripartite inseparability. * (23) Chang-shui Yu and He-shan Song, Phys. Rev. A 72, 022333 (2005). * (24) Alexander Wong and Nelson Christensen, Phys. Rev. A 63, 044301 (2001). * (25) Chang-shui Yu and He-shan Song, Phys. Rev. A 73, 022325 (2006).

arxiv-papers

2024-09-04T02:48:54.430135

0803.2958

# Generalizations of Popoviciu’s inequality Darij Grinberg (20 March 2008) ###### Abstract We establish a general criterion for inequalities of the kind $\displaystyle\text{convex combination of }f\left(x_{1}\right),\text{ }f\left(x_{2}\right),\text{ }...,\text{ }f\left(x_{n}\right)$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \text{ and }f\left(\text{some weighted mean of }x_{1},\text{ }x_{2},\text{ }...,\text{ }x_{n}\right)$ $\displaystyle\geq\text{convex combination of }f\left(\text{some other weighted means of }x_{1},\text{ }x_{2},\text{ }...,\text{ }x_{n}\right),$ where $f$ is a convex function on an interval $I\subseteq\mathbb{R}$ containing the reals $x_{1},$ $x_{2},$ $...,$ $x_{n},$ to hold. Here, the left hand side contains only one weighted mean, while the right hand side may contain as many as possible, as long as there are finitely many. The weighted mean on the left hand side must have positive weights, while those on the right hand side must have nonnegative weights. This criterion entails Vasile Cîrtoaje’s generalization of the Popoviciu inequality (in its standard and in its weighted forms) as well as a cyclic inequality that sharpens another result by Vasile Cîrtoaje. The latter cyclic inequality (in its non-weighted form) states that $2\sum_{i=1}^{n}f\left(x_{i}\right)+n\left(n-2\right)f\left(x\right)\geq n\sum_{s=1}^{n}f\left(x+\dfrac{x_{s}-x_{s+r}}{n}\right),$ where indices are cyclic modulo $n,$ and $x=\dfrac{x_{1}+x_{2}+...+x_{n}}{n}.$ This is the standard version of this note. A ”formal” version with more detailed proofs can be found at http://www.stud.uni-muenchen.de/~darij.grinberg/PopoviciuFormal.pdf However, due to these details, it is longer and much more troublesome to read, so it should be used merely as a resort in case you do not understand the proofs in this standard version. Keywords: Convexity on the real axis, majorization theory, inequalities. 1\. Introduction The last few years saw some activity related to the Popoviciu inequality on convex functions. Some generalizations were conjectured and subsequently proven using majorization theory and (mostly) a lot of computations. In this note I am presenting an apparently new approach that proves these generalizations as well as some additional facts with a lesser amount of computation and avoiding majorization theory (more exactly, avoiding the standard, asymmetric definition of majorization; we will prove a ”symmetric” version of the Karamata inequality on the way, which will not even use the word ”majorize”). The very starting point of the whole theory is the following famous fact: > Theorem 1a, the Jensen inequality. Let $f$ be a convex function from an > interval $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ > $...,$ $x_{n}$ be finitely many points from $I.$ Then, > > > $\dfrac{f\left(x_{1}\right)+f\left(x_{2}\right)+...+f\left(x_{n}\right)}{n}\geq > f\left(\dfrac{x_{1}+x_{2}+...+x_{n}}{n}\right).$ > > In words, the arithmetic mean of the values of $f$ at the points $x_{1},$ > $x_{2},$ $...,$ $x_{n}$ is greater or equal to the value of $f$ at the > arithmetic mean of these points. We can obtain a ”weighted version” of this inequality by replacing arithmetic means by weighted means with some nonnegative weights $w_{1},$ $w_{2},$ $...,$ $w_{n}$: > Theorem 1b, the weighted Jensen inequality. Let $f$ be a convex function > from an interval $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ > $x_{2},$ $...,$ $x_{n}$ be finitely many points from $I.$ Let $w_{1},$ > $w_{2},$ $...,$ $w_{n}$ be $n$ nonnegative reals which are not all equal to > $0.$ Then, > > > $\dfrac{w_{1}f\left(x_{1}\right)+w_{2}f\left(x_{2}\right)+...+w_{n}f\left(x_{n}\right)}{w_{1}+w_{2}+...+w_{n}}\geq > f\left(\dfrac{w_{1}x_{1}+w_{2}x_{2}+...+w_{n}x_{n}}{w_{1}+w_{2}+...+w_{n}}\right).$ Obviously, Theorem 1a follows from Theorem 1b applied to $w_{1}=w_{2}=...=w_{n}=1,$ so that Theorem 1b is more general than Theorem 1a. We won’t stop at discussing equality cases here, since they can depend in various ways on the input (i. e., on the function $f,$ the reals $w_{1},$ $w_{2},$ $...,$ $w_{n}$ and the points $x_{1},$ $x_{2},$ $...,$ $x_{n}$) - but each time we use a result like Theorem 1b, with enough patience we can extract the equality case from the proof of this result and the properties of the input. The Jensen inequality, in both of its versions above, is applied often enough to be called one of the main methods of proving inequalities. Now, in 1965, a similarly styled inequality was found by the Romanian Tiberiu Popoviciu: > Theorem 2a, the Popoviciu inequality. Let $f$ be a convex function from an > interval $I\subseteq\mathbb{R}$ to $\mathbb{R},$ and let $x_{1},$ $x_{2},$ > $x_{3}$ be three points from $I.$ Then, > > > $f\left(x_{1}\right)+f\left(x_{2}\right)+f\left(x_{3}\right)+3f\left(\dfrac{x_{1}+x_{2}+x_{3}}{3}\right)\geq > 2f\left(\dfrac{x_{2}+x_{3}}{2}\right)+2f\left(\dfrac{x_{3}+x_{1}}{2}\right)+2f\left(\dfrac{x_{1}+x_{2}}{2}\right).$ Again, a weighted version can be constructed: > Theorem 2b, the weighted Popoviciu inequality. Let $f$ be a convex function > from an interval $I\subseteq\mathbb{R}$ to $\mathbb{R},$ let $x_{1},$ > $x_{2},$ $x_{3}$ be three points from $I,$ and let $w_{1},$ $w_{2},$ $w_{3}$ > be three nonnegative reals such that $w_{2}+w_{3}\neq 0,$ $w_{3}+w_{1}\neq > 0$ and $w_{1}+w_{2}\neq 0.$ Then, > > $\displaystyle > w_{1}f\left(x_{1}\right)+w_{2}f\left(x_{2}\right)+w_{3}f\left(x_{3}\right)+\left(w_{1}+w_{2}+w_{3}\right)f\left(\dfrac{w_{1}x_{1}+w_{2}x_{2}+w_{3}x_{3}}{w_{1}+w_{2}+w_{3}}\right)$ > $\displaystyle\geq\left(w_{2}+w_{3}\right)f\left(\dfrac{w_{2}x_{2}+w_{3}x_{3}}{w_{2}+w_{3}}\right)+\left(w_{3}+w_{1}\right)f\left(\dfrac{w_{3}x_{3}+w_{1}x_{1}}{w_{3}+w_{1}}\right)+\left(w_{1}+w_{2}\right)f\left(\dfrac{w_{1}x_{1}+w_{2}x_{2}}{w_{1}+w_{2}}\right).$ The really interesting part of the story began when Vasile Cîrtoaje - alias ”Vasc” on the MathLinks forum - proposed the following two generalizations of Theorem 2a ([1] and [2] for Theorem 3a, and [1] and [3] for Theorem 4a): > Theorem 3a (Vasile Cîrtoaje). Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ Then, > > > $\sum_{i=1}^{n}f\left(x_{i}\right)+n\left(n-2\right)f\left(\dfrac{x_{1}+x_{2}+...+x_{n}}{n}\right)\geq\sum_{j=1}^{n}\left(n-1\right)f\left(\dfrac{\sum\limits_{1\leq > i\leq n;\ i\neq j}x_{i}}{n-1}\right).$ > > Theorem 4a (Vasile Cîrtoaje). Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ Then, > > > $\left(n-2\right)\sum_{i=1}^{n}f\left(x_{i}\right)+nf\left(\dfrac{x_{1}+x_{2}+...+x_{n}}{n}\right)\geq\sum_{1\leq > i<j\leq n}2f\left(\dfrac{x_{i}+x_{j}}{2}\right).$ In [1], both of these facts were nicely proven by Cîrtoaje. I gave a different and rather long proof of Theorem 3a in [2]. All of these proofs use the Karamata inequality. Theorem 2a follows from each of the Theorems 3a and 4a upon setting $n=3.$ It is pretty straightforward to obtain generalizations of Theorems 3a and 4a by putting in weights as in Theorems 1b and 2b. A more substantial generalization was given by Yufei Zhao - alias ”Billzhao” on MathLinks - in [3]: > Theorem 5a (Yufei Zhao). Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I,$ and let $m$ be an integer. Then, > > > $\displaystyle\binom{n-2}{m-1}\sum_{i=1}^{n}f\left(x_{i}\right)+\binom{n-2}{m-2}nf\left(\frac{x_{1}+x_{2}+...+x_{n}}{n}\right)$ > $\displaystyle\geq\sum_{1\leq i_{1}<i_{2}<...<i_{m}\leq > n}mf\left(\frac{x_{i_{1}}+x_{i_{2}}+...+x_{i_{m}}}{m}\right).$ Note that if $m\leq 0$ or $m>n,$ the sum $\sum\limits_{1\leq i_{1}<i_{2}<...<i_{m}\leq n}mf\left(\dfrac{x_{i_{1}}+x_{i_{2}}+...+x_{i_{m}}}{m}\right)$ is empty, so that its value is $0.$ Note that Theorems 3a and 4a both are particular cases of Theorem 5a (in fact, set $m=n-1$ to get Theorem 3a and $m=2$ to get Theorem 4a). An rather complicated proof of Theorem 5a was given by myself in [3]. After some time, the MathLinks user ”Zhaobin” proposed a weighted version of this result: > Theorem 5b (Zhaobin). Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I,$ let $w_{1},$ $w_{2},$ $...,$ $w_{n}$ be > nonnegative reals, and let $m$ be an integer. Assume that > $w_{1}+w_{2}+...+w_{n}\neq 0,$ and that > $w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}\neq 0$ for any $m$ integers $i_{1},$ > $i_{2},$ $...,$ $i_{m}$ satisfying $1\leq i_{1}<i_{2}<...<i_{m}\leq n.$ > > Then, > > > $\displaystyle\binom{n-2}{m-1}\sum_{i=1}^{n}w_{i}f\left(x_{i}\right)+\binom{n-2}{m-2}\left(w_{1}+w_{2}+...+w_{n}\right)f\left(\frac{w_{1}x_{1}+w_{2}x_{2}+...+w_{n}x_{n}}{w_{1}+w_{2}+...+w_{n}}\right)$ > $\displaystyle\geq\sum_{1\leq i_{1}<i_{2}<...<i_{m}\leq > n}\left(w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}\right)f\left(\frac{w_{i_{1}}x_{i_{1}}+w_{i_{2}}x_{i_{2}}+...+w_{i_{m}}x_{i_{m}}}{w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}}\right).$ If we set $w_{1}=w_{2}=...=w_{n}=1$ in Theorem 5b, we obtain Theorem 5a. On the other hand, putting $n=3$ and $m=2$ in Theorem 5b, we get Theorem 2b. In this note, I am going to prove Theorem 5b (and therefore also its particular cases - Theorems 2a, 2b, 3a, 4a and 5a). The proof is going to use no preknowledge - in particular, classical majorization theory will be avoided. Then, we are going to discuss an assertion similar to Theorem 5b with its applications. 2\. Absolute values interpolate convex functions We start preparing for our proof by showing a property of convex functions which is definitely not new - it was mentioned by a MathLinks user called ”Fleeting_Guest” in [4], post #18 as a known fact: > Theorem 6. Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ Then, there exist two real constants $u$ > and $v$ and $n$ nonnegative constants $a_{1},$ $a_{2},$ $...,$ $a_{n}$ such > that > > $f\left(t\right)=vt+u+\sum\limits_{i=1}^{n}a_{i}\left|t-x_{i}\right|\text{ > holds for every }t\in\left\\{x_{1},x_{2},...,x_{n}\right\\}.$ In brief, this result states that every convex function $f\left(x\right)$ on $n$ reals $x_{1},$ $x_{2},$ $...,$ $x_{n}$ can be interpolated by a linear combination with nonnegative coefficients of a linear function and the $n$ functions $\left|x-x_{i}\right|.$ The proof of Theorem 6, albeit technical, will be given here for the sake of completeness: First, we need an almost trivial fact which I use to call the $\max\left\\{0,x\right\\}$ formula: For any real number $x,$ we have $\max\left\\{0,x\right\\}=\dfrac{1}{2}\left(x+\left|x\right|\right).$ Furthermore, we denote $f\left[y,z\right]=\dfrac{f\left(y\right)-f\left(z\right)}{y-z}$ for any two points $y$ and $z$ from $I$ satisfying $y\neq z.$ Then, we have $\left(y-z\right)\cdot f\left[y,z\right]=f\left(y\right)-f\left(z\right)$ for any two points $y$ and $z$ from $I$ satisfying $y\neq z.$ We can assume that all points $x_{1},$ $x_{2},$ $...,$ $x_{n}$ are pairwisely distinct (if not, we can remove all superfluous $x_{i}$ and apply Theorem 6 to the remaining points). Therefore, we can WLOG assume that $x_{1}<x_{2}<...<x_{n}.$ Then, for every $j\in\left\\{1,2,...,n\right\\},$ we have $\displaystyle f\left(x_{j}\right)$ $\displaystyle=f\left(x_{1}\right)+\sum_{k=1}^{j-1}\left(f\left(x_{k+1}\right)-f\left(x_{k}\right)\right)=f\left(x_{1}\right)+\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)\cdot f\left[x_{k+1},x_{k}\right]$ $\displaystyle=f\left(x_{1}\right)+\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)\cdot\left(f\left[x_{2},x_{1}\right]+\sum_{i=2}^{k}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)\right)$ $\displaystyle=f\left(x_{1}\right)+\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)\cdot f\left[x_{2},x_{1}\right]+\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)\cdot\sum_{i=2}^{k}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)+\sum_{k=1}^{j-1}\sum_{i=2}^{k}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)\cdot\left(x_{k+1}-x_{k}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)+\sum_{i=2}^{j-1}\sum_{k=i}^{j-1}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)\cdot\left(x_{k+1}-x_{k}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\sum_{k=1}^{j-1}\left(x_{k+1}-x_{k}\right)+\sum_{i=2}^{j-1}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)\cdot\sum_{k=i}^{j-1}\left(x_{k+1}-x_{k}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\sum_{i=2}^{j-1}\left(f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\right)\cdot\left(x_{j}-x_{i}\right).$ Now we set $\displaystyle\alpha_{1}=\alpha_{n}=0;$ $\displaystyle\alpha_{i}=f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]\ \ \ \ \ \ \ \ \ \ \text{for all }i\in\left\\{2,3,...,n-1\right\\}.$ Using these notations, the above computation becomes $\displaystyle f\left(x_{j}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\sum_{i=2}^{j-1}\alpha_{i}\cdot\left(x_{j}-x_{i}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\underbrace{0}_{=\alpha_{1}}\cdot\max\left\\{0,x_{j}-x_{1}\right\\}$ $\displaystyle+\sum_{i=2}^{j-1}\alpha_{i}\cdot\underbrace{\left(x_{j}-x_{i}\right)}_{=\max\left\\{0,x_{j}-x_{i}\right\\},\text{ since }x_{j}-x_{i}\geq 0,\text{ as }x_{i}\leq x_{j}}+\sum_{i=j}^{n}\alpha_{i}\cdot\underbrace{0}_{=\max\left\\{0,x_{j}-x_{i}\right\\},\text{ since }x_{j}-x_{i}\leq 0,\text{ as }x_{j}\leq x_{i}}$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\alpha_{1}\cdot\max\left\\{0,x_{j}-x_{1}\right\\}$ $\displaystyle+\sum_{i=2}^{j-1}\alpha_{i}\cdot\max\left\\{0,x_{j}-x_{i}\right\\}+\sum_{i=j}^{n}\alpha_{i}\cdot\max\left\\{0,x_{j}-x_{i}\right\\}$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\sum_{i=1}^{n}\alpha_{i}\cdot\max\left\\{0,x_{j}-x_{i}\right\\}$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\sum_{i=1}^{n}\alpha_{i}\cdot\dfrac{1}{2}\left(\left(x_{j}-x_{i}\right)+\left|x_{j}-x_{i}\right|\right)$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{since }\max\left\\{0,x_{j}-x_{i}\right\\}=\frac{1}{2}\left(\left(x_{j}-x_{i}\right)+\left|x_{j}-x_{i}\right|\right)\text{ by the }\max\left\\{0,x\right\\}\text{ formula}\right)$ $\displaystyle=f\left(x_{1}\right)+f\left[x_{2},x_{1}\right]\cdot\left(x_{j}-x_{1}\right)+\sum_{i=1}^{n}\alpha_{i}\cdot\dfrac{1}{2}\left(x_{j}-x_{i}\right)+\sum_{i=1}^{n}\alpha_{i}\cdot\dfrac{1}{2}\left|x_{j}-x_{i}\right|$ $\displaystyle=f\left(x_{1}\right)+\left(f\left[x_{2},x_{1}\right]x_{j}-f\left[x_{2},x_{1}\right]x_{1}\right)+\left(\dfrac{1}{2}\sum_{i=1}^{n}\alpha_{i}x_{j}-\dfrac{1}{2}\sum_{i=1}^{n}\alpha_{i}x_{i}\right)+\sum_{i=1}^{n}\dfrac{1}{2}\alpha_{i}\left|x_{j}-x_{i}\right|$ $\displaystyle=\left(f\left[x_{2},x_{1}\right]+\dfrac{1}{2}\sum_{i=1}^{n}\alpha_{i}\right)x_{j}+\left(f\left(x_{1}\right)-f\left[x_{2},x_{1}\right]x_{1}-\dfrac{1}{2}\sum_{i=1}^{n}\alpha_{i}x_{i}\right)+\sum_{i=1}^{n}\dfrac{1}{2}\alpha_{i}\left|x_{j}-x_{i}\right|.$ Thus, if we denote $\displaystyle v$ $\displaystyle=f\left[x_{2},x_{1}\right]+\dfrac{1}{2}\sum\limits_{i=1}^{n}\alpha_{i};\ \ \ \ \ \ \ \ \ \ u=f\left(x_{1}\right)-f\left[x_{2},x_{1}\right]x_{1}-\dfrac{1}{2}\sum_{i=1}^{n}\alpha_{i}x_{i};$ $\displaystyle a_{i}$ $\displaystyle=\dfrac{1}{2}\alpha_{i}\ \ \ \ \ \ \ \ \ \ \text{for all }i\in\left\\{1,2,...,n\right\\},$ then we have $f\left(x_{j}\right)=vx_{j}+u+\sum\limits_{i=1}^{n}a_{i}\left|x_{j}-x_{i}\right|.$ Since we have shown this for every $j\in\left\\{1,2,...,n\right\\},$ we can restate this as follows: We have $f\left(t\right)=vt+u+\sum\limits_{i=1}^{n}a_{i}\left|t-x_{i}\right|\text{ for every }t\in\left\\{x_{1},x_{2},...,x_{n}\right\\}.$ Hence, in order for the proof of Theorem 6 to be complete, it is enough to show that the $n$ reals $a_{1},$ $a_{2},$ $...,$ $a_{n}$ are nonnegative. Since $a_{i}=\dfrac{1}{2}\alpha_{i}$ for every $i\in\left\\{1,2,...,n\right\\},$ this will follow once it is proven that the $n$ reals $\alpha_{1},$ $\alpha_{2},$ $...,$ $\alpha_{n}$ are nonnegative. Thus, we have to show that $\alpha_{i}$ is nonnegative for every $i\in\left\\{1,2,...,n\right\\}.$ This is trivial for $i=1$ and for $i=n$ (since $\alpha_{1}=0$ and $\alpha_{n}=0$), so it remains to prove that $\alpha_{i}$ is nonnegative for every $i\in\left\\{2,3,...,n-1\right\\}.$ Now, since $\alpha_{i}=f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]$ for every $i\in\left\\{2,3,...,n-1\right\\},$ we thus have to show that $f\left[x_{i+1},x_{i}\right]-f\left[x_{i},x_{i-1}\right]$ is nonnegative for every $i\in\left\\{2,3,...,n-1\right\\}.$ In other words, we have to prove that $f\left[x_{i+1},x_{i}\right]\geq f\left[x_{i},x_{i-1}\right]$ for every $i\in\left\\{2,3,...,n-1\right\\}.$ But since $x_{i-1}<x_{i}<x_{i+1},$ this follows from the next lemma: > Lemma 7. Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x,$ $y,$ $z$ be three points > from $I$ satisfying $x<y<z.$ Then, $f\left[z,y\right]\geq > f\left[y,x\right].$ Proof of Lemma 7. Since the function $f$ is convex on $I,$ and since $z$ and $x$ are points from $I,$ the definition of convexity yields $\dfrac{\dfrac{1}{z-y}f\left(z\right)+\dfrac{1}{y-x}f\left(x\right)}{\dfrac{1}{z-y}+\dfrac{1}{y-x}}\geq f\left(\dfrac{\dfrac{1}{z-y}z+\dfrac{1}{y-x}x}{\dfrac{1}{z-y}+\dfrac{1}{y-x}}\right)$ (here we have used that $\dfrac{1}{z-y}>0$ and $\dfrac{1}{y-x}>0,$ what is clear from $x<y<z$). Since $\dfrac{\dfrac{1}{z-y}z+\dfrac{1}{y-x}x}{\dfrac{1}{z-y}+\dfrac{1}{y-x}}=y,$ this simplifies to $\displaystyle\dfrac{\dfrac{1}{z-y}f\left(z\right)+\dfrac{1}{y-x}f\left(x\right)}{\dfrac{1}{z-y}+\dfrac{1}{y-x}}$ $\displaystyle\geq f\left(y\right),\ \ \ \ \ \ \ \ \ \ \text{so that}$ $\displaystyle\dfrac{1}{z-y}f\left(z\right)+\dfrac{1}{y-x}f\left(x\right)$ $\displaystyle\geq\left(\dfrac{1}{z-y}+\dfrac{1}{y-x}\right)f\left(y\right),\ \ \ \ \ \ \ \ \ \ \text{so that}$ $\displaystyle\dfrac{1}{z-y}f\left(z\right)-\dfrac{1}{z-y}f\left(y\right)$ $\displaystyle\geq\dfrac{1}{y-x}f\left(y\right)-\dfrac{1}{y-x}f\left(x\right),\ \ \ \ \ \ \ \ \ \ \text{so that}$ $\displaystyle\dfrac{f\left(z\right)-f\left(y\right)}{z-y}$ $\displaystyle\geq\dfrac{f\left(y\right)-f\left(x\right)}{y-x}.$ This becomes $f\left[z,y\right]\geq f\left[y,x\right],$ and thus Lemma 7 is proven. Thus, the proof of Theorem 6 is completed. 3\. The Karamata inequality in symmetric form Now as Theorem 6 is proven, it becomes easy to prove the Karamata inequality in the following form: > Theorem 8a, the Karamata inequality in symmetric form. Let $f$ be a convex > function from an interval $I\subseteq\mathbb{R}$ to $\mathbb{R},$ and let > $n$ be a positive integer. Let $x_{1},$ $x_{2},$ $...,$ $x_{n},$ $y_{1},$ > $y_{2},$ $...,$ $y_{n}$ be $2n$ points from $I.$ Assume that > > > $\left|x_{1}-t\right|+\left|x_{2}-t\right|+...+\left|x_{n}-t\right|\geq\left|y_{1}-t\right|+\left|y_{2}-t\right|+...+\left|y_{n}-t\right|$ > > holds for every > $t\in\left\\{x_{1},x_{2},...,x_{n},y_{1},y_{2},...,y_{n}\right\\}.$ Then, > > $f\left(x_{1}\right)+f\left(x_{2}\right)+...+f\left(x_{n}\right)\geq > f\left(y_{1}\right)+f\left(y_{2}\right)+...+f\left(y_{n}\right).$ We will not need this result, but we will rather use its weighted version: > Theorem 8b, the weighted Karamata inequality in symmetric form. Let $f$ be a > convex function from an interval $I\subseteq\mathbb{R}$ to $\mathbb{R},$ and > let $N$ be a positive integer. Let $z_{1},$ $z_{2},$ $...,$ $z_{N}$ be $N$ > points from $I,$ and let $w_{1},$ $w_{2},$ $...,$ $w_{N}$ be $N$ reals. > Assume that > > $\sum_{k=1}^{N}w_{k}=0,$ (1) > > and that > > $\sum_{k=1}^{N}w_{k}\left|z_{k}-t\right|\geq 0\text{ holds for every > }t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}.$ (2) > > Then, > > $\sum_{k=1}^{N}w_{k}f\left(z_{k}\right)\geq 0.$ (3) It is very easy to conclude Theorem 8a from Theorem 8b by setting $N=2n$ and $\displaystyle z_{1}=x_{1},\ \ \ \ \ \ \ \ \ \ z_{2}=x_{2},\ \ \ \ \ \ \ \ \ \ ...,\ \ \ \ \ \ \ \ \ \ z_{n}=x_{n};$ $\displaystyle z_{n+1}=y_{1},\ \ \ \ \ \ \ \ \ \ z_{n+2}=y_{2},\ \ \ \ \ \ \ \ \ \ ...,\ \ \ \ \ \ \ \ \ \ z_{2n}=y_{n};$ $\displaystyle w_{1}=w_{2}=...=w_{n}=1;\ \ \ \ \ \ \ \ \ \ w_{n+1}=w_{n+2}=...=w_{2n}=-1,$ but as I said, we will never use Theorem 8a in this paper. Time for a remark to readers familiar with majorization theory. One may wonder why I call the two results above ”Karamata inequalities”. In fact, the Karamata inequality in its most known form claims: > Theorem 9, the Karamata inequality. Let $f$ be a convex function from an > interval $I\subseteq\mathbb{R}$ to $\mathbb{R},$ and let $n$ be a positive > integer. Let $x_{1},$ $x_{2},$ $...,$ $x_{n},$ $y_{1},$ $y_{2},$ $...,$ > $y_{n}$ be $2n$ points from $I$ such that > $\left(x_{1},x_{2},...,x_{n}\right)\succ\left(y_{1},y_{2},...,y_{n}\right).$ > Then, > > $f\left(x_{1}\right)+f\left(x_{2}\right)+...+f\left(x_{n}\right)\geq > f\left(y_{1}\right)+f\left(y_{2}\right)+...+f\left(y_{n}\right).$ According to [2], post #11, Lemma 1, the condition $\left(x_{1},x_{2},...,x_{n}\right)\succ\left(y_{1},y_{2},...,y_{n}\right)$ yields that $\left|x_{1}-t\right|+\left|x_{2}-t\right|+...+\left|x_{n}-t\right|\geq\left|y_{1}-t\right|+\left|y_{2}-t\right|+...+\left|y_{n}-t\right|$ holds for every real $t$ \- and thus, in particular, for every $t\in\left\\{z_{1},z_{2},...,z_{n}\right\\}.$ Hence, whenever the condition of Theorem 9 holds, the condition of Theorem 8a holds as well. Thus, Theorem 9 follows from Theorem 8a. With just a little more work, we could also derive Theorem 8a from Theorem 9, so that Theorems 8a and 9 are equivalent. Note that Theorem 8b is more general than the Fuchs inequality (a more well- known weighted version of the Karamata inequality). See [5] for a generalization of majorization theory to weighted families of points (apparently already known long time ago), with a different approach to this fact. As promised, here is a proof of Theorem 8b: First, substituting $t=\max\left\\{z_{1},z_{2},...,z_{N}\right\\}$ into (2) (it is clear that this $t$ satisfies $t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}$), we get $z_{k}\leq t$ for every $k\in\left\\{1,2,...,N\right\\},$ so that $z_{k}-t\leq 0$ and thus $\left|z_{k}-t\right|=-\left(z_{k}-t\right)=t-z_{k}$ for every $k\in\left\\{1,2,...,N\right\\},$ and thus (2) becomes $\sum\limits_{k=1}^{N}w_{k}\left(t-z_{k}\right)\geq 0.$ In other words, $t\sum\limits_{k=1}^{N}w_{k}-\sum\limits_{k=1}^{N}w_{k}z_{k}\geq 0.$ In sight of $\sum\limits_{k=1}^{N}w_{k}=0,$ this rewrites as $t\cdot 0-\sum\limits_{k=1}^{N}w_{k}z_{k}\geq 0.$ Hence, $\sum\limits_{k=1}^{N}w_{k}z_{k}\leq 0.$ Similarly, substituting $t=\min\left\\{z_{1},z_{2},...,z_{N}\right\\}$ into (2), we get $\sum\limits_{k=1}^{N}w_{k}z_{k}\geq 0.$ Thus, $\sum\limits_{k=1}^{N}w_{k}z_{k}=0.$ The function $f:I\rightarrow\mathbb{R}$ is convex, and $z_{1},$ $z_{2},$ $...,$ $z_{N}$ are finitely many points from $I.$ Hence, Theorem 6 yields the existence of two real constants $u$ and $v$ and $N$ nonnegative constants $a_{1},$ $a_{2},$ $...,$ $a_{N}$ such that $f\left(t\right)=vt+u+\sum\limits_{i=1}^{N}a_{i}\left|t-z_{i}\right|\text{ holds for every }t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}.$ Thus, $f\left(z_{k}\right)=vz_{k}+u+\sum\limits_{i=1}^{N}a_{i}\left|z_{k}-z_{i}\right|\ \ \ \ \ \ \ \ \ \ \text{for every }k\in\left\\{1,2,...,N\right\\}.$ Hence, $\displaystyle\sum_{k=1}^{N}w_{k}f\left(z_{k}\right)$ $\displaystyle=\sum_{k=1}^{N}w_{k}\left(vz_{k}+u+\sum\limits_{i=1}^{N}a_{i}\left|z_{k}-z_{i}\right|\right)=v\underbrace{\sum_{k=1}^{N}w_{k}z_{k}}_{=0}+u\underbrace{\sum_{k=1}^{N}w_{k}}_{=0}+\sum_{k=1}^{N}w_{k}\sum\limits_{i=1}^{N}a_{i}\left|z_{k}-z_{i}\right|$ $\displaystyle=\sum_{k=1}^{N}w_{k}\sum\limits_{i=1}^{N}a_{i}\left|z_{k}-z_{i}\right|=\sum\limits_{i=1}^{N}a_{i}\underbrace{\sum_{k=1}^{N}w_{k}\left|z_{k}-z_{i}\right|}_{\geq 0\text{ according to (2) for }t=z_{i}}\geq 0.$ Thus, Theorem 8b is proven. 4\. A property of zero-sum vectors Next, we are going to show some properties of real vectors. If $k$ is an integer and $v\in\mathbb{R}^{k}$ is a vector, then, for any $i\in\left\\{1,2,...,k\right\\},$ we denote by $v_{i}$ the $i$-th coordinate of the vector $v.$ Then, $v=\left(\begin{array}[]{c}v_{1}\\\ v_{2}\\\ ...\\\ v_{k}\end{array}\right).$ Let $n$ be a positive integer. We consider the vector space $\mathbb{R}^{n}.$ Let $\left(e_{1},e_{2},...,e_{n}\right)$ be the standard basis of this vector space $\mathbb{R}^{n};$ in other words, for every $i\in\left\\{1,2,...,n\right\\},$ let $e_{i}$ be the vector from $\mathbb{R}^{n}$ such that $\left(e_{i}\right)_{i}=1$ and $\left(e_{i}\right)_{j}=0$ for every $j\in\left\\{1,2,...,n\right\\}\setminus\left\\{i\right\\}.$ Let $V_{n}$ be the subspace of $\mathbb{R}^{n}$ defined by $V_{n}=\left\\{x\in\mathbb{R}^{n}\ \mid\ x_{1}+x_{2}+...+x_{n}=0\right\\}.$ For any $u\in\left\\{1,2,...,n\right\\}$ and any two distinct numbers $i$ and $j$ from the set $\left\\{1,2,...,n\right\\},$ we have $\left(e_{i}-e_{j}\right)_{u}=\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ -1,\text{ if }u=j;\\\ 0,\text{ if }u\neq i\text{ and }u\neq j\end{array}\right..$ (4) Clearly, $e_{i}-e_{j}\in V_{n}$ for any two numbers $i$ and $j$ from the set $\left\\{1,2,...,n\right\\}.$ For any vector $t\in\mathbb{R}^{n},$ we denote $I\left(t\right)=\left\\{k\in\left\\{1,2,...,n\right\\}\mid t_{k}>0\right\\}$ and $J\left(t\right)=\left\\{k\in\left\\{1,2,...,n\right\\}\mid t_{k}<0\right\\}.$ Obviously, for every $t\in\mathbb{R}^{n},$ the sets $I\left(t\right)$ and $J\left(t\right)$ are disjoint. Now we are going to show: > Theorem 10. Let $n$ be a positive integer. Let $x\in V_{n}$ be a vector. > Then, there exist nonnegative reals $a_{i,j}$ for all pairs > $\left(i,j\right)\in I\left(x\right)\times J\left(x\right)$ such that > > $x=\sum_{\left(i,j\right)\in I\left(x\right)\times > J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right).$ Proof of Theorem 10. We will prove Theorem 10 by induction over $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|.$ The basis of the induction \- the case when $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|=0$ \- is trivial: If $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|=0,$ then $I\left(x\right)=J\left(x\right)=\varnothing$ and $x=0,$ so that $x=\sum\limits_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)$ holds because both sides of this equation are $0.$ Now we come to the induction step: Let $r$ be a positive integer. Assume that Theorem 10 holds for all $x\in V_{n}$ with $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|<r.$ We have to show that Theorem 10 holds for all $x\in V_{n}$ with $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|=r.$ In order to prove this, we let $z\in V_{n}$ be an arbitrary vector with $\left|I\left(z\right)\right|+\left|J\left(z\right)\right|=r.$ We then have to prove that Theorem 10 holds for $x=z.$ In other words, we have to show that there exist nonnegative reals $a_{i,j}$ for all pairs $\left(i,j\right)\in I\left(z\right)\times J\left(z\right)$ such that $z=\sum_{\left(i,j\right)\in I\left(z\right)\times J\left(z\right)}a_{i,j}\left(e_{i}-e_{j}\right).$ (5) First, $\left|I\left(z\right)\right|+\left|J\left(z\right)\right|=r$ and $r>0$ yield $\left|I\left(z\right)\right|+\left|J\left(z\right)\right|>0.$ Hence, at least one of the sets $I\left(z\right)$ and $J\left(z\right)$ is non-empty. Now, since $z\in V_{n},$ we have $z_{1}+z_{2}+...+z_{n}=0.$ Hence, either $z_{k}=0$ for every $k\in\left\\{1,2,...,n\right\\},$ or there is at least one positive number and at least one negative number in the set $\left\\{z_{1},z_{2},...,z_{n}\right\\}.$ The first case is impossible (since at least one of the sets $I\left(z\right)$ and $J\left(z\right)$ is non- empty). Thus, the second case must hold - i. e., there is at least one positive number and at least one negative number in the set $\left\\{z_{1},z_{2},...,z_{n}\right\\}.$ In other words, there exists a number $u\in\left\\{1,2,...,n\right\\}$ such that $z_{u}>0,$ and a number $v\in\left\\{1,2,...,n\right\\}$ such that $z_{v}<0.$ Of course, $z_{u}>0$ yields $u\in I\left(z\right),$ and $z_{v}<0$ yields $v\in J\left(z\right).$ Needless to say that $u\neq v.$ Now, we distinguish between two cases: the first case will be the case when $z_{u}+z_{v}\geq 0,$ and the second case will be the case when $z_{u}+z_{v}\leq 0.$ Let us consider the first case: In this case, $z_{u}+z_{v}\geq 0.$ Then, let $z^{\prime}=z+z_{v}\left(e_{u}-e_{v}\right).$ Since $z\in V_{n}$ and $e_{u}-e_{v}\in V_{n},$ we have $z+z_{v}\left(e_{u}-e_{v}\right)\in V_{n}$ (since $V_{n}$ is a vector space), so that $z^{\prime}\in V_{n}.$ From $z^{\prime}=z+z_{v}\left(e_{u}-e_{v}\right),$ the coordinate representation of the vector $z^{\prime}$ is easily obtained: $z^{\prime}=\left(\begin{array}[]{c}z_{1}^{\prime}\\\ z_{2}^{\prime}\\\ ...\\\ z_{n}^{\prime}\end{array}\right),\ \ \ \ \ \ \ \ \ \ \text{where }\left\\{\begin{array}[]{c}z_{k}^{\prime}=z_{k}\text{ for all }k\in\left\\{1,2,...,n\right\\}\setminus\left\\{u,v\right\\};\\\ z_{u}^{\prime}=z_{u}+z_{v};\\\ z_{v}^{\prime}=0\end{array}\right..$ It is readily seen from this that $I\left(z^{\prime}\right)\subseteq I\left(z\right)$, so that $\left|I\left(z^{\prime}\right)\right|\leq\left|I\left(z\right)\right|.$ Besides, $J\left(z^{\prime}\right)\subseteq J\left(z\right).$ Moreover, $J\left(z^{\prime}\right)$ is a proper subset of $J\left(z\right),$ because $v\notin J\left(z^{\prime}\right)$ (since $z_{v}^{\prime}$ is not $<0,$ but $=0$) but $v\in J\left(z\right).$ Hence, $\left|J\left(z^{\prime}\right)\right|<\left|J\left(z\right)\right|.$ Combined with $\left|I\left(z^{\prime}\right)\right|\leq\left|I\left(z\right)\right|,$ this yields $\left|I\left(z^{\prime}\right)\right|+\left|J\left(z^{\prime}\right)\right|<\left|I\left(z\right)\right|+\left|J\left(z\right)\right|.$ In view of $\left|I\left(z\right)\right|+\left|J\left(z\right)\right|=r,$ this becomes $\left|I\left(z^{\prime}\right)\right|+\left|J\left(z^{\prime}\right)\right|<r.$ Thus, since we have assumed that Theorem 10 holds for all $x\in V_{n}$ with $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|<r,$ we can apply Theorem 10 to $x=z^{\prime},$ and we see that there exist nonnegative reals $a_{i,j}^{\prime}$ for all pairs $\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)$ such that $z^{\prime}=\sum_{\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)}a_{i,j}^{\prime}\left(e_{i}-e_{j}\right).$ Now, $z^{\prime}=z+z_{v}\left(e_{u}-e_{v}\right)$ yields $z=z^{\prime}-z_{v}\left(e_{u}-e_{v}\right).$ Since $z_{v}<0,$ we have $-z_{v}>0,$ so that, particularly, $-z_{v}$ is nonnegative. Since $I\left(z^{\prime}\right)\subseteq I\left(z\right)$ and $J\left(z^{\prime}\right)\subseteq J\left(z\right),$ we have $I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)\subseteq I\left(z\right)\times J\left(z\right).$ Also, $\left(u,v\right)\in I\left(z\right)\times J\left(z\right)$ (because $u\in I\left(z\right)$ and $v\in J\left(z\right)$) and $\left(u,v\right)\notin I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)$ (because $v\notin J\left(z^{\prime}\right)$). Hence, the sets $I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)$ and $\left\\{\left(u,v\right)\right\\}$ are two disjoint subsets of the set $I\left(z\right)\times J\left(z\right).$ We can thus define nonnegative reals $a_{i,j}$ for all pairs $\left(i,j\right)\in I\left(z\right)\times J\left(z\right)$ by setting $a_{i,j}=\left\\{\begin{array}[]{c}a_{i,j}^{\prime},\text{ if }\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right);\\\ -z_{v},\text{ if }\left(i,j\right)=\left(u,v\right);\\\ 0,\text{ if neither of the two cases above holds}\end{array}\right.$ (these $a_{i,j}$ are all nonnegative because $a_{i,j}^{\prime},$ $-z_{v}$ and $0$ are nonnegative). Then, $\displaystyle\sum_{\left(i,j\right)\in I\left(z\right)\times J\left(z\right)}a_{i,j}\left(e_{i}-e_{j}\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)}a_{i,j}^{\prime}\left(e_{i}-e_{j}\right)+\sum_{\left(i,j\right)=\left(u,v\right)}\left(-z_{v}\right)\left(e_{i}-e_{j}\right)+\left(\text{sum of some }0\text{'s}\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)}a_{i,j}^{\prime}\left(e_{i}-e_{j}\right)+\left(-z_{v}\right)\left(e_{u}-e_{v}\right)+0=z^{\prime}+\left(-z_{v}\right)\left(e_{u}-e_{v}\right)+0$ $\displaystyle=\left(z+z_{v}\left(e_{u}-e_{v}\right)\right)+\left(-z_{v}\right)\left(e_{u}-e_{v}\right)+0=z.$ Thus, (5) is fulfilled. Similarly, we can fulfill (5) in the second case, repeating the arguments we have done for the first case while occasionally interchanging $u$ with $v,$ as well as $I$ with $J,$ as well as $<$ with $>$. Here is a brief outline of how we have to proceed in the second case: Denote $z^{\prime}=z-z_{u}\left(e_{u}-e_{v}\right).$ Show that $z^{\prime}\in V_{n}$ (as in the first case). Notice that $z^{\prime}=\left(\begin{array}[]{c}z_{1}^{\prime}\\\ z_{2}^{\prime}\\\ ...\\\ z_{n}^{\prime}\end{array}\right),\ \ \ \ \ \ \ \ \ \ \text{where }\left\\{\begin{array}[]{c}z_{k}^{\prime}=z_{k}\text{ for all }k\in\left\\{1,2,...,n\right\\}\setminus\left\\{u,v\right\\};\\\ z_{u}^{\prime}=0;\\\ z_{v}^{\prime}=z_{u}+z_{v}\end{array}\right..$ Prove that $u\notin I\left(z^{\prime}\right)$ (as we proved $v\notin J\left(z^{\prime}\right)$ in the first case). Prove that $J\left(z^{\prime}\right)\subseteq J\left(z\right)$ (similarly to the proof of $I\left(z^{\prime}\right)\subseteq I\left(z\right)$ in the first case) and that $I\left(z^{\prime}\right)$ is a proper subset of $I\left(z\right)$ (similarly to the proof that $J\left(z^{\prime}\right)$ is a proper subset of $J\left(z\right)$ in the first case). Show that there exist nonnegative reals $a_{i,j}^{\prime}$ for all pairs $\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)$ such that $z^{\prime}=\sum_{\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)}a_{i,j}^{\prime}\left(e_{i}-e_{j}\right)$ (as in the first case). Note that $z_{u}$ is nonnegative (since $z_{u}>0$). Prove that the sets $I\left(z^{\prime}\right)\times J\left(z^{\prime}\right)$ and $\left\\{\left(u,v\right)\right\\}$ are two disjoint subsets of the set $I\left(z\right)\times J\left(z\right)$ (as in the first case). Define nonnegative reals $a_{i,j}$ for all pairs $\left(i,j\right)\in I\left(z\right)\times J\left(z\right)$ by setting $a_{i,j}=\left\\{\begin{array}[]{c}a_{i,j}^{\prime},\text{ if }\left(i,j\right)\in I\left(z^{\prime}\right)\times J\left(z^{\prime}\right);\\\ z_{u},\text{ if }\left(i,j\right)=\left(u,v\right);\\\ 0,\text{ if neither of the two cases above holds}\end{array}\right..$ Prove that these nonnegative reals $a_{i,j}$ fulfill (5). Thus, in each of the two cases, we have proven that there exist nonnegative reals $a_{i,j}$ for all pairs $\left(i,j\right)\in I\left(z\right)\times J\left(z\right)$ such that (5) holds. Hence, Theorem 10 holds for $x=z.$ Thus, Theorem 10 is proven for all $x\in V_{n}$ with $\left|I\left(x\right)\right|+\left|J\left(x\right)\right|=r.$ This completes the induction step, and therefore, Theorem 10 is proven. As an application of Theorem 10, we can now show: > Theorem 11. Let $n$ be a positive integer. Let $a_{1},$ $a_{2},$ $...,$ > $a_{n}$ be $n$ nonnegative reals. Let $S$ be a finite set. For every $s\in > S,$ let $r_{s}$ be an element of $\left(\mathbb{R}^{n}\right)^{\ast}$ (in > other words, a linear transformation from $\mathbb{R}^{n}$ to $\mathbb{R}$), > and let $b_{s}$ be a nonnegative real. Define a function > $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ by > > $f\left(x\right)=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in > S}b_{s}\left|r_{s}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where > }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ > x_{n}\end{array}\right)\in\mathbb{R}^{n}.$ > > Then, the following two assertions are equivalent: > > Assertion $\mathcal{A}_{1}$: We have $f\left(x\right)\geq 0$ for every $x\in > V_{n}.$ > > Assertion $\mathcal{A}_{2}$: We have $f\left(e_{i}-e_{j}\right)\geq 0$ for > any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\}.$ Proof of Theorem 11. We have to prove that the assertions $\mathcal{A}_{1}$ and $\mathcal{A}_{2}$ are equivalent. In other words, we have to prove that $\mathcal{A}_{1}\Longrightarrow\mathcal{A}_{2}$ and $\mathcal{A}_{2}\Longrightarrow\mathcal{A}_{1}.$ Actually, $\mathcal{A}_{1}\Longrightarrow\mathcal{A}_{2}$ is trivial (we just have to use that $e_{i}-e_{j}\in V_{n}$ for any two numbers $i$ and $j$ from $\left\\{1,2,...,n\right\\}$). It remains to show that $\mathcal{A}_{2}\Longrightarrow\mathcal{A}_{1}.$ So assume that Assertion $\mathcal{A}_{2}$ is valid, i. e. we have $f\left(e_{i}-e_{j}\right)\geq 0$ for any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\}.$ We have to prove that Assertion $\mathcal{A}_{1}$ holds, i. e. that $f\left(x\right)\geq 0$ for every $x\in V_{n}.$ So let $x\in V_{n}$ be some vector. According to Theorem 10, there exist nonnegative reals $a_{i,j}$ for all pairs $\left(i,j\right)\in I\left(x\right)\times J\left(x\right)$ such that $x=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right).$ We will now show that $\left|x_{u}\right|=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|\left(e_{i}-e_{j}\right)_{u}\right|\ \ \ \ \ \ \ \ \ \ \text{for every }u\in\left\\{1,2,...,n\right\\}.$ (6) Here, of course, $\left(e_{i}-e_{j}\right)_{u}$ means the $u$-th coordinate of the vector $e_{i}-e_{j}.$ In fact, two cases are possible: the case when $x_{u}\geq 0,$ and the case when $x_{u}<0.$ We will consider these cases separately. Case 1: We have $x_{u}\geq 0.$ Then, $\left|x_{u}\right|=x_{u}.$ Hence, in this case, we have $\left(e_{i}-e_{j}\right)_{u}\geq 0$ for any two numbers $i\in I\left(x\right)$ and $j\in J\left(x\right)$ (in fact, $j\in J\left(x\right)$ yields $x_{j}<0,$ so that $u\neq j$ (because $x_{j}<0$ and $x_{u}\geq 0$) and thus $\left(e_{j}\right)_{u}=0,$ so that $\left(e_{i}-e_{j}\right)_{u}=\left(e_{i}\right)_{u}-\left(e_{j}\right)_{u}=\left(e_{i}\right)_{u}-0=\left(e_{i}\right)_{u}=\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ 0,\text{ if }u\neq i\end{array}\right.\geq 0$). Thus, $\left(e_{i}-e_{j}\right)_{u}=\left|\left(e_{i}-e_{j}\right)_{u}\right|$ for any two numbers $i\in I\left(x\right)$ and $j\in J\left(x\right).$ Thus, $\displaystyle\left|x_{u}\right|$ $\displaystyle=x_{u}=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)_{u}\ \ \ \ \ \ \ \ \ \ \left(\text{since }x=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|\left(e_{i}-e_{j}\right)_{u}\right|,$ and (6) is proven. Case 2: We have $x_{u}<0.$ Then, $u\in J\left(x\right)$ and $\left|x_{u}\right|=-x_{u}.$ Hence, in this case, we have $\left(e_{i}-e_{j}\right)_{u}\leq 0$ for any two numbers $i\in I\left(x\right)$ and $j\in J\left(x\right)$ (in fact, $i\in I\left(x\right)$ yields $x_{i}>0,$ so that $u\neq i$ (because $x_{i}>0$ and $x_{u}<0$) and thus $\left(e_{i}\right)_{u}=0,$ so that $\left(e_{i}-e_{j}\right)_{u}=\left(e_{i}\right)_{u}-\left(e_{j}\right)_{u}=0-\left(e_{j}\right)_{u}=-\left(e_{j}\right)_{u}=-\left\\{\begin{array}[]{c}1,\text{ if }u=j;\\\ 0,\text{ if }u\neq j\end{array}\right.\leq 0$). Thus, $-\left(e_{i}-e_{j}\right)_{u}=\left|\left(e_{i}-e_{j}\right)_{u}\right|$ for any two numbers $i\in I\left(x\right)$ and $j\in J\left(x\right).$ Thus, $\displaystyle\left|x_{u}\right|$ $\displaystyle=-x_{u}=-\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)_{u}\ \ \ \ \ \ \ \ \ \ \left(\text{since }x=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(-\left(e_{i}-e_{j}\right)_{u}\right)=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|\left(e_{i}-e_{j}\right)_{u}\right|,$ and (6) is proven. Hence, in both cases, (6) is proven. Thus, (6) always holds. Now let us continue our proof of $\mathcal{A}_{2}\Longrightarrow\mathcal{A}_{1}$: We have $\displaystyle\sum_{s\in S}b_{s}\left|r_{s}x\right|$ $\displaystyle=\sum_{s\in S}b_{s}\left|r_{s}\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)\right|\ \ \ \ \ \ \ \ \ \ \left(\text{since }x=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left(e_{i}-e_{j}\right)\right)$ $\displaystyle=\sum_{s\in S}b_{s}\left|\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}r_{s}\left(e_{i}-e_{j}\right)\right|$ $\displaystyle\leq\sum_{s\in S}b_{s}\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|r_{s}\left(e_{i}-e_{j}\right)\right|$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{by the triangle inequality, since all }a_{i,j}\text{ and all }b_{s}\text{ are nonnegative}\right).$ Thus, $\displaystyle f\left(x\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}x\right|\geq\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in S}b_{s}\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|r_{s}\left(e_{i}-e_{j}\right)\right|$ $\displaystyle=\sum_{u=1}^{n}a_{u}\cdot\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|\left(e_{i}-e_{j}\right)_{u}\right|-\sum_{s\in S}b_{s}\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\left|r_{s}\left(e_{i}-e_{j}\right)\right|\ \ \ \ \ \ \ \ \ \ \left(\text{by (6)}\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\sum_{u=1}^{n}a_{u}\left|\left(e_{i}-e_{j}\right)_{u}\right|-\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\sum_{s\in S}b_{s}\left|r_{s}\left(e_{i}-e_{j}\right)\right|$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}a_{i,j}\cdot\left(\sum_{u=1}^{n}a_{u}\left|\left(e_{i}-e_{j}\right)_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\left(e_{i}-e_{j}\right)\right|\right)$ $\displaystyle=\sum_{\left(i,j\right)\in I\left(x\right)\times J\left(x\right)}\underbrace{a_{i,j}}_{\geq 0}\cdot\underbrace{f\left(e_{i}-e_{j}\right)}_{\geq 0}\geq 0.$ (Here, $f\left(e_{i}-e_{j}\right)\geq 0$ because $i$ and $j$ are two distinct integers from $\left\\{1,2,...,n\right\\};$ in fact, $i$ and $j$ are distinct because $i\in I\left(x\right)$ and $j\in J\left(x\right),$ and the sets $I\left(x\right)$ and $J\left(x\right)$ are disjoint.) Hence, we have obtained $f\left(x\right)\geq 0.$ This proves the assertion $\mathcal{A}_{1}.$ Therefore, the implication $\mathcal{A}_{2}\Longrightarrow\mathcal{A}_{1}$ is proven, and the proof of Theorem 11 is complete. 5\. Restating Theorem 11 Now we consider a result which follows from Theorem 11 pretty obviously (although the formalization of the proof is going to be gruelling): > Theorem 12. Let $n$ be a nonnegative integer. Let $a_{1},$ $a_{2},$ $...,$ > $a_{n}$ and $a$ be $n+1$ nonnegative reals. Let $S$ be a finite set. For > every $s\in S,$ let $r_{s}$ be an element of > $\left(\mathbb{R}^{n}\right)^{\ast}$ (in other words, a linear > transformation from $\mathbb{R}^{n}$ to $\mathbb{R}$), and let $b_{s}$ be a > nonnegative real. Define a function $g:\mathbb{R}^{n}\rightarrow\mathbb{R}$ > by > > > $g\left(x\right)=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|+a\left|x_{1}+x_{2}+...+x_{n}\right|-\sum_{s\in > S}b_{s}\left|r_{s}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where > }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ > x_{n}\end{array}\right)\in\mathbb{R}^{n}.$ > > Then, the following two assertions are equivalent: > > Assertion $\mathcal{B}_{1}$: We have $g\left(x\right)\geq 0$ for every > $x\in\mathbb{R}^{n}.$ > > Assertion $\mathcal{B}_{2}$: We have $g\left(e_{i}\right)\geq 0$ for every > integer $i\in\left\\{1,2,...,n\right\\},$ and $g\left(e_{i}-e_{j}\right)\geq > 0$ for any two distinct integers $i$ and $j$ from > $\left\\{1,2,...,n\right\\}.$ Proof of Theorem 12. We are going to restate Theorem 12 before we actually prove it. But first, we introduce a notation: Let $\left(\widetilde{e_{1}},\widetilde{e_{2}},...,\widetilde{e_{n-1}}\right)$ be the standard basis of the vector space $\mathbb{R}^{n-1};$ in other words, for every $i\in\left\\{1,2,...,n-1\right\\},$ let $\widetilde{e_{i}}$ be the vector from $\mathbb{R}^{n-1}$ such that $\left(\widetilde{e_{i}}\right)_{i}=1$ and $\left(\widetilde{e_{i}}\right)_{j}=0$ for every $j\in\left\\{1,2,...,n-1\right\\}\setminus\left\\{i\right\\}.$ Now we will restate Theorem 12 by renaming $n$ into $n-1$ (thus replacing $e_{i}$ by $\widetilde{e_{i}}$ as well) and $a$ into $a_{n}$: > Theorem 12b. Let $n$ be a positive integer. Let $a_{1},$ $a_{2},$ $...,$ > $a_{n-1},$ $a_{n}$ be $n$ nonnegative reals. Let $S$ be a finite set. For > every $s\in S,$ let $r_{s}$ be an element of > $\left(\mathbb{R}^{n-1}\right)^{\ast}$ (in other words, a linear > transformation from $\mathbb{R}^{n-1}$ to $\mathbb{R}$), and let $b_{s}$ be > a nonnegative real. Define a function > $g:\mathbb{R}^{n-1}\rightarrow\mathbb{R}$ by > > > $g\left(x\right)=\sum_{u=1}^{n-1}a_{u}\left|x_{u}\right|+a_{n}\left|x_{1}+x_{2}+...+x_{n-1}\right|-\sum_{s\in > S}b_{s}\left|r_{s}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where > }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ > x_{n-1}\end{array}\right)\in\mathbb{R}^{n-1}.$ > > Then, the following two assertions are equivalent: > > Assertion $\mathcal{C}_{1}$: We have $g\left(x\right)\geq 0$ for every > $x\in\mathbb{R}^{n-1}.$ > > Assertion $\mathcal{C}_{2}$: We have $g\left(\widetilde{e_{i}}\right)\geq 0$ > for every integer $i\in\left\\{1,2,...,n-1\right\\},$ and > $g\left(\widetilde{e_{i}}-\widetilde{e_{j}}\right)\geq 0$ for any two > distinct integers $i$ and $j$ from $\left\\{1,2,...,n-1\right\\}.$ Theorem 12b is equivalent to Theorem 12 (because Theorem 12b is just Theorem 12, applied to $n-1$ instead of $n$). Thus, proving Theorem 12b will be enough to verify Theorem 12. Proof of Theorem 12b. The implication $\mathcal{C}_{1}\Longrightarrow\mathcal{C}_{2}$ is absolutely trivial. Hence, in order to establish Theorem 12b, it only remains to prove the implication $\mathcal{C}_{2}\Longrightarrow\mathcal{C}_{1}.$ So assume that the assertion $\mathcal{C}_{2}$ holds, i. e. that we have $g\left(\widetilde{e_{i}}\right)\geq 0$ for every integer $i\in\left\\{1,2,...,n-1\right\\},$ and $g\left(\widetilde{e_{i}}-\widetilde{e_{j}}\right)\geq 0$ for any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n-1\right\\}.$ We want to show that Assertion $\mathcal{C}_{1}$ holds, i. e. that $g\left(x\right)\geq 0$ is satisfied for every $x\in\mathbb{R}^{n-1}.$ Since $\left(\widetilde{e_{1}},\widetilde{e_{2}},...,\widetilde{e_{n-1}}\right)$ is the standard basis of the vector space $\mathbb{R}^{n-1},$ every vector $x\in\mathbb{R}^{n-1}$ satisfies $x=\sum\limits_{i=1}^{n-1}x_{i}\widetilde{e_{i}}.$ Since $\left(e_{1},e_{2},...,e_{n}\right)$ is the standard basis of the vector space $\mathbb{R}^{n},$ every vector $x\in\mathbb{R}^{n}$ satisfies $x=\sum\limits_{i=1}^{n}x_{i}e_{i}.$ Let $\phi_{n}:\mathbb{R}^{n-1}\rightarrow\mathbb{R}^{n}$ be the linear transformation defined by $\phi_{n}\widetilde{e_{i}}=e_{i}-e_{n}$ for every $i\in\left\\{1,2,...,n-1\right\\}.$ (This linear transformation is uniquely defined this way because $\left(\widetilde{e_{1}},\widetilde{e_{2}},...,\widetilde{e_{n-1}}\right)$ is a basis of $\mathbb{R}^{n-1}.$) For every $x\in\mathbb{R}^{n-1},$ we then have $\displaystyle\phi_{n}x$ $\displaystyle=\phi_{n}\left(\sum\limits_{i=1}^{n-1}x_{i}\widetilde{e_{i}}\right)=\sum\limits_{i=1}^{n-1}x_{i}\phi_{n}\widetilde{e_{i}}\ \ \ \ \ \ \ \ \ \ \left(\text{since }\phi_{n}\text{ is linear}\right)$ $\displaystyle=\sum\limits_{i=1}^{n-1}x_{i}\left(e_{i}-e_{n}\right)=\sum\limits_{i=1}^{n-1}x_{i}e_{i}-\sum\limits_{i=1}^{n-1}x_{i}e_{n}=\sum\limits_{i=1}^{n-1}x_{i}e_{i}-\left(x_{1}+x_{2}+...+x_{n-1}\right)e_{n}$ $\displaystyle=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ x_{n-1}\\\ -\left(x_{1}+x_{2}+...+x_{n-1}\right)\end{array}\right),$ (12) Consequently, $\phi_{n}x\in V_{n}$ for every $x\in\mathbb{R}^{n-1}$. Hence, $\mathop{\mathrm{I}m}\phi_{n}\subseteq V_{n}.$ Let $\psi_{n}:\mathbb{R}^{n}\rightarrow\mathbb{R}^{n-1}$ be the linear transformation defined by $\psi_{n}e_{i}=\left\\{\begin{array}[]{c}\widetilde{e_{i}},\text{ if }i\in\left\\{1,2,...,n-1\right\\};\\\ 0,\text{ if }i=n\end{array}\right.$ for every $i\in\left\\{1,2,...,n\right\\}.$ (This linear transformation is uniquely defined this way because $\left(e_{1},e_{2},...,e_{n}\right)$ is a basis of $\mathbb{R}^{n}.$) For every $x\in\mathbb{R}^{n},$ we then have $\displaystyle\psi_{n}x$ $\displaystyle=\psi_{n}\left(\sum\limits_{i=1}^{n}x_{i}e_{i}\right)=\sum\limits_{i=1}^{n}x_{i}\psi_{n}e_{i}\ \ \ \ \ \ \ \ \ \ \left(\text{since }\psi_{n}\text{ is linear}\right)$ $\displaystyle=\sum\limits_{i=1}^{n}x_{i}\left\\{\begin{array}[]{c}\widetilde{e_{i}},\text{ if }i\in\left\\{1,2,...,n-1\right\\};\\\ 0,\text{ if }i=n\end{array}\right.$ $\displaystyle=\sum_{i=1}^{n-1}x_{i}\widetilde{e_{i}}=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ x_{n-1}\end{array}\right).$ Then, $\psi_{n}\phi_{n}=\mathop{\mathrm{i}d}$ (in fact, for every $i\in\left\\{1,2,...,n-1\right\\},$ we have $\displaystyle\psi_{n}\phi_{n}\widetilde{e_{i}}$ $\displaystyle=\psi_{n}\left(e_{i}-e_{n}\right)=\psi_{n}e_{i}-\psi_{n}e_{n}\ \ \ \ \ \ \ \ \ \ \left(\text{since }\psi_{n}\text{ is linear}\right)$ $\displaystyle=\widetilde{e_{i}}-0=\widetilde{e_{i}};$ thus, for every $x\in\mathbb{R}^{n-1},$ we have $\displaystyle\psi_{n}\phi_{n}x$ $\displaystyle=\psi_{n}\phi_{n}\left(\sum\limits_{i=1}^{n-1}x_{i}\widetilde{e_{i}}\right)=\sum\limits_{i=1}^{n-1}x_{i}\psi_{n}\phi_{n}\widetilde{e_{i}}$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{since the function }\psi_{n}\phi_{n}\text{ is linear, because }\psi_{n}\text{ and }\phi_{n}\text{ are linear}\right)$ $\displaystyle=\sum\limits_{i=1}^{n-1}x_{i}\widetilde{e_{i}}=x,$ and therefore $\psi_{n}\phi_{n}=\mathop{\mathrm{i}d}$). We define a function $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ by $f\left(x\right)=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ x_{n}\end{array}\right)\in\mathbb{R}^{n}.$ Note that $f\left(-x\right)=f\left(x\right)\ \ \ \ \ \ \ \ \ \ \text{for every }x\in\mathbb{R}^{n},$ (13) since $\displaystyle f\left(-x\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left(-x\right)_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}\left(-x\right)\right|=\sum_{u=1}^{n}a_{u}\left|-x_{u}\right|-\sum_{s\in S}b_{s}\left|-r_{s}\psi_{n}x\right|$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{here, we have }r_{s}\psi_{n}\left(-x\right)=-r_{s}\psi_{n}x\text{ since }r_{s}\text{ and }\psi_{n}\text{ are linear functions}\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}x\right|=f\left(x\right).$ Furthermore, I claim that $f\left(\phi_{n}x\right)=g\left(x\right)\ \ \ \ \ \ \ \ \ \ \text{for every }x\in\mathbb{R}^{n-1}.$ (14) In order to prove this, we note that (7) yields $\left(\phi_{n}x\right)_{u}=x_{u}$ for all $u\in\left\\{1,2,...,n-1\right\\}$ and $\left(\phi_{n}x\right)_{n}=-\left(x_{1}+x_{2}+...+x_{n-1}\right),$ while $\psi_{n}\phi_{n}=\mathop{\mathrm{i}d}$ yields $\psi_{n}\phi_{n}x=x,$ so that $\displaystyle f\left(\phi_{n}x\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left(\phi_{n}x\right)_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}\phi_{n}x\right|$ $\displaystyle=\sum_{u=1}^{n-1}a_{u}\left|\left(\phi_{n}x\right)_{u}\right|+a_{n}\left|\left(\phi_{n}x\right)_{n}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}\phi_{n}x\right|$ $\displaystyle=\sum_{u=1}^{n-1}a_{u}\left|x_{u}\right|+a_{n}\left|-\left(x_{1}+x_{2}+...+x_{n-1}\right)\right|-\sum_{s\in S}b_{s}\left|r_{s}x\right|$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{since }\left(\phi_{n}x\right)_{u}=x_{u}\text{ for all }u\in\left\\{1,2,...,n-1\right\\}\text{ and}\right.$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left.\left(\phi_{n}x\right)_{n}=-\left(x_{1}+x_{2}+...+x_{n-1}\right),\text{ and }\psi_{n}\phi_{n}x=x\right)$ $\displaystyle=\sum_{u=1}^{n-1}a_{u}\left|x_{u}\right|+a_{n}\left|x_{1}+x_{2}+...+x_{n-1}\right|-\sum_{s\in S}b_{s}\left|r_{s}x\right|=g\left(x\right),$ and thus (9) is proven. Now, we are going to show that $f\left(e_{i}-e_{j}\right)\geq 0\text{ for any two distinct integers }i\text{ and }j\text{ from }\left\\{1,2,...,n\right\\}.$ (15) In order to prove (10), we distinguish between three different cases: Case 1: We have $i\in\left\\{1,2,...,n-1\right\\}$ and $j\in\left\\{1,2,...,n-1\right\\}.$ Case 2: We have $i\in\left\\{1,2,...,n-1\right\\}$ and $j=n.$ Case 3: We have $i=n$ and $j\in\left\\{1,2,...,n-1\right\\}.$ (In fact, the case when both $i=n$ and $j=n$ cannot occur, since $i$ and $j$ must be distinct). In Case 1, we have $\displaystyle f\left(e_{i}-e_{j}\right)$ $\displaystyle=f\left(\left(e_{i}-e_{n}\right)-\left(e_{j}-e_{n}\right)\right)=f\left(\phi_{n}\widetilde{e_{i}}-\phi_{n}\widetilde{e_{j}}\right)$ $\displaystyle=f\left(\phi_{n}\left(\widetilde{e_{i}}-\widetilde{e_{j}}\right)\right)\ \ \ \ \ \ \ \ \ \ \left(\text{since }\phi_{n}\widetilde{e_{i}}-\phi_{n}\widetilde{e_{j}}=\phi_{n}\left(\widetilde{e_{i}}-\widetilde{e_{j}}\right),\text{ because }\phi_{n}\text{ is linear}\right)$ $\displaystyle=g\left(\widetilde{e_{i}}-\widetilde{e_{j}}\right)\ \ \ \ \ \ \ \ \ \ \left(\text{after (9)}\right)$ $\displaystyle\geq 0\ \ \ \ \ \ \ \ \ \ \left(\text{by assumption}\right).$ In Case 2, we have $\displaystyle f\left(e_{i}-e_{j}\right)$ $\displaystyle=f\left(e_{i}-e_{n}\right)=f\left(\phi_{n}\widetilde{e_{i}}\right)=g\left(\widetilde{e_{i}}\right)\ \ \ \ \ \ \ \ \ \ \left(\text{after (9)}\right)$ $\displaystyle\geq 0\ \ \ \ \ \ \ \ \ \ \left(\text{by assumption}\right).$ In Case 3, we have $\displaystyle f\left(e_{i}-e_{j}\right)$ $\displaystyle=f\left(e_{n}-e_{j}\right)=f\left(-\left(e_{j}-e_{n}\right)\right)=f\left(e_{j}-e_{n}\right)\ \ \ \ \ \ \ \ \ \ \left(\text{after (8)}\right)$ $\displaystyle=f\left(\phi_{n}\widetilde{e_{j}}\right)=g\left(\widetilde{e_{j}}\right)\ \ \ \ \ \ \ \ \ \ \left(\text{after (9)}\right)$ $\displaystyle\geq 0\ \ \ \ \ \ \ \ \ \ \left(\text{by assumption}\right).$ Thus, $f\left(e_{i}-e_{j}\right)\geq 0$ holds in all three possible cases. Hence, (10) is proven. Now, our function $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ is defined by $f\left(x\right)=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|-\sum_{s\in S}b_{s}\left|r_{s}\psi_{n}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ x_{n}\end{array}\right)\in\mathbb{R}^{n}.$ Here, $n$ is a positive integer; the numbers $a_{1},$ $a_{2},$ $...,$ $a_{n}$ are $n$ nonnegative reals; the set $S$ is a finite set; for every $s\in S,$ the function $r_{s}\psi_{n}$ is an element of $\left(\mathbb{R}^{n}\right)^{\ast}$ (in other words, a linear transformation from $\mathbb{R}^{n}$ to $\mathbb{R}$), and $b_{s}$ is a nonnegative real. Hence, we can apply Theorem 11 to our function $f,$ and we obtain that for our function $f,$ the Assertions $\mathcal{A}_{1}$ and $\mathcal{A}_{2}$ are equivalent. In other words, our function $f$ satisfies Assertion $\mathcal{A}_{1}$ if and only if it satisfies Assertion $\mathcal{A}_{2}.$ Now, according to (10), our function $f$ satisfies Assertion $\mathcal{A}_{2}.$ Thus, this function $f$ must also satisfy Assertion $\mathcal{A}_{1}.$ In other words, $f\left(x\right)\geq 0$ holds for every $x\in V_{n}.$ Hence, $f\left(\phi_{n}x\right)\geq 0$ holds for every $x\in\mathbb{R}^{n-1}$ (because $\phi_{n}x\in V_{n},$ since $\mathop{\mathrm{I}m}\phi_{n}\subseteq V_{n}$). Since $f\left(\phi_{n}x\right)=g\left(x\right)$ according to (9), we have therefore proven that $g\left(x\right)\geq 0$ holds for every $x\in\mathbb{R}^{n-1}.$ Hence, Assertion $\mathcal{C}_{1}$ is proven. Thus, we have showed that $\mathcal{C}_{2}\Longrightarrow\mathcal{C}_{1},$ and thus the proof of Theorem 12b is complete. Since Theorem 12b is equivalent to Theorem 12, this also proves Theorem 12. As if this wasn’t enough, here comes a further restatement of Theorem 12: > Theorem 13. Let $n$ be a nonnegative integer. Let $a_{1},$ $a_{2},$ $...,$ > $a_{n}$ and $a$ be $n+1$ nonnegative reals. Let $S$ be a finite set. For > every $s\in S,$ let $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ be $n$ > nonnegative reals, and let $b_{s}$ be a nonnegative real. Assume that the > following two assertions hold: > > $\displaystyle a_{i}+a$ $\displaystyle\geq\sum_{s\in S}b_{s}r_{s,i}\ \ \ \ \ > \ \ \ \ \ \text{for every }i\in\left\\{1,2,...,n\right\\};$ $\displaystyle > a_{i}+a_{j}$ $\displaystyle\geq\sum_{s\in > S}b_{s}\left|r_{s,i}-r_{s,j}\right|\ \ \ \ \ \ \ \ \ \ \text{for any two > distinct integers }i\text{ and }j\text{ from }\left\\{1,2,...,n\right\\}.$ > > Let $y_{1},$ $y_{2},$ $...,$ $y_{n}$ be $n$ reals. Then, > > > $\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|-\sum\limits_{s\in > S}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|\geq 0.$ Proof of Theorem 13. For every $s\in S,$ let $r_{s}=\left(r_{s,1},r_{s,2},...,r_{s,n}\right)\in\left(\mathbb{R}^{n}\right)^{\ast}$ be the $n$-dimensional covector whose $i$-th coordinate is $r_{s,i}$ for every $i\in\left\\{1,2,...,n\right\\}.$ Define a function $g:\mathbb{R}^{n}\rightarrow\mathbb{R}$ by $g\left(x\right)=\sum_{u=1}^{n}a_{u}\left|x_{u}\right|+a\left|x_{1}+x_{2}+...+x_{n}\right|-\sum_{s\in S}b_{s}\left|r_{s}x\right|,\ \ \ \ \ \ \ \ \ \ \text{where }x=\left(\begin{array}[]{c}x_{1}\\\ x_{2}\\\ ...\\\ x_{n}\end{array}\right)\in\mathbb{R}^{n}.$ For every $i\in\left\\{1,2,...,n\right\\},$ we have $\left(e_{i}\right)_{u}=\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ 0,\text{ if }u\neq i\end{array}\right.$ for all $u\in\left\\{1,2,...,n\right\\},$ so that $\left(e_{i}\right)_{1}+\left(e_{i}\right)_{2}+...+\left(e_{i}\right)_{n}=1,$ and for every $s\in S,$ we have $\displaystyle r_{s}e_{i}$ $\displaystyle=\sum_{u=1}^{n}r_{s,u}\left(e_{i}\right)_{u}\ \ \ \ \ \ \ \ \ \ \left(\text{since }r_{s}=\left(r_{s,1},r_{s,2},...,r_{s,n}\right)\right)$ $\displaystyle=\sum_{u=1}^{n}r_{s,u}\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ 0,\text{ if }u\neq i\end{array}\right.=r_{s,i},$ so that $\displaystyle g\left(e_{i}\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left(e_{i}\right)_{u}\right|+a\left|\left(e_{i}\right)_{1}+\left(e_{i}\right)_{2}+...+\left(e_{i}\right)_{n}\right|-\sum_{s\in S}b_{s}\left|r_{s}e_{i}\right|$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ 0,\text{ if }u\neq i\end{array}\right.\right|+a\left|1\right|-\sum_{s\in S}b_{s}\left|r_{s,i}\right|$ $\displaystyle=a_{i}\left|1\right|+a\left|1\right|-\sum_{s\in S}b_{s}\underbrace{\left|r_{s,i}\right|}_{\begin{subarray}{c}=r_{s,i},\\\ \text{since}\\\ r_{s,i}\geq 0\end{subarray}}=a_{i}+a-\sum_{s\in S}b_{s}r_{s,i}\geq 0$ (since $a_{i}+a\geq\sum\limits_{s\in S}b_{s}r_{s,i}$ by the conditions of Theorem 13). For any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\},$ we have $\left(e_{i}-e_{j}\right)_{u}=\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ -1,\text{ if }u=j;\\\ 0,\text{ if }u\neq i\text{ and }u\neq j\end{array}\right.\ $for all $u\in\left\\{1,2,...,n\right\\},$ so that $\left(e_{i}-e_{j}\right)_{1}+\left(e_{i}-e_{j}\right)_{2}+...+\left(e_{i}-e_{j}\right)_{n}=0,$ and for every $s\in S,$ we have $\displaystyle r_{s}\left(e_{i}-e_{j}\right)$ $\displaystyle=\sum_{u=1}^{n}r_{s,u}\left(e_{i}-e_{j}\right)_{u}\ \ \ \ \ \ \ \ \ \ \left(\text{since }r_{s}=\left(r_{s,1},r_{s,2},...,r_{s,n}\right)\right)$ $\displaystyle=\sum_{u=1}^{n}r_{s,u}\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ -1,\text{ if }u=j;\\\ 0,\text{ if }u\neq i\text{ and }u\neq j\end{array}\right.=r_{s,i}-r_{s,j},$ and thus $\displaystyle g\left(e_{i}-e_{j}\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left(e_{i}-e_{j}\right)_{u}\right|+a\left|\left(e_{i}-e_{j}\right)_{1}+\left(e_{i}-e_{j}\right)_{2}+...+\left(e_{i}-e_{j}\right)_{n}\right|-\sum_{s\in S}b_{s}\left|r_{s}\left(e_{i}-e_{j}\right)\right|$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|\left\\{\begin{array}[]{c}1,\text{ if }u=i;\\\ -1,\text{ if }u=j;\\\ 0,\text{ if }u\neq i\text{ and }u\neq j\end{array}\right.\right|+a\left|0\right|-\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|$ $\displaystyle=\left(a_{i}\left|1\right|+a_{j}\left|-1\right|\right)+a\left|0\right|-\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|=\left(a_{i}+a_{j}\right)+0-\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|$ $\displaystyle=a_{i}+a_{j}-\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|\geq 0$ (since $a_{i}+a_{j}\geq\sum\limits_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|$ by the condition of Theorem 13). So we have shown that $g\left(e_{i}\right)\geq 0$ for every integer $i\in\left\\{1,2,...,n\right\\},$ and $g\left(e_{i}-e_{j}\right)\geq 0$ for any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\}.$ Thus, Assertion $\mathcal{B}_{2}$ of Theorem 12 is fulfilled. According to Theorem 12, the assertions $\mathcal{B}_{1}$ and $\mathcal{B}_{2}$ are equivalent, so that Assertion $\mathcal{B}_{1}$ must be fulfilled as well. Hence, $g\left(x\right)\geq 0$ for every $x\in\mathbb{R}^{n}.$ In particular, if we set $x=\left(\begin{array}[]{c}y_{1}\\\ y_{2}\\\ ...\\\ y_{n}\end{array}\right),$ then $r_{s}x=\sum\limits_{v=1}^{n}r_{s,v}y_{v}$ (since $r_{s}=\left(r_{s,1},r_{s,2},...,r_{s,n}\right)$), so that $\displaystyle g\left(x\right)$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|y_{u}\right|+a\left|y_{1}+y_{2}+...+y_{n}\right|-\sum_{s\in S}b_{s}\left|r_{s}x\right|$ $\displaystyle=\sum_{u=1}^{n}a_{u}\left|y_{u}\right|+a\left|y_{1}+y_{2}+...+y_{n}\right|-\sum_{s\in S}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|$ $\displaystyle=\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|-\sum_{s\in S}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|,$ and thus $g\left(x\right)\geq 0$ yields $\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|-\sum_{s\in S}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|\geq 0.$ Theorem 13 is thus proven. 6\. A general condition for Popoviciu-like inequalities Now, we state a result more general than Theorem 5b: > Theorem 14. Let $n$ be a nonnegative integer. Let $a_{1},$ $a_{2},$ $...,$ > $a_{n}$ and $a$ be $n+1$ nonnegative reals. Let $S$ be a finite set. For > every $s\in S,$ let $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ be $n$ > nonnegative reals, and let $b_{s}$ be a nonnegative real. Assume that the > following two assertions hold111The second of these two assertions > ($a_{i}+a_{j}\geq\sum\limits_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|\ $for > any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\}$) is > identic with the second required assertion in Theorem 13, but the first one > ($a_{i}+a=\sum\limits_{s\in S}b_{s}r_{s,i}$ for every > $i\in\left\\{1,2,...,n\right\\}$) is stronger than the first required > assertion in Theorem 13 (which only said that $a_{i}+a\geq\sum\limits_{s\in > S}b_{s}r_{s,i}$ for every $i\in\left\\{1,2,...,n\right\\}$).: > > $\displaystyle a_{i}+a$ $\displaystyle=\sum_{s\in S}b_{s}r_{s,i}\ \ \ \ \ \ > \ \ \ \ \text{for every }i\in\left\\{1,2,...,n\right\\};$ $\displaystyle > a_{i}+a_{j}$ $\displaystyle\geq\sum_{s\in > S}b_{s}\left|r_{s,i}-r_{s,j}\right|\ \ \ \ \ \ \ \ \ \ \text{for any two > distinct integers }i\text{ and }j\text{ from }\left\\{1,2,...,n\right\\}.$ > > Let $f$ be a convex function from an interval $I\subseteq\mathbb{R}$ to > $\mathbb{R}.$ Let $w_{1},$ $w_{2},$ $...,$ $w_{n}$ be nonnegative reals. > Assume that $\sum\limits_{v=1}^{n}w_{v}\neq 0$ and > $\sum\limits_{v=1}^{n}r_{s,v}w_{v}\neq 0$ for all $s\in S.$ > > Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ be $n$ points from the interval $I.$ > Then, the inequality > > > $\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)\geq\sum_{s\in > S}b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right)$ > > holds. > > Remark. Written in a less formal way, this inequality states that > > > $\displaystyle\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(w_{1}+w_{2}+...+w_{n}\right)f\left(\frac{w_{1}x_{1}+w_{2}x_{2}+...+w_{n}x_{n}}{w_{1}+w_{2}+...+w_{n}}\right)$ > $\displaystyle\geq\sum_{s\in > S}b_{s}\left(r_{s,1}w_{1}+r_{s,2}w_{2}+...+r_{s,n}w_{n}\right)f\left(\frac{r_{s,1}w_{1}x_{1}+r_{s,2}w_{2}x_{2}+...+r_{s,n}w_{n}x_{n}}{r_{s,1}w_{1}+r_{s,2}w_{2}+...+r_{s,n}w_{n}}\right).$ Proof of Theorem 14. Since the elements of the finite set $S$ are used as labels only, we can assume without loss of generality that $S=\left\\{n+2,n+3,...,N\right\\}$ for some integer $N\geq n+1$ (we just rename the elements of $S$ into $n+2,$ $n+3,$ $...,$ $N,$ where $N=n+1+\left|S\right|;$ this is possible because the set $S$ is finite222In particular, $N=n+1$ if $S=\varnothing.$). Define $\displaystyle u_{i}$ $\displaystyle=a_{i}w_{i}\ \ \ \ \ \ \ \ \ \ \text{for all }i\in\left\\{1,2,...,n\right\\};$ $\displaystyle u_{n+1}$ $\displaystyle=a\left(\sum_{v=1}^{n}w_{v}\right);$ $\displaystyle u_{s}$ $\displaystyle=-b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)\ \ \ \ \ \ \ \ \ \ \text{for all }s\in\left\\{n+2,n+3,...,N\right\\}\text{ (that is, for all }s\in S\text{).}$ Also define $\displaystyle z_{i}$ $\displaystyle=x_{i}\ \ \ \ \ \ \ \ \ \ \text{for all }i\in\left\\{1,2,...,n\right\\};$ $\displaystyle z_{n+1}$ $\displaystyle=\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}};$ $\displaystyle z_{s}$ $\displaystyle=\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\ \ \ \ \ \ \ \ \ \ \text{for all }s\in\left\\{n+2,n+3,...,N\right\\}\text{ (that is, for all }s\in S\text{).}$ Each of these $N$ reals $z_{1},$ $z_{2},$ $...,$ $z_{N}$ is a weighted mean of the reals $x_{1},$ $x_{2},$ $...,$ $x_{n}$ with nonnegative weights. Since the reals $x_{1},$ $x_{2},$ $...,$ $x_{n}$ lie in the interval $I,$ we can thus conclude that each of the $N$ reals $z_{1},$ $z_{2},$ $...,$ $z_{N}$ lies in the interval $I$ as well. In other words, the points $z_{1},$ $z_{2},$ $...,$ $z_{N}$ are $N$ points from $I.$ Now, $\displaystyle\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)-\sum_{s\in S}b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right)$ $\displaystyle=\sum_{i=1}^{n}\underbrace{a_{i}w_{i}}_{=u_{i}}f\left(\underbrace{x_{i}}_{=z_{i}}\right)+\underbrace{a\left(\sum_{v=1}^{n}w_{v}\right)}_{=u_{n+1}}f\left(\underbrace{\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}}_{=z_{n+1}}\right)+\sum_{s\in S}\left(\underbrace{-b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)}_{=u_{s}}\right)f\left(\underbrace{\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}}_{=z_{s}}\right)$ $\displaystyle=\sum_{i=1}^{n}u_{i}f\left(z_{i}\right)+u_{n+1}f\left(z_{n+1}\right)+\sum_{s\in S}u_{s}f\left(z_{s}\right)=\sum_{i=1}^{n}u_{i}f\left(z_{i}\right)+u_{n+1}f\left(z_{n+1}\right)+\sum_{s=n+2}^{N}u_{s}f\left(z_{s}\right)$ $\displaystyle=\sum_{k=1}^{N}u_{k}f\left(z_{k}\right).$ Hence, once we are able to show that $\sum\limits_{k=1}^{N}u_{k}f\left(z_{k}\right)\geq 0,$ we will obtain $\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)\geq\sum_{s\in S}b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right),$ and thus Theorem 14 will be established. Therefore, in order to prove Theorem 14, it remains to prove the inequality $\sum\limits_{k=1}^{N}u_{k}f\left(z_{k}\right)\geq 0.$ We have $\displaystyle\sum_{k=1}^{N}u_{k}$ $\displaystyle=\sum_{i=1}^{n}u_{i}+u_{n+1}+\sum_{s=n+2}^{N}u_{s}=\sum_{i=1}^{n}u_{i}+u_{n+1}+\sum_{s\in S}u_{s}$ $\displaystyle=\sum_{i=1}^{n}a_{i}w_{i}+a\left(\sum_{v=1}^{n}w_{v}\right)+\sum_{s\in S}\left(-b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)\right)$ $\displaystyle=\sum_{i=1}^{n}a_{i}w_{i}+a\left(\sum_{i=1}^{n}w_{i}\right)+\sum_{s\in S}\left(-b_{s}\left(\sum_{i=1}^{n}r_{s,i}w_{i}\right)\right)$ $\displaystyle=\sum_{i=1}^{n}a_{i}w_{i}+\sum_{i=1}^{n}aw_{i}-\sum_{i=1}^{n}\sum_{s\in S}b_{s}r_{s,i}w_{i}=\sum_{i=1}^{n}\left(a_{i}w_{i}+aw_{i}-\sum_{s\in S}b_{s}r_{s,i}w_{i}\right)$ $\displaystyle=\sum_{i=1}^{n}\left(a_{i}+a-\sum_{s\in S}b_{s}r_{s,i}\right)w_{i}$ $\displaystyle=\sum_{i=1}^{n}0w_{i}\ \ \ \ \ \ \ \ \ \ \left(\begin{array}[]{c}\text{since }a_{i}+a=\sum\limits_{s\in S}b_{s}r_{s,i}\text{ by an assumption of Theorem 14,}\\\ \text{and thus }a_{i}+a-\sum\limits_{s\in S}b_{s}r_{s,i}=0\end{array}\right)$ $\displaystyle=0.$ Next, we are going to prove that $\sum\limits_{k=1}^{N}u_{k}\left|z_{k}-t\right|\geq 0$ holds for every $t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}.$ In fact, let $t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}$ be arbitrary. Set $y_{i}=w_{i}\left(x_{i}-t\right)$ for every $i\in\left\\{1,2,...,n\right\\}.$ Then, for all $i\in\left\\{1,2,...,n\right\\},$ we have $w_{i}\left(z_{i}-t\right)=w_{i}\left(x_{i}-t\right)=y_{i}.$ Furthermore, $z_{n+1}-t=\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}-t=\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}-\sum\limits_{v=1}^{n}w_{v}\cdot t}{\sum\limits_{v=1}^{n}w_{v}}=\frac{\sum\limits_{v=1}^{n}w_{v}\left(x_{v}-t\right)}{\sum\limits_{v=1}^{n}w_{v}}=\frac{\sum\limits_{v=1}^{n}y_{v}}{\sum\limits_{v=1}^{n}w_{v}}.$ Finally, for all $s\in\left\\{n+2,n+3,...,N\right\\}$ (that is, for all $s\in S$), we have $z_{s}-t=\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}-t=\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}-\sum\limits_{v=1}^{n}r_{s,v}w_{v}\cdot t}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}=\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}\left(x_{v}-t\right)}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}=\frac{\sum\limits_{v=1}^{n}r_{s,v}y_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}.$ Hence, $\displaystyle\sum\limits_{k=1}^{N}u_{k}\left|z_{k}-t\right|=\sum\limits_{i=1}^{n}u_{i}\left|z_{i}-t\right|+u_{n+1}\left|z_{n+1}-t\right|+\sum\limits_{s=n+2}^{N}u_{s}\left|z_{s}-t\right|$ $\displaystyle=\sum\limits_{i=1}^{n}a_{i}\underbrace{w_{i}\left|z_{i}-t\right|}_{\begin{subarray}{c}=\left|w_{i}\left(z_{i}-t\right)\right|,\\\ \text{since }w_{i}\geq 0\end{subarray}}+a\left(\sum_{v=1}^{n}w_{v}\right)\left|\frac{\sum\limits_{v=1}^{n}y_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right|+\sum\limits_{s=n+2}^{N}\left(-b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)\right)\left|\frac{\sum\limits_{v=1}^{n}r_{s,v}y_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right|$ $\displaystyle=\sum\limits_{i=1}^{n}a_{i}\left|w_{i}\left(z_{i}-t\right)\right|+a\left(\sum_{v=1}^{n}w_{v}\right)\frac{\left|\sum\limits_{v=1}^{n}y_{v}\right|}{\sum\limits_{v=1}^{n}w_{v}}+\sum\limits_{s=n+2}^{N}\left(-b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)\right)\frac{\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\begin{array}[]{c}\text{here we have pulled the }\sum\limits_{v=1}^{n}w_{v}\text{ and }\sum\limits_{v=1}^{n}r_{s,v}w_{v}\text{ terms out of the modulus}\\\ \text{signs, since they are positive (in fact, they are }\neq 0\text{ by an assumption}\\\ \text{of Theorem 14, and nonnegative because }w_{i}\text{ and }r_{s,i}\text{ are all nonnegative)}\end{array}\right)$ $\displaystyle=\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|+\sum\limits_{s=n+2}^{N}\left(-b_{s}\right)\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|=\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|-\sum\limits_{s=n+2}^{N}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|$ $\displaystyle=\sum\limits_{i=1}^{n}a_{i}\left|y_{i}\right|+a\left|\sum\limits_{v=1}^{n}y_{v}\right|-\sum\limits_{s\in S}b_{s}\left|\sum\limits_{v=1}^{n}r_{s,v}y_{v}\right|\geq 0$ by Theorem 13 (in fact, we were allowed to apply Theorem 13 because all the requirements of Theorem 13 are fulfilled - in particular, we have $a_{i}+a\geq\sum\limits_{s\in S}b_{s}r_{s,i}$ for every $i\in\left\\{1,2,...,n\right\\}$ because we know that $a_{i}+a=\sum\limits_{s\in S}b_{s}r_{s,i}$ for every $i\in\left\\{1,2,...,n\right\\}$ by an assumption of Theorem 14). Altogether, we have now shown the following: The points $z_{1},$ $z_{2},$ $...,$ $z_{N}$ are $N$ points from $I.$ The $N$ reals $u_{1},$ $u_{2},$ $...,$ $u_{N}$ satisfy $\sum\limits_{k=1}^{N}u_{k}=0,$ and $\sum\limits_{k=1}^{N}u_{k}\left|z_{k}-t\right|\geq 0$ holds for every $t\in\left\\{z_{1},z_{2},...,z_{N}\right\\}.$ Hence, according to Theorem 8b, we have $\sum\limits_{k=1}^{N}u_{k}f\left(z_{k}\right)\geq 0.$ And as we have seen above, once $\sum\limits_{k=1}^{N}u_{k}f\left(z_{k}\right)\geq 0$ is shown, the proof of Theorem 14 is complete. Thus, Theorem 14 is proven. Theorem 14 gives a sufficient criterion for the validity of inequalities of the kind $\displaystyle\text{convex combination of }f\left(x_{1}\right),\text{ }f\left(x_{2}\right),\text{ }...,\text{ }f\left(x_{n}\right)$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \text{ and }f\left(\text{some weighted mean of }x_{1},\text{ }x_{2},\text{ }...,\text{ }x_{n}\right)$ $\displaystyle\geq\text{convex combination of finitely many }f\left(\text{some other weighted means of }x_{1},\text{ }x_{2},\text{ }...,\text{ }x_{n}\right)\text{'s,}$ where $f$ is a convex function and $x_{1},$ $x_{2},$ $...,$ $x_{n}$ are $n$ reals in its domain, and where the weights of the weighted mean on the left hand side are positive (those of the weighted means on the right hand side may be $0$ as well, but still have to be nonnegative). This criterion turns out to be necessary as well: > Theorem 14b. Let $n$ be a nonnegative integer. Let $w_{1},$ $w_{2},$ $...,$ > $w_{n}$ be positive reals. Let $a_{1},$ $a_{2},$ $...,$ $a_{n}$ and $a$ be > $n+1$ nonnegative reals. Let $S$ be a finite set. For every $s\in S,$ let > $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ be $n$ nonnegative reals, and let > $b_{s}$ be a nonnegative real. Let $I\subseteq\mathbb{R}$ be an interval. > > Assume that the inequality > > > $\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)\geq\sum_{s\in > S}b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right)$ > > holds for any convex function $f:I\rightarrow\mathbb{R}$ and any $n$ points > $x_{1},$ $x_{2},$ $...,$ $x_{n}$ in the interval $I.$ Then, > > $\displaystyle a_{i}+a$ $\displaystyle=\sum_{s\in S}b_{s}r_{s,i}\ \ \ \ \ \ > \ \ \ \ \text{for every }i\in\left\\{1,2,...,n\right\\};$ $\displaystyle > a_{i}+a_{j}$ $\displaystyle\geq\sum_{s\in > S}b_{s}\left|r_{s,i}-r_{s,j}\right|\ \ \ \ \ \ \ \ \ \ \text{for any two > distinct integers }i\text{ and }j\text{ from }\left\\{1,2,...,n\right\\}.$ Since we are not going to use this fact, we are not proving it either, but the idea of the proof is the following: Assume WLOG that $I=\left[-1,1\right].$ For every $i\in\left\\{1,2,...,n\right\\},$ you get $a_{i}+a\geq\sum\limits_{s\in S}b_{s}r_{s,i}$ (by considering the convex function $f\left(x\right)=x$ and the points $x_{k}=\left\\{\begin{array}[]{c}1,\text{ if }k=i;\\\ 0,\text{ if }k\neq i\end{array}\right.$) and $a_{i}+a\leq\sum\limits_{s\in S}b_{s}r_{s,i}$ (by considering the convex function $f\left(x\right)=-x$ and the same points), so that $a_{i}+a=\sum\limits_{s\in S}b_{s}r_{s,i}.$ For any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\},$ you get $a_{i}+a_{j}\geq\sum\limits_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|$ (by considering the convex function $f\left(x\right)=\left|x\right|$ and the points $x_{k}=\left\\{\begin{array}[]{c}1,\text{ if }k=i;\\\ -1,\text{ if }k=j;\\\ 0,\text{ if }k\neq i\text{ and }k\neq j\end{array}\right.$). This altogether proves Theorem 14b. 7\. Proving the Popoviciu inequality Now we can finally step to the proof of Theorem 5b: We assume that $n\geq 2,$ because all cases where $n<2$ (that is, $n=1$ or $n=0$) can be checked manually (and are uninteresting). Let $a_{i}=\dbinom{n-2}{m-1}$ for every $i\in\left\\{1,2,...,n\right\\}.$ Let $a=\dbinom{n-2}{m-2}.$ These reals $a_{1},$ $a_{2},$ $...,$ $a_{n}$ and $a$ are all nonnegative (since $n\geq 2$ yields $n-2\geq 0$ and thus $\dbinom{n-2}{t}\geq 0$ for all integers $t$). Let $S=\left\\{s\subseteq\left\\{1,2,...,n\right\\}\mid\left|s\right|=m\right\\};$ that is, we denote by $S$ the set of all $m$-element subsets of the set $\left\\{1,2,...,n\right\\}.$ This set $S$ is obviously finite. For every $s\in S,$ define $n$ reals $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ as follows: $r_{s,i}=\left\\{\begin{array}[]{c}1,\text{ if }i\in s;\\\ 0,\text{ if }i\notin s\end{array}\right.\ \ \ \ \ \ \ \ \ \ \text{for every }i\in\left\\{1,2,...,n\right\\}.$ Obviously, these reals $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ are all nonnegative. Also, for every $s\in S,$ set $b_{s}=1;$ then, $b_{s}$ is a nonnegative real as well. For every $i\in\left\\{1,2,...,n\right\\},$ we have $\displaystyle\sum_{s\in S}b_{s}r_{s,i}$ $\displaystyle=\sum_{s\in S}1r_{s,i}=\sum_{s\in S}r_{s,i}=\sum_{s\in S}\left\\{\begin{array}[]{c}1,\text{ if }i\in s;\\\ 0,\text{ if }i\notin s\end{array}\right.=\sum_{\begin{subarray}{c}s\subseteq\left\\{1,2,...,n\right\\};\\\ \left|s\right|=m\end{subarray}}\left\\{\begin{array}[]{c}1,\text{ if }i\in s;\\\ 0,\text{ if }i\notin s\end{array}\right.$ $\displaystyle=\left(\text{number of }m\text{-element subsets }s\text{ of the set }\left\\{1,2,...,n\right\\}\text{ that contain }i\right)$ $\displaystyle=\dbinom{n-1}{m-1},$ so that $\displaystyle a_{i}+a$ $\displaystyle=\dbinom{n-2}{m-1}+\dbinom{n-2}{m-2}=\dbinom{n-1}{m-1}$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{by the recurrence relation of the binomial coefficients}\right)$ $\displaystyle=\sum_{s\in S}b_{s}r_{s,i}.$ (16) For any two distinct integers $i$ and $j$ from $\left\\{1,2,...,n\right\\},$ we have $\displaystyle\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|=\sum_{s\in S}1\left|r_{s,i}-r_{s,j}\right|=\sum_{s\in S}\left|r_{s,i}-r_{s,j}\right|$ $\displaystyle=\sum_{s\in S}\left|\left\\{\begin{array}[]{c}1,\text{ if }i\in s;\\\ 0,\text{ if }i\notin s\end{array}\right.-\left\\{\begin{array}[]{c}1,\text{ if }j\in s;\\\ 0,\text{ if }j\notin s\end{array}\right.\right|=\sum_{s\in S}\left\\{\begin{array}[]{c}0,\text{ if }i\in s\text{ and }j\in s;\\\ 1,\text{ if }i\in s\text{ and }j\notin s;\\\ 1,\text{ if }i\notin s\text{ and }j\in s;\\\ 0,\text{ if }i\notin s\text{ and }j\notin s\end{array}\right.$ $\displaystyle=\sum_{s\in S}\left(\left\\{\begin{array}[]{c}1,\text{ if }i\in s\text{ and }j\notin s;\\\ 0\text{ otherwise}\end{array}\right.+\left\\{\begin{array}[]{c}1,\text{ if }i\notin s\text{ and }j\in s;\\\ 0\text{ otherwise}\end{array}\right.\right)$ $\displaystyle=\sum_{s\in S}\left\\{\begin{array}[]{c}1,\text{ if }i\in s\text{ and }j\notin s;\\\ 0\text{ otherwise}\end{array}\right.+\sum_{s\in S}\left\\{\begin{array}[]{c}1,\text{ if }i\notin s\text{ and }j\in s;\\\ 0\text{ otherwise}\end{array}\right.$ $\displaystyle=\sum_{\begin{subarray}{c}s\subseteq\left\\{1,2,...,n\right\\};\\\ \left|s\right|=m\end{subarray}}\left\\{\begin{array}[]{c}1,\text{ if }i\in s\text{ and }j\notin s;\\\ 0\text{ otherwise}\end{array}\right.+\sum_{\begin{subarray}{c}s\subseteq\left\\{1,2,...,n\right\\};\\\ \left|s\right|=m\end{subarray}}\left\\{\begin{array}[]{c}1,\text{ if }i\notin s\text{ and }j\in s;\\\ 0\text{ otherwise}\end{array}\right.$ $\displaystyle=\left(\text{number of }m\text{-element subsets }s\text{ of the set }\left\\{1,2,...,n\right\\}\text{ that contain }i\text{ but not }j\right)$ $\displaystyle+\left(\text{number of }m\text{-element subsets }s\text{ of the set }\left\\{1,2,...,n\right\\}\text{ that contain }j\text{ but not }i\right)$ $\displaystyle=\dbinom{n-2}{m-1}+\dbinom{n-2}{m-1}=a_{i}+a_{j},$ so that $a_{i}+a_{j}=\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|.$ (17) Also, $\sum_{v=1}^{n}w_{v}=w_{1}+w_{2}+...+w_{n}\neq 0$ (18) (by an assumption of Theorem 5b). The elements of $S$ are all the $m$-element subsets of $\left\\{1,2,...,n\right\\}.$ Hence, to every element $s\in S$ uniquely correspond $m$ integers $i_{1},$ $i_{2},$ $...,$ $i_{m}$ satisfying $1\leq i_{1}<i_{2}<...<i_{m}\leq n$ and $s=\left\\{i_{1},i_{2},...,i_{m}\right\\}$ (these $m$ integers $i_{1},$ $i_{2},$ $...,$ $i_{m}$ are the $m$ elements of $s$ in increasing order). And conversely, any $m$ integers $i_{1},$ $i_{2},$ $...,$ $i_{m}$ satisfying $1\leq i_{1}<i_{2}<...<i_{m}\leq n$ can be obtained this way - in fact, they correspond to the $m$-element set $s=\left\\{i_{1},i_{2},...,i_{m}\right\\}\in S.$ Given an element $s\in S$ and the corresponding $m$ integers $i_{1},$ $i_{2},$ $...,$ $i_{m},$ we can write $\displaystyle\sum\limits_{v=1}^{n}r_{s,v}w_{v}$ $\displaystyle=\sum\limits_{v=1}^{n}\left\\{\begin{array}[]{c}1,\text{ if }v\in s;\\\ 0,\text{ if }v\notin s\end{array}\right.\cdot w_{v}=\sum\limits_{v\in s}w_{v}=\sum_{v\in\left\\{i_{1},i_{2},...,i_{m}\right\\}}w_{v}=w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}};$ $\displaystyle\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}$ $\displaystyle=\sum\limits_{v=1}^{n}\left\\{\begin{array}[]{c}1,\text{ if }v\in s;\\\ 0,\text{ if }v\notin s\end{array}\right.\cdot w_{v}x_{v}=\sum\limits_{v\in s}w_{v}x_{v}$ $\displaystyle=\sum_{v\in\left\\{i_{1},i_{2},...,i_{m}\right\\}}w_{v}x_{v}=w_{i_{1}}x_{i_{1}}+w_{i_{2}}x_{i_{2}}+...+w_{i_{m}}x_{i_{m}}.$ From this, we can conclude that $\sum\limits_{v=1}^{n}r_{s,v}w_{v}\neq 0\ \ \ \ \ \ \ \ \ \ \text{for every }s\in S$ (19) (because $\sum\limits_{v=1}^{n}r_{s,v}w_{v}=w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}},$ and $w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}\neq 0$ by an assumption of Theorem 5b), and we can also conclude that $\displaystyle\sum_{s\in S}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right)$ $\displaystyle=\sum_{1\leq i_{1}<i_{2}<...<i_{m}\leq n}\left(w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}\right)f\left(\frac{w_{i_{1}}x_{i_{1}}+w_{i_{2}}x_{i_{2}}+...+w_{i_{m}}x_{i_{m}}}{w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}}\right).$ Using the conditions of Theorem 5b and the relations (11), (12), (13) and (14), we see that all conditions of Theorem 14 are fulfilled. Thus, we can apply Theorem 14, and obtain $\sum_{i=1}^{n}a_{i}w_{i}f\left(x_{i}\right)+a\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)\geq\sum_{s\in S}b_{s}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right).$ This rewrites as $\displaystyle\sum_{i=1}^{n}\dbinom{n-2}{m-1}w_{i}f\left(x_{i}\right)+\dbinom{n-2}{m-2}\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)$ $\displaystyle\geq\sum_{s\in S}1\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right).$ In other words, $\displaystyle\dbinom{n-2}{m-1}\sum_{i=1}^{n}w_{i}f\left(x_{i}\right)+\dbinom{n-2}{m-2}\left(\sum_{v=1}^{n}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}\right)$ $\displaystyle\geq\sum_{s\in S}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right).$ Using (15) and the obvious relations $\displaystyle\sum_{v=1}^{n}w_{v}$ $\displaystyle=w_{1}+w_{2}+...+w_{n};$ $\displaystyle\sum\limits_{v=1}^{n}w_{v}x_{v}$ $\displaystyle=w_{1}x_{1}+w_{2}x_{2}+...+w_{n}x_{n},$ we can rewrite this as $\displaystyle\dbinom{n-2}{m-1}\sum_{i=1}^{n}w_{i}f\left(x_{i}\right)+\dbinom{n-2}{m-2}\left(w_{1}+w_{2}+...+w_{n}\right)f\left(\frac{w_{1}x_{1}+w_{2}x_{2}+...+w_{n}x_{n}}{w_{1}+w_{2}+...+w_{n}}\right)$ $\displaystyle\geq\sum_{1\leq i_{1}<i_{2}<...<i_{m}\leq n}\left(w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}\right)f\left(\frac{w_{i_{1}}x_{i_{1}}+w_{i_{2}}x_{i_{2}}+...+w_{i_{m}}x_{i_{m}}}{w_{i_{1}}+w_{i_{2}}+...+w_{i_{m}}}\right).$ This proves Theorem 5b. 8\. A cyclic inequality The most general form of the Popoviciu inequality is now proven. But this is not the end to the applications of Theorem 14. We will now apply it to show a cyclic inequality similar to Popoviciu’s: > Theorem 15a. Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ > > We extend the indices in $x_{1},$ $x_{2},$ $...,$ $x_{n}$ cyclically modulo > $n$; this means that for any integer $i\notin\left\\{1,2,...,n\right\\},$ we > define a real $x_{i}$ by setting $x_{i}=x_{j},$ where $j$ is the integer > from the set $\left\\{1,2,...,n\right\\}$ such that $i\equiv > j\mathop{\mathrm{m}od}n.$ (For instance, this means that $x_{n+3}=x_{3}.$) > > Let $x=\dfrac{x_{1}+x_{2}+...+x_{n}}{n}$. Let $r$ be an integer. Then, > > $2\sum_{i=1}^{n}f\left(x_{i}\right)+n\left(n-2\right)f\left(x\right)\geq > n\sum_{s=1}^{n}f\left(x+\dfrac{x_{s}-x_{s+r}}{n}\right).$ A weighted version of this inequality is: > Theorem 15b. Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ Let $r$ be an integer. > > Let $w_{1},$ $w_{2},$ $...,$ $w_{n}$ be nonnegative reals. Let > $x=\dfrac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}$ and > $w=\sum\limits_{v=1}^{n}w_{v}$. Assume that $w\neq 0$ and that > $w+\left(w_{s}-w_{s+r}\right)\neq 0$ for every $s\in S.$ > > We extend the indices in $x_{1},$ $x_{2},$ $...,$ $x_{n}$ and in $w_{1},$ > $w_{2},$ $...,$ $w_{n}$ cyclically modulo $n$; this means that for any > integer $i\notin\left\\{1,2,...,n\right\\},$ we define reals $x_{i}$ and > $w_{i}$ by setting $x_{i}=x_{j}$ and $w_{i}=w_{j},$ where $j$ is the integer > from the set $\left\\{1,2,...,n\right\\}$ such that $i\equiv > j\mathop{\mathrm{m}od}n.$ (For instance, this means that $x_{n+3}=x_{3}$ and > $w_{n+2}=w_{2}.$) > > Then, > > > $2\sum_{i=1}^{n}w_{i}f\left(x_{i}\right)+\left(n-2\right)wf\left(x\right)\geq\sum_{s=1}^{n}\left(w+\left(w_{s}-w_{s+r}\right)\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}+\left(w_{s}x_{s}-w_{s+r}x_{s+r}\right)}{w+\left(w_{s}-w_{s+r}\right)}\right).$ Proof of Theorem 15b. We assume that $n\geq 2,$ because all cases where $n<2$ (that is, $n=1$ or $n=0$) can be checked manually (and are uninteresting). Before we continue with the proof, let us introduce a simple notation: For any assertion $\mathcal{A}$, we denote by $\left[\mathcal{A}\right]$ the Boolean value of the assertion $\mathcal{A}$ (that is, $\left[\mathcal{A}\right]=\left\\{\begin{array}[]{c}1\text{, if }\mathcal{A}\text{ is true;}\\\ 0\text{, if }\mathcal{A}\text{ is false}\end{array}\right.$). Therefore, $0\leq\left[\mathcal{A}\right]\leq 1$ for every assertion $\mathcal{A}.$ Let $a_{i}=2$ for every $i\in\left\\{1,2,...,n\right\\}.$ Let $a=n-2.$ These reals $a_{1},$ $a_{2},$ $...,$ $a_{n}$ and $a$ are all nonnegative (since $n\geq 2$ yields $n-2\geq 0$). Let $S=\left\\{1,2,...,n\right\\}.$ This set $S$ is obviously finite. For every $s\in S,$ define $n$ reals $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ as follows: $r_{s,i}=1+\left[i=s\right]-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\ \ \ \ \ \ \ \ \ \ \text{for every }i\in\left\\{1,2,...,n\right\\}.$ These reals $r_{s,1},$ $r_{s,2},$ $...,$ $r_{s,n}$ are all nonnegative (because $r_{s,i}=1+\underbrace{\left[i=s\right]}_{\geq 0}-\underbrace{\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]}_{\leq 1}\geq 1+0-1=0$ for every $i\in\left\\{1,2,...,n\right\\}$). Also, for every $s\in S,$ set $b_{s}=1;$ then, $b_{s}$ is a nonnegative real as well. For every $i\in\left\\{1,2,...,n\right\\},$ we have $\sum\limits_{s=1}^{n}\left[i=s\right]=1$ (because there exists one and only one $s\in\left\\{1,2,...,n\right\\}$ satisfying $i=s$). Also, for every $i\in\left\\{1,2,...,n\right\\},$ we have $\sum\limits_{s=1}^{n}\left[s\equiv i-r\mathop{\mathrm{m}od}n\right]=1$ (because there exists one and only one $s\in\left\\{1,2,...,n\right\\}$ satisfying $s\equiv i-r\mathop{\mathrm{m}od}n$). In other words, $\sum\limits_{s=1}^{n}\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]=1$ (because $\left[s\equiv i-r\mathop{\mathrm{m}od}n\right]=\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]$). For every $i\in\left\\{1,2,...,n\right\\},$ we have $\displaystyle\sum_{s\in S}b_{s}r_{s,i}$ $\displaystyle=\sum_{s=1}^{n}\underbrace{b_{s}}_{=1}r_{s,i}=\sum_{s=1}^{n}r_{s,i}=\sum_{s=1}^{n}\left(1+\left[i=s\right]-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right)$ $\displaystyle=\sum_{s=1}^{n}1+\sum_{s=1}^{n}\left[i=s\right]-\sum_{s=1}^{n}\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]=n+1-1=n=2+\left(n-2\right)=a_{i}+a,$ so that $a_{i}+a=\sum_{s\in S}b_{s}r_{s,i}.$ (21) For any two integers $i$ and $j$ from $\left\\{1,2,...,n\right\\},$ we have $\displaystyle\sum_{s=1}^{n}\left|r_{s,i}-1\right|$ $\displaystyle=\sum_{s=1}^{n}\left|\left(1+\left[i=s\right]-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right)-1\right|$ $\displaystyle=\sum_{s=1}^{n}\left|\left[i=s\right]+\left(-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right)\right|\leq\sum_{s=1}^{n}\left(\left|\left[i=s\right]\right|+\left|-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right|\right)$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\text{by the triangle inequality}\right)$ $\displaystyle=\sum_{s=1}^{n}\left(\left[i=s\right]+\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right)$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left(\begin{array}[]{c}\text{because }\left[i=s\right]\text{ and }\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\text{ are nonnegative, so that}\\\ \left|\left[i=s\right]\right|=\left[i=s\right]\text{ and }\left|-\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\right|=\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]\end{array}\right)$ $\displaystyle=\sum_{s=1}^{n}\left[i=s\right]+\sum_{s=1}^{n}\left[i\equiv s+r\mathop{\mathrm{m}od}n\right]=1+1=2$ and similarly $\sum\limits_{s=1}^{n}\left|r_{s,j}-1\right|\leq 2,$ so that $\displaystyle\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|$ $\displaystyle=\sum_{s=1}^{n}\underbrace{b_{s}}_{=1}\left|r_{s,i}-r_{s,j}\right|=\sum_{s=1}^{n}\left|r_{s,i}-r_{s,j}\right|=\sum_{s=1}^{n}\left|\left(r_{s,i}-1\right)+\left(1-r_{s,j}\right)\right|$ $\displaystyle\leq\sum_{s=1}^{n}\left(\left|r_{s,i}-1\right|+\left|1-r_{s,j}\right|\right)\ \ \ \ \ \ \ \ \ \ \left(\text{by the triangle inequality}\right)$ $\displaystyle=\sum_{s=1}^{n}\left(\left|r_{s,i}-1\right|+\left|r_{s,j}-1\right|\right)=\sum_{s=1}^{n}\left|r_{s,i}-1\right|+\sum_{s=1}^{n}\left|r_{s,j}-1\right|$ $\displaystyle\leq 2+2=a_{i}+a_{j},$ and thus $a_{i}+a_{j}\geq\sum_{s\in S}b_{s}\left|r_{s,i}-r_{s,j}\right|.$ (22) For every $s\in S$ (that is, for every $s\in\left\\{1,2,...,n\right\\}$), we have $\sum\limits_{v=1}^{n}\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\cdot w_{v}=\sum\limits_{v=1}^{n}\left\\{\begin{array}[]{c}w_{v},\text{ if }v\equiv s+r\mathop{\mathrm{m}od}n;\\\ 0\text{ otherwise}\end{array}\right.=w_{s+r}$ (because there is one and only one element $v\in\left\\{1,2,...,n\right\\}$ that satisfies $v\equiv s+r\mathop{\mathrm{m}od}n,$ and for this element $v,$ we have $w_{v}=w_{s+r}$), so that $\displaystyle\sum\limits_{v=1}^{n}r_{s,v}w_{v}$ $\displaystyle=\sum\limits_{v=1}^{n}\left(1+\left[v=s\right]-\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\right)\cdot w_{v}$ $\displaystyle=\underbrace{\sum\limits_{v=1}^{n}w_{v}}_{=w}+\underbrace{\sum\limits_{v=1}^{n}\left[v=s\right]\cdot w_{v}}_{=w_{s}}-\underbrace{\sum\limits_{v=1}^{n}\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\cdot w_{v}}_{=w_{s+r}}$ $\displaystyle=w+w_{s}-w_{s+r}=w+\left(w_{s}-w_{s+r}\right).$ Also, for every $s\in S$ (that is, for every $s\in\left\\{1,2,...,n\right\\}$), we have $\sum\limits_{v=1}^{n}\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\cdot w_{v}x_{v}=\sum\limits_{v=1}^{n}\left\\{\begin{array}[]{c}w_{v}x_{v},\text{ if }v\equiv s+r\mathop{\mathrm{m}od}n;\\\ 0\text{ otherwise}\end{array}\right.=w_{s+r}x_{s+r}$ (because there is one and only one element $v\in\left\\{1,2,...,n\right\\}$ that satisfies $v\equiv s+r\mathop{\mathrm{m}od}n,$ and for this element $v,$ we have $w_{v}=w_{s+r}$ and $x_{v}=x_{s+r}$), and thus $\displaystyle\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}$ $\displaystyle=\sum\limits_{v=1}^{n}\left(1+\left[v=s\right]-\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\right)\cdot w_{v}x_{v}$ $\displaystyle=\sum\limits_{v=1}^{n}w_{v}x_{v}+\underbrace{\sum\limits_{v=1}^{n}\left[v=s\right]\cdot w_{v}x_{v}}_{=w_{s}x_{s}}-\underbrace{\sum\limits_{v=1}^{n}\left[v\equiv s+r\mathop{\mathrm{m}od}n\right]\cdot w_{v}x_{v}}_{=w_{s+r}x_{s+r}}$ $\displaystyle=\sum\limits_{v=1}^{n}w_{v}x_{v}+w_{s}x_{s}-w_{s+r}x_{s+r}=\sum\limits_{v=1}^{n}w_{v}x_{v}+\left(w_{s}x_{s}-w_{s+r}x_{s+r}\right).$ Now it is clear that $\sum\limits_{v=1}^{n}r_{s,v}w_{v}\neq 0$ for all $s\in S$ (because $\sum\limits_{v=1}^{n}r_{s,v}w_{v}=w+\left(w_{s}-w_{s+r}\right)$ and $w+\left(w_{s}-w_{s+r}\right)\neq 0$). Also, $\sum\limits_{v=1}^{n}w_{v}\neq 0$ (since $\sum\limits_{v=1}^{n}w_{v}=w$ and $w\neq 0$). Using these two relations, the conditions of Theorem 15b and the relations (16) and (17), we see that all conditions of Theorem 14 are fulfilled. Hence, we can apply Theorem 14 and obtain $\sum_{i=1}^{n}\underbrace{a_{i}}_{=2}w_{i}f\left(x_{i}\right)+\underbrace{a}_{=n-2}\left(\underbrace{\sum_{v=1}^{n}w_{v}}_{=w}\right)f\left(\underbrace{\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}}{\sum\limits_{v=1}^{n}w_{v}}}_{=x}\right)\geq\sum_{s\in S}\underbrace{b_{s}}_{=1}\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right).$ This immediately simplifies to $\sum_{i=1}^{n}2w_{i}f\left(x_{i}\right)+\left(n-2\right)wf\left(x\right)\geq\sum_{s\in S}1\left(\sum_{v=1}^{n}r_{s,v}w_{v}\right)f\left(\frac{\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}}{\sum\limits_{v=1}^{n}r_{s,v}w_{v}}\right).$ Recalling that for every $s\in S,$ we have $\sum\limits_{v=1}^{n}r_{s,v}w_{v}=w+\left(w_{s}-w_{s+r}\right)$ and $\sum\limits_{v=1}^{n}r_{s,v}w_{v}x_{v}=\sum\limits_{v=1}^{n}w_{v}x_{v}+\left(w_{s}x_{s}-w_{s+r}x_{s+r}\right),$ we can rewrite this as $\sum_{i=1}^{n}2w_{i}f\left(x_{i}\right)+\left(n-2\right)wf\left(x\right)\geq\sum_{s\in S}1\left(w+\left(w_{s}-w_{s+r}\right)\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}+\left(w_{s}x_{s}-w_{s+r}x_{s+r}\right)}{w+\left(w_{s}-w_{s+r}\right)}\right).$ In other words, $2\sum_{i=1}^{n}w_{i}f\left(x_{i}\right)+\left(n-2\right)wf\left(x\right)\geq\sum_{s=1}^{n}\left(w+\left(w_{s}-w_{s+r}\right)\right)f\left(\frac{\sum\limits_{v=1}^{n}w_{v}x_{v}+\left(w_{s}x_{s}-w_{s+r}x_{s+r}\right)}{w+\left(w_{s}-w_{s+r}\right)}\right).$ This proves Theorem 15b. Theorem 15a follows from Theorem 15b if we set $w_{1}=w_{2}=...=w_{n}=1.$ Theorem 15a generalizes two inequalities that have appeared on the MathLinks forum. The first of these results if we apply Theorem 15a to $r=1,$ to $r=2,$ to $r=3,$ and so on up to $r=n-1,$ and sum up the $n-1$ inequalities obtained: > Theorem 16. Let $f$ be a convex function from an interval > $I\subseteq\mathbb{R}$ to $\mathbb{R}.$ Let $x_{1},$ $x_{2},$ $...,$ $x_{n}$ > be finitely many points from $I.$ > > Let $x=\dfrac{x_{1}+x_{2}+...+x_{n}}{n}$. Then, > > > $2\left(n-1\right)\sum_{i=1}^{n}f\left(x_{i}\right)+n\left(n-1\right)\left(n-2\right)f\left(x\right)\geq > n\sum_{\begin{subarray}{c}1\leq i\leq n;\\\ 1\leq j\leq n;\\\ i\neq > j\end{subarray}}f\left(x+\dfrac{x_{i}-x_{j}}{n}\right).$ This inequality occured in [6], post #4 as a result by Vasile Cîrtoaje (Vasc). Our Theorem 15a is therefore a strengthening of this result. The next inequality was proposed by Michael Rozenberg (aka ”Arqady”) in [7]: > Theorem 17. Let $a,$ $b,$ $c,$ $d$ be four nonnegative reals. Then, > > $a^{4}+b^{4}+c^{4}+d^{4}+4abcd\geq > 2\left(a^{2}bc+b^{2}cd+c^{2}da+d^{2}ab\right).$ Proof of Theorem 17. The case when at least one of the reals $a,$ $b,$ $c,$ $d$ equals $0$ is easy (and a limiting case). Hence, we can assume for the rest of this proof that none of the reals $a,$ $b,$ $c,$ $d$ equals $0.$ Since $a,$ $b,$ $c,$ $d$ are nonnegative, this means that $a,$ $b,$ $c,$ $d$ are positive. Let $A=\ln\left(a^{4}\right),$ $B=\ln\left(b^{4}\right),$ $C=\ln\left(c^{4}\right),$ $D=\ln\left(d^{4}\right).$ Then, $\exp A=a^{4},$ $\exp B=b^{4},$ $\exp C=c^{4},$ $\exp D=d^{4}.$ Let $I\subseteq\mathbb{R}$ be an interval containing the reals $A,$ $B,$ $C,$ $D$ (for instance, $I=\mathbb{R}$). Let $f:I\rightarrow\mathbb{R}$ be the function defined by $f\left(x\right)=\exp x$ for all $x\in I.$ Then, it is known that this function $f$ is convex. Applying Theorem 15a to $n=4,$ $x_{1}=A,$ $x_{2}=B,$ $x_{3}=C,$ $x_{4}=D,$ and $r=3,$ we obtain $\displaystyle 2\left(f\left(A\right)+f\left(B\right)+f\left(C\right)+f\left(D\right)\right)+4\left(4-2\right)f\left(\dfrac{A+B+C+D}{4}\right)$ $\displaystyle\geq 4\left(f\left(\dfrac{A+B+C+D}{4}+\dfrac{A-D}{4}\right)+f\left(\dfrac{A+B+C+D}{4}+\dfrac{B-A}{4}\right)\right.$ $\displaystyle\ \ \ \ \ \ \ \ \ \ \left.+f\left(\dfrac{A+B+C+D}{4}+\dfrac{C-B}{4}\right)+f\left(\dfrac{A+B+C+D}{4}+\dfrac{D-C}{4}\right)\right).$ Dividing this by $2$ and simplifying, we obtain $\displaystyle f\left(A\right)+f\left(B\right)+f\left(C\right)+f\left(D\right)+4f\left(\dfrac{A+B+C+D}{4}\right)$ $\displaystyle\geq 2\left(f\left(\dfrac{2A+B+C}{4}\right)+f\left(\dfrac{2B+C+D}{4}\right)+f\left(\dfrac{2C+D+A}{4}\right)+f\left(\dfrac{2D+A+B}{4}\right)\right).$ Since we have $\displaystyle f\left(A\right)$ $\displaystyle=\exp A=a^{4}\ \ \ \ \ \ \ \ \ \ \text{and similarly}$ $\displaystyle f\left(B\right)$ $\displaystyle=b^{4},\text{ }f\left(C\right)=c^{4},\text{ and }f\left(D\right)=d^{4};$ $\displaystyle f\left(\dfrac{A+B+C+D}{4}\right)$ $\displaystyle=\exp\dfrac{A+B+C+D}{4}=\sqrt[4]{\exp A\cdot\exp B\cdot\exp C\cdot\exp D}$ $\displaystyle=\sqrt[4]{a^{4}\cdot b^{4}\cdot c^{4}\cdot d^{4}}=abcd;$ $\displaystyle f\left(\dfrac{2A+B+C}{4}\right)$ $\displaystyle=\exp\dfrac{2A+B+C}{4}=\sqrt[4]{\left(\exp A\right)^{2}\cdot\exp B\cdot\exp C}$ $\displaystyle=\sqrt[4]{\left(a^{4}\right)^{2}\cdot b^{4}\cdot c^{4}}=a^{2}bc\ \ \ \ \ \ \ \ \ \ \text{and similarly}$ $\displaystyle f\left(\dfrac{2B+C+D}{4}\right)$ $\displaystyle=b^{2}cd,\text{ }f\left(\dfrac{2C+D+A}{4}\right)=c^{2}da,\text{ and }f\left(\dfrac{2D+A+B}{4}\right)=d^{2}ab,$ this becomes $a^{4}+b^{4}+c^{4}+d^{4}+4abcd\geq 2\left(a^{2}bc+b^{2}cd+c^{2}da+d^{2}ab\right).$ This proves Theorem 17. References [1] Vasile Cîrtoaje, Two Generalizations of Popoviciu’s Inequality, Crux Mathematicorum 5/2001 (volume 31), pp. 313-318. http://journals.cms.math.ca/CRUX/ [2] Billzhao et al., Generalized Popoviciu - MathLinks topic #19097. http://www.mathlinks.ro/Forum/viewtopic.php?t=19097 [3] Billzhao et al., Like Popoviciu - MathLinks topic #21786. http://www.mathlinks.ro/Forum/viewtopic.php?t=21786 [4] Darij Grinberg et al., The Karamata Inequality - MathLinks topic #14975. http://www.mathlinks.ro/Forum/viewtopic.php?t=14975 [5] Darij Grinberg et al., Weighted majorization and a result stronger than Fuchs - MathLinks topic #104714. http://www.mathlinks.ro/Forum/viewtopic.php?t=104714 [6] Harazi et al., improvement of Popoviciu’s inequality in a particular case - MathLinks topic #22364. http://www.mathlinks.ro/Forum/viewtopic.php?t=22364 [7] Arqady et al., New, old inequality - MathLinks topic #56040. http://www.mathlinks.ro/Forum/viewtopic.php?t=56040

arxiv-papers

2024-09-04T02:48:54.435168

0803.3008

# A characterization of surfaces whose universal cover is the bidisk Fabrizio Catanese, Marco Franciosi (Date: March 25, 2008) ###### Abstract. We show that the universal cover of a compact complex surface $X$ is the bidisk $\mathbb{H}\times\mathbb{H}$, or $X$ is biholomorphic to $\mathbb{P}^{1}\times\mathbb{P}^{1}$, if and only if $K_{X}^{2}>0$ and there exists an invertible sheaf $\eta$ such that $\eta^{2}\cong\mathcal{O}_{X}$ and $H^{0}(X,S^{2}\Omega^{1}_{X}(-K_{X})\otimes\eta)\neq 0$. The two cases are distinguished by the second plurigenus, $P_{2}(X)\geq 2$ in the former case, $P_{2}(X)=0$ in the latter. We also discuss related questions. ## 1\. Introduction The beauty of the theory of algebraic curves is deeply related to the manifold implications of the: ###### Theorem 1.1 (Uniformization theorem of Koebe and Poincaré). A connected and simply connected complex curve $\tilde{C}$ is biholomorphic to: $\tilde{C}\cong\left\\{\begin{array}[]{ll}{\mathbb{P}}^{1}&\mbox{ if }\ g=0\\\ \mathbb{C}&\mbox{ if }\ g=1\\\ \mathbb{H}&\mbox{ if }\ g\geq 2\\\ \end{array}\right.$ ($\mathbb{H}$ denotes as usual the Poincaré upper half-plane $\mathbb{H}=\\{\tau\in\mathbb{C}:Im(\tau)>0\\}$, but we shall often refer to it as the ‘disk’ since it is biholomorphic to $\\{z\in\mathbb{C}:||z||<1\\}$). Hence a smooth (connected) compact complex curve $C$ of genus $g\geq 1$ admits a uniformization in the strong sense (iii) of the following definition: ###### Definition 1.2. A connected complex space $X$ of complex dimension $n$ admits a uniformization if one of the following conditions hold: 1. (i) there is a connected open set $\Omega\subset\mathbb{C}^{n}$ and a surjective holomorphic map $f\colon\Omega\rightarrow X$ (weak uniformization); 2. (ii) there is a connected open set $\Omega\subset\mathbb{C}^{n}$ and a properly discontinuous group ${\Gamma}\subset Aut(\Omega)$ such that $\Omega/{\Gamma}\cong X$ (Galois uniformization). If $X$ is a complex manifold, there are two stronger properties: 1. (iii) there is a connected open set $\Omega\subset\mathbb{C}^{n}$ and a surjective holomorphic submersion $f\colon\Omega\rightarrow X$ (étale uniformization); 2. (iv) there is a connected open set $\Omega\subset\mathbb{C}^{n}$ biholomorphic to the universal cover of $X$ (strong uniformization). Hence the universal cover of a compact complex curve is completely determined by its genus; in particular $\tilde{C}\cong\mathbb{H}$ if and only if $g\geq 2$, i.e., “$C$ is of general type”, and we get then an isomorphism of $\pi_{1}(C)$ with a Fuchsian group ${\Gamma}\subset\operatorname{Aut}(\mathbb{H})\cong\operatorname{PSL}(2,\mathbb{R})$. In higher dimension the condition that the universal cover be biholomorphic to a bounded domain $\Omega$ is quite exceptional; but still in the Galois étale case, where $\Omega/{\Gamma}\cong X$ and ${\Gamma}$ acts freely with compact quotient, we have, if $\Omega$ is bounded, that the complex manifold $X$ has ample canonical bundle $K_{X}$ (see [Sieg73]), in particular it is a projective manifold of general type. Even more exceptional is the case where the universal cover is biholomorphic to a bounded symmetric domain $\Omega$, or where there is Galois uniformization (ii) of definition 1.2) with source a bounded symmetric domain, and there is a vast literature on a characterization of these properties (cf. [Yau77], [Yau88], [Yau93], [Bea00]). The basic result in this direction is S.T. Yau’s uniformization theorem (explained in [Yau88] and [Yau93]), and for which a very readable exposition is contained in the first section of [V-Z05], enphasyzing the role of polystability of the cotangent bundle for varieties of general type. One would wish nevertheless for more precise characterizations of the various possible cases. For the sake of simplicity, we shall stick here to the case of smooth complex surfaces, where the former problem boils down to two very specific questions. Question. When is the universal cover of a compact complex surface $X$ biholomorphic to the two dimensional ball ${\mathbb{B}}_{2}:=\\{z\in\mathbb{C}^{2}:||z||<1\\}$, respectively to the bidisk $\mathbb{H}\times\mathbb{H}$ ? The first part of this question is fully answered by the well-known inequality by Miyaoka and Yau (cf. [Miy77], [Yau77] [Miy82]). Setting, as usual, $K_{X}=$ the canonical divisor, $\chi(X):=\chi(\mathcal{O}_{X})$ the holomorphic Euler characteristic and $P_{2}(X)=h^{0}(X,2K_{X})$ the second plurigenus of $X$, we have the following characterization: ###### Theorem 1.3 (Miyaoka-Yau). Let $X$ be a compact complex surface. Then $X\cong{\mathbb{B}}_{2}/\Gamma$ (with $\Gamma$ a cocompact discrete subgroup of $\operatorname{Aut}({\mathbb{B}}_{2})$ acting freely on ${\mathbb{B}}_{2}$) if and only if 1. (1) $K_{X}^{2}=9\chi(S)$; 2. (2) the second plurigenus $P_{2}(X)>0$. The above well known characterization is obtained combining Miyaoka’s result ([Miy82]) that these two conditions imply the ampleness of $K_{X}$, with Yau’s uniformization result ([Yau77]) which uses the existence of a Kähler Einstein metric; quite remarkably, it is given purely in terms of certain numbers which are either bimeromorphic or topological invariants. In the case where $X=\mathbb{H}\times\mathbb{H}/\Gamma$, with $\Gamma$ a discrete cocompact subgroup of $\operatorname{Aut}(\mathbb{H}\times\mathbb{H})$ acting freely, one has $K_{X}^{2}=8\chi(X)$. But Moishezon and Teicher in [MT87] showed the existence of a simply connected surface of general type (whence with $P_{2}(X)>0$) having $K_{X}^{2}=8\chi(X)$, so that the above conditions are necessary, but not sufficient. We observe however that (and our contribution here is a by-product of our attempt to answer the latter question) it is still unknown if there exists a surface of general type with $\chi(X)=1,K_{X}^{2}=8$ which is not uniformized by $\mathbb{H}\times\mathbb{H}$. The purpose of this note is to point out a precise characterization of compact complex surfaces whose universal cover is the bidisk, and of the quadric $\mathbb{P}^{1}\times\mathbb{P}^{1}$, discussing whether some hypotheses can be dispensed with, and to pose an analogous question in higher dimension. Our characterization, which is of course based on Yau’s results, relies on the following crucial ###### Definition 1.4. Let $X$ be a complex manifold of complex dimension $n$. Then a special tensor is a non zero section $0\neq\omega\in H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X}))$, while a semi special tensor is a non zero section $0\neq\omega\in H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X})\otimes\eta)$, where $\eta$ is an invertible sheaf such that $\eta^{2}\cong\mathcal{O}_{X}$. We shall say that $X$ admits a unique semi special tensor if moreover $dim(H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X})\otimes\eta))=1.$ In fact, the existence of such tensors is a fundamental property of manifolds strongly uniformized by the polydisk as we are now going to see. Recall that the group of automorphism of $\mathbb{H}^{n}$, $\operatorname{Aut}(\mathbb{H}^{n})$, is the semidirect product of $(\operatorname{Aut}(\mathbb{H}))^{n}$ with the symmetric group ${\mathfrak{S}}_{n}$, hence for every subgroup ${\Gamma}$ of $\operatorname{Aut}(\mathbb{H}^{n})$ we have a diagram: $\begin{array}[]{ccccccc}1\rightarrow&(\operatorname{Aut}(\mathbb{H}))^{n}&\rightarrow&\operatorname{Aut}(\mathbb{H}^{n})&\rightarrow{\mathfrak{S}}_{n}&\rightarrow 1\\\ &\bigcup&&\bigcup&\bigcup&\\\ 1\rightarrow&\Gamma^{0}&\hookrightarrow&\Gamma&\rightarrow H&\rightarrow 1.\\\ \end{array}$ ###### Proposition 1.5. Let $X=\mathbb{H}^{n}/{\Gamma}$ be a compact complex manifold whose universal covering is the polydisk $\mathbb{H}^{n}$: then $X$ admits a semi special tensor and $K_{X}$ is ample, in particular $K_{X}^{n}>0$. ###### Proof. In $\mathbb{H}^{n}$ take coordinates $\\{z_{1},\dots,z_{n}\\}$ and define $\tilde{\omega}:=\frac{\operatorname{d}z_{1}\otimes\dots\otimes\operatorname{d}z_{n}}{\operatorname{d}z_{1}\wedge\dots\wedge\operatorname{d}z_{n}}.$ Observe that $\tilde{\omega}$ is clearly invariant for $(\operatorname{Aut}(\mathbb{H}))^{n}$ and for the alternating subgroup. Let $\eta$ be the 2-torsion invertible sheaf associated to the signature character of ${\mathfrak{S}}_{n}$ restricted to $H$. Then clearly $\tilde{\omega}$ descends to a semi special tensor $\omega\in H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X})\otimes\eta)$. The other assertions are well known (cf. [Sieg73] and [K-M71]). ∎ ###### Remark 1.6. We observe that also $(\mathbb{P}^{1})^{n}$ admits the following special tensor $\omega$, given on $\mathbb{C}^{n}\subset(\mathbb{P}^{1})^{n}$ by $\omega:=\frac{\operatorname{d}z_{1}\otimes\dots\otimes\operatorname{d}z_{n}}{\operatorname{d}z_{1}\wedge\dots\wedge\operatorname{d}z_{n}}.$ In dimension two we have then the following ###### Theorem 1.7. Let $X$ be a compact complex surface. Then the following two conditions: 1. (1) $X$ admits a semi special tensor; 2. (2) $K_{X}^{2}>0$ hold if and only if either 1. (i) $X\cong\mathbb{P}^{1}\times\mathbb{P}^{1}$; or 2. (ii) $X\cong\mathbb{H}\times\mathbb{H}/\Gamma$ (where $\Gamma$ is a cocompact discrete subgroup of $\operatorname{Aut}(\mathbb{H}\times\mathbb{H})$ acting freely ). In particular one has the following reformulation of a theorem of S.T. Yau (theorem 2.5 of [Yau93], giving sufficient conditions for (ii) to hold). ###### Theorem 1.8. (Yau) $X$ is strongly uniformized by the bidisk if and only if 1. (1) $X$ admits a semi special tensor; 2. (2) $K_{X}^{2}>0$; 3. (3) the second plurigenus $P_{2}(X)\geq 1$. One can indeed be even more precise: ###### Theorem 1.9. $X$ is strongly uniformized by the bidisk if and only if 1. (1*) $X$ admits a unique semi special tensor; 2. (2) $K_{X}^{2}>0$; 3. (3*) the second plurigenus $P_{2}(X)\geq 2$. $X$ is biholomorphic to $\mathbb{P}^{1}\times\mathbb{P}^{1}$ if and only if (1*), (2) hold and $P_{2}(X)=0.$ It is interesting to see that none of the above hypotheses can be dispensed with. ###### Remark 1.10. The following examples show the existence of surfaces which satisfy two of the three conditions stated in Thm. 1.8, respectively in Thm. 1.9, but are not uniformized by the bidisk 1. (i) $\mathbb{P}^{1}\times\mathbb{P}^{1}$ satisfies (1*) and (2); 2. (ii) A complex torus $X=\mathbb{C}^{2}/\Lambda$ satisfies (1) and (3), but neither (1*) nor (3*) (obviously, it does not satisfy (2)); 3. (iii) $X=C_{1}\times C_{2}$ with $g(C_{1})=1$, $g(C_{2})=2$ satisfies (1*) and (3*), but its universal cover is $\tilde{X}\cong\mathbb{C}\times\mathbb{H}$. The most intriguing examples are provided by ###### Proposition 1.11. There do exist properly elliptic surfaces $X$ satisfying * • (1) $X$ admits a special tensor; * • (3*) the second plurigenus $P_{2}(X)\geq 2$; * • $q(X):=dim(H^{1}(\mathcal{O}_{X}))>0$; * • $K_{X}^{2}=0$; * • $X$ is not birational to a product. We would like to pose then the following Question. Let $X$ be a surface with $q(X)=0$ and satisfying (1*) and (3*): is then $X$ strongly uniformized by the bidisk? Concerning the above question, recall the following ###### Definition 1.12. $\Gamma\subset\operatorname{Aut}(\mathbb{H}^{n})$ is said to be reducible if there exists $\Gamma^{0}$ as above (i.e., such that $\gamma(z_{1},...,z_{n})=(\gamma_{1}(z_{1}),...,\gamma_{n}(z_{n}))$ for every $\gamma\in\Gamma^{0}$) and a decomposition $\mathbb{H}^{n}=\mathbb{H}^{k}\times\mathbb{H}^{h}$ (with $h>0$) such that the action of $\Gamma^{0}$ on $\mathbb{H}^{k}$ is discrete. For $n=2$ there are then only two alternatives: ###### Remark 1.13. Let $\Gamma\subset\operatorname{Aut}(\mathbb{H}^{2})$ be a discrete cocompact subgroup acting freely and let $X=\mathbb{H}^{2}/\Gamma$. Then * • $\Gamma$ is reducible if and only if $X$ is isogenous to a product of curves, i.e., there is a finite group $G$ and two curves of genera at least 2 such that $X\cong C_{1}\times C_{2}/G$. Both cases $q(X)\neq 0$, $q(X)=0$ can occur here. * • $\Gamma$ is irreducible and $q(X)=0$ ( this result holds in all dimensions and is a well-known result of Matsushima [Ma62]). Let us try to explain the main idea of our main result. In order to do this, it is important to make the following ###### Remark 1.14. A complex manifold $X$ admits a semi special tensor if and only if it has an unramified cover $X^{\prime}$ of degree at most two which admits a special tensor. ###### Proof. Assume that we have an invertible sheaf $\eta$ such that $\eta^{2}\cong\mathcal{O}_{X}$, $\eta\not\cong\mathcal{O}_{X}$. Take the corresponding double connected étale covering $\pi:X^{\prime}\rightarrow X$ and observe that $H^{0}(X^{\prime},S^{n}\Omega^{1}_{X^{\prime}}(-K_{X^{\prime}}))\cong H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X}))\oplus H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X})\otimes\eta).$ Whence, there is a special tensor on $X^{\prime}$ if and only if there is a semi special tensor on $X$. ∎ In dimension $n=2$ things are easier, since the existence of a special tensor $\omega$ is equivalent to the existence of a trace free endomorphism $\epsilon$ of the tangent bundle of $X$. Our proof of Theorem 1.7 consists essentially in finding a decomposition of the tangent bundle $T_{X}$ as a direct sum of two line bundles $L_{1}$ and $L_{2}$, which are the eigenbundles of an invertible endomorphism $\epsilon\in\operatorname{End}(T_{X})$ (see §2 and §3 for details), and then applying the results on surfaces with split tangent bundles as given in [Bea00]. Since the results on manifolds with split tangent bundles hold in dimension $n\geq 3$, one has a characterization of compact manifolds strongly uniformized by the polydisk under a very strong condition on the semi special tensor $\omega\in H^{0}(X,S^{n}\Omega^{1}_{X}(-K_{X})\otimes\eta)$, which essentially corresponds to ask for the local splitting of $\omega$ as the product of $n$ 1-forms which are linearly independent at each point. There remains the problem of finding a simpler characterization. ## 2\. Preliminaries and remarks Notation. $X$ denotes throughout a compact complex surface. We use standard notation of algebraic geometry: $\Omega^{1}_{X}$ is the cotangent sheaf, $T_{X}$ is the holomorphic tangent bundle (locally free sheaf), $c_{1}(X)$, $c_{2}(X)$ are the Chern classes of $X$; $K_{X}$ is the canonical divisor, and $P_{n}:=h^{0}(X,nK_{X})$ is called the $n$-th plurigenus, in particular for $n=1$ we have the geometric genus of $X$ $p_{g}(X):=h^{0}(X,K_{X})$, while $q:=h^{1}(X,\mathcal{O}_{X})$ is classically called the irregularity of $X$. Finally, $\chi(X):=\chi(\mathcal{O}_{X})=1+p_{g}-q$ is the holomorphic Euler characteristic. With a slight abuse of notation, we do not distinguish between invertible sheaves, line bundles and divisors, while the symbol $\equiv$ denotes linear equivalence of divisors. First of all let us recall a result of Beauville which characterizes compact complex surfaces whose universal cover is a product of two complex curves (cf. [Bea00, Thm. C]). ###### Theorem 2.1 (Beauville). Let $X$ be a compact complex surface. The tangent bundle $T_{X}$ splits as a direct sum of two line bundles if and only if either $X$ is a special Hopf surface or the universal covering space of $X$ is a product $U\times V$ of two complex curves and the group $\pi_{1}(X)$ acts diagonally on $U\times V$. Given a direct sum decomposition of the cotangent bundle $\Omega^{1}_{X}\cong L_{1}\oplus L_{2}$, Beauville shows that $(L_{1})^{2}=(L_{2})^{2}=0$ (cf. [Bea00, 4.1, 4.2]) hence $K_{X}\equiv L_{1}+L_{2}\hskip 28.45274ptc_{1}(X)^{2}=2\cdot(L_{1}\cdot L_{2})=2\cdot c_{2}(X)$ The last equality corresponds to $K_{X}^{2}=8\chi(X)$. Let us now consider the bundle $\operatorname{End}(T_{X})$ of endomorphisms of the tangent bundle. We can write $\operatorname{End}(T_{X})=\Omega^{1}_{X}\otimes T_{X}$ and from the nondegenerate bilinear map $\Omega^{1}_{X}\times\Omega^{1}_{X}\longrightarrow\Omega^{2}_{X}\cong K_{X}$ we see that $T_{X}=(\Omega^{1}_{X})^{\vee}\cong\Omega^{1}_{X}(-K_{X})$. This exactly means that we have an isomorphism $\operatorname{End}(T_{X})\cong\Omega^{1}_{X}\otimes\Omega^{1}_{X}(-K_{X}).$ Let us see how this isomorphism works in local coordinates $(z_{1},z_{2})$. I.e., let us see how an element $\frac{\operatorname{d}z_{i}\otimes\operatorname{d}z_{j}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}$ in $\Omega^{1}_{X}\otimes\Omega^{1}_{X}(-K_{X})$ acts on a vector of the form $\frac{\partial}{\partial z_{h}}$. We have $\frac{\operatorname{d}z_{i}\otimes\operatorname{d}z_{j}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}\bigl{(}\frac{\partial}{\partial z_{h}}\bigr{)}=\left\\{\begin{array}[]{cl}\frac{\operatorname{d}z_{j}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}&\mbox{ if }h=i\\\ 0&\mbox{ if }h\neq i\\\ \end{array}\right.$ In turn, $\displaystyle{\frac{\operatorname{d}z_{j}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}}$ evaluated on $\operatorname{d}z_{k}$ gives $\displaystyle{\frac{\operatorname{d}z_{j}\wedge\operatorname{d}z_{k}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}}$. Therefore a generic element $\displaystyle{\sum_{i,j}a_{ij}\frac{\operatorname{d}z_{i}\otimes\operatorname{d}z_{j}}{\operatorname{d}z_{1}\wedge\operatorname{d}z_{2}}}$ corresponds to an endomorphism, which, with respect to the basis $\bigl{\\{}\frac{\partial}{\partial z_{1}},\frac{\partial}{\partial z_{2}}\bigr{\\}}$ is expressed by the matrix $\begin{pmatrix}-a_{12}&-a_{22}\\\ a_{11}&a_{21}\\\ \end{pmatrix}$ In particular for the symmetric tensors (i.e., $a_{12}=a_{21}$), respectively for the skewsymmetric tensors (i.e., $a_{12}=-a_{21},a_{11}=a_{22}=0$) the following isomorphisms hold: $S^{2}(\Omega^{1}_{X})(-K_{X})\cong\bigg{\\{}\begin{pmatrix}-a&-a_{22}\\\ a_{11}&a\\\ \end{pmatrix}\bigg{\\}}\ ;\ \ \hskip 14.22636pt{\bigwedge}^{2}(\Omega^{1}_{X})(-K_{X})\cong\bigg{\\{}\begin{pmatrix}b&0\\\ 0&b\\\ \end{pmatrix}\bigg{\\}}$ We can summarize the above discussion in the following ###### Lemma 2.2. If $X$ is a complex surface there is a natural isomorphism between the sheaf $S^{2}(\Omega^{1}_{X})(-K_{X})$ and the sheaf of trace zero endomorphisms of the (co)tangent sheaf $\operatorname{End}^{0}(T_{X})\cong\operatorname{End}^{0}(\Omega^{1}_{X})$. A special tensor $\omega\in H^{0}(S^{2}(\Omega^{1}_{X})(-K_{X}))$ with nonzero determinant $det(\omega)\in\mathbb{C}$ yields an eigenbundle splitting $\Omega^{1}_{X}\cong L_{1}\bigoplus L_{2}$ of the cotangent bundle. If instead $det(\omega)=0\in\mathbb{C}$, the corresponding endomorphism $\epsilon$ is nilpotent and yields an exact sequence of sheaves $0\rightarrow L\rightarrow\Omega^{1}_{X}\rightarrow{\mathcal{I}}_{Z}L(-\Delta)\rightarrow 0$ where $L:=ker(\epsilon)$ is invertible, $\Delta$ is an effective divisor, and $Z$ is a 0-dimensional subscheme(which is a local complete intersection). We have in particular $K_{X}\equiv 2L-\Delta$ and $c_{2}(X)={length}(Z)+L\cdot(L-\Delta)$. Proof. We need only to observe that $det(\omega)$ is a constant, since $det(\operatorname{End}(T_{X}))=det(\operatorname{End}(\Omega^{1}_{X}))\cong\mathcal{O}_{X}$. If $det(\omega)\neq 0$, there is a constant $c\in\mathbb{C}\setminus\\{0\\}$ such that $det(\omega)=c^{2}$, hence at every point of $X$ the endomorphism $\epsilon$ corresponding to the special tensor $\omega$ has two distinct eigenvalues $\pm c$. Let $\omega\in H^{0}(S^{2}\Omega^{1}_{X}(-K_{X}))$, $\omega\neq 0$, be such that $\det(\omega)=0$. Then the corresponding endomorphism $\epsilon$ is nilpotent of order 2, and there exists an open nonempty subset $U\subseteq X$ such that $\operatorname{Ker}(\epsilon_{|U})=\operatorname{Im}(\epsilon_{|U})$. At a point $p$ where $\operatorname{rank}(\epsilon)=0$, in local coordinates the endomorphism $\epsilon$ may be expressed by $\begin{pmatrix}a&b\\\ c&-a\\\ \end{pmatrix}\ \ a,b,c\mbox{ regular functions such that }a^{2}=-b\cdot c$ Let $\delta:=\operatorname{G.C.D.}(a,b,c)$. After dividing by $\delta$, every prime factor of $a$ is either not in $b$, or not in $c$, thus we can write $-b=\beta^{2}\hskip 28.45274ptc=\gamma^{2}\hskip 28.45274pta=\beta\cdot\gamma$ Therefore we obtain $\begin{pmatrix}u\\\ v\\\ \end{pmatrix}\in\operatorname{Ker}{\epsilon}\Longleftrightarrow\left\\{\begin{array}[]{l}a\cdot u+b\cdot v=0\\\ c\cdot u-a\cdot v=0\end{array}\right.\Longleftrightarrow\gamma\cdot u-\beta\cdot v=0\Longleftrightarrow\begin{pmatrix}u\\\ v\\\ \end{pmatrix}=\begin{pmatrix}\beta\cdot f\\\ \gamma\cdot f\\\ \end{pmatrix}$ and, writing our endomorphism $\epsilon$ as $\epsilon=\delta\cdot\alpha$, we have $\operatorname{Im}(\alpha)=\left\\{\begin{array}[]{l}\beta\cdot\gamma\cdot u-\beta^{2}\cdot v=\beta\cdot(\gamma\cdot u-\beta\cdot v)\\\ \gamma^{2}\cdot u-\gamma\cdot\beta\cdot v=\gamma\cdot(\gamma\cdot u-\beta\cdot v)\end{array}\right.$ Let $Z$ be the 0-dimensional scheme defined by $\\{\beta=\gamma=0\\}$ and $\Delta$ be the Cartier divisor defined by $\\{\delta=0\\}$. From the above description we deduce that the kernel of $\epsilon$ is a line bundle $L$ which fits in the following exact sequence: $0\rightarrow L\rightarrow\Omega^{1}_{X}\rightarrow{\mathcal{I}}_{Z}L(-\Delta)\rightarrow 0.$ Taking the total Chern classes we infer that: $K_{X}\equiv 2L-\Delta$ as divisors on $X$ and $c_{2}(X)={length}(Z)+L\cdot(L-\Delta)$. ∎ ###### Lemma 2.3. Let $X$ be a complex surface and let $X^{\prime}$ be the blow up of $X$ at a point $p$. Then a special tensor $\omega^{\prime}$ on $X^{\prime}$ induces a special tensor $\omega$ on $X$, and the converse only holds if and only if $\omega$ vanishes at $p$ (in particular, it must hold : $det(\omega)=0$). Proof. First of all, $\omega^{\prime}$ induces a special tensor on $X\setminus\\{p\\}$, and by Hartogs’ theorem the latter extends to a special tensor $\omega$ on $X$. Conversely, choose local coordinates $(x,y)$ for $X$ around $p$ and take a local chart of the blow up with coordinates $(x,u)$ where $y=ux$. Locally around $p$ we can write $\omega=\frac{a(\operatorname{d}x)^{2}+b(\operatorname{d}y)^{2}+c(\operatorname{d}x\operatorname{d}y)}{\operatorname{d}x\wedge\operatorname{d}y}.$ The pull back $\omega^{\prime}$ of $\omega$ is given by the following expression: $\frac{a(\operatorname{d}x)^{2}+b(u\operatorname{d}x+x\operatorname{d}u)^{2}+c(u\operatorname{d}x+x\operatorname{d}u)\operatorname{d}x}{x\operatorname{d}x\wedge\operatorname{d}u}=$ $=\frac{\operatorname{d}x^{2}(a+bu^{2}+cu)+bx^{2}\operatorname{d}u^{2}+(2bux+cx)\operatorname{d}x\operatorname{d}u}{x\operatorname{d}x\wedge\operatorname{d}u},$ hence $\omega^{\prime}$ is regular if and only if $\frac{a+bu^{2}+cu}{x}$ is a regular function. This is obvious if $a,b,c$ vanish at $p$, since then their pull back is divisible by $x$. Assume on the other side that $a,b,c$ are constant: then we get a rational function which is only regular if $a=b=c=0.$ ∎ ###### Lemma 2.4. Let $X$ be a compact minimal rational surface admitting a special tensor $\omega$. Then $X\cong\mathbb{P}^{1}\times\mathbb{P}^{1}$ if $det(\omega)\neq 0$. Proof. Assume that $X$ is a $\mathbb{P}^{1}$ bundle over a curve $B\cong\mathbb{P}^{1}$, i.e., a ruled surface ${\bf F}_{n}$ with $n\geq 0$. Let $\pi\colon X\rightarrow B$ the projection. By the exact sequence $0\rightarrow\pi^{*}\Omega^{1}_{B}\rightarrow\Omega^{1}_{X}\rightarrow\Omega^{1}_{X|B}\rightarrow 0$ and since on a general fibre $F$ the subsheaf $\pi^{*}\Omega^{1}_{B}$ is trivial, while the quotient sheaf $\Omega^{1}_{X|B}$ is negative, we conclude that any endomorphism $\epsilon$ carries $\pi^{*}\Omega^{1}_{B}$ to itself. If it has non zero determinant we can conclude by Theorem 2.1 that $X\cong\mathbb{P}^{1}\times\mathbb{P}^{1}$. Otherwise, $\epsilon$ is nilpotent and we have a nonzero element in ${\rm Hom}(\Omega^{1}_{X|B},\pi^{*}\Omega^{1}_{B})$. Since these are invertible sheaves, it suffices to see when $H^{0}(\mathcal{O}_{X}(2\pi^{*}K_{B}-K_{X}))\neq 0.$ But, letting $\Sigma$ be the section with selfintersection $\Sigma^{2}=-n$, our vector space equals $H^{0}(\mathcal{O}_{X}(2\Sigma-(n+2)F)).$ Intersecting this divisor with $\Sigma$ we see that (since each time the intersection number with $\Sigma$ is negative) $H^{0}(\mathcal{O}_{X}(2\Sigma-(n+2)F))=H^{0}(\mathcal{O}_{X}(\Sigma-(n+2)F))=H^{0}(\mathcal{O}_{X}(-(n+2)F))=0.$ There remains the case where $X$ is $\mathbb{P}^{2}$. In this case $\epsilon$ must be a nilpotent endomorphism by Theorem 2.1, and it cannot vanish at any point by our previous result on ${\bf F}_{1}$. Therefore the rank of $\epsilon$ equals 1 at each point. By lemma 2.2 it follows that there is a divisor $L$ such that $K_{X}=2L$, a contradiction. ∎ ## 3\. Proof of Theorems 1.7 and 1.9 ###### Proof. If $X$ is strongly uniformized by the bidisk, then $K_{X}$ is ample, in particular $K^{2}_{X}\geq 1$ and, since by Castelnuovo’s theorem $\chi(X)\geq 1$, by the vanishing theorem of Kodaira and Mumford it follows that $P_{2}(X)\geq 2$ (see [Bom73]). Thus one direction follows from proposition 1.5, except that we shall show only later that (1*) holds. Assume conversely that $(1),(2)$ hold. Without loss of generality we may assume by lemma 2.3 that $X$ is minimal, since $K_{X}^{2}$ can only decrease via a blowup and the bigenus is a birational invariant. $K^{2}_{X}\geq 1$ implies that either the surface $X$ is of general type, or it is a rational surface. In the latter case we conclude by lemma 2.4. Observe that the further hypothesis (3) (obviously implied by (3*)) guarantees that $X$ is of general type. Thus, from now on, we may assume that $X$ is of general type and, passing to an étale double cover if necessary, that $X$ admits a special tensor. By the cited Theorem 2.1 of [Bea00] it suffices to find a decomposition of the cotangent bundle $\Omega^{1}_{X}$ as a direct sum of two line bundles $L_{1}$ and $L_{2}$. The two line bundles $L_{1}$, $L_{2}$ will be given as eigenbundles of a diagonizable endomorphism $\epsilon\in\operatorname{End}(\Omega^{1}_{X})$. Our previous discussion shows then that it is sufficient to show that any special tensor cannot yield a nilpotent endomorphism. Otherwise, by lemma 2.2, we can write $2L\equiv K_{X}+\Delta$ and then deduce that $L$ is a big divisor since $\Delta$ is effective by construction and $K_{X}$ is big because $X$ is of general type. This assertion gives the required contradiction since by the Bogomolov-Castelnuovo-de Franchis Theorem (cf. [Bog77]) for an invertible subsheaf $L$ of $\Omega^{1}_{X}$ it is $h^{0}(X,mL)\leq O(m)$, contradicting the bigness of $L$. There remains to show (1*). But if $h^{0}(X,S^{2}\Omega^{1}_{X}(-K_{X}))\geq 2$ then, given a point $p\in X$, there is a special tensor which is not invertible in $p$, hence a special tensor with vanishing determinant, a contradiction. ∎ ## 4\. Proof of proposition 1.11 In this section we consider surfaces $X$ with bigenus $P_{2}(X)\geq 2$ (property (3*)), therefore their Kodaira dimension equals 1 or 2, hence either they are properly (canonically) elliptic, or they are of general type. Since we took already care of the latter case in the main theorems 1.7 and 1.9, we restrict our attention here to the former case, and try to see when does a properly elliptic surface admit a special tensor (we can reduce to this situation in view of remark 1.14). We can moreover assume that the associated endomorphism $\epsilon$ is nilpotent by theorem 2.1. Again without loss of generality we may assume that $X$ is minimal by virtue of lemma 2.3. Proof. Let $X$ be a minimal properly elliptic surface and let $f:X\rightarrow B$ be its (multi)canonical elliptic fibration. Write any fibre $f^{-1}(p)$ as $F_{p}=\sum_{i=1}^{h_{p}}m_{i}C_{i}$ and, setting $n_{p}:=G.C.D.(m_{i})$, $F_{p}=n_{p}F^{\prime}_{p}$, we say that a fibre is multiple if $n_{p}>1$. By Kodaira’s classification ([Kod60]) of the singular fibres we know that in this case $m_{i}=n_{p},\forall i.$ Assume that the multiple fibres of the elliptic fibration are $n_{1}F_{1}^{\prime},\dots,n_{r}F_{r}^{\prime}$, and consider the divisorial part of the critical locus $\mathcal{S}_{p}:=\sum_{i=1}^{h_{p}}(m_{i}-1)C_{i},\ \ \mathcal{S}:=\sum_{p\in B}\mathcal{S}_{p}$ so that we have then the exact sequence $0\rightarrow f^{*}\Omega^{1}_{B}(\mathcal{S})\rightarrow\Omega^{1}_{X}\rightarrow\mathcal{I}_{{\mathcal{C}}}\ \omega_{X|B}\rightarrow 0,$ where ${\mathcal{C}}$ is a 0-dimensional (l.c.i.) subscheme. For further calculations we separate the divisorial part of the critical locus as the sum of two disjoint effective divisors, the multiple fibre contribution and the rest: $\mathcal{S}_{m}:=\sum_{i=1}^{r}(n_{i}-1)F^{\prime}_{i},\ \hat{\mathcal{S}}:=\mathcal{S}-\mathcal{S}_{m}.$ Let us assume that we have a nilpotent endomorphism corresponding to another exact sequence $0\rightarrow L\rightarrow\Omega^{1}_{X}\rightarrow{\mathcal{I}}_{Z}L(-\Delta)\rightarrow 0,$ in turn determined by a homomorphism $\epsilon^{\prime}:{\mathcal{I}}_{Z}L(-\Delta)\rightarrow L,$ i.e., by a section $s\in H^{0}(\mathcal{O}_{X}(\Delta))=$ $=H^{0}(\mathcal{O}_{X}(2L-K_{X}))=H^{0}(S^{2}(L)(-K_{X}))\subset H^{0}(S^{2}(\Omega^{1}_{X})(-K_{X})).$ We observe that, since $2L\equiv K_{X}+\Delta$, it follows that, if $F$ is a general fibre, then $L\cdot F=\Delta\cdot F=0,$ hence the effective divisor $\Delta$ is contained in a finite union of fibres. The first candidate to try with is the choice of $L=L^{\prime}$, where we set $L^{\prime}:=f^{*}\Omega^{1}_{B}(\mathcal{S})$. To this purpose we recall Kodaira’s canonical bundle formula: $K_{X}\equiv\mathcal{S}_{m}+f^{*}(\delta)=\sum_{i=1}^{r}(n_{i}-1)F^{\prime}_{i}+f^{*}(\delta),\ deg(\delta)=\chi(X)-2+2b,$ where $b$ is the genus of the base curve $B$. Then $H^{0}(\mathcal{O}_{X}(2L^{\prime}-K_{X}))=H^{0}(\mathcal{O}_{X}(f^{*}(2K_{B}-\delta)+2\mathcal{S}-\mathcal{S}_{m})$, and we search for an effective divisor linearly equivalent to $f^{*}(2K_{B}-\delta)+2\mathcal{S}-\mathcal{S}_{m}=f^{*}(2K_{B}-\delta)+2\hat{\mathcal{S}}+\mathcal{S}_{m}.$ We claim that $H^{0}(\mathcal{O}_{X}(2L^{\prime}-K_{X}))=H^{0}(\mathcal{O}_{X}(f^{*}(2K_{B}-\delta))$: it will then suffice to have examples where $|2K_{B}-\delta|\neq\emptyset.$ Proof of the claim It suffices to show that $f_{*}\mathcal{O}_{X}(2\hat{\mathcal{S}}+\mathcal{S}_{m})=\mathcal{O}_{B}$. Since the divisor $2\hat{\mathcal{S}}+\mathcal{S}_{m}$ is supported on the singular fibres, and it is effective, we have to show that, for each singular fibre $F_{p}=\sum_{i=1}^{h_{p}}m_{i}C_{i}$, neither $2\hat{\mathcal{S}}_{p}\geq F_{p}$ nor ${\mathcal{S}_{m}}_{,p}\geq F_{p}$. The latter case is obvious since ${\mathcal{S}_{m}}_{,p}=(n_{p}-1)F^{\prime}_{p}<F_{p}=n_{p}F^{\prime}_{p}$. In the former case, $2\hat{\mathcal{S}}_{p}=\sum_{i=1}^{h_{p}}2(m_{i}-1)C_{i}$, but it is not possible that $\forall i$ one has $2(m_{i}-1)\geq m_{i}$, since there is always an irreducible curve $C_{i}$ with multiplicity $m_{i}=1$. $Q.E.D.$for the claim Assume that the elliptic fibration is not a product (in this case there is no special tensor with vanishing determinant): then the irregularity of $X$ equals the genus of $B$, whence our divisor on the curve $B$ has degree equal to $2b-2-(1-b+p_{g}(X))=3b-3-p_{g}.$ Since $\chi(X)\geq 1$, $p_{g}:=p_{g}(X)\geq b$, and there exist an elliptic surface $X$ with any $p_{g}\geq b$ ([Cat07]). Since any divisor on $B$ of degree $\geq b$ is effective, it suffices to choose $b\leq p_{g}\leq 2b-3$ and we get a special tensor with trivial determinant, provided that $b\geq 3$. Take now a Jacobian elliptic surface in Weierstrass normal form $ZY^{2}-4X^{3}-g_{2}XZ^{2}-g_{3}Z^{3}=0,$ where $g_{2}\in H^{0}(\mathcal{O}_{B}(4M))$, $g_{3}\in H^{0}(\mathcal{O}_{B}(6M))$, and assume that all the fibres are irreducible. Then the space of special tensors corresponding to our choice of $L$ corresponds to the vector space $H^{0}(\mathcal{O}_{B}(2K_{B}-\delta))=H^{0}(\mathcal{O}_{B}(K_{B}-6M))$. It suffices to take a hyperelliptic curve $B$ of genus $b=6h+1$, and, denoting by $H$ the hyperelliptic divisor, set $M:=hH$, so that $K_{B}-6M\equiv 0$ and we have $h^{0}(\mathcal{O}_{X}(2L-K_{X}))=1$. We leave aside for the time being the question whether the surface $X$ admits a unique special tensor. ∎ Acknowledgements. These research was performed in the realm of the SCHWERPUNKT ”Globale Methoden in der komplexen Geometrie”, and of the FORSCHERGRUPPE 790 ‘Classification of algebraic surfaces and compact complex manifolds’. The second author thanks the Universität Bayreuth for its warm hospitality in the months of november and december 2006 (where the research was begun) and the DFG for supporting his visit. We would like to thank Eckart Viehweg and Kang Zuo for pointing out some aspects of Yau’s uniformization theorem that we had not properly credited in the first version. ## References * [BPV84] W. Barth, C. Peters, A. Van de Ven, Compact complex surfaces. Ergebnisse der Mathematik und ihrer Grenzgebiete (3). Springer-Verlag, Berlin,(1984). * [Bea00] A. Beauville, Complex manifolds with split tangent bundle Complex analysis and algebraic geometry, de Gruyter, Berlin, 2000, 61–70. * [Bog77] F. A. Bogomolov Families of curves on a surface of general type (Russian) Dokl. Akad. Nauk SSSR, 236 no. 5, 1977 , p.1041–1044. * [Bom73] E. Bombieri, Canonical models of surfaces of general type, Publ. Math. I.H.E.S., 42 (1973), 173–219. * [Cat07] F. Catanese, “Q.E.D. for algebraic varieties”, Jour. Diff. Geom. 77 no. 1 (2007) 43–75 * [Kod60] K. Kodaira, On compact complex analytic surfaces, I, Ann. of Math. 71 (1960), 111–152. * [K-M71] Kodaira, K. Morrow, J. Complex manifolds Holt, Rinehart and Winston, Inc., New-York-Montreal, Que.-London, 1971 * [Ma62] Y . Matsushima, On the first Betti number of compact quotient spaces of higher-dimensional symmetric spaces Ann. of Math. (2) 75, 1962, 312–330 * [MT87] B.Moishezon, M. Teicher, Simply-connected algebraic surfaces of positive index Invent. Math. 89 (1987), no. 3, 601–643. * [Miy77] Y. Miyaoka, On the Chern numbers of surfaces of general type Invent. Math. 42 (1977), 225–237. * [Miy82] Y. Miyaoka, Algebraic surfaces with positive indices Classification of algebraic and analytic manifolds (Katata, 1982), Progr. Math. 39 Birkh user Boston, Boston, MA, (1983) 281–301 * [Mu79] D. Mumford, An algebraic surface with $K$ ample, $K^{2}=9$, $p_{g}=q=0$ Amer. J. Math. 101, no. 1, (1979), 233 244. * [Sieg73] C.L. Siegel, Topics in complex function theory. Vol. III. Abelian functions and modular functions of several variables. Translated from the German by E. Gottschling and M. Tretkoff. With a preface by Wilhelm Magnus. Reprint of the 1973 original. Wiley Classics Library. A Wiley-Interscience Publication. John Wiley and Sons, Inc., New York (1989), x+244 pp. * [V-Z05] E. Viehweg, K. Zuo, Arakelov inequalities and the uniformization of certain rigid Shimura varieties, arXiv:math/0503339. * [Yau77] S. T. Yau, Calabi’ s conjecture and some new results in algebraic geometry Proc. Nat. Acad. Sci. USA 74 (1977), 1798–1799 * [Yau88] S. T. Yau, Uniformization of geometric structures. The mathematical heritage of Hermann Weyl (Durham, NC, 1987), Proc. Sympos. Pure Math., 48, Amer. Math. Soc., Providence, RI (1988), 265–274. * [Yau93] S. T. Yau, A splitting theorem and an algebraic geometric characterization of locally hermitian symmetric spaces Comm. in Analysis and Geometry 1 (1993), 473–486 Authors’ addresses: Prof. Fabrizio Catanese Lehrstuhl Mathematik VIII, Universität Bayreuth, NWII D-95440 Bayreuth, Germany e-mail: Fabrizio.Catanese@uni-bayreuth.de Marco Franciosi Dipartimento di Matemativa Applicata ”U. Dini”, Università di Pisa via Buonarroti 1C, I-56127, Pisa, Italy e-mail: franciosi@dma.unipi.it

arxiv-papers

2024-09-04T02:48:54.444327

0803.3074

# Fundamental Solutions for the Klein-Gordon Equation in de Sitter Spacetime Karen Yagdjian, Anahit Galstian ###### Abstract In this article we construct the fundamental solutions for the Klein-Gordon equation in de Sitter spacetime. We use these fundamental solutions to represent solutions of the Cauchy problem and to prove $L^{p}-L^{q}$ estimates for the solutions of the equation with and without a source term. Department of Mathematics, University of Texas-Pan American, 1201 W. University Drive, Edinburg, TX 78541-2999, USA ## 0 Introduction and Statement of Results In this paper we construct the fundamental solutions for the Klein-Gordon equation in the de Sitter spacetime and use these fundamental solutions to find representations of the solutions to the Cauchy problem as well as $L^{p}-L^{q}$ estimates for them. After averaging on a suitable scale, our universe is homogeneous and isotropic; therefore, the properties of the universe can be properly described by treating the matter as a perfect homogeneous fluid. In the models of the universe proposed by Einstein [12] and de Sitter [11] the line element is connected with the proper mass density and the proper pressure in the universe by the field equations for a perfect fluid. There are two alternatives which lead to the solutions of Einstein and de Sitter, respectively [22, Sec.132]. In the models proposed by Einstein [12] and de Sitter [11] the universe is assumed to be a static system, which means that we can introduce a system of coordinates $x^{i}=(r,\theta,\phi,ct)$ in which the line element has the static and spherically symmetric form $ds^{2}=-b(r)c^{2}dt^{2}+a(r)dr^{2}+r^{2}(d\theta^{2}+\sin^{2}d\phi^{2}),$ where $a$ and $b$ are functions of $r$ only. On account of the assumed homogeneity of the universe any point in the space may be taken as the origin $r=0$ of the spatial system of coordinates. The functions $a(r)$ and $b(r)$ are connected with the proper mass density $\mu$ and the proper pressure $p$ in the universe by the field equations for a perfect fluid $\displaystyle(\mu c^{2}+p)b^{\prime}=0,$ (0.1) $\displaystyle\frac{b^{\prime}}{abr}-\frac{1}{r^{2}}\left(1-\frac{1}{a}\right)+\Lambda=\kappa p,$ (0.2) $\displaystyle\frac{a^{\prime}}{a^{2}r}+\frac{1}{r^{2}}\left(1-\frac{1}{a}\right)-\Lambda=\kappa\mu c^{2},$ (0.3) where $\Lambda$ is cosmological constant, while $p$ and $\mu$ are constants. The general solution of the equation (0.3) is $a(r)=\left(1-\frac{2M_{bh}}{r}-\frac{\Lambda r^{2}}{3}\right)^{-1}.$ (0.4) The constant of integration $M_{bh}$ may have a meaning of the “mass of the black holes”. There are two alternatives $b^{\prime}=0$ or $\mu c^{2}+p=0$, which lead to the solutions of Einstein and de Sitter, respectively. In the case of de Sitter universe $ab=constant$. By a trivial change of scale of the time variable, this constant can, of course, always be made equal to 1, which means $b(r)=1-\frac{2M_{bh}}{r}-\frac{\Lambda r^{2}}{3}\,.$ The corresponding metric with the line element $ds^{2}=-\left(1-\frac{2M_{bh}}{r}-\frac{\Lambda r^{2}}{3}\right)c^{2}\,dt^{2}+\left(1-\frac{2M_{bh}}{r}-\frac{\Lambda r^{2}}{3}\right)^{-1}dr^{2}+r^{2}(d\theta^{2}+\sin^{2}\theta\,d\phi^{2})$ is called the Schwarzschild - de Sitter metric. The Cauchy problem for the linear wave equation without source term on the maximally extended Schwarzschild - de Sitter spacetime in the case of non-extremal black-hole corresponding to parameter values $0<M_{bh}<\frac{1}{3\sqrt{\Lambda}},$ is considered by Dafermos and Rodnianski [8]. They proved that in the region bounded by a set of black/white hole horizons and cosmological horizons, solutions converge pointwise to a constant faster than any given polynomial rate, where the decay is measured with respect to natural future-directed advanced and retarded time coordinates. There is an important question of local energy decay for the solution of the wave equation and Klein-Gordon equation in black hole spacetime. Results on the decay of local energy can provide a proof of the global nonlinear stability of the spacetime. The global nonlinear stability we believe is only known for Minkowski spacetime. Bony and Hafner [4] describe an expansion of the solution of the wave equation in the Schwarzschild - de Sitter metric in terms of resonances. The resonances correspond to the frequencies and rates of dumping of signals emitted by the black hole in the presence of perturbations (see [9, Chapter 4.35]). The main term in the expansion obtained in [4] is due to a zero resonance. The error term decays polynomially if one permits a logarithmic derivative loss in the angular directions and exponentially if one permits a small derivative loss in the angular directions. In the present paper we set $M_{bh}=0$ to exclude black holes. The case of the Klein-Gordon equation in the presence of the black holes will be discussed in the forthcoming paper. Thus, the de Sitter line element has the form $ds^{2}=-\left(1-\frac{r^{2}}{R^{2}}\right)c^{2}\,dt^{2}+\left(1-\frac{r^{2}}{R^{2}}\right)^{-1}dr^{2}+r^{2}(d\theta^{2}+\sin^{2}\theta\,d\phi^{2})\,.$ The transformation $\displaystyle r^{\prime}=\frac{r}{\sqrt{1-r^{2}/R^{2}}}e^{-ct/R}\,,\quad t^{\prime}=t+\frac{R}{2c}\ln\left(1-\frac{r^{2}}{R^{2}}\right)\,,\quad\theta^{\prime}=\theta\,,\quad\phi^{\prime}=\phi$ leads to the following form for the line element: $ds^{2}=-c^{2}\,d{t^{\prime}}^{2}+e^{2ct^{\prime}/R}(d{r^{\prime}}^{2}+r^{\prime 2}\,d{\theta^{\prime}}^{2}+r^{\prime 2}\sin^{2}\theta^{\prime}\,d{\phi^{\prime}}^{2})\,.$ (0.5) Finally, defining new space coordinates $x^{\prime}$, $y^{\prime}$, $z^{\prime}$ connected with $r^{\prime}$, $\theta^{\prime}$, $\phi^{\prime}$ by the usual equations connecting Cartesian coordinates and polar coordinates in a Euclidean space, (0.5) may be written [22, Sec.134] $ds^{2}=-c^{2}\,d{t^{\prime}}^{2}+e^{2ct^{\prime}/R}(d{x^{\prime}}^{2}+d{y^{\prime}}^{2}+d{z^{\prime}}^{2})\,.$ (0.6) The new coordinates $x^{\prime}$, $y^{\prime}$, $z^{\prime}$, $t^{\prime}$ can take all values from $-\infty$ to $\infty$. Here $R$ is the “radius” of the universe. The de Sitter model allows us to get an explanation of the actual red shift of spectral lines observed by Hubble and Humanson [22]. The de Sitter model also enjoys the advantage of being the only time-dependent cosmological model for which both particle creation and vacuum stress have been explicitly evaluated by all known techniques [3]. In a certain sense all solutions look like the de Sitter solution at late times [16]. The homogeneous and isotropic cosmological models possess highest symmetry that makes them more amenable to rigorous study. Among them we mention FLRW (Friedmann-Lematre-Robertson-Walker) models. The simplest class of cosmological models can be obtained if we assume additionally that the metric of the slices of constant time is flat and that the spacetime metric can be written in the form $ds^{2}=-dt^{2}+a^{2}(t)(d{x}^{2}+d{y}^{2}+d{z}^{2})$ with an appropriate scale factor $a(t)$. Although on the made assumptions, the spatially flat FLRW models appear to give a good explanation of our universe. The assumption that the universe is expanding leads to the positivity of the time derivative $\frac{d}{dt}a(t)$. A further assumption that the universe obeys the accelerated expansion suggests that the second derivative $\frac{d^{2}}{dt^{2}}a(t)$ is positive. A substantial amount of the observational material can be satisfactorily interpreted in terms of the models which take into account existing acceleration of the recession of distant galaxies. The time dependence of the function $a(t)$ is determined by the Einstein field equations for gravity. The Einstein equations with the cosmological constant $\Lambda$ have form $R_{\mu\nu}-\frac{1}{2}g_{\mu\nu}R=-8\pi GT_{\mu\nu}-\Lambda g_{\mu\nu},$ where term $\Lambda g_{\mu\nu}$ can be interpreted as an energy-momentum of the vacuum. Even a small value of $\Lambda$ could have drastic effects on the evolution of the universe. Under the assumption of FLRW symmetry the equation of motion in the case of positive cosmological constant $\Lambda$ leads to the solution $a(t)=a(0)e^{\sqrt{\frac{\Lambda}{3}}t},$ which produces models with exponentially accelerated expansion. The model described by the last equation is usually called the de Sitter model. The unknown of principal importance in the Einstein equations is a metric $g$. It comprises the basic geometrical feature of the gravitational field, and consequently explains the phenomenon of the mutual gravitational attraction of substance. In the presence of matter these equations contain a non-vanishing right hand side $-8\pi GT_{\mu\nu}$. In general the matter fields described by the function $\phi$ must satisfy equations of motion and in the case of the massive scalar field the equation of motion is that $\phi$ should satisfy the Klein-Gordon equation generated by the metric $g$. In the de Sitter universe the equation for the scalar field with mass $m$ and potential function $V$ written out explicitly in coordinates is (See, e.g. [14, 26].) $\phi_{tt}+nH\phi_{t}-e^{-2Ht}\bigtriangleup\phi+m^{2}\phi=-V^{\prime}(\phi)\,.$ (0.7) Here $x\in{\mathbb{R}}^{n}$, $t\in{\mathbb{R}}$, and $\bigtriangleup$ is the Laplace operator on the flat metric, $\bigtriangleup:=\sum_{j=1}^{n}\frac{\partial^{2}}{\partial x_{j}^{2}}$, while $H=\sqrt{\Lambda/3}$ is Hubble constant. If we introduce new unknown function $u=e^{-\frac{n}{2}Ht}\phi$, then the semilinear Klein-Gordon equation for $u$ on de Sitter spacetime takes the form $u_{tt}-e^{-2Ht}\bigtriangleup u+M^{2}u=-e^{\frac{n}{2}t}V^{\prime}(e^{-\frac{n}{2}t}u),$ (0.8) where $M^{2}:=m^{2}-\frac{n^{2}}{4}H^{2}$. Henceforth the quantity $M$, with nonnegative real part $\Re M\geq 0$, defined by $M^{2}:=m^{2}-\frac{n^{2}}{4}H^{2}\,,$ will be called the “curved mass” of particle. We extract a linear part of the equation (0.8) as an initial model that must be treated first: $u_{tt}-e^{-2Ht}\bigtriangleup u+M^{2}u=0\,.$ (0.9) The fundamental solutions and the Cauchy problem for the equation with $M=0$ in the backward direction of time are considered in [33]. The de Sitter line element in the higher dimensional analogue of de Sitter space is $ds^{2}=-dt^{2}+e^{2Ht}\big{(}(dx^{1})^{2}+\ldots+(dx^{n})^{2}\big{)}\,.$ It is a simplified version of the multidimensional cosmological models with the metric tensor given by $g=-e^{2\gamma(t)}dt^{2}+e^{2\phi^{1}(t)}g_{1}+\ldots+e^{2\phi^{n}(t)}g_{n}\,,$ and can be chosen as a starting point for the study. The multidimensional cosmological models have attracted a lot of attention during recent years in constructing mathematical models of an anisotropic universe (see, e.g. [7, 16] and references therein). The equation (0.9) is strictly hyperbolic. That implies the well-posedness of the Cauchy problem for (0.9) in the different functional spaces. The coefficient of the equation is an analytic function and Holmgren’s theorem implies a local uniqueness in the space of distributions. Moreover, the speed of propagation is finite, namely, it is equal to $e^{-Ht}$ for every $t\in{\mathbb{R}}$. The second-order strictly hyperbolic equation (0.9) possesses two fundamental solutions resolving the Cauchy problem. They can be written microlocally in terms of the Fourier integral operators [17], which give a complete description of the wave front sets of the solutions. The distance between two characteristic roots $\lambda_{1}(t,\xi)$ and $\lambda_{2}(t,\xi)$ of the equation (0.9) is $|\lambda_{1}(t,\xi)-\lambda_{2}(t,\xi)|=e^{-Ht}|\xi|$, $t\in{\mathbb{R}}$, $\xi\in{\mathbb{R}}^{n}$. It tends to zero as $t$ approaches $\infty$. Thus, the operator is not uniformly (that is for all $t\in{\mathbb{R}}$) strictly hyperbolic. Moreover, the finite integrability of the characteristic roots, $\int_{0}^{\infty}|\lambda_{i}(t,\xi)|dt<\infty$, leads to the existence of so-called “horizon” for that equation. More precisely, any signal emitted from the spatial point $x_{0}\in{\mathbb{R}}^{n}$ at time $t_{0}\in{\mathbb{R}}$ remains inside the ball $|x-x_{0}|<\frac{1}{H}e^{-Ht_{0}}$ for all time $t\in(t_{0},\infty)$. The equation (0.9) is neither Lorentz invariant nor invariant with respect to usual scaling and that brings additional difficulties. In particular, it can cause a nonexistence of the $L^{p}-L^{q}$ decay for the solutions in the direction of time. In [30] it is mentioned the model equation with permanently bounded domain of influence, power decay of characteristic roots, and without $L^{p}-L^{q}$ decay for the solutions that illustrates that phenomenon. The above mentioned $L^{p}-L^{q}$ decay estimates are one of the important tools for studying nonlinear equations (see, e.g. [25, 27]). The time inversion transformation $t\to-t$ reduces the equation (0.9) to the mathematically equivalent equation $u_{tt}-e^{2Ht}\bigtriangleup u+M^{2}u=0\,.$ (0.10) The wave equation, that is equation (0.10) with $M=0$, was investigated in [15] by the second author. More precisely, in [15] the resolving operator for the Cauchy problem $u_{tt}-e^{2t}\bigtriangleup u=0,\qquad u(x,0)=\varphi_{0}(x),\quad u_{t}(x,0)=\varphi_{1}(x)\,,$ (0.11) is written as a sum of the Fourier integral operators with the amplitudes given in terms of the Bessel functions and in terms of confluent hypergeometric functions. In particular, it is proved in [15] that for $t>0$ the solution of the Cauchy problem (0.11) is given by $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle-i\frac{2}{(2\pi)^{n}}\int_{{\mathbb{R}}^{n}}\Big{\\{}e^{i[x\cdot\xi+(e^{t}-1)|\xi|]}H_{+}\big{(}\frac{1}{2};1;2ie^{t}|\xi|\big{)}H_{-}\big{(}\frac{3}{2};3;2i|\xi|\big{)}$ $\displaystyle\hskip 51.21504pt-\,e^{i[x\cdot\xi-(e^{t}-1)|\xi|]}H_{-}\big{(}\frac{1}{2};1;2ie^{t}|\xi|\big{)}H_{+}\big{(}\frac{3}{2};3;2i|\xi|\big{)}\Big{\\}}|\xi|^{2}\hat{\varphi}_{0}(\xi)d\xi$ $\displaystyle-i\frac{1}{(2\pi)^{n}}\int_{{\mathbb{R}}^{n}}\Big{\\{}e^{i[x\cdot\xi+(e^{t}-1)|\xi|]}H_{+}\big{(}\frac{1}{2};1;2ie^{t}|\xi|\big{)}H_{-}\big{(}\frac{1}{2};1;2i|\xi|\big{)}$ $\displaystyle\hskip 51.21504pt-\,e^{i[x\cdot\xi-(e^{t}-1)|\xi|]}H_{-}(\frac{1}{2};1;2ie^{t}|\xi|)H_{+}\big{(}\frac{1}{2};1;2i|\xi|\big{)}\Big{\\}}\hat{\varphi}_{1}(\xi)d\xi\,.$ In the notations of [2] the last functions are $H_{-}(\alpha;\gamma;z)=e^{i\alpha\pi}\Psi(\alpha;\gamma;z)$ and $H_{+}(\alpha;\gamma;z)=e^{i\alpha\pi}\Psi(\gamma-\alpha;\gamma;-z)$, where function $\Psi(a;c;z)$ is defined in [2, Sec.6.5]. Here $\hat{\varphi}(\xi)$ is a Fourier transform of $\varphi(x)$. The typical $L^{p}-L^{q}$ decay estimates obtained in [15] by dyadic decomposition of the phase space contain some derivative loss. More precisely, it is proved that for the solution $u=u(x,t)$ to the Cauchy problem (0.11) with $n\geq 2$, $\varphi_{0}(x)\in C_{0}^{\infty}({\mathbb{R}}^{n})$ and $\varphi_{1}(x)=0$ for all large $t\geq T>0$, the following estimate is satisfied $\|u(x,t)\|_{L^{q}({\mathbb{R}}^{n})}\leq C(1+e^{t})^{-\frac{1}{2}(n-1)(\frac{1}{p}-\frac{1}{q})}\|\varphi_{0}\|_{W^{N}_{p}({\mathbb{R}}^{n})},$ (0.12) where $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, and $\frac{1}{2}(n+1)(\frac{1}{p}-\frac{1}{q})\leq N<\frac{1}{2}(n+1)(\frac{1}{p}-\frac{1}{q})+1$ and $W^{N}_{p}({\mathbb{R}}^{n})$ is the Sobolev space. In particular, the derivative loss, $N$, is positive, unless $p=q=2$. This derivative loss phenomenon exists for the classical wave equation as well. Indeed, it is well- known (see, e.g., [19, 20, 24]) that for the Cauchy problem $u_{tt}-\bigtriangleup u=0$, $u(x,0)=\varphi(x)$, $u_{t}(x,0)=0$, the estimate $\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x}^{n})}\leq C\|\varphi(x)\|_{L^{q}({\mathbb{R}}_{x}^{n})}$ fails to fulfill even for small positive $t$ unless $q=2$. The obstacle is created by the distinguishing feature of the (different from translation) Fourier integral operators of order zero, which compose a resolving operator. According to Theorem 1 [15], for the solution $u=u(x,t)$ to the Cauchy problem (0.11) with $n\geq 2$, $\varphi_{0}(x)=0$ and $\varphi_{1}(x)\in C_{0}^{\infty}({\mathbb{R}}^{n})$ for all large $t\geq T>0$ and for any small $\varepsilon>0$, the following estimate is satisfied $\|u(x,t)\|_{L^{q}({\mathbb{R}}^{n})}\leq C_{\varepsilon}(1+t)(1+e^{t})^{r_{0}-n(\frac{1}{p}-\frac{1}{q})}\|\varphi_{1}\|_{W^{N}_{p}({\mathbb{R}}^{n})},\,$ where $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $r_{0}=\max\\{\varepsilon;\frac{(n+1)}{2}(\frac{1}{p}-\frac{1}{q})-\frac{1}{q}\\}$, $\frac{n+1}{2}(\frac{1}{p}-\frac{1}{q})-\frac{1}{q}\leq N<\frac{n+1}{2}(\frac{1}{p}-\frac{1}{q})+\frac{1}{p}$. The nonlinear equations (0.7) and (0.8) are those we would like to solve, but the linear problem is a natural first step. Exceptionally efficient tool for the studying nonlinear equations is a fundamental solution of the associate linear operator. The fundamental solutions for the operator of the equation (0.11) are constructed in [33] and the representations of the solutions of the Cauchy problem $u_{tt}-e^{2t}\bigtriangleup u=f(x,t),\qquad u(x,0)=\varphi_{0}(x),\quad u_{t}(x,0)=\varphi_{1}(x)\,,$ (0.13) are given in the terms of the solutions of wave equation in Minkowski spacetime. Then in [33] for $n\geq 2$ the following decay estimate $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}\\!\\!$ $\displaystyle\\!\\!\leq$ $\displaystyle\\!\\!\\!\\!Ce^{t(2s-n(\frac{1}{p}-\frac{1}{q}))}\int_{0}^{t}(1+t-b)\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}\,db$ $\displaystyle+C(e^{t}-1)^{2s-n(\frac{1}{p}-\frac{1}{q})}\left\\{\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}+\|\varphi_{1}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}(1+t)(1-e^{-t})\right\\}$ is proven, provided that $s\geq 0$, $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $\frac{1}{2}(n+1)\left(\frac{1}{p}-\frac{1}{q}\right)\leq 2s\leq n\left(\frac{1}{p}-\frac{1}{q}\right)<2s+1$. Moreover, this estimate is fulfilled for $n=1$ and $s=0$ as well as if $\varphi_{0}(x)=0$ and $\varphi_{1}(x)=0$. Case of $n=1$, $f(x,t)=0$, and non-vanishing $\varphi_{1}(x)$ and $\varphi_{1}(x)$ also is discussed in Section 8 [33]. In the construction of the fundamental solutions for the operator (0.9) we follow the approach proposed in [29] that allows us to represent the fundamental solutions as some integral of the family of the fundamental solutions of the Cauchy problem for the wave equation without source term. The kernel of that integral contains the Gauss’s hypergeometric function. In that way, many properties of the wave equation can be extended to the hyperbolic equations with the time dependent speed of propagation. That approach was successfully applied in [31, 32] by the first author to investigate the semilinear Tricomi-type equations. Thus, in the present paper we consider Klein-Gordon operator in de Sitter model of the universe, that is ${\mathcal{S}}:=\partial_{t}^{2}-e^{-2t}\bigtriangleup+M^{2}\,,$ where $M\geq 0$ is the reduced mass, $x\in{\mathbb{R}}^{n}$, $t\in{\mathbb{R}}$. We look for the fundamental solution $E=E(x,t;x_{0},t_{0})$, $E_{tt}-e^{-2t}\Delta E-M^{2}E=\delta(x-x_{0},t-t_{0}),$ with a support in the “forward light cone” $D_{+}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, and for the fundamental solution with a support in the “backward light cone” $D_{-}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, defined as follows $\displaystyle D_{\pm}(x_{0},t_{0})$ $\displaystyle:=$ $\displaystyle\Big{\\{}(x,t)\in{\mathbb{R}}^{n+1}\,;\,|x-x_{0}|\leq\pm(e^{-t_{0}}-e^{-t})\,\Big{\\}}\,.$ (0.14) In fact, any intersection of $D_{-}(x_{0},t_{0})$ with the hyperplane $t=const<t_{0}$ determines the so-called dependence domain for the point $(x_{0},t_{0})$, while the intersection of $D_{+}(x_{0},t_{0})$ with the hyperplane $t=const>t_{0}$ is the so-called domain of influence of the point $(x_{0},t_{0})$. The equation (0.9) is non-invariant with respect to time inversion. Moreover, the dependence domain is wider than any given ball if time $const>t_{0}$ is sufficiently large, while the domain of influence is permanently, for all time $const<t_{0}$, in the ball of the radius $e^{t_{0}}$. Define for $t_{0}\in{\mathbb{R}}$ in the domain $D_{+}(x_{0},t_{0})\cup D_{-}(x_{0},t_{0})$ the function $\displaystyle E(x,t;x_{0},t_{0})$ $\displaystyle=$ $\displaystyle(4e^{-t_{0}-t})^{iM}\Big{(}(e^{-t}+e^{-t_{0}})^{2}-(x-x_{0})^{2}\Big{)}^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-t_{0}}-e^{-t})^{2}-(x-x_{0})^{2}}{(e^{-t_{0}}+e^{-t})^{2}-(x-x_{0})^{2}}\Big{)},$ where $F\big{(}a,b;c;\zeta\big{)}$ is the hypergeometric function (See, e.g. [2].). Let $E(x,t;0,b)$ be function ( ‣ 0 Introduction and Statement of Results), and set ${\mathcal{E}}_{+}(x,t;0,t_{0}):=\cases{E(x,t;0,t_{0})\quad\mbox{\rm in}\,\,D_{+}(0,t_{0}),\cr 0\hskip 56.9055pt\mbox{\rm elsewhere}}\,,\qquad{\mathcal{E}}_{-}(x,t;0,t_{0}):=\cases{E(x,t;0,t_{0})\quad\mbox{\rm in}\,\,D_{-}(0,t_{0}),\cr 0\hskip 56.9055pt\mbox{\rm elsewhere}}\,.$ Since function $E=E(x,t;0,t_{0})$ is smooth in $D_{\pm}(0,t_{0})$ and is locally integrable, it follows that ${\mathcal{E}}_{+}(x,t;0,$ $t_{0})$ and ${\mathcal{E}}_{-}(x,t;0,$ $t_{0})$ are distributions whose supports are in $D_{+}(0,t_{0})$ and $D_{-}(0,t_{0})$, respectively. The next theorem gives our first result. ###### Theorem 0.1 Suppose that $n=1$. The distributions ${\mathcal{E}}_{+}(x,t;0,t_{0})$ and ${\mathcal{E}}_{-}(x,t;0,t_{0})$ are the fundamental solutions for the operator ${\mathcal{S}}=\partial_{t}^{2}-e^{-2t}\bigtriangleup+M^{2}$ relative to the point $(0,t_{0})$, that is ${\mathcal{S}}{\mathcal{E}}_{\pm}(x,t;0,t_{0})=\delta(x,t-t_{0}),$ or $\frac{\partial^{2}}{\partial t^{2}}{\mathcal{E}}_{\pm}(x,t;0,t_{0})-e^{-2t}\frac{\partial^{2}}{\partial x^{2}}{\mathcal{E}}_{\pm}(x,t;0,t_{0})+M^{2}{\mathcal{E}}_{\pm}(x,t;0,t_{0})=\delta(x,t-t_{0}).$ To motivate one construction for the higher dimensional case $n\geq 2$ we follow the approach suggested in [29] and represent fundamental solution ${\mathcal{E}}_{+}(x,t;0,t_{0})$ as follows $\displaystyle{\mathcal{E}}_{+}(x,t;0,t_{0})$ $\displaystyle=$ $\displaystyle\int_{e^{-t}-e^{-t_{0}}}^{e^{-t_{0}}-e^{-t}}{\mathcal{E}}^{string}(x,r)(4e^{-t_{0}-t})^{iM}\Big{(}(e^{-t_{0}}+e^{-t})^{2}-r^{2}\Big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\hskip 71.13188pt\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-t_{0}}-e^{-t})^{2}-r^{2}}{(e^{-t_{0}}+e^{-t})^{2}-r^{2}}\Big{)}\,dr,\quad t>t_{0},$ where the distribution ${\mathcal{E}}^{string}(x,t)$ is the fundamental solution of the Cauchy problem for the string equation: $\frac{\partial^{2}}{\partial t^{2}}{\mathcal{E}}^{string}-\frac{\partial^{2}}{\partial x^{2}}{\mathcal{E}}^{string}=0,\qquad{\mathcal{E}}^{string}(x,0)=\delta(x),\,\,\,{\mathcal{E}}^{string}_{t}(x,0)=0\,.$ Hence, ${\mathcal{E}}^{string}(x,t)=\frac{1}{2}\\{\delta(x+t)+\delta(x-t)\\}$. The integral makes sense in the topology of the space of distributions. The fundamental solution ${\mathcal{E}}_{-}(x,t;0,t_{0})$ for $t<t_{0}$ admits a similar representation. We appeal to the wave equation in Minkowski spacetime to obtain in the next theorem very similar representations of the fundamental solutions of the higher dimensional equation in de Sitter spacetime with $n\geq 2$. ###### Theorem 0.2 If $x\in{\mathbb{R}}^{n}$, $n\geq 2$, then for the operator ${\mathcal{S}}=\partial_{t}^{2}-e^{-2t}\bigtriangleup+M^{2}$ the fundamental solution ${\mathcal{E}}_{+,n}(x,t;x_{0},t_{0})$ $(={\mathcal{E}}_{+,n}(x-x_{0},t;0,t_{0}))$ with a support in the forward cone $D_{+}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, supp$\,{\mathcal{E}}_{+,n}\subseteq D_{+}(x_{0},t_{0})$, is given by the following integral $(t>t_{0})$ $\displaystyle{\mathcal{E}}_{+,n}(x-x_{0},t;0,t_{0})$ $\displaystyle=$ $\displaystyle 2\int_{0}^{e^{-t_{0}}-e^{-t}}{\mathcal{E}}^{w}(x-x_{0},r)(4e^{-t_{0}-t})^{iM}\Big{(}(e^{-t_{0}}+e^{-t})^{2}-r^{2}\Big{)}^{-\frac{1}{2}-iM}$ (0.16) $\displaystyle\hskip 71.13188pt\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-t_{0}}-e^{-t})^{2}-r^{2}}{(e^{-t_{0}}+e^{-t})^{2}-r^{2}}\Big{)}\,dr.$ Here the function ${\mathcal{E}}^{w}(x,t;b)$ is a fundamental solution to the Cauchy problem for the wave equation ${\mathcal{E}}^{w}_{tt}-\bigtriangleup{\mathcal{E}}^{w}=0\,,\quad{\mathcal{E}}^{w}(x,0)=\delta(x)\,,\quad{\mathcal{E}}^{w}_{t}(x,0)=0\,.$ The fundamental solution ${\mathcal{E}}_{-,n}(x,t;x_{0},t_{0})$ $(={\mathcal{E}}_{-,n}(x-x_{0},t;0,t_{0}))$ with a support in the backward cone $D_{-}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, supp$\,{\mathcal{E}}_{-,n}\subseteq D_{-}(x_{0},t_{0})$, is given by the following integral $(t<t_{0})$ $\displaystyle{\mathcal{E}}_{+,n}(x-x_{0},t;0,t_{0})$ $\displaystyle=$ $\displaystyle-2\int_{e^{-t_{0}}-e^{-t}}^{0}{\mathcal{E}}^{w}(x-x_{0},r)(4e^{-t_{0}-t})^{iM}\Big{(}(e^{-t_{0}}+e^{-t})^{2}-r^{2}\Big{)}^{-\frac{1}{2}-iM}$ (0.17) $\displaystyle\hskip 71.13188pt\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-t_{0}}-e^{-t})^{2}-r^{2}}{(e^{-t_{0}}+e^{-t})^{2}-r^{2}}\Big{)}\,dr.$ In particular, the formula (0.16) shows that Huygens’s Principle is not valid for the waves propagating in de Sitter model of the universe. Fields satisfying a wave equation in de Sitter model of the universe can be accompanied by tails propagating inside the light cone. This phenomenon will be discussed in the spirit of [28] in the forthcoming paper. Next we use Theorem 0.1 to solve the Cauchy problem for the one-dimensional equation $u_{tt}-e^{-2t}u_{xx}+M^{2}u=f(x,t)\,,\qquad t>0\,,\quad x\in{\mathbb{R}}\,,$ (0.18) with vanishing initial data, $u(x,0)=u_{t}(x,0)=0\,.$ (0.19) ###### Theorem 0.3 Assume that the function $f$ is continuous along with its all second order derivatives, and that for every fixed $t$ it has a compact support, supp$f(\cdot,t)\subset{\mathbb{R}}$. Then the function $u=u(x,t)$ defined by $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,f(y,b)(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-(x-y)^{2}\Big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\hskip 71.13188pt\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-(x-y)^{2}}{(e^{-b}+e^{-t})^{2}-(x-y)^{2}}\Big{)}$ is a $C^{2}$-solution to the Cauchy problem for the equation (0.18) with vanishing initial data, (0.19). The representation of the solution of the Cauchy problem for the one- dimensional case ($n=1$) of the equation (0.9) without source term is given by the next theorem. ###### Theorem 0.4 The solution $u=u(x,t)$ of the Cauchy problem $u_{tt}-e^{-2t}u_{xx}+M^{2}u=0\,,\qquad u(x,0)=\varphi_{0}(x)\,,\qquad u_{t}(x,0)=\varphi_{1}(x)\,,$ (0.20) with $\varphi_{0},\varphi_{1}\in C_{0}^{\infty}({\mathbb{R}})$ can be represented as follows $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}+\int_{0}^{1-e^{-t}}\big{[}\varphi_{0}(x-z)+\varphi_{0}(x+z)\big{]}K_{0}(z,t)\,dz$ $\displaystyle+\,\,\int_{0}^{1-e^{-t}}\,\Big{[}\varphi_{1}(x-z)+\varphi_{1}(x+z)\Big{]}K_{1}(z,t)dz\,,$ where the kernels $K_{0}(z,t)$ and $K_{1}(z,t)$ are defined by $\displaystyle K_{0}(z,t)$ $\displaystyle:=$ $\displaystyle-\left[\frac{\partial}{\partial b}E(z,t;0,b)\right]_{b=0}$ $\displaystyle=$ $\displaystyle(4e^{-t})^{iM}\big{(}(e^{-t}+1)^{2}-z^{2}\big{)}^{-iM}\frac{1}{[(1-e^{-t})^{2}-z^{2}]\sqrt{(e^{-t}+1)^{2}-z^{2}}}$ $\displaystyle\times\Bigg{[}\big{(}e^{-t}-1-iM(e^{-2t}-1-z^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(1-e^{-t})^{2}-z^{2}}{(1+e^{-t})^{2}-z^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}1-e^{-2t}+z^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(1-e^{-t})^{2}-z^{2}}{(1+e^{-t})^{2}-z^{2}}\Big{)}\Bigg{]}$ and $\displaystyle K_{1}(z,t)$ $\displaystyle:=$ $\displaystyle E(z,t;0,0)$ $\displaystyle=$ $\displaystyle(4e^{-t})^{iM}\big{(}(1+e^{-t})^{2}-z^{2}\big{)}^{-\frac{1}{2}-iM}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(1-e^{-t})^{2}-z^{2}}{(1+e^{-t})^{2}-z^{2}}\right),\,0\leq z\leq 1-e^{-t},$ respectively. The kernels $K_{0}(z,t)$ and $K_{1}(z,t)$ play leading roles in the derivation of $L^{p}-L^{q}$ estimates. Their main properties are listed and proved in Section 8. Next we turn to the higher-dimensional equation with $n\geq 2$. ###### Theorem 0.5 If $n$ is odd, $n=2m+1$, $m\in{\mathbb{N}}$, then the solution $u=u(x,t)$ to the Cauchy problem $u_{tt}-e^{-2t}\Delta u+M^{2}u=f,\quad u(x,0)=0,\quad u_{t}(x,0)=0,$ (0.21) with $f\in C^{\infty}({\mathbb{R}}^{n+1})$ and with the vanishing initial data is given by the next expression $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle 2\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dr_{1}\,\left(\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{\frac{n-3}{2}}\frac{r^{n-2}}{\omega_{n-1}c_{0}^{(n)}}\\!\\!\int_{S^{n-1}}f(x+ry,b)\,dS_{y}\right)_{r=r_{1}}$ (0.22) $\displaystyle\times(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-r_{1}^{2}\right)^{-\frac{1}{2}-iM}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-r_{1}^{2}}{(e^{-b}+e^{-t})^{2}-r_{1}^{2}}\right)\\!\\!,$ where $c_{0}^{(n)}=1\cdot 3\cdot\ldots\cdot(n-2)$. Constant $\omega_{n-1}$ is the area of the unit sphere $S^{n-1}\subset{\mathbb{R}}^{n}$. If $n$ is even, $n=2m$, $m\in{\mathbb{N}}$, then the solution $u=u(x,t)$ is given by the next expression $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle 2\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dr_{1}\,\left(\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{\frac{n-2}{2}}\frac{2r^{n-1}}{\omega_{n-1}c_{0}^{(n)}}\\!\\!\int_{B_{1}^{n}(0)}\frac{f(x+ry,b)}{\sqrt{1-|y|^{2}}}\,dV_{y}\right)_{r=r_{1}}$ (0.23) $\displaystyle\times(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-r_{1}^{2}\right)^{-\frac{1}{2}-iM}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-r_{1}^{2}}{(e^{-b}+e^{-t})^{2}-r_{1}^{2}}\right)\\!\\!.$ Here $B_{1}^{n}(0):=\\{|y|\leq 1\\}$ is the unit ball in ${\mathbb{R}}^{n}$, while $c_{0}^{(n)}=1\cdot 3\cdot\ldots\cdot(n-1)$. Thus, in both cases, of even and odd $n$, one can write $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle 2\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dr\,v(x,r;b)(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-r^{2}\right)^{-\frac{1}{2}-iM}$ (0.24) $\displaystyle\hskip 71.13188pt\times F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-r^{2}}{(e^{-b}+e^{-t})^{2}-r^{2}}\right),$ where the function $v(x,t;b)$ is a solution to the Cauchy problem for the wave equation $v_{tt}-\bigtriangleup v=0\,,\quad v(x,0;b)=f(x,b)\,,\quad v_{t}(x,0;b)=0\,.$ The next theorem gives representation of the solutions of equation (0.9) with the initial data prescribed at $t=0$. ###### Theorem 0.6 The solution $u=u(x,t)$ to the Cauchy problem $u_{tt}-e^{-2t}\bigtriangleup u+M^{2}u=0\,,\quad u(x,0)=\varphi_{0}(x)\,,\quad u_{t}(x,0)=\varphi_{1}(x)\,,$ (0.25) with $\varphi_{0}$, $\varphi_{1}\in C_{0}^{\infty}({\mathbb{R}}^{n})$, $n\geq 2$, can be represented as follows: $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle e^{\frac{t}{2}}v_{\varphi_{0}}(x,\phi(t))+\,2\int_{0}^{1}v_{\varphi_{0}}(x,\phi(t)s)K_{0}(\phi(t)s,t)\phi(t)\,ds$ (0.26) $\displaystyle+\,2\int_{0}^{1}v_{\varphi_{1}}(x,\phi(t)s)K_{1}(\phi(t)s,t)\phi(t)\,ds,\quad x\in{\mathbb{R}}^{n},\,\,t>0\,,$ $\phi(t):=1-e^{-t}$, by means of the kernels $K_{0}$ and $K_{1}$ have been defined in Theorem 0.4. Here for $\varphi\in C_{0}^{\infty}({\mathbb{R}}^{n})$ and for $x\in{\mathbb{R}}^{n}$, $n=2m+1$, $m\in{\mathbb{N}}$, $\displaystyle v_{\varphi}(x,\phi(t)s):=\Bigg{(}\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{\frac{n-3}{2}}\frac{r^{n-2}}{\omega_{n-1}c_{0}^{(n)}}\int_{S^{n-1}}\varphi(x+ry)\,dS_{y}\Bigg{)}_{r=\phi(t)s}$ while for $x\in{\mathbb{R}}^{n}$, $n=2m$, $m\in{\mathbb{N}}$, $v_{\varphi}(x,\phi(t)s):=\Bigg{(}\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{\frac{n-2}{2}}\frac{2r^{n-1}}{\omega_{n-1}c_{0}^{(n)}}\int_{B_{1}^{n}(0)}\frac{1}{\sqrt{1-|y|^{2}}}\varphi(x+ry)\,dV_{y}\Bigg{)}_{r=s\phi(t)}\,.$ The function $v_{\varphi}(x,\phi(t)s)$ coincides with the value $v(x,\phi(t)s)$ of the solution $v(x,t)$ of the Cauchy problem $v_{tt}-\bigtriangleup v=0,\quad v(x,0)=\varphi(x),\quad v_{t}(x,0)=0\,.$ As a consequence of the above theorems we obtain in Sections 9-10 for $n\geq 2$ and for the particles with “large” mass $m$, $m\geq n/2$, that is, with nonnegative curved mass $M\geq 0$, the following $L^{p}-L^{q}$ estimate $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}\\!\\!$ $\displaystyle\\!\\!\leq\\!\\!$ $\displaystyle\\!\\!C\int_{0}^{t}\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}e^{-b}\left(e^{-b}-e^{-t}\right)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\left(1+t-b\right)db$ (0.27) $\displaystyle+C(1+t)(1-e^{-t})^{2s-n(\frac{1}{p}-\frac{1}{q})}\Big{\\{}e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}+(1-e^{-t})\|\varphi_{1}\|_{{L}^{p}({\mathbb{R}}^{n})}\Big{\\}}$ provided that $s\geq 0$, $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $\frac{1}{2}(n+1)\left(\frac{1}{p}-\frac{1}{q}\right)\leq 2s\leq n\left(\frac{1}{p}-\frac{1}{q}\right)<2s+1$. Moreover, according to Theorem 7.1 the estimate (0.27) is valid for $n=1$ and $s=0$ as well as if $\varphi_{0}(x)=0$ and $\varphi_{1}(x)=0$. Case of $n=1$, $f(x,t)=0$, and non- vanishing $\varphi_{1}(x)$ and $\varphi_{1}(x)$ is discussed in Section 8. The case of particles with small mass $m<n/2$ will be discussed in the forthcoming paper. The paper is organized as follows. In Section 1 we construct the Riemann function of the operator of (0.9) in the characteristic coordinates for the case of $n=1$. That Riemann function used in Section 2 to prove Theorem 0.1. Then in Section 3 we apply the fundamental solutions to solve the Cauchy problem with the source term and with the vanishing initial data given at $t=0$. More precisely, we give a representation formula for the solutions. In that section we also prove several basic properties of the function $E(x,t;y,b)$. In Sections 4-5 we use formulas of Section 3 to derive and to complete the list of representation formulas for the solutions of the Cauchy problem for the case of one-dimensional spatial variable. The higher- dimensional equation with the source term is considered in Section 6, where we derive a representation formula for the solutions of the Cauchy problem with the source term and with the vanishing initial data given at $t=0$. In the same section this formula is used to derive the fundamental solutions of the operator and to complete the proof of Theorem 0.6. Then in Sections 7-10 we establish the $L^{p}-L^{q}$ decay estimates. Applications of all these results to the nonlinear equations will be done in the forthcoming paper. ## 1 The Riemann Function In the characteristic coordinates $l$ and $m$, $l=x+e^{-t},\qquad m=x-e^{-t},$ (1.1) one has $\displaystyle\frac{\partial^{2}}{\partial t^{2}}$ $\displaystyle=$ $\displaystyle\frac{1}{2}(l-m)\left(\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\right)+\frac{1}{4}(l-m)^{2}\left(\frac{\partial^{2}}{\partial l^{2}}-2\frac{\partial^{2}}{\partial l\,\partial m}+\frac{\partial^{2}}{\partial m^{2}}\right)$ and $\displaystyle e^{-2t}\frac{\partial^{2}}{\partial x^{2}}=\frac{1}{4}(l-m)^{2}\left(\frac{\partial^{2}}{\partial l^{2}}+2\frac{\partial^{2}}{\partial l\,\partial m}+\frac{\partial^{2}}{\partial m^{2}}\right).$ Then the operator ${\mathcal{S}}$ of the equation (0.18) reads ${\mathcal{S}}:=\frac{\partial^{2}}{\partial t^{2}}-e^{-2t}\frac{\partial^{2}}{\partial x^{2}}+M^{2}=-(l-m)^{2}\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}M^{2}\Bigg{\\}}\,.$ (1.2) In particular, in the new variables the equation $\left(\frac{\partial^{2}}{\partial t^{2}}-e^{-2t}\frac{\partial^{2}}{\partial x^{2}}+M^{2}\right)u=0\quad\mbox{\rm implies}\quad\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}\Bigg{\\}}u-\frac{1}{(l-m)^{2}}M^{2}u=0\,.$ We need the following lemma with $\displaystyle\gamma=\frac{1}{2}+iM$. ###### Lemma 1.1 The function $V(l,m;a,b)=(l-b)^{-\gamma}(a-m)^{-\gamma}F\Big{(}\gamma,\gamma;1;\frac{(l-a)(m-b)}{(l-b)(m-a)}\Big{)}$ solves the equation $\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{(l-m)}\gamma\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}\Bigg{\\}}V(l,m;a,b)=0\,.$ (1.3) Proof. We denote the argument of the hypergeometric function by $z$, and evaluate its derivatives, $z:=\frac{(l-a)(m-b)}{(l-b)(m-a)}\,,\qquad\frac{\partial}{\partial l}z=\frac{(a-b)(b-m)}{(l-b)^{2}(a-m)}\,,\quad\frac{\partial}{\partial m}z=-\frac{(a-b)(a-l)}{(b-l)(a-m)^{2}}\,.$ Further, we obtain $\frac{\partial}{\partial l}V(l,m;a,b)=(a-m)^{-\gamma}(l-b)^{-\gamma-1}\Bigg{\\{}-\gamma F\Big{(}\gamma,\gamma;1;z\Big{)}+\frac{(a-b)(b-m)}{(l-b)(a-m)}F^{\prime}_{z}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}\,.$ Next $\frac{\partial}{\partial m}V(l,m;a,b)=(a-m)^{-\gamma}(l-b)^{-\gamma}\Bigg{\\{}\gamma(a-m)^{-1}F\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)^{2}}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}\,.$ Then $\displaystyle\left(\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\right)V(l,m;a,b)$ $\displaystyle=$ $\displaystyle(a-m)^{-\gamma}(l-b)^{-\gamma}\left(-\gamma\right)F\big{(}\gamma,\gamma;1;z\big{)}\Big{\\{}\frac{1}{(l-b)}+\frac{1}{(a-m)}\Big{\\}}$ $\displaystyle+(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}F_{z}^{\prime}\big{(}\gamma,\gamma;1;z\big{)}(a-b)\left\\{\frac{(b-m)}{(l-b)}-\frac{(a-l)}{(a-m)}\right\\}\,.$ Furthermore, $\displaystyle\frac{\partial^{2}}{\partial l\,\partial m}V(l,m;a,b)$ $\displaystyle=$ $\displaystyle(a-m)^{-\gamma-1}\Bigg{[}-\gamma(l-b)^{-\gamma-1}\Bigg{\\{}\gamma F\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}\Bigg{]}$ $\displaystyle+\,(a-m)^{-\gamma-1}\Bigg{[}(l-b)^{-\gamma}\frac{\partial}{\partial l}\Bigg{\\{}\gamma F\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}\Bigg{]}.$ We calculate $\displaystyle\frac{\partial}{\partial l}\Bigg{\\{}\gamma F\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}$ $\displaystyle=$ $\displaystyle\gamma F^{\prime}_{z}\Big{(}\gamma,\gamma;1;z\Big{)}\frac{(a-b)(b-m)}{(l-b)^{2}(a-m)}-\frac{(a-b)^{2}}{(b-l)^{2}(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}$ $\displaystyle-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{zz}{}^{\prime}{}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\frac{(a-b)(b-m)}{(l-b)^{2}(a-m)}\,.$ Here $\frac{\partial}{\partial l}\frac{(a-b)(a-l)}{(b-l)(a-m)}=\frac{(a-b)^{2}}{(b-l)^{2}(a-m)}$. Finally, $\displaystyle\frac{\partial^{2}}{\partial l\,\partial m}V(l,m;a,b)$ $\displaystyle=$ $\displaystyle(a-m)^{-\gamma-1}\Bigg{[}-\gamma(l-b)^{-\gamma-1}\Bigg{\\{}\gamma F\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\Bigg{\\}}\Bigg{]}$ $\displaystyle+\,(a-m)^{-\gamma-1}(l-b)^{-\gamma}\Bigg{\\{}\gamma F^{\prime}_{z}\Big{(}\gamma,\gamma;1;z\Big{)}\frac{(a-b)(b-m)}{(l-b)^{2}(a-m)}$ $\displaystyle-\frac{(a-b)^{2}}{(b-l)^{2}(a-m)}F_{z}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}-\frac{(a-b)(a-l)}{(b-l)(a-m)}F_{zz}{}^{\prime}{}^{\prime}\Big{(}\gamma,\gamma;1;z\Big{)}\frac{(a-b)(b-m)}{(l-b)^{2}(a-m)}\Bigg{\\}}.$ The coefficients of the derivatives of the hypergeometric function, $F_{zz}{}^{\prime}{}^{\prime}$, $F_{z}{}^{\prime}$, and of $F$ in the expression for $\displaystyle\frac{\partial^{2}}{\partial l\,\partial m}V(l,m;a,b)$ are $\displaystyle(a-m)^{-\gamma-3}(l-b)^{-\gamma-3}(a-b)^{2}(a-l)(b-m)\,,$ $\displaystyle(a-m)^{-\gamma-2}(l-b)^{-\gamma-2}(a-b)\Big{\\{}\gamma(b-m-a+l)-(a-b)\Big{\\}}\,,$ $\displaystyle(a-m)^{-\gamma-1}\left(-\gamma\right)(l-b)^{-\gamma-1}\\{-\left(-\gamma\right)\\}\,=\,-(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}\gamma^{2}\,,$ respectively. The coefficients of $F_{z}{}^{\prime}$ and $F$ in the expression for $\displaystyle\frac{1}{(l-m)}\gamma\left(\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\right)V(l,m;a,b)$ are $\frac{1}{(l-m)}\gamma(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(a-b)\left\\{\frac{(b-m)}{(l-b)}-\frac{(a-l)}{(a-m)}\right\\}$ and $-\,\frac{1}{(l-m)}\gamma(a-m)^{-\gamma}(l-b)^{-\gamma}\gamma\Big{\\{}\frac{1}{(l-b)}+\frac{1}{(a-m)}\Big{\\}}\,.$ Now we turn to the equation (1.3). The coefficients of $F$ and $F_{z}{}^{\prime}$ in that equation are $\frac{1}{(m-l)}(a-b)(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(-\gamma^{2}),$ and $\displaystyle\frac{1}{(m-l)}(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(a-b)$ $\displaystyle\times\Bigg{[}\gamma\frac{(m-l)}{(a-m)(l-b)}(b-m-a+l)+\gamma\left\\{\frac{(b-m)}{(l-b)}-\frac{(a-l)}{(a-m)}\right\\}-\frac{(m-l)}{(a-m)(l-b)}(a-b)\Bigg{]}.$ The first two terms in the brackets can be written as follows $\displaystyle\gamma\frac{(m-l)}{(a-m)(l-b)}(b-m-a+l)+\gamma\left\\{\frac{(b-m)}{(l-b)}-\frac{(a-l)}{(a-m)}\right\\}=-2\gamma z\,,$ while the last term can be transformed to $\displaystyle-\frac{(m-l)}{(a-m)(l-b)}(a-b)=1-z\,.$ Thus, the coefficient of $F_{z}{}^{\prime}$ in the equation (1.3) is $\displaystyle\frac{1}{(m-l)}(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(a-b)\left[1-(1+2\gamma)z\right]\,.$ Finally, the coefficient of $F_{zz}{}^{\prime}{}^{\prime}$ in the equation (1.3) is $\displaystyle\frac{1}{(m-l)}(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(a-b)\Big{[}\frac{(a-b)(a-l)(b-m)(m-l)}{(a-m)^{2}(l-b)^{2}}\Big{]}\,,$ where $\displaystyle\frac{(a-b)(a-l)(b-m)(m-l)}{(a-m)^{2}(l-b)^{2}}=z(1-z)\,.$ Hence, the left-hand side of (1.3) reads $\displaystyle\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{(l-m)}\gamma\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}\Bigg{\\}}V(l,m;a,b)$ $\displaystyle=$ $\displaystyle\frac{1}{(m-l)}(a-m)^{-\gamma-1}(l-b)^{-\gamma-1}(a-b)\Big{[}z(1-z)F_{zz}+\big{(}1-(1+2\gamma)z\big{)}F_{z}-\gamma^{2}F\Big{]}=0\,,$ and vanishes, since $F$ solves the Gauss hypergeometic equation with $c=1$, $a=\gamma$, and $b=\gamma$. Lemma is proven. $\square$ ###### Lemma 1.2 For $\gamma\in{\mathbb{C}}$ such that $F\big{(}\gamma,\gamma;1;z\big{)}$ is well defined, the function $\displaystyle E(l,m;a,b)$ $\displaystyle:=$ $\displaystyle(a-b)^{\gamma-\frac{1}{2}}(l-m)^{\gamma-\frac{1}{2}}V(l,m;a,b)$ $\displaystyle=$ $\displaystyle(a-b)^{\gamma-\frac{1}{2}}(l-m)^{\gamma-\frac{1}{2}}(l-b)^{-\gamma}(a-m)^{-\gamma}F\Big{(}\gamma,\gamma;1;\frac{(l-a)(m-b)}{(l-b)(m-a)}\Big{)}$ solves the equation $\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}\Bigg{\\}}E(l,m;a,b)+\frac{1}{(l-m)^{2}}\left(\frac{1}{2}-\gamma\right)^{2}E(l,m;a,b)=0\,.$ (1.4) Proof. Indeed, straightforward calculations show $\displaystyle(a-b)^{-\gamma+\frac{1}{2}}\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}+\frac{1}{(l-m)^{2}}\left(\frac{1}{2}-\gamma\right)^{2}\Bigg{\\}}E$ $\displaystyle=$ $\displaystyle(l-m)^{\gamma-\frac{1}{2}}\left[V_{lm}-\frac{1}{(l-m)}\gamma\left(V_{l}-V_{m}\right)\right]=0\,.$ Lemma is proven. $\square$ Consider now the operator ${\mathcal{S}}_{ch}^{*}:=\frac{\partial^{2}\,}{\partial l\,\partial m}+\frac{1}{2(l-m)}\Big{(}\frac{\partial\,}{\partial l}-\frac{\partial\,}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}(M^{2}+1)\,,$ which is a formally adjoint to the operator ${\mathcal{S}}_{ch}:=\frac{\partial^{2}\,}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial\,}{\partial l}-\frac{\partial\,}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}M^{2}\,.$ In the next lemma the Riemann function is presented. ###### Proposition 1.3 The function $\displaystyle R(l,m;a,b)$ $\displaystyle=$ $\displaystyle(l-m)E(l,m;a,b)$ $\displaystyle=$ $\displaystyle(a-b)^{iM}(l-m)^{1+iM}(l-b)^{-\frac{1}{2}-iM}(a-m)^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(l-a)(m-b)}{(l-b)(m-a)}\Big{)}$ defined for all $l$, $m$, $a$, $b\in{\mathbb{R}}$, such that $l>m$, is a unique solution of the equation ${\mathcal{S}}_{ch}^{*}R=0$, which satisfies the following conditions: $(\mbox{\rm i})$ $\displaystyle{R_{l}=\frac{1}{2(l-m)}R\quad}$ along the line $m=b$; $(\mbox{\rm ii})$ $\displaystyle{R_{m}=-\frac{1}{2(l-m)}R\quad}$ along the line $l=a$; $(\mbox{\rm iii})$ $R(a,b;a,b)=1$. Proof. Indeed, if we denote $\gamma=\frac{1}{2}+iM$, then for the Riemann function we have $R(l,m;a,b)=(a-b)^{\gamma-\frac{1}{2}}(l-m)^{\gamma+\frac{1}{2}}V(l,m;a,b)=(l-m)E(l,m;a,b)\,.$ The operators ${\mathcal{S}}_{ch}$ and ${\mathcal{S}}_{ch}^{*}$ can be written as follows: $\displaystyle{\mathcal{S}}_{ch}$ $\displaystyle=$ $\displaystyle\frac{\partial^{2}\,}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial\,}{\partial l}-\frac{\partial\,}{\partial m}\Big{)}+\frac{1}{(l-m)^{2}}\left(\gamma-\frac{1}{2}\right)^{2}\,,$ $\displaystyle{\mathcal{S}}_{ch}^{*}$ $\displaystyle=$ $\displaystyle\frac{\partial^{2}\,}{\partial l\,\partial m}+\frac{1}{2(l-m)}\Big{(}\frac{\partial\,}{\partial l}-\frac{\partial\,}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}\left(1-\left(\gamma-\frac{1}{2}\right)^{2}\right)\,.$ The direct calculations show that, if function $u$ solves the equation ${\mathcal{S}}_{ch}u=0$, then the function $v=(l-m)u$ solves the equation ${\mathcal{S}}_{ch}^{*}v=0$, and vice versa. Then Lemma 1.2 completes the proof. Lemma is proven. $\square$ ## 2 Proof of Theorem 0.1 Next we use Riemann function $R(l,m;a,b)$ and function $E(x,t;x_{0},t_{0})$ defined by ( ‣ 0 Introduction and Statement of Results) to complete the proof of Theorem 0.1, which gives the fundamental solution with a support in the forward cone $D_{+}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, and the fundamental solution with a support in the backward cone $D_{-}(x_{0},t_{0})$, $x_{0}\in{\mathbb{R}}^{n}$, $t_{0}\in{\mathbb{R}}$, defined by (0.14) with plus and minus, respectively. We present a proof for ${\mathcal{E}}_{+}(x,t;0,b)$ since for ${\mathcal{E}}_{-}(x,t;0,b)$ it is similar. First, we note that the operator ${\mathcal{S}}$ is formally self-adjoint, ${\mathcal{S}}={\mathcal{S}}^{*}$. We must show that $<{\mathcal{E}}_{+},{\mathcal{S}}\varphi>=\varphi(0,b)\,,\qquad\mbox{\rm for every}\quad\varphi\in C_{0}^{\infty}({\mathbb{R}}^{2})\,.$ Since $E(x,t;0,b)$ is locally integrable in ${\mathbb{R}}^{2}$, this is equivalent to showing that $\int\\!\\!\int_{{\mathbb{R}}^{2}}{\mathcal{E}}_{+}(x,t;0,b){\mathcal{S}}\varphi(x,t)\,dx\,dt=\varphi(0,b),\quad\mbox{\rm for every}\quad\varphi\in C_{0}^{\infty}({\mathbb{R}}^{2}).$ (2.1) In the mean time ${D(x,t)}/{D(l,m)}=(l-m)^{-1}$ is the Jacobian of the transformation (1.1). Hence the integral in the left-hand side of (2.1) is equal to $\displaystyle\int\\!\\!\int_{{\mathbb{R}}^{2}}{\mathcal{E}}_{+}(x,t;0,b){\mathcal{S}}\varphi(x,t)\,dx\,dt=\int_{b}^{\infty}dt\int_{e^{-t}-e^{-b}}^{e^{-b}-e^{-t}}E(x,t;0,b){\mathcal{S}}\varphi(x,t)\,dx$ $\displaystyle=$ $\displaystyle-\int_{-e^{-b}}^{\infty}\int_{-\infty}^{e^{-b}}R(l,m;e^{-b},-e^{-b})\,dl\,dm\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}M^{2}\Bigg{\\}}\varphi.$ We consider the first term of the right hand side, and integrate it by parts $\displaystyle\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}dl\,\,R(l,m;e^{-b},-e^{-b})\frac{\partial^{2}}{\partial l\,\partial m}\varphi$ $\displaystyle=$ $\displaystyle\int_{-e^{-b}}^{\infty}R(e^{-b},m;e^{-b},-e^{-b})\frac{\partial\varphi}{\partial m}\Bigg{|}_{l=e^{-b}}\,dm-\Bigg{[}-\int_{-\infty}^{e^{-b}}\,dl\left(\frac{\partial}{\partial l}R(l,-e^{-b};e^{-b},-e^{-b})\right)\varphi\Bigg{|}_{m=-e^{-b}}$ $\displaystyle\hskip 56.9055pt-\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}\,dl\left(\frac{\partial^{2}}{\partial l\,\partial m}R(l,m;e^{-b},-e^{-b})\right)\varphi\Bigg{]}$ $\displaystyle=$ $\displaystyle-\varphi(e^{-b},-e^{-b})-\int_{-e^{-b}}^{\infty}dm\,\left(\frac{\partial}{\partial m}R(e^{-b},m;e^{-b},-e^{-b})\right)\varphi(e^{-b},m)$ $\displaystyle+\int_{-\infty}^{e^{-b}}\,dl\left(\frac{\partial}{\partial l}R(l,-e^{-b};e^{-b},-e^{-b})\right)\,\varphi(l,-e^{-b})+\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}\,dl\left(\frac{\partial^{2}}{\partial l\,\partial m}R(l,m;e^{-b},-e^{-b})\right)\varphi.$ Then, for the second term we obtain $\displaystyle-\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}\,dl\,R(l,m;e^{-b},-e^{-b})\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}\varphi$ $\displaystyle=$ $\displaystyle-\int_{-\infty}^{e^{-b}}R(l,-e^{-b};e^{-b},-e^{-b})\frac{1}{2(l+e^{-b})}\varphi(l,-e^{-b})\,dl$ $\displaystyle-\int_{-e^{-b}}^{\infty}R(e^{-b},m;e^{-b},-e^{-b})\frac{1}{2(e^{-b}-m)}\varphi(e^{-b},m)\,dm$ $\displaystyle-\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}dl\,\frac{1}{(l-m)^{2}}R(l,m;e^{-b},-e^{-b})\varphi(l,m)\,$ $\displaystyle+\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}\,dl\,\Big{[}\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}R(l,m;e^{-b},-e^{-b})\Big{]}\varphi(l,m)\,.$ Using properties of the Riemann function we derive $\displaystyle\int_{-e^{-b}}^{\infty}dm\int_{-\infty}^{e^{-b}}R(l,m;e^{-b},-e^{-b})\,dl\,\Bigg{\\{}\frac{\partial^{2}}{\partial l\,\partial m}-\frac{1}{2(l-m)}\Big{(}\frac{\partial}{\partial l}-\frac{\partial}{\partial m}\Big{)}-\frac{1}{(l-m)^{2}}M^{2}\Bigg{\\}}\varphi$ $\displaystyle=$ $\displaystyle-\varphi(e^{-b},-e^{-b})-\int_{-e^{-b}}^{\infty}\left(\frac{\partial}{\partial m}R(e^{-b},m;e^{-b},-e^{-b})\right)\varphi(e^{-b},m)\,dm$ $\displaystyle+\int_{-\infty}^{e^{-b}}\left(\frac{\partial}{\partial l}R(l,-e^{-b};e^{-b},-e^{-b})\right)\,\varphi(l,-e^{-b})\,dl$ $\displaystyle-\int_{-\infty}^{e^{-b}}R(l,-e^{-b};e^{-b},-e^{-b})\frac{1}{2(l+e^{-b})}\varphi(l,-e^{-b})\,dl$ $\displaystyle-\int_{-e^{-b}}^{\infty}R(e^{-b},m;e^{-b},-e^{-b})\frac{1}{2(e^{-b}-m)}\varphi(e^{-b},m)\,dm$ $\displaystyle=$ $\displaystyle-\varphi(e^{-b},-e^{-b})\,.$ Theorem is proven. $\square$ ## 3 Application to the Cauchy Problem: Source Term and $n=1$ Consider now the Cauchy problem for the equation (0.18) with vanishing initial data (0.19). For every $(x,t)\in D_{+}(0,b)$ one has $e^{-t}-e^{-b}\leq x\leq e^{-b}-e^{-t}$, so that $\displaystyle E(x,t;0,b)$ $\displaystyle=$ $\displaystyle(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}.$ The coefficient of the equation (0.9) is independent of $x$, therefore ${\mathcal{E}}_{+}(x,t;$ $y,b)$ $={\mathcal{E}}_{+}(x-y,t;0,b)$. Using the fundamental solution from Theorem 0.1 one can write the convolution $u(x,t)=\int_{-\infty}^{\infty}\\!\int_{-\infty}^{\infty}{\mathcal{E}}_{+}(x,t;y,b)f(y,b)\,db\,dy=\int_{0}^{t}\\!db\\!\int_{-\infty}^{\infty}{\mathcal{E}}_{+}(x-y,t;0,b)f(y,b)\,dy,$ (3.1) which is well-defined since supp$f\subset\\{t\geq 0\\}$. Then according to the definition of the distribution ${\mathcal{E}}_{+}$ we obtain the statement of the Theorem 0.3. Thus, Theorem 0.3 is proven. ###### Remark 3.1 The argument of the hypergeometric function is nonnegative and bounded, $0\leq\frac{(e^{-b}-e^{-t})^{2}-z^{2}}{(e^{-b}+e^{-t})^{2}-z^{2}}<1\quad\mbox{ for all}\quad b\in(0,t),\,\,z\in(e^{-t}-e^{-b},e^{-b}-e^{-t})\,.$ The following corollary is a manifestation of the time-speed transformation principle introduced in [29]. It implies the existence of an operator transforming the solutions of the Cauchy problem for the string equation to the solutions of the Cauchy problem for the inhomogeneous equation with time- dependent speed of propagation. One may think of this transformation as a “two-stage” Duhamel’s principal, but unlike the last one, it reduces the equation with the time-dependent speed of propagation to the one with the speed of propagation independent of time. ###### Corollary 3.2 The solution $u=u(x,t)$ of the Cauchy problem (0.18)-(0.19) can be represented by (0.24), where the functions $v(x,t;\tau):=\frac{1}{2}(f(x+t,\tau)+f(x-t,\tau))$, $\tau\in[0,\infty)$, form a one-parameter family of solutions to the Cauchy problem for the string equation, that is, $v_{tt}-v_{xx}=0$, $v(x,0;\tau)=f(x,\tau)$, $v_{t}(x,0;\tau)=0$. Proof. From the convolution (3.1) we derive $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\int_{-(e^{-b}-e^{-t})}^{e^{-b}-e^{-t}}dz\,f(z+x,b)(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-z^{2}\Big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-z^{2}}{(e^{-b}+e^{-t})^{2}-z^{2}}\Big{)}$ $\displaystyle=$ $\displaystyle 2\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dz\,\frac{1}{2}\\{f(x+z,b)+f(x-z,b)\\}(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-z^{2}\Big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-z^{2}}{(e^{-b}+e^{-t})^{2}-z^{2}}\Big{)}\,.$ Corollary is proven. $\square$ Some Properties of the Function ${\mathbf{E}(x,t;y,b)}$. In this section we collect some elementary auxiliary formulas in order to make the proofs of main theorems more transparent. ###### Proposition 3.3 Let $E(x,t;x_{0},t_{0})$ be function defined by ( ‣ 0 Introduction and Statement of Results). One has $\displaystyle E(x,t;y,b)$ $\displaystyle=$ $\displaystyle E(y,b;x,t),\,\,$ (3.2) $\displaystyle E(x,t;y,b)=E(x-y,t;0,b)$ , $\displaystyle E(x,t;0,b)=E(-x,t;0,b),$ (3.3) $\displaystyle E(x,t;0,-\ln(x+e^{-t}))$ $\displaystyle=$ $\displaystyle\frac{1}{2}\frac{1}{\sqrt{e^{-t}}\sqrt{x+e^{-t}}},$ (3.4) $\displaystyle\frac{\partial}{\partial b}\Big{(}e^{-b}E(e^{-b}-e^{-t},t;0,b)\Big{)}$ $\displaystyle=$ $\displaystyle-\frac{1}{4}e^{t/2}e^{-b/2},$ (3.5) $\displaystyle\frac{\partial}{\partial b}\Big{(}be^{-b}E(-e^{-b}+e^{-t},t;0,b)\Big{)}$ $\displaystyle=$ $\displaystyle\frac{\partial}{\partial b}\Big{(}be^{-b}E(e^{-b}-e^{-t},t;0,b)\Big{)}=\frac{1}{4}e^{t/2}e^{-b/2}(2-b),$ (3.6) $\displaystyle\lim_{y\rightarrow x+e^{-b}-e^{-t}}\frac{\partial}{\partial x}E(x-y,t;0,b)$ $\displaystyle=$ $\displaystyle-\frac{1}{16}(1+4M^{2})e^{t/2}e^{b/2}(e^{t}-e^{b}),$ (3.7) $\displaystyle\lim_{y\rightarrow x-e^{-b}+e^{-t}}\frac{\partial}{\partial x}E(x-y,t;0,b)$ $\displaystyle=$ $\displaystyle\frac{1}{16}(1+4M^{2})e^{t/2}e^{b/2}(e^{t}-e^{b}),$ (3.8) $\displaystyle\left[\frac{\partial}{\partial b}E(x,t;0,b)\right]_{b=-\ln(x+e^{-t})}$ $\displaystyle=$ $\displaystyle\frac{1}{16}e^{t}\frac{4+e^{t}x(1+4M^{2})}{\sqrt{1+e^{t}x}},$ (3.9) $\displaystyle\Bigg{[}\frac{\partial}{\partial b}E(x,t;0,b)\Bigg{]}_{b=0}\\!\\!$ $\displaystyle\\!\\!=\\!\\!$ $\displaystyle-(4e^{-t})^{iM}\big{(}(e^{-t}+1)^{2}-x^{2}\big{)}^{-iM}\frac{1}{[(1-e^{-t})^{2}-x^{2}]\sqrt{(e^{-t}+1)^{2}-x^{2}}}$ (3.10) $\displaystyle\times\Bigg{[}\big{(}e^{-t}-1-iM(e^{-2t}-1-x^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(1-e^{-t})^{2}-x^{2}}{(1+e^{-t})^{2}-x^{2}}\Big{)}$ $\displaystyle\hskip 14.22636pt+\big{(}1-e^{-2t}+x^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(1-e^{-t})^{2}-x^{2}}{(1+e^{-t})^{2}-x^{2}}\Big{)}\Bigg{]}.$ Proof. The properties (3.2),(3.3), and (3.4) are evident. To prove (3.5) and (3.6) we write $E(e^{-b}-e^{-t},t;0,b)=(4e^{-b-t})^{-\frac{1}{2}},$ that implies both (3.5) and (3.6). To prove (3.7) we denote $\displaystyle z:=\frac{(e^{-b}-e^{-t})^{2}-(x-y)^{2}}{(e^{-b}+e^{-t})^{2}-(x-y)^{2}}\,,$ so that, $\displaystyle\frac{\partial z}{\partial x}$ $\displaystyle=$ $\displaystyle-2(x-y)\frac{4e^{-b}e^{-t}}{[(e^{-b}+e^{-t})^{2}-(x-y)^{2}]^{2}}\,,\qquad\frac{\partial z}{\partial b}=-\frac{4e^{-b}e^{-t}\left(-e^{-2t}+e^{-2b}+(x-y)^{2}\right)}{\left[(e^{-b}+e^{-t})^{2}-(x-y)^{2}\right]^{2}}\,.$ Then we obtain $\displaystyle\frac{\partial}{\partial x}E(x-y,t;0,b)$ $\displaystyle=$ $\displaystyle(4e^{-b-t})^{iM}\Bigg{[}(-2(x-y))(-\frac{1}{2}-iM)\Big{(}(e^{-t}+e^{-b})^{2}-(x-y)^{2}\Big{)}^{-\frac{3}{2}-iM}$ (3.11) $\displaystyle\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\Big{)}$ $\displaystyle+\Big{(}(e^{-t}+e^{-b})^{2}-(x-y)^{2}\Big{)}^{-\frac{1}{2}-iM}\Big{(}\frac{\partial z}{\partial x}\Big{)}F_{z}\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\Big{)}\Bigg{]}\,.$ It is easily seen that $\lim_{y\to x+e^{-b}-e^{-t}}z=0,\qquad\lim_{y\to x+e^{-b}-e^{-t}}\frac{\partial z}{\partial x}=\frac{1}{2}(e^{t}-e^{b}),$ while according to (20) [2, Sec.2.8 v.1], we have $\displaystyle\lim_{y\to x+e^{-b}-e^{-t}}\partial_{z}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\right)$ $\displaystyle=$ $\displaystyle\frac{\Gamma^{2}(\frac{3}{2}+iM)}{\Gamma^{2}(\frac{1}{2}+iM)}=\left(\frac{1}{2}+iM\right)^{2}\,.$ (3.12) Consequently, from (3.11) we obtain $\displaystyle\lim_{y\to x+e^{-b}-e^{-t}}\frac{\partial}{\partial x}E(x-y,t;0,b)$ $\displaystyle=$ $\displaystyle(4e^{-b-t})^{iM}\Bigg{[}2(e^{-b}-e^{-t})\left(-\frac{1}{2}-iM\right)\Big{(}4e^{-t}e^{-b}\Big{)}^{-\frac{3}{2}-iM}+\Big{(}4e^{-t}e^{-b}\Big{)}^{-\frac{1}{2}-iM}\frac{1}{2}(e^{t}-e^{b})\left(\frac{1}{2}+iM\right)^{2}\Bigg{]}$ $\displaystyle=$ $\displaystyle-e^{\frac{t}{2}}e^{\frac{b}{2}}\frac{1}{16}(e^{t}-e^{b})\Big{[}1+4M^{2}\Big{]}\,.$ The proof of (3.8) is similar. To prove (3.9) we write $\displaystyle\frac{\partial}{\partial b}E(x,t;0,b)$ (3.13) $\displaystyle=$ $\displaystyle(-iM)(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}$ $\displaystyle-\Big{(}\frac{1}{2}+iM\Big{)}(4e^{-b-t})^{iM}(-e^{-b})2(e^{-t}+e^{-b})\Big{(}(e^{-t}+e^{-b})^{2}-x^{2}\Big{)}^{-\frac{3}{2}-iM}$ $\displaystyle\times F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}$ $\displaystyle+(4e^{-b-t})^{iM}\Big{(}(e^{-t}+e^{-b})^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}\frac{\partial}{\partial b}\Bigg{(}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}\Bigg{)}.$ On the other hand (3.12) and (3.13) imply $\displaystyle\left[\frac{\partial}{\partial b}E(x,t;0,b)\right]_{b=-\ln(x+e^{-t})}$ $\displaystyle=$ $\displaystyle(4e^{-b-t})^{iM}\Big{(}4e^{-b-t}\Big{)}^{-\frac{1}{2}-iM}\Bigg{[}(-iM)+2\Big{(}\frac{1}{2}+iM\Big{)}e^{-b}(e^{-t}+e^{-b})\Big{(}4e^{-b-t}\Big{)}^{-1}$ $\displaystyle\hskip 142.26378pt+\frac{\partial}{\partial b}\Bigg{(}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}\Bigg{)}\Bigg{]}_{b=-\ln(x+e^{-t})}$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{1}{2}b+\frac{1}{2}t}\Bigg{[}(-iM)+\frac{1}{2}\Big{(}\frac{1}{2}+iM\Big{)}(e^{-t}+e^{-b})e^{t}-\frac{4e^{-b}e^{-t}\left(e^{-2b}-e^{-2t}+x^{2}\right)}{\left[(e^{-b}+e^{-t})^{2}-x^{2}\right]^{2}}\Big{(}\frac{1}{2}+iM\Big{)}^{2}\Bigg{]}_{b=-\ln(x+e^{-t})}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\frac{1}{\sqrt{1+xe^{t}}}e^{t}\Bigg{[}-iM+\frac{1}{2}\Big{(}\frac{1}{2}+iM\Big{)}(2+xe^{t})-\frac{xe^{t}}{2}\Big{(}\frac{1}{2}+iM\Big{)}^{2}\Bigg{]}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\frac{1}{\sqrt{1+xe^{t}}}e^{t}\frac{4+xe^{t}(1+4M^{2})}{8}\,.$ Thus, (3.9) is proven. To prove (3.10) we appeal to (23)[2, v.1, Sec.2.8], that reads with $a=b=\frac{1}{2}+iM$, $c=1$, $\displaystyle F_{z}\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\Big{)}$ $\displaystyle=$ $\displaystyle\frac{1}{z(1-z)}\Bigg{\\{}\Big{[}-\Big{(}\frac{1}{2}-iM\Big{)}+\Big{(}\frac{1}{2}+iM\Big{)}z\Big{]}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\Big{)}$ $\displaystyle\hskip 56.9055pt+\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;z\Big{)}\Bigg{\\}}\,.$ Then we plug the last relation in (3.13) and obtain $\displaystyle\Bigg{[}\frac{\partial}{\partial b}E(x,t;0,b)\Bigg{]}_{b=0}$ $\displaystyle=$ $\displaystyle- iM(4e^{-t})^{iM}\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle-\Big{(}\frac{1}{2}+iM\Big{)}(4e^{-t})^{iM}(-1)2(e^{-t}+1)\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{3}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle+(4e^{-t})^{iM}\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}\left[\frac{\partial}{\partial b}\left(F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta\Big{)}\right)\right]_{b=0}\,,$ where we have denoted $\zeta:=\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\,,\qquad\zeta_{0}:=\frac{(1-e^{-t})^{2}-x^{2}}{(1+e^{-t})^{2}-x^{2}}\,,\qquad\zeta_{0}(1-\zeta_{0})=\frac{4e^{-t}[(1-e^{-t})^{2}-x^{2}]}{[(1+e^{-t})^{2}-x^{2}]^{2}}\,,$ with $\frac{\partial}{\partial b}\zeta=-\frac{4e^{-b}e^{-t}\left(e^{-2b}-e^{-2t}+x^{2}\right)}{\left[(e^{-b}+e^{-t})^{2}-x^{2}\right]^{2}}\,,\qquad\frac{\partial\zeta}{\partial b}\Big{|}_{b=0}=-\frac{4e^{-t}\left(1-e^{-2t}+x^{2}\right)}{\left[(1+e^{-t})^{2}-x^{2}\right]^{2}}\,.$ Hence, due to (20) [2, v.1, Sec.2.8], we obtain $\displaystyle\Bigg{[}\frac{\partial}{\partial b}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-x^{2}}{(e^{-b}+e^{-t})^{2}-x^{2}}\Big{)}\Bigg{]}_{b=0}$ $\displaystyle=$ $\displaystyle-\frac{1-e^{-2t}+x^{2}}{(1-e^{-t})^{2}-x^{2}}\Bigg{\\{}\Big{[}-\Big{(}\frac{1}{2}-iM\Big{)}+\Big{(}\frac{1}{2}+iM\Big{)}\zeta_{0}\Big{]}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle\hskip 170.71652pt+\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}\Bigg{\\}}\,.$ Hence, $\displaystyle\Bigg{[}\frac{\partial}{\partial b}E(x,t;0,b)\Bigg{]}_{b=0}$ $\displaystyle=$ $\displaystyle- iM(4e^{-t})^{iM}\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle-\Big{(}\frac{1}{2}+iM\Big{)}(4e^{-t})^{iM}(-1)2(e^{-t}+1)\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{3}{2}-iM}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle+(4e^{-t})^{iM}\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\times\Bigg{[}-\frac{1-e^{-2t}+x^{2}}{(1-e^{-t})^{2}-x^{2}}\Bigg{\\{}\Big{[}-\Big{(}\frac{1}{2}-iM\Big{)}+\Big{(}\frac{1}{2}+iM\Big{)}\zeta_{0}\Big{]}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle+\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}\Bigg{\\}}\Bigg{]}$ $\displaystyle=$ $\displaystyle(4e^{-t})^{iM}\Big{(}(e^{-t}+1)^{2}-x^{2}\Big{)}^{-\frac{1}{2}-iM}\Bigg{[}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}\Big{\\{}-iM+(1+2iM)\frac{e^{-t}+1}{(e^{-t}+1)^{2}-x^{2}}$ $\displaystyle-\frac{1-e^{-2t}+x^{2}}{(1-e^{-t})^{2}-x^{2}}\Big{[}-\Big{(}\frac{1}{2}-iM\Big{)}+\Big{(}\frac{1}{2}+iM\Big{)}\frac{(1-e^{-t})^{2}-x^{2}}{(1+e^{-t})^{2}-x^{2}}\Big{]}\Big{\\}}$ $\displaystyle-\frac{1-e^{-2t}+x^{2}}{(1-e^{-t})^{2}-x^{2}}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}\Bigg{]}.$ Finally, $\displaystyle\Bigg{[}\frac{\partial}{\partial b}E(x,t;0,b)\Bigg{]}_{b=0}$ $\displaystyle=$ $\displaystyle(4e^{-t})^{iM}\big{(}(e^{-t}+1)^{2}-x^{2}\big{)}^{-\frac{1}{2}-iM}$ $\displaystyle\times\Bigg{[}\frac{e^{t}-iM+e^{2t}\big{(}iM(1+x^{2})-1\big{)}}{e^{t}\big{(}2+e^{t}(x^{2}-1)\big{)}-1}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}$ $\displaystyle\hskip 28.45274pt-\frac{1-e^{-2t}+x^{2}}{(1-e^{-t})^{2}-x^{2}}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta_{0}\Big{)}\Bigg{]}\,.$ The formula (3.10) and, consequently, the proposition are proven. $\square$ ## 4 The Cauchy Problem: Second Datum and $n=1$ In this section we prove Theorem 0.4 in the case of $\varphi_{0}(x)=0$. More precisely, we have to prove that the solution $u(x,t)$ of the Cauchy problem (0.20) with $\varphi_{0}(x)=0$ and $\varphi_{1}(x)=\varphi(x)$ can be represented as follows $u(x,t)=\int_{0}^{1-e^{-t}}\,\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}K_{1}(z,t)dz=\int_{0}^{1}\,\Big{[}\varphi(x+\phi(t)s)+\varphi(x-\phi(t)s)\Big{]}K_{1}(\phi(t)s,t)\phi(t)ds,$ (4.1) where $\phi(t)=1-e^{-t}$. The proof of the theorem is splitted into several steps. ###### Proposition 4.1 The solution $u=u(x,t)$ of the Cauchy problem (0.20) with $\varphi_{0}(x)=0$ and $\varphi_{1}(x)=\varphi(x)$ can be represented as follows $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\,\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}$ (4.2) $\displaystyle\hskip 28.45274pt\times\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}$ $\displaystyle+\int_{0}^{t}db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)b\Big{[}e^{-2b}\,\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)-M^{2}E(x-y,t;0,b)\Big{]}\,\,.$ Proof. We look for the solution $u=u(x,t)$ of the form $u(x,t)=w(x,t)+t\varphi(x)$. Then (0.20) implies $\displaystyle w_{tt}-e^{-2t}w_{xx}+M^{2}w=te^{-2t}\varphi^{(2)}(x)-M^{2}t\varphi(x),\qquad w(x,0)=0,\quad w_{t}(x,0)=0\,.$ We set $f(x,t)=te^{-2t}\varphi^{(2)}(x)-M^{2}t\varphi(x)$ and due to Theorem 0.3 obtain $w(x,t)=\widetilde{w(x,t)}-M^{2}\int_{0}^{t}b\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)E(x-y,t;0,b),$ where $\displaystyle\widetilde{w(x,t)}$ $\displaystyle:=$ $\displaystyle\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi^{(2)}(y)E(x-y,t;0,b)\,.$ Then we integrate by parts: $\displaystyle\widetilde{w(x,t)}\\!\\!$ $\displaystyle\\!\\!=\\!\\!$ $\displaystyle\\!\\!\int_{0}^{t}be^{-2b}\,db\Bigg{[}\varphi^{(1)}(x+e^{-b}-e^{-t})E(-e^{-b}+e^{-t},t;0,b)-\varphi^{(1)}(x-e^{-b}+e^{-t})E(e^{-b}-e^{-t},t;0,b)\Bigg{]}$ $\displaystyle-\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ But $\varphi^{(1)}(x+e^{-b}-e^{-t})=-e^{b}\frac{\partial}{\partial b}\varphi(x+e^{-b}-e^{-t})\quad\mbox{\rm and}\quad\varphi^{(1)}(x-e^{-b}+e^{-t})=e^{b}\frac{\partial}{\partial b}\varphi(x-e^{-b}+e^{-t})\,.$ Then, $E(e^{-b}-e^{-t},t;0,b)=E(-e^{-b}+e^{-t},t;0,b)$ due to (3.3), and we obtain $\displaystyle\widetilde{w(x,t)}$ $\displaystyle=$ $\displaystyle\int_{0}^{t}be^{-2b}\,db\Bigg{[}-e^{b}\frac{\partial}{\partial b}\varphi(x+e^{-b}-e^{-t})-e^{b}\frac{\partial}{\partial b}\varphi(x-e^{-b}+e^{-t})\Bigg{]}E(e^{-b}-e^{-t},t;0,b)$ $\displaystyle-\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ One more integration by parts leads to $\displaystyle\widetilde{w(x,t)}$ $\displaystyle=$ $\displaystyle-2te^{-t}\varphi(x)E(0,t;0,t)$ $\displaystyle+\int_{0}^{t}\,db\Big{(}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{)}\frac{\partial}{\partial b}\Big{(}be^{-b}E(e^{-b}-e^{-t},t;0,b)\Big{)}$ $\displaystyle-\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ Since $E(0,t;0,t)=e^{t}/2$ we use (3.6) of Proposition 3.3 to derive the next representation $\displaystyle\widetilde{w(x,t)}+t\varphi(x)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)\Big{(}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{)}\,db$ $\displaystyle-\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ The integration by parts in the second term leads to $\displaystyle\widetilde{w(x,t)}+t\varphi(x)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle-\int_{0}^{t}be^{-2b}\,db\,\varphi(x+e^{-b}-e^{-t})\left[\frac{\partial}{\partial y}E(x-y,t;0,b)\right]_{y=x+e^{-b}-e^{-t}}$ $\displaystyle+\int_{0}^{t}be^{-2b}\,db\,\varphi(x-e^{-b}+e^{-t})\left[\frac{\partial}{\partial y}E(x-y,t;0,b)\right]_{y=x-e^{-b}+e^{-t}}$ $\displaystyle+\int_{0}^{t}be^{-2b}\,db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)\,.$ The application of (3.7) and (3.8) of Proposition 3.3 and $\frac{\partial}{\partial y}E(x-y,t;0,b)=-\frac{\partial}{\partial x}E(x-y,t;0,b)$ imply $\displaystyle\widetilde{w(x,t)}+t\varphi(x)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle+\int_{0}^{t}be^{-2b}\,db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)\,.$ Thus, for the function $u(x,t)=w(x,t)+t\varphi(x)$ we have obtained $\displaystyle u(x,t)$ (4.3) $\displaystyle=$ $\displaystyle\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle+\int_{0}^{t}be^{-2b}\,db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)$ $\displaystyle-M^{2}\int_{0}^{t}b\,db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,\varphi(y)E(x-y,t;0,b)\,\,.$ The proposition is proven. $\square$ ###### Corollary 4.2 The solution $u=u(x,t)$ of the Cauchy problem (0.20) with $\varphi_{0}(x)=0$ and $\varphi_{1}(x)=\varphi(x)$ can be represented as follows $\displaystyle u(x,t)$ (4.4) $\displaystyle=$ $\displaystyle\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle+\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dz\,\Big{[}\varphi(x-z)+\varphi(x+z)\Big{]}b\left[e^{-2b}\,\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]\,,$ as well as by (4.1), where $\displaystyle K_{1}(z,t)$ $\displaystyle=$ $\displaystyle\Big{[}\frac{1}{4}e^{t/2}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{3t/2}z\ln(z+e^{-t})\Big{]}\frac{1}{\sqrt{z+e^{-t}}}$ (4.5) $\displaystyle+\int_{0}^{-\ln(z+e^{-t})}b\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]db\,.$ Proof of Corollary. By means of the statement (4.2) of Proposition 4.1 and (3.3) we obtain $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\,\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}$ $\displaystyle\hskip 28.45274pt\times\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}$ $\displaystyle+\int_{0}^{t}db\int_{0}^{-(e^{-b}-e^{-t})}(-1)dz\,\varphi(x-z)\left[be^{-2b}\,\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}bE(z,t;0,b)\right]$ $\displaystyle+\int_{0}^{t}db\int_{-(e^{-b}-e^{-t})}^{0}dz\,\varphi(x+z)\left[be^{-2b}\,\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}bE(z,t;0,b)\right]\,.$ To prove (4.1) with $K_{1}(z,t)$ defined by (4.5) we apply (3.3) and write $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\,\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}$ $\displaystyle\hskip 28.45274pt\times\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}$ $\displaystyle+\int_{0}^{t}\,db\int_{0}^{e^{-b}-e^{-t}}dz\,\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}\left[be^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}bE(z,t;0,b)\right].$ Next we make change $z=e^{-b}-e^{-t}$, $dz=-e^{-b}db$, $db=-(z+e^{-t})^{-1}dz$, and $b=-\ln(z+e^{-t})$ in the first integral: $\displaystyle\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}(2-b)+\frac{1}{16}be^{-3b/2}e^{t/2}(e^{b}-e^{t})(1+4M^{2})\Big{]}\Big{[}\varphi(x+e^{-b}-e^{-t})+\varphi(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}\Big{[}\frac{1}{4}e^{t/2}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{3t/2}z\ln(z+e^{-t})\Big{]}\frac{1}{\sqrt{z+e^{-t}}}$ $\displaystyle\hskip 56.9055pt\times\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}\,dz\,.$ Then $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}\Big{[}\frac{1}{4}e^{t/2}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{3t/2}z\ln(z+e^{-t})\Big{]}\frac{1}{\sqrt{z+e^{-t}}}$ $\displaystyle\hskip 56.9055pt\times\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}\,dz$ $\displaystyle+\int_{0}^{t}\,db\int_{0}^{e^{-b}-e^{-t}}dz\,\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}b\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right].$ In the last integral we change the order of integration, $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}\Big{[}\frac{1}{4}e^{t/2}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{3t/2}z\ln(z+e^{-t})\Big{]}\frac{1}{\sqrt{z+e^{-t}}}\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}\,dz$ $\displaystyle+\int_{0}^{1-e^{-t}}\,dz\Big{[}\varphi(x+z)+\varphi(x-z)\Big{]}\int_{0}^{-\ln(z+e^{-t})}db\,b\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right],$ and obtain (4.1), where $K_{1}(z,t)$ is defined by (4.5). Corollary is proven. $\square$ Proof of Theorem 0.4 with $\varphi_{0}=0$. The next lemma completes the proof of Theorem 0.4. ###### Lemma 4.3 The kernel $K_{1}(z,t)$ defined by (4.5) coincides with one given in Theorem 0.4. Proof. We have due to Lemma 1.2, (3.3), and by integration by parts $\displaystyle\int_{0}^{-\ln(z+e^{-t})}b\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]\,db=\int_{0}^{-\ln(z+e^{-t})}b\Big{(}\frac{\partial}{\partial b}\Big{)}^{2}E(z,t;0,b)db$ $\displaystyle=$ $\displaystyle(-\ln(z+e^{-t}))\Big{[}\frac{\partial}{\partial b}E(z,t;0,b)\Big{]}_{b=-\ln(z+e^{-t})}-E(z,t;0,-\ln(z+e^{-t}))+E(z,t;0,0)\,.$ On the other hand, (3.4) and (3.9) of Proposition 3.3 imply $\displaystyle\int_{0}^{-\ln(z+e^{-t})}db\,b\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]$ $\displaystyle=$ $\displaystyle-\ln(z+e^{-t})\frac{1}{16}\frac{4+e^{t}z(1+4M^{2})}{\sqrt{e^{-t}}\sqrt{z+e^{-t}}}-\frac{1}{2}\frac{1}{\sqrt{e^{-t}}\sqrt{z+e^{-t}}}+E(z,t;0,0).$ Thus, for the kernel $K_{1}(z,t)$ defined by (4.5) we have $\displaystyle K_{1}(z,t)$ $\displaystyle=$ $\displaystyle\Big{[}\frac{1}{4}e^{t/2}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{3t/2}z\ln(z+e^{-t})\Big{]}\frac{1}{\sqrt{z+e^{-t}}}$ $\displaystyle-\ln(z+e^{-t})\frac{1}{16}\frac{4+e^{t}z(1+4M^{2})}{\sqrt{e^{-t}}\sqrt{z+e^{-t}}}-\frac{1}{2}\frac{1}{\sqrt{e^{-t}}\sqrt{z+e^{-t}}}+E(z,t;0,0)$ $\displaystyle=$ $\displaystyle e^{t/2}\Big{[}\frac{1}{4}\big{(}2+\ln(z+e^{-t})\big{)}+\frac{1}{16}(1+4M^{2})e^{t}z\ln(z+e^{-t})$ $\displaystyle-\ln(z+e^{-t})\frac{1}{16}(4+e^{t}z(1+4M^{2}))-\frac{1}{2}\Big{]}\frac{1}{\sqrt{z+e^{-t}}}+E(z,t;0,0)$ $\displaystyle=$ $\displaystyle E(z,t;0,0)\,.$ The last line coincides with $K_{1}(z,t)$ of Theorem 0.4. Lemma is proven. $\square$ ## 5 The Cauchy Problem: First Datum and $n=1$ In this section we prove Theorem 0.4 in the case of $\varphi_{1}(x)=0$. Thus, we have to prove the representation given by Theorem 0.4 for the solution $u=u(x,t)$ of the Cauchy problem (0.20) with $\varphi_{1}(x)=0$, that is equivalent to $u(x,t)=\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}+\,\int_{0}^{1}\big{[}\varphi_{0}(x-\phi(t)s)+\varphi_{0}(x+\phi(t)s)\big{]}K_{0}(\phi(t)s,t)\phi(t)\,ds,$ where $\phi(t)=1-e^{-t}$. The proof of this case consists of the several steps. ###### Proposition 5.1 The solution $u=u(x,t)$ of the Cauchy problem (0.20) can be represented as follows $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}$ $\displaystyle-\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}+\frac{1}{16}(1+4M^{2})e^{t/2}e^{-3b/2}(e^{t}-e^{b})\Big{]}\Big{[}\varphi_{0}(x+e^{-b}-e^{-t})+\varphi_{0}(x-e^{-b}+e^{-t}))\Big{]}\,db$ $\displaystyle+\int_{0}^{t}db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi_{0}(y)\left[e^{-2b}\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)-M^{2}E(x-y,t;0,b)\right]\,.$ Proof. We set $u(x,t)=w(x,t)+\varphi_{0}(x)$, then $w_{tt}-e^{-2t}w_{xx}+M^{2}w=e^{-2t}\varphi_{0,xx}-M^{2}\varphi_{0}(x)\,,\qquad w(x,0)=0\,,\qquad w_{t}(x,0)=0\,.$ Next we plug $f(x,t)=e^{-2t}\varphi_{0,xx}(x)-M^{2}\varphi_{0}(x)$ in the formula given by Theorem 0.3 and obtain $\displaystyle w(x,t)$ $\displaystyle=$ $\displaystyle\widetilde{w(x,t)}-\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,M^{2}\varphi_{0}(y)E(x-y,t;0,b),$ (5.2) where we have denoted $\displaystyle\widetilde{w(x,t)}$ $\displaystyle:=$ $\displaystyle\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,e^{-2b}\varphi_{0}^{(2)}(y)E(x-y,t;0,b)\,.$ Next we integrate by parts and apply (3.3): $\displaystyle\widetilde{w(x,t)}$ $\displaystyle=$ $\displaystyle\int_{0}^{t}e^{-2b}\Big{[}\varphi_{0}^{(1)}(x+e^{-b}-e^{-t})-\varphi_{0}^{(1)}(x-e^{-b}+e^{-t})\Big{]}E(e^{-b}-e^{-t},t;0,b)\,db$ $\displaystyle-\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,e^{-2b}\varphi_{0}^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ On the other hand, $\varphi_{0}^{(1)}(x+e^{-b}-e^{-t})=-e^{b}\frac{\partial}{\partial b}\varphi_{0}(x+e^{-b}-e^{-t})\quad\mbox{\rm and}\quad\varphi_{0}^{(1)}(x-e^{-b}+e^{-t})=e^{b}\frac{\partial}{\partial b}\varphi_{0}(x-e^{-b}+e^{-t})$ imply $\displaystyle\widetilde{w(x,t)}$ $\displaystyle=$ $\displaystyle-\int_{0}^{t}e^{-b}\Big{[}\frac{\partial}{\partial b}\varphi_{0}(x+e^{-b}-e^{-t})+\frac{\partial}{\partial b}\varphi_{0}(x-e^{-b}+e^{-t})\Big{]}E(e^{-b}-e^{-t},t;0,b)\,db$ $\displaystyle-\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,e^{-2b}\varphi_{0}^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,.$ One more integration by parts leads to $\displaystyle\widetilde{w(x,t)}+\varphi_{0}(x)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}$ $\displaystyle+\int_{0}^{t}\Big{[}\varphi_{0}(x+e^{-b}-e^{-t})+\varphi_{0}(x-e^{-b}+e^{-t})\Big{]}\frac{\partial}{\partial b}\Big{(}e^{-b}E(e^{-b}-e^{-t},t;0,b)\Big{)}\,db$ $\displaystyle-\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,e^{-2b}\varphi_{0}^{(1)}(y)\frac{\partial}{\partial y}E(x-y,t;0,b)\,,$ where $E(0,t;0,t)=\frac{1}{2}e^{t}$ and $E(1-e^{-t},t;0,0)=\frac{1}{2}e^{\frac{t}{2}}$ have been used. Next we apply (3.5) of Proposition 3.3 and an integration by parts to obtain $\displaystyle\widetilde{w(x,t)}+\varphi_{0}(x)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}$ $\displaystyle-\int_{0}^{t}\frac{1}{4}e^{t/2}e^{-b/2}\Big{[}\varphi_{0}(x+e^{-b}-e^{-t})+\varphi_{0}(x-e^{-b}+e^{-t})\Big{]}\,db$ $\displaystyle+\int_{0}^{t}db\,e^{-2b}\Big{[}\varphi_{0}(y)\frac{\partial}{\partial x}E(x-y,t;0,b)\Big{]}_{y=x-e^{-b}+e^{-t}}^{y=x+e^{-b}-e^{-t}}$ $\displaystyle+\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,\varphi_{0}(y)e^{-2b}\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)\,.$ We have due to (3.7) and (3.8) of Proposition 3.3 $\displaystyle\widetilde{w(x,t)}+\varphi_{0}(x)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}$ $\displaystyle-\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}+\frac{1}{16}(1+4M^{2})e^{t/2}e^{-3b/2}(e^{t}-e^{b})\Big{]}\Big{[}\varphi_{0}(x+e^{-b}-e^{-t})+\varphi_{0}(x-e^{-b}+e^{-t}))\Big{]}\,db$ $\displaystyle+\int_{0}^{t}db\int_{x-e^{-b}+e^{-t}}^{x+e^{-b}-e^{-t}}dy\,\varphi_{0}(y)e^{-2b}\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)\,.$ Then the last equation together with (5.2) proves the desired representation. Proposition is proven. $\square$ Completion of the proof of Theorem 0.4. We make change $z=e^{-b}-e^{-t}$, $dz=-e^{-b}db$, and $b=-\ln(z+e^{t})$ in the second term of the representation given by the previous proposition: $\displaystyle\int_{0}^{t}\Big{[}\frac{1}{4}e^{t/2}e^{-b/2}+\frac{1}{16}(1+4M^{2})e^{t/2}e^{-3b/2}(e^{t}-e^{b})\Big{]}\Big{[}\varphi_{0}(x+e^{-b}-e^{-t})+\varphi_{0}(x-e^{-b}+e^{-t}))\Big{]}\,db$ $\displaystyle=$ $\displaystyle\int^{1-e^{-t}}_{0}\Big{[}\frac{1}{4}e^{t/2}\sqrt{z+e^{-t}}+\frac{1}{16}(1+4M^{2})e^{t/2}(\sqrt{z+e^{-t}})^{3}\frac{ze^{t}}{z+e^{-t}}\Big{]}\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\frac{1}{z+e^{-t}}\,dz$ $\displaystyle=$ $\displaystyle\int^{1-e^{-t}}_{0}e^{t}\Big{[}\frac{1}{4}+\frac{1}{16}(1+4M^{2})ze^{t}\Big{]}\frac{1}{\sqrt{e^{t}z+1}}\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\,dz\,.$ Next we apply (3.3) to the last term of that representation, and then we change the order of integration: $\displaystyle\int_{0}^{t}db\int_{x-(e^{-b}-e^{-t})}^{x+e^{-b}-e^{-t}}dy\,\varphi_{0}(y)\left[e^{-2b}\left(\frac{\partial}{\partial y}\right)^{2}E(x-y,t;0,b)-M^{2}E(x-y,t;0,b)\right]$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dz\,\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}dz\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\int_{0}^{-\ln(z+e^{-t})}db\,\left[e^{-2b}\left(\frac{\partial}{\partial z}\right)^{2}E(z,t;0,b)-M^{2}E(z,t;0,b)\right]\,.$ On the other hand, since the function $E(z,t;0,b)$ solves Klein-Gordon equation, the last integral is equal to $\displaystyle\int_{0}^{1-e^{-t}}dz\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\int_{0}^{-\ln(z+e^{-t})}db\,\left[\left(\frac{\partial}{\partial t}\right)^{2}E(z,t;0,b)\right]$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}dz\Big{[}\varphi_{0}(x+z)+\varphi_{0}(x-z)\Big{]}\left[\frac{\partial}{\partial b}E(z,t;0,b)\right]_{b=0}^{b=-\ln(z+e^{-t})}\,.$ Application of (3.9) and (3.10) completes the proof. Theorem 0.4 is proven. $\square$ ## 6 n-Dimensional Case, $n\geq 2$ Proof of Theorem 0.5. Let us consider the case of $x\in{\mathbb{R}}^{n}$, where $n=2m+1$, $m\in{\mathbb{N}}$. First for the given function $u=u(x,t)$ we define the spherical means of $u$ about point $x$: $\displaystyle I_{u}(x,r,t)$ $\displaystyle=$ $\displaystyle\frac{1}{\omega_{n-1}}\int_{S^{n-1}}u(x+ry,t)\,dS_{y}\,,$ where $\omega_{n-1}$ denotes the area of the unit sphere $S^{n-1}\subset{\mathbb{R}}^{n}$. Then we define an operator $\Omega_{r}$ by $\Omega_{r}(u)(x,t):=\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{m-1}r^{2m-1}I_{u}(x,r,t)\,.$ One can show that there are constants $c_{j}^{(n)}$, $j=0,\ldots,m-1$, where $n=2m+1$, with $c_{0}^{(n)}=1\cdot 3\cdot 5\cdots(n-2)$, such that $\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{m-1}r^{2m-1}\varphi(r)=r\sum_{j=0}^{m-1}c_{j}^{(n)}r^{j}\frac{\partial^{j}}{\partial r^{j}}\varphi(r)\,.$ One can recover the functions according to $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\lim_{r\to 0}I_{u}(x,r,t)=\lim_{r\to 0}\frac{1}{c_{0}^{(n)}r}\Omega_{r}(u)(x,t)\,,$ (6.1) $\displaystyle u(x,0)$ $\displaystyle=$ $\displaystyle\lim_{r\to 0}\frac{1}{c_{0}^{(n)}r}\Omega_{r}(u)(x,0)\,,\quad u_{t}(x,0)=\lim_{r\to 0}\frac{1}{c_{0}^{(n)}r}\Omega_{r}(\partial_{t}u)(x,0)\,.$ (6.2) It is well known that $\Delta_{x}\Omega_{r}h=\frac{\partial^{2}}{\partial\,r^{2}}\Omega_{r}h$ for every function $h\in C^{2}({\mathbb{R}}^{n})$. Therefore we arrive at the following mixed problem for the function $v(x,r,t):=\Omega_{r}(u)(x,r,t)$: $\displaystyle\cases{v_{tt}(x,r,t)-e^{-2t}v_{rr}(x,r,t)+M^{2}v(x,r,t)=F(x,r,t)\quad\mbox{\rm for all}\quad t\geq 0\,,\,\,r\geq 0\,,\,\,x\in{\mathbb{R}}^{n}\,,\cr v(x,0,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,\quad x\in{\mathbb{R}}^{n},\cr v(x,r,0)=0\,,\quad v_{t}(x,r,0)=0\quad\mbox{\rm for all}\quad r\geq 0\,,\quad x\in{\mathbb{R}}^{n}\,,\cr F(x,r,t):=\Omega_{r}(f)(x,t)\,,\quad F(x,0,t)=0\,,\quad\mbox{\rm for all}\quad x\in{\mathbb{R}}^{n}\,.}$ It must be noted here that the spherical mean $I_{u}$ defined for $r>0$ has an extension as even function for $r<0$ and hence $\Omega_{r}(u)$ has a natural extension as an odd function. That allows replacing the mixed problem with the Cauchy problem. Namely, let functions $\widetilde{v}$ and $\widetilde{F}$ be the continuations of the functions $v$ and $F$, respectively, by $\widetilde{v}(x,r,t)=\cases{\,v(x,r,t),\,\,if\,\,r\geq 0\cr-v(x,-r,t),\,\,if\,\,r\leq 0}\,,\quad\widetilde{F}(x,r,t)=\cases{\,F(x,r,t),\,\,if\,\,r\geq 0\cr-F(x,-r,t),\,\,if\,\,r\leq 0}\,.$ Then $\widetilde{v}$ solves the Cauchy problem $\displaystyle\cases{\widetilde{v}_{tt}(x,r,t)-e^{-2t}\widetilde{v}_{rr}(x,r,t)+M^{2}\widetilde{v}(x,r,t)=\widetilde{F}(x,r,t)\quad\mbox{\rm for all}\quad t\geq 0\,,\quad r\in{\mathbb{R}}\,,\quad x\in{\mathbb{R}}^{n}\,,\cr\widetilde{v}(x,r,0)=0\,,\quad\widetilde{v}_{t}(x,r,0)=0\quad\mbox{\rm for all}\quad r\in{\mathbb{R}}\,,\quad x\in{\mathbb{R}}^{n}.}$ Hence, according to Theorem 0.3, one has the representation $\displaystyle\widetilde{v}(x,r,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{t}db\int_{r-(e^{-b}-e^{-t})}^{r+e^{-b}-e^{-t}}dr_{1}\,\widetilde{F}(x,r_{1},b)(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-(r-r_{1})^{2}\right)^{-\frac{1}{2}-iM}$ $\displaystyle\hskip 85.35826pt\times F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-(r-r_{1})^{2}}{(e^{-b}+e^{-t})^{2}-(r-r_{1})^{2}}\right).$ Since $u(x,t)=\lim_{r\to 0}\big{(}\widetilde{v}(x,r,t)/(c_{0}^{(n)}r)\big{)}$, we consider the case of $r<t$ in the above representation to obtain: $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{c_{0}^{(n)}}\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dr_{1}\,\lim_{r\to 0}\frac{1}{r}\left\\{\widetilde{F}(x,r+r_{1},b)+\widetilde{F}(x,r-r_{1},b)\right\\}$ $\displaystyle\times(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-r_{1}^{2}\right)^{-\frac{1}{2}-iM}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-r_{1}^{2}}{(e^{-b}+e^{-t})^{2}-r_{1}^{2}}\right)\,.$ Then by definition of the function $\widetilde{F}$, we replace $\lim_{r\to 0}\frac{1}{r}\Big{\\{}\widetilde{F}(x,r-r_{1},b)+\widetilde{F}(x,r+r_{1},b)\Big{\\}}$ with $2\Big{(}\frac{\partial}{\partial r}F(x,r,b)\Big{)}_{r=r_{1}}$ in the last formula. The definitions of $F(x,r,t)$ and of the operator $\Omega_{r}$ yield: $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{2}{c_{0}^{(n)}}\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}dr_{1}\,\left(\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{m-1}r^{2m-1}I_{f}(x,r,t)\right)_{r=r_{1}}$ $\displaystyle\times(4e^{-b-t})^{iM}\left((e^{-t}+e^{-b})^{2}-r_{1}^{2}\right)^{-\frac{1}{2}-iM}F\left(\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{-b}-e^{-t})^{2}-r_{1}^{2}}{(e^{-b}+e^{-t})^{2}-r_{1}^{2}}\right)\,,$ where $x\in{\mathbb{R}}^{n}$, $n=2m+1$, $m\in{\mathbb{N}}$. Thus, the solution to the Cauchy problem is given by (0.22). We employ the method of descent to complete the proof for the case of even $n$, $n=2m$, $m\in{\mathbb{N}}$. Theorem 0.5 is proven. $\square$ Proof of (0.16) and (0.17). We set $f(x,b)=\delta(x)\delta(t-t_{0})$ in (0.22) and (0.23), and we obtain (0.16) and (0.17), where if $n$ is odd, then $E^{w}(x,t):=\frac{1}{\omega_{n-1}1\cdot 3\cdot 5\ldots\cdot(n-2)}\frac{\partial}{\partial t}\Big{(}\frac{1}{t}\frac{\partial}{\partial t}\Big{)}^{\frac{n-3}{2}}\frac{1}{t}\delta(|x|-t)\,,$ while for $n$ even we have $E^{w}(x,t):=\frac{2}{\omega_{n-1}1\cdot 3\cdot 5\ldots\cdot(n-1)}\frac{\partial}{\partial t}\Big{(}\frac{1}{t}\frac{\partial}{\partial t}\Big{)}^{\frac{n-2}{2}}\frac{1}{\sqrt{t^{2}-|x|^{2}}}\chi_{B_{t}(x)}\,.$ Here $\chi_{B_{t}(x)}$ denotes the characteristic function of the ball $B_{t}(x):=\\{x\in{\mathbb{R}}^{n};\,|x|$ $\leq t\\}$. Constant $\omega_{n-1}$ is the area of the unit sphere $S^{n-1}\subset{\mathbb{R}}^{n}$. The distribution $\delta(|x|-t)$ is defined by $<\delta(|\cdot|-t),f(\cdot)>=\int_{|x|=t}f(x)\,dx$ for $f\in C^{\infty}_{0}({\mathbb{R}}^{n})$. Proof of Theorem 0.6. First we consider case of $\varphi_{0}(x)=0$. More precisely, we have to prove that the solution $u(x,t)$ of the Cauchy problem (0.25) with $\varphi_{0}(x)=0$ can be represented by (0.26) with $\varphi_{0}(x)=0$. The next lemma will be used in both cases. ###### Lemma 6.1 Consider the mixed problem $\displaystyle\cases{v_{tt}-e^{-2t}v_{rr}+M^{2}v=0\quad\mbox{\rm for all}\quad t\geq 0\,,\quad r\geq 0\,,\quad\cr v(r,0)=\tau_{0}(r)\,,\quad v_{t}(r,0)=\tau_{1}(r)\quad\mbox{\rm for all}\quad r\geq 0\,,\cr v(0,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,}$ and denote by $\widetilde{\tau}_{0}(r)$ and $\widetilde{\tau}_{1}(r)$ the continuations of the functions $\tau_{0}(r)$ and $\tau_{1}(r)$ for negative $r$ as odd functions: $\widetilde{\tau}_{0}(-r)=-\tau_{0}(r)$ and $\widetilde{\tau}_{1}(-r)=-\tau_{1}(r)$ for all $r\geq 0$, respectively. Then solution $v(r,t)$ to the mixed problem is given by the restriction of (4.1) to $r\geq 0$: $\displaystyle v(r,t)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\widetilde{\tau}_{0}(r+1-e^{-t})+\widetilde{\tau}_{0}(r-1+e^{-t})\Big{]}+\,\int_{0}^{1}\big{[}\widetilde{\tau}_{0}(r-\phi(t)s)+\widetilde{\tau}_{0}(r+\phi(t)s)\big{]}K_{0}(\phi(t)s,t)\phi(t)\,ds$ $\displaystyle+\int_{0}^{1}\Big{[}\widetilde{\tau}_{1}\Big{(}r+\phi(t)s\Big{)}+\widetilde{\tau}_{1}\Big{(}r-\phi(t)s\Big{)}\Big{]}K_{1}(\phi(t)s,t)\phi(t)\,ds\,,$ where $K_{0}(z,t)$ and $K_{1}(z,t)$ are defined in Theorem 0.4 and $\phi(t)=1-e^{-t}$. Proof. This lemma is a direct consequence of Theorem 0.4. $\square$ Now let us consider the case of $x\in{\mathbb{R}}^{n}$, where $n=2m+1$. First for the given function $u=u(x,t)$ we define the spherical means of $u$ about point $x$. One can recover the functions by means of (6.1), (6.2), and $\displaystyle\varphi_{i}(x)$ $\displaystyle=$ $\displaystyle\lim_{r\to 0}I_{\varphi_{i}}(x,r)=\lim_{r\to 0}\frac{1}{c_{0}^{(n)}r}\Omega_{r}(\varphi_{i})(x)\,,\quad i=0,1\,.$ Then we arrive at the following mixed problem $\displaystyle\cases{v_{tt}(x,r,t)-e^{-2t}v_{rr}(x,r,t)+M^{2}v(x,r,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,\,\,r\geq 0\,,\,\,x\in{\mathbb{R}}^{n}\,,\cr v(x,0,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,\quad x\in{\mathbb{R}}^{n}\,,\cr v(x,r,0)=0\,,\quad v_{t}(x,r,0)=\Phi_{1}(x,r)\quad\mbox{\rm for all}\quad r\geq 0\,,\quad x\in{\mathbb{R}}^{n}\,,}$ with the unknown function $v(x,r,t):=\Omega_{r}(u)(x,r,t)$, where $\displaystyle\Phi_{i}(x,r):=\Omega_{r}(\varphi_{i})(x)=\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{m-1}r^{2m-1}\frac{1}{\omega_{n-1}}\int_{S^{n-1}}\varphi_{i}(x+ry)\,dS_{y}\,,$ (6.3) $\displaystyle\Phi_{i}(x,0)=0\,,\quad i=0,1,\quad\quad\mbox{\rm for all}\quad x\in{\mathbb{R}}^{n}\,.$ (6.4) Then, according to Lemma 6.1 and to $u(x,t)=\lim_{r\to 0}\big{(}v(x,r,t)/(c_{0}^{(n)}r)\big{)}$, we obtain: $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{c_{0}^{(n)}}\lim_{r\to 0}\frac{1}{r}\int_{0}^{1}\Big{[}\widetilde{\Phi}_{1}\big{(}x,r+\phi(t)s\big{)}+\widetilde{\Phi}_{1}\big{(}x,r-\phi(t)s\big{)}\Big{]}K_{1}(\phi(t)s,t)\phi(t)\,ds\,.$ The last limit is equal to $\displaystyle 2\int_{0}^{1}\left(\frac{\partial}{\partial r}\Phi_{1}(x,r)\right)_{r=\phi(t)s}K_{1}(\phi(t)s,t)\phi(t)\,ds$ $\displaystyle=$ $\displaystyle 2\int_{0}^{1}\left(\frac{\partial}{\partial r}\Big{(}\frac{1}{r}\frac{\partial}{\partial r}\Big{)}^{\frac{n-3}{2}}\frac{r^{n-2}}{\omega_{n-1}}\int_{S^{n-1}}\varphi_{1}(x+ry)\,dS_{y}\right)_{r=\phi(t)s}K_{1}(\phi(t)s,t)\phi(t)\,ds\,.$ Thus, Theorem 0.6 in the case of $\varphi_{0}(x)=0$ is proven. Now we turn to the case of $\varphi_{1}(x)=0$. Thus, we arrive at the following mixed problem $\displaystyle\cases{v_{tt}(x,r,t)-e^{-2t}v_{rr}(x,r,t)+M^{2}v(x,r,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,\,\,r\geq 0\,,\,\,x\in{\mathbb{R}}^{n}\,,\cr v(x,r,0)=\Phi_{0}(x,r)\,,\quad v_{t}(x,r,0)=0\quad\mbox{\rm for all}\quad r\geq 0\,,\quad x\in{\mathbb{R}}^{n}\,,\cr v(x,0,t)=0\quad\mbox{\rm for all}\quad t\geq 0\,,\quad x\in{\mathbb{R}}^{n}\,,}$ with the unknown function $v(x,r,t):=\Omega_{r}(u)(x,r,t)$ defined by (6.3), (6.4). Then, according to Lemma 6.1 and to $u(x,t)=\lim_{r\to 0}\big{(}v(x,r,t)/(c_{0}^{(n)}r)\big{)}$, we obtain: $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{c_{0}^{(n)}}e^{\frac{t}{2}}\lim_{r\to 0}\frac{1}{2r}\Big{[}\widetilde{\Phi}_{0}(x,r+e^{t}-1)+\widetilde{\Phi}_{0}(x,r-e^{t}+1)\Big{]}$ $\displaystyle+\,\frac{2}{c_{0}^{(n)}}\int_{0}^{1}\lim_{r\to 0}\frac{1}{2r}\big{[}\widetilde{\Phi}_{0}(x,r-\phi(t)s)+\widetilde{\Phi}_{0}(x,r+\phi(t)s)\big{]}K_{0}(\phi(t)s,t)\phi(t)\,ds\,,$ $\displaystyle=$ $\displaystyle\frac{1}{c_{0}^{(n)}}e^{\frac{t}{2}}\left(\frac{\partial}{\partial r}\Phi_{0}(x,r)\right)_{r=\phi(t)}+\,\frac{2}{c_{0}^{(n)}}\int_{0}^{1}\left(\frac{\partial}{\partial r}\Phi_{0}(x,r)\right)_{r=\phi(t)s}K_{0}(\phi(t)s,t)\phi(t)\,ds$ $\displaystyle=$ $\displaystyle e^{\frac{t}{2}}v_{\varphi_{0}}(x,\phi(t))+\,2\int_{0}^{1}v_{\varphi_{0}}(x,\phi(t)s)K_{0}(\phi(t)s,t)\phi(t)\,ds\,.$ Theorem 0.6 is proven. $\square$ ## 7 $L^{p}-L^{q}$ and $L^{q}-L^{q}$ Estimates for the Solutions of One- dimensional Equation, $n=1$ Consider now the Cauchy problem for the equation (0.18) with the source term and with vanishing initial data (0.19). In this and next sections we restrict ourselves to the particles with “large” mass $m\geq n/2$, that is, with nonnegative curved mass $M\geq 0$, to make presentation more transparent. The case of $m<n/2$ will be discussed in the forthcoming paper. ###### Theorem 7.1 For every function $f\in C^{2}({\mathbb{R}}\times[0,\infty))$ such that $f(\cdot,t)\in C_{0}^{\infty}({\mathbb{R}}_{x})$ the solution $u=u(x,t)$ of the Cauchy problem (0.18), (0.19) satisfies inequality $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle C_{M}e^{t(1-1/\rho)}\int_{0}^{t}(1+t-b)(e^{t-b}-1)^{1/\rho}(e^{t-b}+1)^{-1}\|f(x,b)\|_{L^{p}({\mathbb{R}}_{x})}\,db$ for all $t>0$, where $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\rho<2$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$. Proof. Using the fundamental solution from Theorem 0.1 one can write the convolution $u(x,t)=\int_{-\infty}^{\infty}\\!\int_{-\infty}^{\infty}{\mathcal{E}}_{+}(x,t;y,b)f(y,b)\,db\,dy=\int_{0}^{t}\\!db\\!\int_{-\infty}^{\infty}{\mathcal{E}}_{+}(x-y,t;0,b)f(y,b)\,dy\,.$ Due to Young’s inequality we have $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle c_{k}\int_{0}^{t}db\Bigg{(}\int_{-(\phi(t)-\phi(b))}^{\phi(t)-\phi(b)}|E(x,t;0,b)|^{\rho}dx\Bigg{)}^{1/\rho}\|f(x,b)\|_{L^{p}({\mathbb{R}}_{x})},$ where $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$, $\phi(t)=1-e^{-t}$. The integral in parentheses can be transformed as follows $\displaystyle\int_{-(\phi(t)-\phi(b))}^{\phi(t)-\phi(b)}|E(x,t;0,b)|^{\rho}dx$ $\displaystyle=$ $\displaystyle 2e^{-t+t\rho}\int_{0}^{e^{t-b}-1}\Big{(}(e^{t-b}+1)^{2}-y^{2}\Big{)}^{-\frac{\rho}{2}}\left|F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{t-b}-1)^{2}-y^{2}}{(e^{t-b}+1)^{2}-y^{2}}\Big{)}\right|^{\rho}dy.$ On the other hand, due to integral representation for the hypergeometric function (1)[2, v.1, Sec.2.12] for $\zeta\in[0,1)$ one has $\displaystyle\left|F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\zeta\Big{)}\right|$ $\displaystyle\leq$ $\displaystyle\frac{1}{|\Gamma\left(\frac{1}{2}+iM\right)|^{2}}\pi F\Big{(}\frac{1}{2},\frac{1}{2};1;\zeta\Big{)}\,.$ (7.1) Thus, $\int_{-(\phi(t)-\phi(b))}^{\phi(t)-\phi(b)}|E(x,t;0,b)|^{\rho}dx\leq C_{M}e^{-t+t\rho}\int_{0}^{e^{t-b}-1}\\!\left((e^{t-b}+1)^{2}-y^{2}\right)^{-\frac{\rho}{2}}\left|F\Big{(}\frac{1}{2},\frac{1}{2};1;\frac{(e^{t-b}-1)^{2}-y^{2}}{(e^{t-b}+1)^{2}-y^{2}}\Big{)}\right|^{\rho}\\!dy.$ ###### Lemma 7.2 [33] For all $z>1$ the following estimate $\int_{0}^{z-1}((z+1)^{2}-r^{2})^{-\frac{\rho}{2}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)^{\rho}dr\leq C(1+\ln z)^{\rho}(z-1)(z+1)^{-\rho}F\Big{(}\frac{1}{2},\frac{\rho}{2};\frac{3}{2};\frac{(z-1)^{2}}{(z+1)^{2}}\Big{)}$ is fulfilled, provided that $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$. In particular, if $\rho<2$, then $\displaystyle\int_{0}^{z-1}((z+1)^{2}-r^{2})^{-\frac{\rho}{2}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)^{\rho}dr$ $\displaystyle\leq$ $\displaystyle C(1+\ln z)^{\rho}(z-1)(z+1)^{-\rho}\,.$ Completion of the proof of Theorem 7.1. Thus for $\rho<2$ and $z=e^{t-b}$ we have $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle C_{M}e^{t(1-1/\rho)}\int_{0}^{t}(1+t-b)(e^{t-b}-1)^{1/\rho}(e^{t-b}+1)^{-1}\|f(x,b)\|_{L^{p}({\mathbb{R}}_{x})}\,db\,.$ The last inequality implies the estimate of the statement of theorem. Theorem 7.1 is proven. $\square$ ###### Proposition 7.3 The solution $u=u(x,t)$ of the Cauchy problem $u_{tt}-e^{-2t}u_{xx}+M^{2}u=0\,,\qquad u(x,0)=\varphi_{0}(x)\,,\qquad u_{t}(x,0)=\varphi_{1}(x)\,,$ with $\varphi_{0}$, $\varphi_{1}\in C_{0}^{\infty}({\mathbb{R}})$ satisfies the following estimate $\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}\leq C(1+t)\Big{(}e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{L^{q}({\mathbb{R}}_{x})}+(e^{t}-1)e^{-t}\|\varphi_{1}(x)\|_{L^{q}({\mathbb{R}}_{x})}\Big{)}\,\quad\mbox{for all}\,\quad t\in(0,\infty).$ (7.2) Proof. First we consider the equation without source term but with the second datum that is the case of $\varphi_{0}=0$. For the convenience we drop subindex of $\varphi_{1}$. Then we apply the representation given by Theorem 0.4 for the solution $u=u(x,t)$ of the Cauchy problem with $\varphi_{0}=0$, and obtain $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\int_{0}^{1-e^{-t}}\,\Big{[}\varphi_{1}(x-z)+\varphi_{1}(x+z)\Big{]}K_{1}(z,t)dz\,,$ where the kernel $K_{1}(z,t)$ is defined in Theorem 0.4. Hence, we arrive at inequality $\|u(x,t)\|_{{L}^{q}({\mathbb{R}})}\leq C_{M}\|\varphi(x)\|_{{L}^{q}({\mathbb{R}})}\int_{0}^{1-e^{-t}}\,|K_{1}(r,t)|dr\,.$ To estimate the last integral we denote it by $I_{1}$, $\displaystyle I_{1}(z)$ $\displaystyle:=$ $\displaystyle\int_{0}^{1-e^{-t}}\,|K_{1}(r,t)|dr\,,$ and with $z=e^{t}>1$ due to (7.1) we write $\displaystyle I_{1}(z)$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{z-1}\,\frac{1}{\sqrt{(1+z)^{2}-y^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{y^{2}-(1-z)^{2}}{y^{2}-(1+z)^{2}}\right)dy\,.$ (7.3) Then, according to Lemma 7.2 (the case of $\rho=1$) we have for that integral the following estimate $I_{1}(e^{t})\leq C(1+t)(e^{t}-1)(e^{t}+1)^{-1}\,.$ (7.4) Finally, (7.3) and (7.4) imply the $L^{q}-L^{q}$ estimate (7.2) for the case of $\varphi_{0}=0$. Next we consider the equation without source but with the first datum, that is, the case of $\varphi_{1}=0$. We apply the representation given by Theorem 0.4 for the solution $u=u(x,t)$ of the Cauchy problem with $\varphi_{1}=0$, and obtain $\displaystyle u(x,t)$ $\displaystyle=$ $\displaystyle\frac{1}{2}e^{\frac{t}{2}}\Big{[}\varphi_{0}(x+1-e^{-t})+\varphi_{0}(x-1+e^{-t})\Big{]}+\int_{0}^{1-e^{-t}}\big{[}\varphi_{0}(x-r)+\varphi_{0}(x+r)\big{]}K_{0}(r,t)\,dr\,,$ where the kernel $K_{0}(r,t)$ is defined in Theorem 0.4. Then we easily obtain the following two estimates: $\displaystyle\|u(x,t)-\int_{0}^{1-e^{-t}}\big{[}\varphi_{0}(x-r)+\varphi_{0}(x+r)\big{]}K_{0}(r,t)\,dr\|_{{L}^{q}({\mathbb{R}})}$ $\displaystyle\leq$ $\displaystyle e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{{L}^{q}({\mathbb{R}})}$ and $\displaystyle\|u(x,t)\|_{{L}^{q}({\mathbb{R}})}$ $\displaystyle\leq$ $\displaystyle e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{{L}^{q}({\mathbb{R}})}+2\|\varphi_{0}(x)\|_{{L}^{q}({\mathbb{R}})}\int_{0}^{1-e^{-t}}\left|K_{0}(r,t)\right|dr\,.$ Finally, the following lemma completes the proof of proposition. ###### Lemma 7.4 The kernel $K_{0}(r,t)$ has an integrable singularity at $r=e^{t}-1$, more precisely, one has $\displaystyle\int_{0}^{1-e^{-t}}\left|K_{0}(r,t)\right|dr\leq C(e^{t}-1)e^{-\frac{1}{2}t}(1+t)\quad\mbox{for all}\quad t\in[0,\infty)\,.$ Proof. For the integral we obtain $\displaystyle\int_{0}^{1-e^{-t}}\left|K_{0}(r,t)\right|dr$ $\displaystyle\leq$ $\displaystyle\int_{0}^{z-1}\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{[(z+1)^{2}-y^{2}]}}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 19.91684pt+\big{(}z^{2}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\Bigg{|}dy$ for all $z:=e^{t}>1$. We divide the domain of integration into two zones, $\displaystyle Z_{1}(\varepsilon,z)$ $\displaystyle:=$ $\displaystyle\left\\{(z,r)\,\Big{|}\,\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\leq\varepsilon,\,\,0\leq r\leq z-1\right\\},$ (7.5) $\displaystyle Z_{2}(\varepsilon,z)$ $\displaystyle:=$ $\displaystyle\left\\{(z,r)\,\Big{|}\,\varepsilon\leq\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}},\,\,0\leq r\leq z-1\right\\},$ (7.6) and split the integral into conformable two parts, $\displaystyle\int_{0}^{e^{t}-1}\left|K_{0}(r,t)\right|dr$ $\displaystyle=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}\left|K_{0}(r,t)\right|dr+\int_{(z,r)\in Z_{2}(\varepsilon,z)}\left|K_{0}(r,t)\right|dr\,.$ In the first zone we have $\displaystyle F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\\!\\!$ $\displaystyle\\!\\!=\\!\\!$ $\displaystyle\\!\\!1+\left(\frac{1}{2}+iM\right)^{2}\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}+O\left(\left(\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\right)^{2}\right),$ (7.7) $\displaystyle F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\\!\\!$ $\displaystyle\\!\\!=\\!\\!$ $\displaystyle\\!\\!1-\left(\frac{1}{4}+M^{2}\right)\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}+O\left(\left(\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\right)^{2}\right).$ (7.8) We use the last formulas to estimate the term containing the hypergeometric functions: $\displaystyle\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-r^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}$ (7.9) $\displaystyle\hskip 28.45274pt+\big{(}z^{2}-1+r^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}\Bigg{|}$ $\displaystyle\leq$ $\displaystyle\frac{1}{2}\big{[}(z-1)^{2}-r^{2}\big{]}$ $\displaystyle+\left|\big{(}z-z^{2}-iM(1-z^{2}-r^{2})\big{)}\left(\frac{1}{2}+iM\right)^{2}-\big{(}z^{2}-1+r^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}\left(\frac{1}{4}+M^{2}\right)\right|\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}$ $\displaystyle+\Big{(}\left|z-z^{2}-iM(1-z^{2}-r^{2})\right|+\left|z^{2}-1+y^{2}\right|\Big{)}O\left(\left(\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)^{2}\right)$ $\displaystyle=$ $\displaystyle\frac{1}{2}\big{[}(z-1)^{2}-r^{2}\big{]}$ $\displaystyle+\frac{1}{8}\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\left|(1-2iM)(-1+4M^{2})(y^{2}+z^{2}-1)+2(1+2iM)^{2}(-z^{2}+z+iM(y^{2}+z^{2}-1))\right|$ $\displaystyle+\Big{(}\left|z-z^{2}-iM(1-z^{2}-r^{2})\right|+\left|z^{2}-1+y^{2}\right|\Big{)}O\left(\left(\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\right)^{2}\right).$ Hence, we have to consider the following three integrals, which can be easily evaluated and estimated, $\displaystyle A_{1}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}dr\leq\mbox{\rm Arctan}\left(\frac{z-1}{2\sqrt{z}}\right)\leq\frac{\pi}{2}\,,$ $\displaystyle A_{2}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}\frac{z^{2}}{((z+1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}dr\leq(z+1)^{-1/2}(z-1),$ and $A_{3}:=\int_{(z,r)\in Z_{1}(\varepsilon,z)}\frac{\left|z-z^{2}-iM(1-z^{2}-r^{2})\right|+\left|z^{2}-1+r^{2}\right|}{\sqrt{(z+1)^{2}-r^{2}}}\frac{(z-1)^{2}-r^{2}}{((z+1)^{2}-r^{2})^{2}}dr\leq C_{M}(z+1)^{-1/2}(z-1)$ for all $z\in[1,\infty)$. Finally, for the integral over the first zone we have obtained $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}dr\frac{1}{[(z-1)^{2}-r^{2}]\sqrt{[(z+1)^{2}-r^{2}]}}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-r^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}z^{2}-1+r^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}\Bigg{|}\,\leq\,C_{M}(z+1)^{-1/2}(z-1)$ for all $z\in[1,\infty)$. In the second zone we have $\varepsilon\leq\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\leq 1\quad\mbox{\rm and}\quad\frac{1}{(z-1)^{2}-r^{2}}\leq\frac{1}{\varepsilon[(z+1)^{2}-r^{2}]}\,.$ (7.10) According to the formula 15.3.10 of [2, Ch.15] the hypergeometric functions obey the estimate $\left|F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;x\Big{)}\right|\leq C\,\,\mbox{\rm and}\,\,\left|F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;x\Big{)}\right|\leq C\big{(}1-\ln(1-x))\,\,\,\forall x\in[\varepsilon,1).$ (7.11) This allows to estimate the integral over the second zone: $\displaystyle\int_{(z,r)\in Z_{2}(\varepsilon,z)}dr\frac{1}{[(z-1)^{2}-r^{2}]\sqrt{(z+1)^{2}-r^{2}}}$ (7.12) $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-r^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}$ $\displaystyle\hskip 14.22636pt+\big{(}z^{2}-1+r^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}\Bigg{|}\leq C_{M}(z+1)^{-1/2}(z-1)$ for all $z\in[1,\infty)$. Indeed, for the argument of the hypergeometric functions we have $\varepsilon\leq\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}=1-\frac{4z}{(z+1)^{2}-r^{2}}<1,\quad\frac{4z}{(z+1)^{2}-r^{2}}<1-\varepsilon\quad\mbox{\rm for all}\quad(z,r)\in Z_{2}(\varepsilon,z)\,.$ (7.13) Hence, $\left|F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\Big{)}\right|\leq C\left(1+\ln z\right),\,(z,r)\in Z_{2}(\varepsilon,z).$ (7.14) To prove (7.12) we estimate the following integral $\displaystyle\int_{(z,r)\in Z_{2}(\varepsilon,z)}\frac{z^{2}}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}dr$ $\displaystyle\leq$ $\displaystyle C_{\varepsilon}z^{2}\int_{0}^{z-1}\frac{1}{((z+1)^{2}-r^{2})^{3/2}}dr\,\leq\,C_{\varepsilon}\frac{(z-1)}{\sqrt{z}}\,.$ Thus, the lemma is proven. $\square$ ## 8 $L^{p}-L^{q}$ Estimates for Equation with $n=1$ and without Source Term. Some Estimates of Kernels $K_{0}$ and $K_{1}$ ###### Theorem 8.1 Let $u=u(x,t)$ be a solution of the Cauchy problem $u_{tt}-e^{-2t}u_{xx}+M^{2}u=0\,,\qquad u(x,0)=\varphi_{0}(x)\,,\qquad u_{t}(x,0)=\varphi_{1}(x)\,,$ with $\varphi_{0}$, $\varphi_{1}\in C_{0}^{\infty}({\mathbb{R}})$. If $\rho\in(1,2)$, then $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{L^{q}({\mathbb{R}}_{x})}+C_{\rho}(1+t)(e^{t}-1)^{\frac{1}{\rho}}e^{t[\frac{1}{2}-\frac{1}{\rho}]}\|\varphi_{0}(x)\|_{L^{p}({\mathbb{R}}_{x})}$ $\displaystyle+\,C_{\rho}(1+t)(e^{t}-1)^{\frac{1}{\rho}}e^{-\frac{t}{\rho}}\|\varphi_{1}(x)\|_{L^{p}({\mathbb{R}}_{x})}\,,$ for all $t\in(0,\infty)$. Here $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$. If $\rho=1$, then $\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}\leq C(1+t)\Big{(}e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{L^{q}({\mathbb{R}}_{x})}+(e^{t}-1)e^{-t}\|\varphi_{1}(x)\|_{L^{q}({\mathbb{R}}_{x})}\Big{)}\,,$ (8.1) for all $t\in(0,\infty)$. Proof. For $\rho=1$ we just apply Proposition 7.3. To prove this theorem for $\rho>1$ we need some auxiliary estimates for the kernels $K_{0}$ and $K_{1}$. We start with the case of $\varphi_{0}=0$, where the kernel $K_{1}$ appears. The application of Theorem 0.4 and Young’s inequality lead to $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle 2\Bigg{(}\int_{0}^{1-e^{-t}}|K_{1}(x,t)|^{\rho}dx\Bigg{)}^{1/\rho}\|\varphi_{1}(x)\|_{L^{p}({\mathbb{R}}_{x})},$ where $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$. Now we have to estimate the last integral. ###### Proposition 8.2 We have $\displaystyle\left(\int_{0}^{1-e^{-t}}|K_{1}(x,t)|^{\rho}dx\right)^{1/\rho}$ $\displaystyle\leq$ $\displaystyle C(1+t)(1-e^{-t})^{1/\rho}\quad\mbox{for all}\,\quad t\in(0,\infty)\,.$ Proof. One can write $\left(\int_{0}^{1-e^{-t}}|K_{1}(x,t)|^{\rho}dx\right)^{1/\rho}\leq C_{M}e^{t(1-1/\rho)}\Bigg{(}\int_{0}^{e^{t}-1}\big{(}(e^{t}+1)^{2}-y^{2}\big{)}^{-\frac{\rho}{2}}\Big{|}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\right)\Big{|}^{\rho}dy\Bigg{)}^{1/\rho}.$ Denote $z:=e^{t}>1$ and consider the integral $\displaystyle\int_{0}^{z-1}\left|\frac{1}{\sqrt{(1+z)^{2}-x^{2}}}F\Big{(}\frac{1}{2},\frac{1}{2};1;\frac{(z-1)^{2}-x^{2}}{(z+1)^{2}-x^{2}}\Big{)}\right|^{\rho}dx$ of the right-hand side. Then we apply Lemma 7.2 and obtain $\left(\int_{0}^{1-e^{-t}}|K_{1}(x,t)|^{\rho}dx\right)^{1/\rho}\leq Ce^{t(1-1/\rho)}(1+\ln e^{t})(e^{t}-1)^{1/\rho}(e^{t}+1)^{-1}\leq C(1+t)(1-e^{-t})^{1/\rho}\,.$ Proposition is proven. $\square$ Thus, the theorem in the case of $\varphi_{0}=0$ is proven. Now we turn to the case of $\varphi_{1}=0$, where the kernel $K_{0}$ appears. The application of Theorem 0.4 leads to $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{L^{q}({\mathbb{R}}_{x})}+\,\left\|\int_{0}^{1-e^{-t}}\left[\varphi_{0}(x-z)+\varphi_{0}(x+z)\right]K_{0}(z,t)\,dz\right\|_{L^{q}({\mathbb{R}}_{x})}\,.$ Similarly to the case of the second datum we arrive at $\displaystyle\|u(x,t)\|_{L^{q}({\mathbb{R}}_{x})}$ $\displaystyle\leq$ $\displaystyle e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{L^{q}({\mathbb{R}}_{x})}+\,\|\varphi_{0}(x)\|_{L^{p}({\mathbb{R}}_{x})}\left(\int_{0}^{1-e^{-t}}|K_{0}(r,t)|^{\rho}dr\right)^{1/\rho}\,.$ The next proposition gives an estimate for the integral of the last inequality. ###### Proposition 8.3 Let $1<p<\rho^{\prime}$, $\frac{1}{q}=\frac{1}{p}-\frac{1}{\rho^{\prime}}$, $\frac{1}{\rho}+\frac{1}{\rho^{\prime}}=1$, and $\rho\in[1,2)$. We have $\displaystyle\left(\int_{0}^{1-e^{-t}}|K_{0}(r,t)|^{\rho}dr\right)^{1/\rho}$ $\displaystyle\leq$ $\displaystyle C_{\rho}(1+t)(e^{t}-1)^{\frac{1}{\rho}}e^{t(\frac{1}{2}-\frac{1}{\rho})}\quad\mbox{ for all}\quad t\in(0,\infty)\,.$ Proof. We turn to the integral ($z:=e^{t}>1$) $\displaystyle\left(\int_{0}^{1-e^{-t}}|K_{0}(r,t)|^{\rho}dr\right)^{1/\rho}\\!\\!$ $\displaystyle\\!\\!=\\!\\!$ $\displaystyle\\!\\!\Bigg{(}\int_{0}^{z-1}dy\left(\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{(z+1)^{2}-y^{2}}}\right)^{\rho}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}$ $\displaystyle+\big{(}z^{2}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\Bigg{|}^{\rho}\Bigg{)}^{1/\rho}.$ The formulas (7.7) and (7.8) describe the behavior of the hypergeometric functions in the neighbourhood of zero. Consider therefore two zones, $Z_{1}(\varepsilon,z)$ and $Z_{2}(\varepsilon,z)$, defined in (7.5) and (7.6), respectively. We split integral into two parts: $\displaystyle\int_{0}^{1-e^{-t}}\left|K_{0}(r,t)\right|^{\rho}dr$ $\displaystyle=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}\left|K_{0}(r,t)\right|^{\rho}dr+\int_{(z,r)\in Z_{2}(\varepsilon,z)}\left|K_{0}(r,t)\right|^{\rho}dr\,.$ In the proof of Lemma 7.4 the relation (7.9) was checked in the first zone. If $1\leq z\leq N$ with some constant $N$, then the argument of the hypergeometric functions is bounded, $\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\leq\frac{(z-1)^{2}}{(z+1)^{2}}\leq\frac{(N-1)^{2}}{(N+1)^{2}}<1\quad\mbox{for all}\quad r\in(0,z-1),$ (8.2) and we obtain with $z=e^{t}$, $\displaystyle\Bigg{(}\int_{0}^{1-e^{-t}}\left|K_{0}(r,t)\right|^{\rho}dr\Bigg{)}^{1/\rho}$ $\displaystyle\leq$ $\displaystyle C_{M,N}\Bigg{(}\int_{0}^{z-1}\Bigg{[}\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{(z+1)^{2}-y^{2}}}$ $\displaystyle\hskip 2.84544pt\times\Bigg{\\{}\frac{1}{2}[(z-1)^{2}-y^{2}]+z^{2}\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}+z^{2}\left(\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\right)^{2}\Bigg{\\}}\Bigg{]}^{\rho}dy\Bigg{)}^{1/\rho}$ $\displaystyle\leq$ $\displaystyle C_{M,N}\Bigg{(}\int_{0}^{z-1}\Bigg{[}\frac{1}{\sqrt{(z+1)^{2}-y^{2}}}\Bigg{\\{}1+z^{2}\frac{1}{(z+1)^{2}-y^{2}}\Bigg{\\}}\Bigg{]}^{\rho}dy\Bigg{)}^{1/\rho}$ $\displaystyle\leq$ $\displaystyle C_{M,N}(z-1)^{1/\rho}(z+1)^{-1}\,.$ Thus, we can restrict ourselves to the case of large $z\geq N$ in both zones. Consider therefore for $\rho\in(1,2)$ the following integrals over the first zone $\displaystyle A_{4}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon)}\left(\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\right)^{\rho}dr\,\,\leq\,\,\int_{0}^{z-1}\left(\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\right)^{\rho}dr$ $\displaystyle\leq$ $\displaystyle C(z-1)(z+1)^{-\rho}F\Big{(}\frac{1}{2},\frac{\rho}{2};\frac{3}{2};\frac{(z-1)^{2}}{(z+1)^{2}}\Big{)}$ $\displaystyle\leq$ $\displaystyle C(z-1)(z+1)^{-\rho}\,,$ $\displaystyle A_{5}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon)}\left(\frac{z^{2}}{[(z+1)^{2}-r^{2}]\sqrt{(z+1)^{2}-r^{2}}}\right)^{\rho}dr\leq Cz^{2\rho}(z-1)(z+1)^{-3\rho}F\Big{(}\frac{1}{2},\frac{3\rho}{2};\frac{3}{2};\frac{(z-1)^{2}}{(z+1)^{2}}\Big{)}.$ Then, according to (15.3.6) of Ch.15[1] and [2], $\displaystyle F\left(a,b;c;\zeta\right)$ $\displaystyle=$ $\displaystyle\frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}F\left(a,b;a+b-c+1;1-\zeta\right)$ $\displaystyle+(1-z)^{c-a-b}\frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)}F\left(c-a,c-b;c-a-b+1;1-\zeta\right)$ for all $\zeta\in{\mathbb{C}}$, $|\arg(1-\zeta)|<\pi$. We use (8) with $\zeta=\frac{(z-1)^{2}}{(z+1)^{2}},\qquad 1-\zeta=\frac{4z}{(z+1)^{2}}$ to obtain for $\rho<2$ and large $z\geq N$ the following estimate for the hypergeometric function, $F\Big{(}\frac{1}{2},\frac{3\rho}{2};\frac{3}{2};\frac{(z-1)^{2}}{(z+1)^{2}}\Big{)}\leq C(z+1)^{-1+\frac{3\rho}{2}}\,.$ (8.4) Thus, $\displaystyle A_{5}$ $\displaystyle\leq$ $\displaystyle C(z-1)(z+1)^{-1+\frac{\rho}{2}}\,.$ For the next term we obtain a similar estimate, $\displaystyle A_{6}:=\int_{(z,r)\in Z_{1}(\varepsilon)}\left|\frac{z^{2}}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}\left(\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)^{2}\right|^{\rho}dr$ $\displaystyle\leq$ $\displaystyle C(z-1)(z+1)^{-1+\frac{\rho}{2}}\,.$ Hence, $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}\left|K_{0}(r,t)\right|^{\rho}dr\leq C(z-1)(z+1)^{-1+\frac{\rho}{2}}\,.$ In the second zone $Z_{2}(\varepsilon,z)$ for the argument of the hypergeometric functions we have (7.10), (7.13), and (7.14). We have to estimate the following integral $\displaystyle A_{7}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{2}(\varepsilon,z)}\left|\frac{z^{2}\left(1+\ln z\right)}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}\right|^{\rho}dr\,.$ We apply (7.10) and (8.4) to obtain $\displaystyle A_{7}$ $\displaystyle\leq$ $\displaystyle C\left(1+\ln z\right)^{\rho}(z-1)(z+1)^{-1+\frac{\rho}{2}}\,.$ Hence, $\displaystyle\int_{(z,r)\in Z_{2}(\varepsilon,z)}\left|K_{0}(r,t)\right|^{\rho}dr\leq C\left(1+\ln z\right)^{\rho}(z-1)(z+1)^{-1+\frac{\rho}{2}}\qquad\mbox{\rm for all}\quad z\geq N\,.$ Proposition is proven. $\square$ ## 9 $L^{p}-L^{q}$ Estimates for the Equation with Source, $n\geq 2$ For the wave equation the Duhamel’s principle allows to reduce the case of source term to the case of the Cauchy problem without source term and consequently to derive the $L^{p}-L^{q}$-decay estimates for the equation. For (0.9) the Duhamel’s principle is not applicable straightforward and we have to appeal to the representation formula of Theorem 0.5. In fact, one can regard that formula as an expansion of the two-stage Duhamel’s principle. In this section we consider the Cauchy problem (0.21) for the equation with the source term with zero initial data. ###### Theorem 9.1 Let $u=u(x,t)$ be solution of the Cauchy problem (0.21). Then for $n>1$ one has the following decay estimate $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}db\,\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}\int_{0}^{e^{-b}-e^{-t}}dr\,r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(e^{-t}+e^{-b})^{2}-r^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{-b}-e^{-t})^{2}-r^{2}}{(e^{-b}+e^{-t})^{2}-r^{2}}\right)$ provided that $s\geq 0$, $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $\frac{1}{2}(n+1)\left(\frac{1}{p}-\frac{1}{q}\right)\leq 2s\leq n\left(\frac{1}{p}-\frac{1}{q}\right)<2s+1$. Proof. In both cases, of even and odd $n$, one can write the representation (0.24). Due to the results of [5, 23] for the wave equation, we have $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}db\int_{0}^{e^{-b}-e^{-t}}\|(-\bigtriangleup)^{-s}v(x,r;b)\|_{{L}^{q}({\mathbb{R}}^{n})}\frac{1}{\sqrt{(e^{-t}+e^{-b})^{2}-r^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{-b}-e^{-t})^{2}-r^{2}}{(e^{-b}+e^{-t})^{2}-r^{2}}\right)dr$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}db\,\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}\int_{0}^{e^{-b}-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(e^{-t}+e^{-b})^{2}-r^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{-b}-e^{-t})^{2}-r^{2}}{(e^{-b}+e^{-t})^{2}-r^{2}}\right)dr\,.$ The theorem is proven. $\square$ We are going to transform the estimate of the last theorem to more cosy form. To this aim we estimate for $n(\frac{1}{p}-\frac{1}{q})<2s+1$ the last integral of the right hand side. If we replace $e^{-b}/e^{-t}>1$ with $z:=e^{-b}/e^{-t}>1$, then the integral will be simplified. $\displaystyle\int_{0}^{e^{-b}-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(e^{-t}+e^{-b})^{2}-r^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{-b}-e^{-t})^{2}-r^{2}}{(e^{-b}+e^{-t})^{2}-r^{2}}\right)dr$ $\displaystyle=$ $\displaystyle e^{-t[2s-n(\frac{1}{p}-\frac{1}{q})]}\int_{0}^{z-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-y^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\right)dy$ ###### Lemma 9.2 [33, Lemma 9.2] Assume that $0\geq 2s-n(\frac{1}{p}-\frac{1}{q})>-1$. Then $\int_{0}^{z-1}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)\,dr\leq Cz^{-1}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}(1+\ln z),$ for all $z>1$. ###### Corollary 9.3 Let $u=u(x,t)$ be solution of the Cauchy problem (0.21). Then for $n\geq 2$ one has the following decay estimate $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}e^{-b}\left(e^{-b}-e^{-t}\right)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\left(1+t-b\right)\,db$ (9.1) provided that $s\geq 0$, $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $\frac{1}{2}(n+1)\left(\frac{1}{p}-\frac{1}{q}\right)\leq 2s\leq n\left(\frac{1}{p}-\frac{1}{q}\right)<2s+1$. Proof. Indeed, we apply Lemma 9.2 with $z=e^{t-b}$ to the right-hand side of the estimate given by Theorem 9.1 : $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}db\,\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}e^{-t[2s-n(\frac{1}{p}-\frac{1}{q})]}z^{-1}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}(1+\ln z)$ $\displaystyle\leq$ $\displaystyle C\int_{0}^{t}\|f(x,b)\|_{{L}^{p}({\mathbb{R}}^{n})}e^{-b}\left(e^{-b}-e^{-t}\right)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\left(1+t-b\right)\,db\,.$ Corollary is proven. $\square$ ## 10 $L^{p}-L^{q}$ Estimates for Equation without Source, $n\geq 2$ The $L^{p}-L^{q}$-decay estimates for the energy of the solution of the Cauchy problem for the wave equation without source can be proved by the representation formula, $L_{1}-L_{\infty}$ and $L_{2}-L_{2}$ estimates, and interpolation argument. (See, e.g., [25, Theorem 2.1].) There is also a proof of the $L^{p}-L^{q}$-decay estimates that is based on the microlocal consideration and dyadic decomposition of the phase space. (See, e.g., [5, 23].) To avoid the derivative loss and obtain more sharp estimates we appeal to the representation formula provided by Theorem 0.6. ###### Theorem 10.1 The solution $u=u(x,t)$ of the Cauchy problem (0.25) satisfies the following $L^{p}-L^{q}$ estimate $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C(1+t)(1-e^{-t})^{2s-n(\frac{1}{p}-\frac{1}{q})}\Big{\\{}e^{\frac{t}{2}}\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}+(1-e^{-t})\|\varphi_{1}\|_{{L}^{p}({\mathbb{R}}^{n})}\Big{\\}}$ for all $t\in(0,\infty)$, provided that $s\geq 0$, $1<p\leq 2$, $\frac{1}{p}+\frac{1}{q}=1$, $\frac{1}{2}(n+1)\left(\frac{1}{p}-\frac{1}{q}\right)\leq 2s\leq n\left(\frac{1}{p}-\frac{1}{q}\right)<2s+1$. Proof. We start with the case of $\varphi_{0}=0$. Due to Theorem 0.6 for the solution $u=u(x,t)$ of the Cauchy problem (0.25) with $\varphi_{0}=0$ and to the results of [5, 23] we have: $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\|\varphi_{1}\|_{{L}^{p}({\mathbb{R}}^{n})}\int_{0}^{1-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\left|K_{1}(r,t)\right|\,dr$ $\displaystyle\leq$ $\displaystyle C\|\varphi_{1}\|_{{L}^{p}({\mathbb{R}}^{n})}e^{-t[2s-n(\frac{1}{p}-\frac{1}{q})]}\int_{0}^{e^{t}-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\big{(}(e^{t}+1)^{2}-y^{2}\big{)}^{-\frac{1}{2}}F\left(\frac{1}{2},\frac{1}{2};1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\right)\,dy\,.$ To continue we apply Lemma 9.2 and obtain $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\|\varphi_{1}\|_{{L}^{p}({\mathbb{R}}^{n})}(1+t)(1-e^{-t})^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\,.$ Thus, in the case of $\varphi_{0}=0$ the theorem is proven. Next we turn to the case of $\varphi_{1}=0$. Due to Theorem 0.6 for the solution $u=u(x,t)$ of the Cauchy problem (0.25) with $\varphi_{1}=0$ and to the results of [5, 23] we have: $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\Bigg{(}e^{\frac{t}{2}}(1-e^{-t})^{2s-n(\frac{1}{p}-\frac{1}{q})}+\int_{0}^{1-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}|K_{0}(r,t)|\,dr\Bigg{)}\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}.$ One can estimate the last integral $\displaystyle\int_{0}^{1-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}|K_{0}(r,t)|\,dr$ $\displaystyle\leq$ $\displaystyle e^{-t[2s-n(\frac{1}{p}-\frac{1}{q})]}\int_{0}^{e^{t}-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(e^{t}-1)^{2}-y^{2}]\sqrt{(e^{t}+1)^{2}-y^{2}}}$ $\displaystyle\times\Bigg{|}\big{(}e^{t}-e^{2t}-iM(1-e^{2t}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}e^{2t}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\Big{)}\Bigg{|}\,dy\,.$ The following proposition gives the remaining estimate for that integral and completes the proof of the theorem. ###### Proposition 10.2 If $2s-n(\frac{1}{p}-\frac{1}{q})>-1$, then $\displaystyle\int_{0}^{z-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{(z+1)^{2}-y^{2}}}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}z^{2}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\Bigg{|}\,dy$ $\displaystyle\leq$ $\displaystyle Cz^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\left(1+\ln z\right)\quad\mbox{ for all}\quad z>1.$ Proof. We follow the arguments have been used in the proof of Proposition 8.3. If $1\leq z\leq N$ with some constant $N$, then the argument of the hypergeometric functions is bounded (8.2), and the integral can be estimated by: $\displaystyle\int_{0}^{z-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{(z+1)^{2}-y^{2}}}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}z^{2}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\Bigg{|}\,dy$ $\displaystyle\leq$ $\displaystyle C_{M}\int_{0}^{z-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\Bigg{[}\frac{1}{\sqrt{(z+1)^{2}-y^{2}}}\Bigg{\\{}1+z^{2}\frac{1}{(z+1)^{2}-y^{2}}\Bigg{\\}}\Bigg{]}\,dy$ $\displaystyle\leq$ $\displaystyle C_{M}z^{-1}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\quad\mbox{\rm for all}\quad z\in[1,N]\,.$ Thus, we can restrict ourselves to the case of large $z\geq M$ in both zones $Z_{1}(\varepsilon,z)$ and $Z_{2}(\varepsilon,z)$, defined in (7.5) and (7.6), respectively. In the first zone we have (7.9). Consider therefore the following inequalities, $\displaystyle A_{8}$ $\displaystyle:=$ $\displaystyle\int_{(z,r)\in Z_{1}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\,dr$ $\displaystyle\leq$ $\displaystyle Cz^{-1}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\quad\mbox{\rm for all}\quad z\in[N,\infty)\,.$ For $0\geq a>-1$ and $z\geq N$ the following integral can be easily estimated: $\displaystyle\int_{0}^{z-1}r^{a}\frac{1}{((z+1)^{2}-r^{2})^{3/2}}dr$ $\displaystyle=$ $\displaystyle\int_{0}^{z/2}r^{a}\frac{1}{((z+1)^{2}-r^{2})^{3/2}}dr+\int_{z/2}^{z-1}r^{a}\frac{1}{((z+1)^{2}-r^{2})^{3/2}}dr$ $\displaystyle\leq$ $\displaystyle\frac{16}{9}z^{-3}\int_{0}^{z/2}r^{a}dr+\frac{z^{a}}{4^{a}}\int_{z/2}^{z-1}\frac{1}{((z+1)^{2}-r^{2})^{3/2}}dr$ $\displaystyle\leq$ $\displaystyle Cz^{a-3/2}\quad\mbox{\rm for all}\quad z\in[N,\infty)\,.$ Hence, $\displaystyle A_{9}$ $\displaystyle:=$ $\displaystyle z^{2}\int_{(z,r)\in Z_{1}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\frac{1}{(z+1)^{2}-r^{2}}dr$ $\displaystyle\leq$ $\displaystyle z^{2}\int_{0}^{z-1}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\frac{1}{(z+1)^{2}-r^{2}}dr$ $\displaystyle\leq$ $\displaystyle Cz^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\quad\mbox{\rm for all}\quad z\in[N,\infty)\,,$ and $\displaystyle A_{10}$ $\displaystyle:=$ $\displaystyle z^{2}\int_{(z,r)\in Z_{1}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}\left(\frac{(z-1)^{2}-r^{2}}{(z+1)^{2}-r^{2}}\right)^{2}dr$ $\displaystyle\leq$ $\displaystyle z^{2}\int_{(z,r)\in Z_{1}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\frac{1}{(z+1)^{2}-r^{2}}dr$ $\displaystyle\leq$ $\displaystyle Cz^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\quad\mbox{\rm for all}\quad z\in[N,\infty)\,.$ Finally, $\displaystyle\int_{(z,y)\in Z_{1}(\varepsilon,z)}\,y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(z-1)^{2}-y^{2}]\sqrt{(z+1)^{2}-y^{2}}}$ $\displaystyle\times\Bigg{|}\big{(}z-z^{2}-iM(1-z^{2}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}z^{2}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(z-1)^{2}-y^{2}}{(z+1)^{2}-y^{2}}\Big{)}\Bigg{|}\,dy$ $\displaystyle\leq$ $\displaystyle Cz^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\quad\mbox{\rm for all}\quad z\in[1,\infty)\,.$ In the second zone we use (7.10), (7.11), and (7.14). Thus, we have to estimate the next two integrals: $\displaystyle A_{11}$ $\displaystyle:=$ $\displaystyle z^{2}\int_{(z,r)\in Z_{2}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}\,dr\,,$ $\displaystyle A_{12}$ $\displaystyle:=$ $\displaystyle z^{2}\left(1+\ln z\right)\int_{(z,r)\in Z_{2}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{((z-1)^{2}-r^{2})\sqrt{(z+1)^{2}-r^{2}}}\,dr\,.$ We apply (7.10) to $A_{11}$ and obtain $\displaystyle A_{11}\leq C_{\varepsilon}z^{2}\int_{(z,r)\in Z_{2}(\varepsilon,z)}r^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(z+1)^{2}-r^{2}]}\frac{1}{\sqrt{(z+1)^{2}-r^{2}}}\,dr\leq C_{\varepsilon}z^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}$ for all $z\in[1,\infty)$, while $\displaystyle A_{12}$ $\displaystyle\leq$ $\displaystyle C_{\varepsilon}z^{-\frac{1}{2}}(z-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}\left(1+\ln z\right)\quad\mbox{\rm for all}\quad z\in[1,\infty)\,.$ Proposition is proven. $\square$ To complete the proof of the theorem we write $\displaystyle\int_{0}^{1-e^{-t}}r^{2s-n(\frac{1}{p}-\frac{1}{q})}|K_{0}(r,t)|\,dr$ $\displaystyle\leq$ $\displaystyle e^{-t[2s-n(\frac{1}{p}-\frac{1}{q})]}\int_{0}^{e^{t}-1}y^{2s-n(\frac{1}{p}-\frac{1}{q})}\frac{1}{[(e^{t}-1)^{2}-y^{2}]\sqrt{(e^{t}+1)^{2}-y^{2}}}$ $\displaystyle\times\Bigg{|}\big{(}e^{t}-e^{2t}-iM(1-e^{2t}-y^{2})\big{)}F\Big{(}\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\Big{)}$ $\displaystyle\hskip 28.45274pt+\big{(}e^{2t}-1+y^{2}\big{)}\Big{(}\frac{1}{2}-iM\Big{)}F\Big{(}-\frac{1}{2}+iM,\frac{1}{2}+iM;1;\frac{(e^{t}-1)^{2}-y^{2}}{(e^{t}+1)^{2}-y^{2}}\Big{)}\Bigg{|}\,dy$ $\displaystyle\leq$ $\displaystyle Ce^{-t[\frac{1}{2}+2s-n(\frac{1}{p}-\frac{1}{q})]}(e^{t}-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}(1+t)\,.$ Thus, $\displaystyle\|(-\bigtriangleup)^{-s}u(x,t)\|_{{L}^{q}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C\Big{(}e^{\frac{t}{2}}(1-e^{-t})^{2s-n(\frac{1}{p}-\frac{1}{q})}+e^{-t[\frac{1}{2}+2s-n(\frac{1}{p}-\frac{1}{q})]}(e^{t}-1)^{1+2s-n(\frac{1}{p}-\frac{1}{q})}(1+t)\Big{)}\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}$ $\displaystyle\leq$ $\displaystyle C(1+t)e^{\frac{t}{2}}(1-e^{-t})^{2s-n(\frac{1}{p}-\frac{1}{q})}\|\varphi_{0}(x)\|_{{L}^{p}({\mathbb{R}}^{n})}.$ Theorem is proven. $\square$ ## References * [1] Abramowitz, M., Stegun, I.A.: Handbook of mathematical functions with formulas, graphs, and mathematical tables. National Bureau of Standards Applied Mathematics Series, 55, Washington, DC, 1964 * [2] Bateman, H., Erdelyi, A.: Higher Transcendental Functions. v.1,2, McGraw-Hill, New York, 1953. * [3] Birrell, N. D., Davies, P.C.W. : Quantum fields in curved space, Cambridge ; New York : Cambridge University Press, 1984. * [4] Bony, J.-F., Hafner, D.: Decay and non-decay of the local energy for the wave equation in the De Sitter - Schwarzschild metric, arXiv:0706.0350v1 * [5] Brenner, P.: On $\,L^{p}-L^{q}\,$ estimates for the wave-equation. Math. Zeitschrift 145 (1975) 251-254. * [6] Brevik, I., Simonsen, B.: The scalar field equation in Schwarzschild-de Sitter space. Gen. Relativity Gravitation 33 (2001), no. 10, 1839–1861. * [7] Brozos-Vázquez, M., García-Río, E., Vázquez-Lorenzo, R.: Locally conformally flat multidimensional cosmological models and generalized Friedmann-Robertson-Walker spacetimes, J. Cosmol. Astropart. Phys. JCAP12(2004)008 doi:10.1088/1475-7516/2004/12/008. * [8] Dafermos, M., Rodnianski, I.: The wave equation on Schwarzschild-de Sitter spacetimes,arXiv:0709.2766. * [9] Chandrasekhar, S.: The Mathematical Theory of Black Holes, Oxford,Clarendon Press ; New York : Oxford University Press, 1998. * [10] Chrusciel, P.T., Pollack,D.: Singular Yamabe metrics and initial data with exactly Kottler-Schwarzschild-de Sitter ends, arXiv:0710.3365v1 * [11] De Sitter, W.: On Einstein’s Theory of Gravitation, and its astronomical consequences.II,III. Royal Astronimcal Society. 77 (1917) 155-184; 78 (1917) 3-28. * [12] Einstein, A.: Kosmologische Betrachtungen zur allgemeinen Relativitätstheorie, Sitzungsber Preuss. Akad. Wiss. Berlin (1917) 142-152. * [13] Finster, F., Kamran, N., Smoller, J., Yau, S.-T.: Decay of solutions of the wave equation in the Kerr geometry. Comm. Math. Phys. 264 (2006), no. 2, 465–503. * [14] Friedrich, H., Rendall, A.: The Cauchy problem for the Einstein equations. Einstein’s field equations and their physical implications. Lecture Notes in Phys., 540, Springer, Berlin (2000) 127–223. * [15] Galstian, A.: $L_{p}$-$L_{q}$ decay estimates for the wave equations with exponentially growing speed of propagation. Appl. Anal. 82 (3) (2003) 197–214. * [16] Heinzle, J. M., Rendall, A., Power-law inflation in spacetimes without symmetry. Commun. Math. Phys. 269 (2007) 1-15. * [17] Hörmander, L.: The analysis of linear partial differential operators. IV. Fourier integral operators. Grundlehren der Mathematischen Wissenschaften, 275. Springer-Verlag, Berlin, 1994. * [18] Kronthaler, J.: The Cauchy problem for the wave equation in the Schwarzschild geometry. J. Math. Phys. 47 (2006), no. 4, 042501, 29 pp. * [19] Littman, W.: The wave operator and $L_{p}$ norms. J. Math. Mech. 12 (1963) 55–68. * [20] Littman, W., McCarthy, C., Rivière, N.: The non-existence of $L^{p}$ estimates for certain translation-invariant operators. Studia Math. 30 (1968) 219–229. * [21] López-Ortega, A.: Quasinormal modes of $D$-dimensional de Sitter spacetime. Gen. Relativity Gravitation 38 (2006), no. 11, 1565–1591. * [22] M$\o$ller, C.: The theory of relativity. Oxford, Clarendon Press, 1952. * [23] Pecher, H.: $L^{p}$-Abschätzungen und klassische Lösungen für nichtlineare Wellengleichungen.I, Math. Zeitschrift. 150 (1976) 159-183. * [24] Peral, J.C.: $L^{p}$ estimates for the wave equation. J. Funct. Anal. 36 (1) (1980) 114-145. * [25] Racke, R.: Lectures on Nonlinear Evolution Equations. Aspects of Mathematics, Vieweg, Braunschweig/Wiesbaden, 1992. * [26] Rendall, A.: Asymptotics of solutions of the Einstein equations with positive cosmological constant. Ann. Henri Poincaré 5 (6) (2004) 1041-1064. * [27] Shatah, J., Struwe, M.: Geometric wave equations. Courant Lecture Notes in Mathematics, 2. New York University, Courant Institute of Mathematical Sciences, New York; American Mathematical Society, Providence, RI, 1998. * [28] Sonego, S., Faraoni, V.: Huygens’ principle and characteristic propagation property for waves in curved space-times, J. Math. Phys. 33(2) (1992) 625-632. * [29] Yagdjian, K.: A note on the fundamental solution for the Tricomi-type equation in the hyperbolic domain, J. Differential Equations 206 (2004) 227-252. * [30] Yagdjian, K.: Global existence in the Cauchy problem for nonlinear wave equations with variable speed of propagation, New trends in the theory of hyperbolic equations, 301-385, Oper. Theory Adv. Appl., 159, Birkh$\ddot{\rm a}$user, Basel, 2005. * [31] Yagdjian, K.: Global existence for the $n$-dimensional semilinear Tricomi-type equations, Comm. Partial Diff. Equations 31 (2006) 907-944. * [32] Yagdjian, K.: Self-similar solutions of semilinear wave equation with variable speed of propagation. J. Math. Anal. Appl. 336 (2007) 1259-1286.. * [33] Yagdjian, K. Galstian, A.: Fundamental Solutions for Wave Equation in de Sitter Model of Universe, ISSN 1437-739X, University of Potsdam, August, Preprint 2007/06.

arxiv-papers

2024-09-04T02:48:54.452881

0803.3106

# A Note on Walk versus Wait: Lazy Mathematician Wins111J.G. Chen, S.D. Kominers, and R.W. Sinnott. Walk versus wait: The lazy mathematician wins. arXiv.org Mathematics, January 2008. http://arVix.org/abs/0801.0297 Ramnik Arora Indian Institute of Technology Kanpur It seems that the distance term in the given equation $4$ and the expression above it in the article [1] is not generalised enough and is partially incorrect: The expression and the equation are reproduced here: $\frac{d_{2}}{v_{w}}+\underbrace{\int^{t_{w}}_{0}(\frac{1}{t_{b}}(\frac{d-d_{2}}{v_{b}}))dt}_{1}+\underbrace{(1-\int^{t_{w}}_{0}p(t)dt)(\frac{d}{v_{w}}+t_{w})}_{2}$ and $\underbrace{\int^{t_{w}}_{0}(\frac{1}{t_{b}}(\frac{d-d_{2}}{v_{b}}+t))dt}_{1}+\underbrace{(1-\int^{t_{w}}_{0}p(t)dt)(\frac{d}{v_{w}}+t_{w})}_{2}=\frac{d-d_{2}}{v_{w}}\\\ $ ## Distance The distance ($d$) and the time used in the term 2 is not correct. Thus, on correcting the distance, the $2^{nd}$ term would be: $(1-\int^{t_{w}}_{0}p(t)dt)(\frac{d-d_{2}}{v_{w}}+t_{w})$ It is because if we use $d$ instead of $d-d_{2}$, then we are in effect double counting the distance $d_{2}$ when the mathematician is walking all the way to the destination. ## Time The probability density function of the bus reaching the second bus stop will not be only $p(t)$ as is suggested in the article [1]. It can be seen that the probability of the bus reaching the second bus stop after waiting time $t=t_{\circ}$ is actually given by $p(t_{\circ}-\frac{d_{2}}{v_{b}}+\frac{d_{2}}{v_{w}})$. This can be explained by seeing that we have walked for $\frac{d_{2}}{v_{w}}$ hours before coming to the second bus stop and that for a bus to reach the second bus stop at time $t_{\circ}^{\prime}$, it needs to be at the first bus stop at time $t_{\circ}^{\prime}-\frac{d_{2}}{v_{b}}$. This is keeping in mind that $p(t)$ was defined as the probability of the bus arriving at the first bus stop at time $t$. ## Generalisation Also, it seems that in equation 4 (the second equation here), the result can be generalised by using $p(t_{corrected})$ probability distribution instead of the very specific $\frac{1}{t_{d}}$. This makes the $1^{st}$ term: $\int^{t_{w}}_{0}p(t_{corrected})(\frac{d-d_{2}}{v_{b}}+t)dt$ where $t_{corrected}$ is defined as $t-\frac{d_{2}}{v_{b}}+\frac{d_{2}}{v_{w}}$, as is suggested above. ## Residual term Moreover, it seems, there will be another factor that has not been considered in the expressions/equations. Appearance of the bus at stop $1$ is a sort of a periodic event (stochastic process). Hence, in the uniform distribution case, there will be significant dependence of the waiting time on the moment the bus passes us by while we are en route to the second bus stop. This term would come out to be (under the assumption of uniform distribution): $\int^{\frac{d_{2}}{v_{w}}}_{0}\frac{1}{t_{b}}[\underbrace{(t_{b}-t)}_{1}-\underbrace{(\frac{d_{2}-v_{w}t}{v_{w}})}_{2}]dt$ This is the expected time that he has to wait additionally if the bus happens to pass him at while he was on his way. In case of uniform distribution we know that one bus is expected from $[0,t_{b}]$ and the next from $[t_{b},2t_{b}]$. The $1^{st}$ term is the ’dead’ time for which the next bus is not expected and the $2^{nd}$ term is the time that he spends walking to the bus stop, which is subtracted from the $1^{st}$ term. This is under the assumption that $\frac{d_{2}}{v_{w}}<t_{b}$ as if this is not the case, then the Mathematician would always choose to wait for the bus as it will necessarily pass him before he gets to his destination. This is just for the simple case of uniform distribution, and the result for the general distribution can also be worked out using similar arguments. Nevertheless, these changes will not make a difference to the validity of the result. ## Acknowledgement I have to thank Utkarsh Upadhyay for his helpful comments and backspaces. ## References * [1] J.G. Chen, S.D. Kominers, and R.W. Sinnott. Walk versus wait: The lazy mathematician wins. arXiv.org Mathematics, January 2008. http://arxiv.org/abs/0801.0297.

arxiv-papers

2024-09-04T02:48:54.461781

0803.3109

Voronoi Diagrams for Quantum States and Its Application to a Numerical Estimation of a Quantum Channel Capacity Kimikazu Kato ††thanks: This version is slightly chaged from the original thesis. In the original thesis, the title and the abstract are also written in Japanese. They are omitted to comply with arXiv’s policy, and so that this can be complied by a usual latex without Japanese capability. Hiroshi Imai Professor In quantum information theory, a geometric approach, known as “quantum information geometry,” has been considered as a powerful method. In this thesis, we give a computational geometric interpretation to the geometric structure of a quantum system. Especially we introduce the concept of the Voronoi diagram and the smallest enclosing ball problem to the space of quantum states. With those tools in computational geometry, we analyze the adjacency structure of a point set in the quantum state space. Additionally, as an application, we show an effective method to compute the capacity of a quantum channel. In the first part of this thesis, we show some coincidences of Voronoi diagrams in a quantum state space with respect to some distances. That helps us to reinterpret the structure of the space of quantum pure states as a subspace of the whole space. More properly, we investigate the Voronoi diagrams with respect to the divergence, Fubini-Study distance, Bures distance, geodesic distance and Euclidean distance. For one qubit (or two level) states, the whole space is expressed as the Bloch ball. In the Bloch ball, we analyze the coincidence of the Voronoi diagrams in two different settings: 1)the diagram in pure states when Voronoi sites are taken as pure states and 2)the diagram in mixed states when sites are taken as pure states. We show that in both cases, all the diagrams coincide. This clear result is because of the symmetry specific for one qubit states. For three or higher level systems, we investigate the diagrams in pure states. The natural embedding of the quantum state space into a Euclidean space is no longer symmetric as in one qubit case. Consequently the coincidence of the Euclidean Voronoi diagram and the divergence-Voronoi diagram does not hold in a higher level system. However the coincidence of the divergence-, Fubini-Study- and Bures-Voronoi diagrams still holds. In the second part, we propose a method to compute a capacity of a quantum communication channel and show the result of the actual computation. We show that our method is sufficiently effective not only for one qubit states but for three level states. It is a practical application of the theoretical result shown in the first part; the theorems in the first part guarantee the correctness of the algorithm used in the second part. The algorithm uses Welzl’s algorithm to compute a smallest enclosing ball. Although the original algorithm introduced by Welzl is only for the Euclidean space, we show that the same method is useful for non-Euclidean space. We also implement the algorithm and experiment it to prove it is practical. First of all, my Ph.D work is totally financially supported by Nihon Unisys, Ltd., which I belong to as an employee. I am very grateful to the company for providing me such an opportunity. I am also grateful to my adviser Hiroshi Imai for his advice and continuous support, and above all, for accepting me as a new member of his lab. I thank coauthors of the papers which consists most of this dissertation. In particular, the very start of my research project is the discussion with Mayumi Oto. I owe her very much. I am also grateful to Keiko Imai and Jiro Nishitoba, who are also coauthors of the papers. Discussions with my colleagues and related researchers are of course, essential to progress my work. Although I cannot list up all of them, an incomplete list includes François Le Gall, Jun Hasegawa, Masahito Hayashi, Tsuyosi Ito, Masaki Owari, Toshiyuki Shimono. I am grateful to them for their advice and fruitful discussions. I am also thankful to Kokichi Sugihara, who advised me about figural presentation of some Voronoi diagrams; and Hidetoshi Muta, who provided some figures of Voronoi diagrams. Last but not least, I would like to thank my family. I am specially thankful to my wife, Masumi Kato who supported my decision to go back to the university, and also grateful to my two daughters; Koko and Koto, who supported me by just being there.

arxiv-papers

2024-09-04T02:48:54.464523

0803.3119

# The property of the set of the real numbers generated by a Gelfond-Schneider operator and the countability of all real numbers Slavica Vlahovic Branislav Vlahovic vlahovic@nccu.edu Corresponding author. Gunduliceva 2, 44000 Sisak, Croatia North Carolina Central University, Durham, NC 27707, USA ###### Abstract Considered will be properties of the set of real numbers $\Re$ generated by an operator that has form of an exponential function of Gelfond-Schneider type with rational arguments. It will be shown that such created set has cardinal number equal to ${\aleph_{0}}^{\aleph_{0}}=c$. It will be also shown that the same set is countable. The implication of this contradiction to the countability of the set of real numbers will be discussed. ###### keywords: denumerability , real numbers , countability , cardinal numbers MSC: 11B05 and ## 1 Introduction In 1900 D. Hilbert announced a list of twenty-three outstanding unsolved problems. The seventh problem was settled in 1934 by A. O. Gelfond and an independent proof by Th. Schneider in 1935. They proved that if $\alpha$ and $\beta$ are algebraic numbers with $\alpha\neq 0,\alpha\neq 1$, and if $\beta$ is not a real rational number, then any value of ${\alpha}^{\beta}$ is transcendental [1, 2]. For instance transcendental number is $2^{\sqrt{2}}$. This can be written in the form of ${({2\over 1})}^{[({2\over 1})^{1\over 2}]}={({m_{i1}\over n_{i1}})}^{[({{m_{i2}\over n_{i2}})}^{({m_{i3}\over n_{i3}})}]}$ (1) where $m_{i},n_{i}\in N$. We can ask ourselves a following question which ”class” of transcendental numbers can be presented this way? Or can any transcendental number be expressed in the form (1). Answer is obvious; some transcendental numbers cannot be expressed this way, for instance number $e$ cannot be presented by $e={m_{1}\over n_{1}}^{[{m_{2}\over n_{2}}^{m_{3}\over n_{3}}]}$ (2) because after taking logarithm from both sides $1={m_{2}\over n_{2}}^{m_{3}\over n_{3}}ln{m_{1}\over n_{1}},$ (3) and this cannot be, because $ln{m_{1}\over n_{1}}$ is always transcendental [3-5] for $m_{1},n_{1}\in N$. However, one can take more freedom and try to express the number $e$ in the form ${[{m_{1}\over n_{1}}^{m_{2}\over n_{2}}]}^{[{m_{3}\over n_{3}}^{m_{4}\over n_{4}}]}$ (4) or even more freedom and try to present the number $e$ in the form ${a_{1}}^{a_{2}}$, where both ${a_{1}}$ and ${a_{2}}$ can have the form (4). Obviously, the argument such as shown in (3), that number $e$ cannot be presented in such way, cannot be applied anymore since both ${a_{1}}$ and ${a_{2}}$ can be now transcendental numbers. One can go even further (as it is done in [6]) and take much more freedom in generating the numbers or a set of numbers, which elements will be generated through a general element of the sequence that has the form: ${a_{1}}^{{a_{2}}^{{a_{3}}^{.^{.^{.^{{a_{n}}^{.^{.^{.}}}}}}}}}$ (5) where in (5) each element $a_{i}$ of bases and exponents has the following form: $a_{i}={[({{m_{i1}\over n_{i1}})}^{({m_{i2}\over n_{i2}})}]}^{[({{m_{i3}\over n_{i3}})}^{({m_{i4}\over n_{i4}})}]}$ (6) where $m_{ij},n_{ij}\in N,i=1,2,3,...n,j=1,2,3,4$. The question remains: which class of the transcendental numbers can be or can not be represented in this way? Can majority of the transcendental numbers be presented or can not be presented in this way? If some transcendental numbers can not be presented, is that set countable or not? First let us note that the set of numbers generated through the operator (5) looks similar to the set of the real numbers. Such set does not have the first and last element, it has subset of all rational numbers, and it is dense everywhere in rational, algebraic and transcendental numbers. However, it may not be equal to the set of the real numbers since it is harder to prove that it is dense in Dedekind’s sense, since this would require proof that it does not have holes, i.e. that all numbers can be represented in this way. To avoid that difficulty, let as assume that some numbers can not be presented in this way and let us focus here only on estimating the number of the elements in such set, i.e. on determining the cardinality of such set of numbers. ## 2 The cardinality of the generated set of numbers Let us generate the set of the real numbers through relation (5) where each base and exponent element $a_{i}$ has the form (6). The mechanism to generate the elements of the set is to write (5) for all possible combinations of arguments, with the sum of all bases and exponents equal to 2, 3, 4,… and so on. As the sum increases the number of the exponents will expand. The sample of such generated set with a procedure to avoid double counting of the same numbers is given in [6]. However, let us focus here on our main task which is to estimate the cardinality of such generated set when the process described above continues to infinity. For each particular number the general element (5) that corresponds to that number will have a final number of the exponents $a_{i}$. However, since the process of generating new numbers continues to infinity there is no an upper limit for the number of the exponents $a_{i}$ that will be generated by the general element (5), which will also go to infinity. Each of the elements of $a_{i}$ will have $\aleph_{0}$ possible combinations. This is obvious, since for any arbitrary large value $n\in N$ which one could take for the number of combinations, that value will be exceeded in this described process. The same is true for the number of the exponents. The number of the exponents will also be $\aleph_{0}$, since again any arbitrary taken number that one could chose for the value of the number of the exponents (does not matter how large is the number) will be exceeded in the described process, which continues to infinity. Therefore, the above described set will have ${\aleph_{0}}^{\aleph_{0}}=c$ elements which makes it equivalent in the cardinal number to the set of the real numbers. A one to one correspondence between such produced set and the set of natural numbers $N$ can be easily obtained by arranging the set elements by the sum of the exponents, as it is done for instance in [6]. We will not here proceed to discuss what could be wrong with the Cantor’s famous diagonal proof of countability of the set of real numbers; some of the relevant remarks are done in [6, 7]. Let us note here that that proof could be wrong since it uses the method of induction which, as it is well known [8, 9, 10], can not be applied on the infinite sets. With that method one can only prove that a number created by the diagonal procedure can be different from any $n$ numbers in the set. The method can not prove that that number is different from any number in the assumed denumerable set, which has infinite number of the elements. So, one can move through that set using the diagonal procedure to higher positions numbers $n$ in the sequence, but can not go through all the set elements. At least it cannot be done by using the induction method. ## 3 Conclusion It is proven that the set generated by the general element (5) has cardinal number equal to ${\aleph_{0}}^{\aleph_{0}}=c$. The same set is also denumerable, the elements can be ordered by the sum of the bases and exponents in (5). Therefore it is proven that the cardinality of the real and natural set of numbers are the same, i.e. that ${\aleph_{0}}^{\aleph_{0}}=c=\aleph_{0}$. ## References * [1] A. O. Gelfond, Doklady Akad. Nauk. S.S.S.R., 2 (1934), 1-6. * [2] Th. Schneider, J. Reine angew. Math., 172 (1035) 65-69. * [3] A. Baker, Linear forms in the logarithms of algebraic numbers I, II, III, IV, Mathematika, 13(1966),204-216; 14(1967),102-107, 220-228; 15(1968),204-216. * [4] Ch. Hermite, Sur la fonction exponentialle, Oeuvres III, 150-181. * [5] A. Baker and D. W. Masser, Transcendence Theory: Advances and Applications, Academic Press London New York San Francisco, 1977. * [6] S. Vlahovic and B. Vlahovic, Countability of the Real Numbers arXiv:math/0403169v1 * [7] S. Vlahovic and B. Vlahovic, Remarks on Cantor’s diagonalization proof of 1891, arXiv:math/0403288 * [8] A.A. Fraenkel and Y. Bar-Hillel, Foundation of Set Theory, Amsterdam 1958, chapter IV. * [9] A. Frankel, Y. Bar-Hillel and A. Levy, Foundation of set theory, North Holland, Amsterdam 1973, van Dalen’s remarks p. 268. * [10] M. Hallett, Cantorian Set Theory and Limitation of Size, Clarendon Press Oxford, 1984.

arxiv-papers

2024-09-04T02:48:54.467804

0803.3275

# $\eta/s$ and Phase Transitions Antonio Dobado, Felipe J. Llanes-Estrada and Juan M. Torres-Rincon Departamento de Física Teórica I, Universidad Complutense de Madrid, 28040 Madrid, Spain ###### Abstract We present a calculation of $\eta/s$ for the meson gas (zero baryon number), with the viscosity computed within unitarized NLO chiral perturbation theory, and confirm the observation that $\eta/s$ decreases towards the possible phase transition to a quark-gluon plasma/liquid. The value is somewhat higher than previously estimated in LO $\chi$PT. We also examine the case of atomic Argon gas to check the discontinuity of $\eta/s$ across a first order phase transition. Our results suggest employing this dimensionless number, sometimes called KSS number (in analogy with other ratios in fluid mechanics such as Reynolds number or Prandtl number) to pin down the phase transition and critical end point to a cross-over in strongly interacting nuclear matter between the hadron gas and quark and gluon plasma/liquid. ## I Introduction It has been recently pointed out Csernai:2006zz that the ratio of the shear viscosity to entropy density, $\eta/s$, has an extremum at a phase transition, based on empirical information for several common fluids, and follow-up calculations by us Dobado:2006hw and other groups Chen:2007xe , ChenNakano have suggested that $\eta/s$ in a hadron gas does indeed fall slightly with the temperature towards the predicted transition to a quark and gluon plasma or liquid phase. Renewed interest in this quantity arose after the KSS conjecture Kovtun:2003wp about a possible lower bound $1/(4\pi)$ (the existence of a bound had already been put forward, on the basis of simple physical arguments, in Danielewicz:1984ww ) and it is the subject of much current research in Heavy Ion Collisions. The precise reach of the bound has been under recent discussion, Cohen:2007zc ; Son:2007xw and there is much interest in finding theoretical or laboratory fluids that reach the minimum possible value of $\eta/s$ Cohen:2007qr ; Schafer:2007pr . There is good hope that $\eta/s$ and even $\eta$ by itself can be derived from particle and momentum distributions in heavy ion collisions Gavin ; Lacey:2006bc ; Lacey:2007na . It has been shown through several examples that, empirically, $\eta/s$ seems to have a discontinuity at a first order phase transition, but is continuous and has an extremum at a second order phase transition or at a crossover. Based on lattice data Karsch:2000kv it is believed that the phase transition between a gas of hadrons and a quark-gluon phase at zero baryon chemical potential is actually a cross over. The result of Csernai:2006zz however presents a clear discontinuity. This is of course not a serious claim of that paper, but simply an artifact of the very crude approximations there employed. We here revisit the issue, improving as far as feasible on the hadron-side estimate, and further motivating the proposed behavior of $\eta/s$. ## II Inverse Amplitude Method in $\chi$PT and hadron phase transition We here improve the very rough calculation of Csernai:2006zz for $\eta/s$ on the hadron phase. We have calculated in Dobado:2003wr the shear viscosity of a meson gas (that is, the hadron gas as a function of the temperature and approximate meson chemical potentials, at zero baryon chemical potential). That work employed the Inverse Amplitude Method (IAM) Dobado:1989qm that gives a good fit to the elastic phase shifts for meson-meson scattering at low momentum, respects unitarity, and is consistent with chiral perturbation theory at NLO ChPT . The only explicit degrees of freedom are light pseudoscalar mesons ($\pi$, $K$, $\eta$), but elastic meson-meson resonances below $1\ GeV$ appear through the phase shifts GomezNicola:2001as . It is an elementary exercise to divide the calculated viscosity from that work by the entropy density of the free Bose gas, for $N$ species $s=\frac{S}{V}=\frac{N}{6\pi^{2}T^{2}}\int_{0}^{\infty}dpp^{4}\frac{E-\mu}{E}\frac{e^{\beta(E-\mu)}}{\left[e^{\beta(E-\mu)}-1\right]^{2}},$ (1) and plot the result in Fig. 1. Figure 1: The viscosity over entropy density of a meson gas in chiral perturbation theory unitarized by means of the IAM. ($z$ represents the relativistic fugacity $e^{\beta(\mu-m)}$). Incidently, it can be seen in the figure that the holographic bound $\frac{\eta}{s}>1/4\pi$ is not violated, which had been claimed in the literature (we reported this in Dobado:2006hw ) but independently confirmed by FernandezFraile . The reason is that in chiral perturbation theory alone the cross section grows unchecked, eventually violating the unitarity bound, which induces a very small viscosity. Of more interest for our discussion in this work is to examine the possible behavior across the phase transition. We take the simple estimate for $\eta/s$ in the quark-gluon plasma from Csernai:2006zz , but we use our much improved calculation for the low-temperature hadron side (those authors employ LO chiral perturbation theory without unitarization). The result is plotted in Fig. 2. In addition we plot also the phase-shift based phenomenological calculation of ChenNakano , that is consistent with ours but somewhat smaller. Figure 2: We improve the hadron-side (low $T$) estimate of Csernai:2006zz that showed the jump in the $\eta/s$ ratio in the transition from the hadron gas to the quark-gluon plasma, substituting the Low-Energy-Theorem of those authors (first order chiral perturbation theory) by the Inverse Amplitude Method, that agrees with Chiral Perturbation Theory at NLO, and satisfies elastic unitarity. We confirm the result of those authors, although the actual numerical value of $\eta/s$ is quite different (as should be expected from their calculation reaching temperatures $T\simeq 150\ MeV$ but with only the first order interaction). One should note that, the calculation being performed at zero baryon chemical potential, based on lattice data that suggest a cross-over between the hadron gas and the quark-gluon plasma, and from simple phenomenology this would suggest that $\eta/s$ should be continuous. The calculation that Csernai:2006zz reported shows a discontinuous jump between the QGP and the hadron gas, whereas simple-minded non-relativistic phenomenology would make us expect a continuous function with a minimum. Our improved hadron calculation still shows a discontinuity, although now the jump at the discontinuity has opposite sign (our viscosity is larger since the meson-meson cross section is smaller due to unitarity, instead of being an LO-$\chi$PT polynomial). Since our estimate for $\eta/s$ is now approximate only because of our use of the first order Chapman-Enskog expansion and the quantum Boltzmann equation, both of which are reasonable approximations, we feel further improvement on the hadron side will not restore continuity, and future work needs to concentrate on evaluating the viscosity from the QGP side. Figure 3: We plot the dependence of $\eta/s$ on the quark mass from quark- gluon plasma side by adapting the results from Aarts . Note that given the non-trivial calculation there, we have slightly simplified by taking a constant $g$ in the mass correction. The band in the figure corresponds to the interval $g\in[1,2]$. We have also taken all quarks of equal mass $m_{s}=120MeV$ as the maximum possible variation. As can be seen, the dependence is small and positive, bringing about even better agreement with the hadron side Inverse Amplitude Method evaluation. To calculate the viscosity in a field theory, a possible and popular approach is to employ Kubo’s formula in terms of field correlators. Another method, based on the Wigner function, is to write-down the hierarchy of BBGKY equations. In either case one can perform a low-density expansion, leading to the use of the Boltzmann equation. Employing this on the hadron-side. as opposed to the full hierarchy of BBGKY equations of kinetic theory, presumes the “molecular chaos” hypothesis of Boltzmann, which is tantamount to neglecting correlations between sucessive collisions. This requires the collisions to be well separated over the path of the particle, and induces a systematic error in the calculation of order $2=(m_{\pi}\lambda)$, where $1/m_{\pi}$ is the typical reach of the strong interaction, and $\lambda$ the mean free path (controlled by the density). To keep this number below one requires small densities $n(T)<\frac{m_{\pi}}{2\sigma}$. If we take as a cross-section estimate 100 $mbarn$ we see that the criterion is satisfied up to temperatures of order 140 $MeV$ (where we stop our plot in Fig. 2). We have also estimated the change in $\eta/s$ caused by a small quark mass, by adapting the results of Aarts . Those authors provide, within a 2PI formalism, the shear viscosity of the quark and gluon plasma of one fermion species as a function of the fermion mass divided by the temperature, for fixed coupling constant. Although we are employing, as Csernai et al. do, a coupling that runs with the scale (the temperature), the mass correction is small, so we can take $g=2$ as fixed for a quick eyeball estimate. We normalize the viscosity of Aarts to the value plotted in Figure 2 at zero fermion mass, and then allow the fermion mass to vary. The results are now plotted in Figure 3. We plot the extreme case of all three light quarks equally massive and with mass equaling $m_{s}=120MeV$. As can be seen, the difference to the massless case is irrelevant at current precision and does not change the fact that we cannot conclude as of yet whether the transition between a hadron gas and a quark- gluon plasma/liquid has a discontinuity in $\eta/s$ or not. The reason that the fermion mass is not so relevant in the calculation is twofold. First, even at $m_{s}$, we have $T/m_{s}>1$ for any value of $T>T_{c}$. Since kinetic momentum transport in a gas is dominated by the fraction of molecules (here, partons) with the largest energy allowed by the Boltzmann tail of the distribution $E>T$, and the momentum varies as $p=E\left(1-\frac{m^{2}}{2E^{2}}+\cdots\right),$ we see that the small parton mass makes just a correction to the momentum of each (efficient) carrier. The second reason is that the cross-section, in a regime where perturbation theory is of any use, is weakly dependent on the fermion mass, with a slight dependence brought about by the logarithmic running of the quark-gluon vertex. Still, given the large uncertainties in our knowledge of the quark-gluon medium created in heavy-ion collisions, that make difficult to match with the hadron side, we also study a simple non-relativistic system where the jump in $\eta/s$ at the phase transition is very clear. ## III Liquid-gas phase transition in atomic Argon In prior works it has been pointed out that experimental data suggest that first order phase transitions present a discontinuity in $\eta/s$ and second order phase transitions (and maybe crossovers) present a minimum. We will examine one case a little closer, for a liquid-gas phase transition in the atomic Argon gas, where we will calculate the $\eta/s$ ratio theoretically and compare to data. The empirical data that has been brought forward was based on atomic Helium and molecular Nitrogen and Water. Quantum effects are very strong in the first at low temperatures where the phase transition occurs, and the later have relatively strong interactions. Instead we choose Argon due to its sphericity and closed-shell atomic structure, that make it a case very close to a hard-sphere system. Thus, Argon is the perfect theoretical laboratory, and sufficient data has been tabulated due to its use as a cryogenic fluid. The gas phase is therefore well described in terms of hard-sphere interactions. In elementary kinetic theory one neglects any correlation between sucessive scatterings. The viscosity follows then the formula $\eta_{\rm gas}=\frac{5}{16\,d^{2}}\sqrt{\frac{mT}{\pi}}\ \ ,$ (2) where $d=3.42\times 10^{5}\,fm$ is the viscosity diameter of the Argon atom and $m=37.3\,GeV$ its mass 111Note that this formula follows, up to the numerical factor, from considering a classical non-relativistic gas $\eta=\frac{1}{3}n(m\bar{v})\lambda$ in terms of the mean free path $\lambda$, the particle density $n$, and average momentum. The numerical factor requires a little more work with a transport equation and can be found, for example, in L. D. Landau and E. M. Lifshitz, “Physical Kinetics”, Pergamon Press, Elmsford, N.Y. 1981.. Experimental data is quoted as function of the temperature for fixed pressure. The particle density is then fixed by the equation of state; therefore a chemical potential needs to be introduced. In order to calculate the entropy density we again use Eq. (1) with $N=1$. As said, we keep the pressure $P$ constant, and the chemical potential $\mu$ varies then within the temperature range. In order to obtain $\mu$ we simply invert (numerically) the function $P(T=1/\beta,m,\mu)$ at fixed temperature. The expression for the pressure consistent with the entropy above is $P(T,m,\mu)=-\frac{T}{2\pi^{2}}\int_{0}^{\infty}dp\,p^{2}\log[1-e^{\beta(\mu-E)}]$ (3) (we have neglected in both cases the effect of the Bose-Einstein condensate since the gas liquefies before this is relevant). The problem has then been reduced to computing the viscosity at the given temperature and chemical potential, which we do employing our computer program for the meson gas in the Chapman-Enskog approximation, with minimum modifications. We change variables to absorb the scale and make the integrand of order 1 to: $\bar{\mu}\equiv\frac{\mu-m}{T},\quad x\equiv\frac{p^{2}}{mT}.$ (4) Thus, the final expressions for the entropy density and pressure from Eqs. (1) and (3) (once integrated by parts) are $P=\frac{1}{12\pi^{2}}m^{3/2}T^{5/2}\int_{0}^{\infty}dxx^{3/2}\frac{1}{e^{x/2-\bar{\mu}}-1},$ (5) $\displaystyle s_{gas}=\frac{1}{12\pi^{2}}(mT)^{3/2}\int_{0}^{\infty}dx\,x^{3/2}\left(\frac{x}{2}-\bar{\mu}\right)$ $\displaystyle\times\frac{e^{\,x/2-\bar{\mu}}}{\left(e^{\,x/2-\bar{\mu}}-1\right)^{2}}.$ (6) To treat the liquid Ar phase there is not a very rigorous theory. This is because in liquids the momentum transfer mechanism is quite complex and does involve the interaction between molecules. Here, our choice of a noble gas is of help since long-range interactions are absent. It is common to resort to semiempirical formulas with unclear theory support, or work with formal expressions of difficult applicability. We compromise by combining the Van der Waals equation of state (that ultimately encodes the Lennard-Jones theory for the interatomic potential), and use the Eyring liquid theory Eyring:1961 . The Eyring theory is a vacancy theory of liquids. Each molecule composing the liquid has gas-like degrees of freedom when it jumps into a vacant hole, and solid-like degrees of freedom when fully surrounded by other molecules. This model approach yields a partition function $Z$ for a one-species liquid (in natural units) $\displaystyle Z=\left\\{\frac{e^{E_{s}/N_{A}T}}{(1-e^{-\theta/T})^{3}}\left(1+n\frac{V-V_{s}}{V_{s}}e^{-\frac{aE_{s}V_{s}}{(V-V_{s})N_{A}T}}\right)\right\\}^{\frac{N_{A}V_{s}}{V}}$ $\displaystyle\times\left\\{\frac{e(2\pi mT)^{3/2}V}{(2\pi)^{3}N_{A}}\right\\}^{\frac{N_{A}(V-V_{s})}{V}},\quad$ (7) from which one can derive complete statistical information about the system 222The meaning of the various variables can be found in Eyring:1961 and is as follows. $e=2.71828\dots$ is Neper’s number (the presence of a single $e$ factor in the gas partition function comes from the Stirling’s approximation). $E_{s}$ is the sublimation energy of Argon (that we express in $eV$/particle). $\theta$ is the Einstein characteristic temperature of the solid defined in any textbook. Here $a=a^{\prime}$ (not to be confused with Van der Waals constant) is a model parameter, a pure-number, controlling the molecular “jump” between sites, or activation energy. $nV/V_{s}$ is the number of nearest vacancies to which an atom can jump.. One can recognize in the second line the partition function of a non-relativistic gas for the fraction of atoms with gas-like behavior. The first line corresponds to the solid-like behavior. The first factor is the partition function of a three-dimensional harmonic oscillator. The second term is a correction due to the translation degree of freedom, by which an atom can displace to a neighboring vacancy. The shear viscosity, (like $Z$ itself), turns out to be a weighted average between the viscosity of solid-like (first line) and gas-like (second line) degrees of freedom of the liquid’s particles: $\displaystyle\eta_{liq}=\frac{N_{A}2\pi}{V}\frac{1}{(1-e^{-\theta/T})}\frac{6}{n\kappa}\frac{V}{V-V_{s}}e^{\frac{a^{\prime}E_{s}V_{s}}{(V-V_{s})N_{A}T}}$ $\displaystyle+\frac{V-V_{s}}{V}\frac{5}{16\,d^{2}}\sqrt{\frac{mT}{\pi}}\ \ ,$ (8) and to complete the model, $\theta,n,a,a^{\prime},\kappa,E_{s}$ and $V_{s}$ are given in Table 1 for gaseous Argon. $N_{A}$ is Avogadro’s number. The $V_{s}/V$ solid-like volume fraction controls the weighted average. Note that if this ratio approaches 1, the viscosity diverges as appropriate for a rigid solid. 333In this formula $\kappa$ is an ad-hoc model “transmission coefficient” of order 1 related to the loss of momentum to a crystal wave upon displacing an atom. Here we take it to be independent of the pressure but this could be lifted to further improve the fit in Figure 4. Table 1: Liquid Argon parameters which appears in Eqs. (III) and (9). All these constants are given in Eyring:1961 . However, $\kappa$ has been modified because we use Eq. (2) instead of the formula that appears in Eyring:1961 for the hard-sphere gas case. Parameter | Value ---|--- $\theta$ | $5.17\ meV$ $n$ | $10.80$ $a=a^{\prime}$ | $0.00534$ $\kappa$ | 0.667 $E_{s}$ | $0.082\ eV/particle$ $V_{s}$ | $4.16\times 10^{16}\ fm^{3}/particle$ The entropy is calculated as usual taking a derivative of the Helmholtz free energy ($A\equiv-T\log Z$), $S=\frac{\partial(T\log Z)}{\partial T}\ \ .$ (9) For our purposes we also need the liquid density which is easily estimated by means of the Van der Waals equation of state, that is of some applicability in the liquid phase. This equation takes into account the volume excluded by the particles and the attraction between them. In the simplest form the Van der Waals equation is: $\left(n_{gas}+n_{liq}^{2}\frac{a}{T}\right)(1-n_{liq}b)=n_{liq},$ (10) where $n_{gas}$ and $n_{liq}$ are the particle density of gas and liquid Argon, respectively; $T$ is the temperature, $2b=4\pi d^{3}/3$ is the covolume, that is, the excluded volume by the particle (here we take $d$ as a mean value of the viscosity radius and the gas radius) and $a=27T_{c}/64P_{c}$ is a measure of the particles attraction ($T_{c}=150.87\ K$, $P_{c}=4.898\ MPa$). Eq. (10) is a cubic equation in $n_{liq}$ which gives reasonably good results despite its simplicity. For this reason, we think that it is not necessary to derive a new state equation from the Helmholtz free energy. Putting all together we are able to calculate the $\eta/s$ ratio in both liquid and gas states. The final result is plotted in Fig. 4 where a good agreement with the experimental data of CRC is shown. One can see how the KSS bound is maintained. Moreover, one can observe that for the liquid-gas phase transition $\eta/s$ presents a minimum and discontinuity at the phase transition (below the critical pressure, $P_{c}$). Above this pressure, a minimum is still seen but the function is continuous. Figure 4: $\eta/s$ (a pure number in natural units) for atomic Argon in the liquid and gas phases near the phase transition. Solid lines correspond to theoretical calculation described in the text, dashed lines are the experimental values given from CRC . Note that $\eta/s$ is quite independent of the pressure in the liquid phase, and that the theoretical curves calculated from the liquid side and gas side do get closer together with increasing pressure, suggesting as the data that indeed, $\eta/s$ will be continuous in the cross-over regime. ## IV Conclusions and outlook In this article we have argued, in agreement with previous authors, how it is likely that $\eta/s$ is a reasonable derived observable in relativistic heavy ion collisions to pin down the phase transition and possible critical end point between a hadron gas and the quark and gluon plasma/liquid. We have contributed an evaluation of the hadron-side $\eta/s$ that simultaneously encodes basic theoretical principles such as chiral symmetry and unitarity, and simultaneously produces a practical and good fit of the pion scattering phase shifts, by means of the Inverse Amplitude Method. In so doing we have updated our past meson gas work. Our conclusions are in qualitative agreement with those of Chen:2007jq . Since our lack of understanding of the non-perturbative dynamics on the high-$T$ side of the phase transition to the quark-gluon phase prevents us from matching asymptotic behavior of $\eta/s$ at high $T$ with the hadron gas, we have studied this KSS number in a related Sigma Model. We find numerically, and confirm with an analytical estimate, that keeping the $s$-channel amplitude one can isolate a minimum, and within reasonable calculational uncertainties, this coincides with the known phase transition of the model. A complete analysis is to be reported elsewhere. Since we are not in possession of a good program that can proceed to finite baryon density, we leave this for further investigation. Meanwhile we have investigated the past observation that in going from a cross-over to a first order phase transition, $\eta/s$ changes behavior, from having a continuous minimum to presenting a discontinuity. We choose, as very apt for theoretical study, atomic Argon. We employ standard gas kinetic theory above the critical temperature and the Eyring theory of liquids in the liquid phase. Whereas the discontinuity in $\eta/s$ is very clear for low pressures, theory and data are close to matching (showing continuity) at high pressures where a crossover between the two phases is seen in the phase diagram. The conclusion is that indeed the minimum of the $\eta/s$ and the temperature of the phase transition might well be proportional. Whether the proportionality constant is exactly one could only be established by an exact calculation of the viscosity which is not theoretically at hand. As a consequence, we provide further theory hints to the currently proposed method to search for the critical end point in hot hadron matter. If, as lattice gauge theory suggests, a smooth crossover occurs between the hadron phase and the quark-gluon phase, at least under the conditions in the Relativistic Heavy Ion Collider where the baryon number is small at small rapidity, then one expects to see a minimum of viscosity over entropy density. In the FAIR experimental program however it might be possible to reach the critical end point given the higher baryon density (since the energy per nucleon will be smaller), and whether the phase transition is then first or second order can be inferred from the possibility of a discontinuity of $\eta/s$. ###### Acknowledgements. We thank useful conversations and exchanges on $\eta/s$ with Jochen Wambach, Juan Maldacena, Dam Son, and Tom Cohen. This work has been supported by grants FPA 2004-02602, 2005-02327, BSCH-PR34/07-15875 (Spain) ## References * (1) L. P. Csernai, J. I. Kapusta and L. D. McLerran, Phys. Rev. Lett. 97 (2006) 152303 [arXiv:nucl-th/0604032]. * (2) A. Dobado and F. J. Llanes-Estrada, Eur. Phys. J. C 49 (2007) 1011 [arXiv:hep-ph/0609255]. * (3) J. W. Chen, Y. H. Li, Y. F. Liu and E. Nakano, Phys. Rev. D 76, 114011 (2007) [arXiv:hep-ph/0703230]. * (4) J. W. Chen and E. Nakano, Phys. Lett. B647 371 (2007). * (5) P. Kovtun, D. T. Son and A. O. Starinets, JHEP 0310 (2003) 064 [arXiv:hep-th/0309213]; P. Kovtun, D. T. Son and A. O. Starinets, Phys. Rev. Lett. 94 (2005) 111601. * (6) P. Danielewicz and M. Gyulassy, Phys. Rev. D 31 (1985) 53. * (7) T. D. Cohen, [arXiv:hep-th/0711.2664]. * (8) D. T. Son, Phys. Rev. Lett. 100, 029101 (2008) [arXiv:hep-th/0709.4651]. * (9) T. D. Cohen, Phys. Rev. Lett. 99, 021602 (2007) [arXiv:hep-th/0702136]. * (10) T. Schafer, Phys. Rev. A 76, 063618 (2007) [arXiv:cond-mat/0701251]. * (11) S. Gavin and M. Abdel-Aziz, Phys. Rev. Lett. 97 (2006) 162302 [arXiv:nucl-th/0606061]. * (12) R. A. Lacey et al. Phys. Rev. Lett. 98, 092301 (2007) [arXiv:nucl-ex/0609025]. * (13) R. A. Lacey et al. [arXiv:nucl-ex/0708.3512]. * (14) F. Karsch, E. Laermann and A. Peikert, Nucl. Phys. B605, 579 (2001) [arXiv:hep-lat/0012023]. * (15) A. Dobado and F. J. Llanes-Estrada, Phys. Rev. D 69, 116004 (2004) [arXiv:hep-ph/0309324]. * (16) A. Dobado, M. J. Herrero and T. N. Truong, Phys. Lett. B235, 134 (1990). * (17) S. Weinberg, Physica A96:327 (1979) J. Gasser and H. Leutwyler, Annals Phys. 158:142 (1984) * (18) A. Gomez Nicola and J. R. Pelaez, Phys. Rev. D65, 054009 (2002) [arXiv:hep-ph/0109056]. * (19) D. Fernandez-Fraile and A. Gomez Nicola Int. J. Mod. Phys. E16, 3010-3013 (2007) [arXiv:hep-ph/0706.3561]. D. Fernandez-Fraile and A. Gomez Nicola Eur. Phys. J. A31, 848-850 (2007) [arXiv:hep-ph/0610197]. D. Fernandez-Fraile and A. Gomez Nicola Phys. Rev. D73, 045025 (2006) [arXiv:hep-ph/0512283]. * (20) G. Aarts and J. M. Martinez Resco, JHEP 0503, 074 (2005) [arXiv:hep-ph/0503161]. * (21) J. W. Chen, M. Huang, Y. H. Li, E. Nakano and D. L. Yang, [arXiv:hep-ph/0709.3434]. * (22) H. Eyring and T. Ree, Proc. Nat. Acad. Sci., 47, 526-537 (1961) * (23) D.R. Lide (ed.), CRC Handbook of Chemistry and Physics (75th ed.), CRC Press, Boca Raton, FL (1994)

arxiv-papers

2024-09-04T02:48:54.473846

0803.3299

# ${{~{}~{}~{}}^{{}^{{}^{Published~{}in\,:\,~{}Celestial~{}Mechanics~{}and~{}Dynamical~{}Astronomy\,,~{}Vol.\,104\,,~{}pp.\,257-289~{}\,(2009)}}}}$ Tidal torques. A critical review of some techniques Michael Efroimsky US Naval Observatory, Washington DC 20392 USA e-mail: me @ usno.navy.mil and James G. Williams Jet Propulsion Laboratory, California Institute of Technology, Pasadena CA 91109 USA e-mail: james.g.williams @ jpl.nasa.gov ###### Abstract We review some techniques employed in the studies of torques due to bodily tides, and explain why the MacDonald formula for the tidal torque is valid only in the zeroth order of the eccentricity divided by the quality factor, while its time-average is valid in the first order. As a result, the formula cannot be used for analysis in higher orders of $\,e/Q\,$. This necessitates some corrections in the current theory of tidal despinning and libration damping (though the qualitative conclusions of that theory may largely remain correct). We demonstrate that in the case when the inclinations are small and the phase lags of the tidal harmonics are proportional to the frequency, the Darwin- Kaula expansion is equivalent to a corrected version of the MacDonald method. The latter method rests on the assumption of existence of one total double bulge. The necessary correction to MacDonald’s approach would be to assert (following Singer 1968) that the phase lag of this integral bulge is not constant, but is proportional to the instantaneous synodal frequency (which is twice the difference between the evolution rates of the true anomaly and the sidereal angle). This equivalence of two descriptions becomes violated by a nonlinear dependence of the phase lag upon the tidal frequency. It remains unclear whether it is violated at higher inclinations. Another goal of our paper is to compare two derivations of a popular formula for the tidal despinning rate, and to emphasise that both are strongly limited to the case of a vanishing inclination and a certain (sadly, unrealistic) law of frequency-dependence of the quality factor $\,Q\,$ – the law that follows from the phase lag being proportional to frequency. One of the said derivations is based on the MacDonald torque, the other on the Darwin torque. Fortunately, the second approach is general enough to accommodate both a finite inclination and the actual rheology. We also address the rheological models with the $Q$ factor scaling as the tidal frequency to a positive fractional power, and disprove the popular belief that these models introduce discontinuities into the equations and thus are unrealistic at low frequencies. Although such models indeed make the conventional expressions for the torque diverge at vanishing frequencies, the emerging infinities reveal not the impossible nature of one or another rheology, but a subtle flaw in the underlying mathematical model of friction. Flawed is the common misassumption that damping merely provides phase lags to the terms of the Fourier series for the tidal potential. A careful hydrodynamical treatment by Sir George Darwin (1879), with viscosity explicitly included, had demonstrated that the magnitudes of the terms, too, get changed – a fine detail later neglected as “irrelevant”. Reinstating of this detail tames the fake infinities and rehabilitates the “impossible” scaling law (which happens to be the actual law the terrestrial planets obey at low frequencies). Finally, we explore the limitations of the popular formula interconnecting the quality factor and the phase lag. It turns out that, for low values of Q, the quality factor is no longer equal to the cotangent of the lag. ## 1 Prologue ${\left.~{}~{}~{}~{}~{}~{}\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}\,\right.}^{\mbox{\small\it When~{}it~{}shall~{}be~{}found~{}that~{}much~{}is~{}omitted,}}$ ${\left.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}\,\right.}^{\mbox{\small\it let~{}it~{}not~{}be~{}forgotten~{}that~{}much~{}likewise~{}is~{}performed}}$ ${\left.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}\right.}^{\mbox{\small\it Samuel~{}Johnson, ~{}1755}}$ In his short work _“Untersuchung der Frage … ,”_ known among the historians also as the _“Spin-Cycle essay,”_ Immanuel Kant (1754) stated that the Moon not only pulls the Earth, but also exerts a retarding torque upon its surface; this torque slows down the Earth’s rotation and lets go only when terrestrial days become as long as lunar months. Although Kant had in mind only the ocean tides, not the bodily ones, we may say that, qualitatively, he predicted the celebrated $\,1:1\,$ spin-orbit resonance, the _pas de deux_ wherein Pluto and Charon are locked. For the first time, the idea of tidal action not being confined only to the fluid portion of the planet but affecting also the solid, so as to induce a state of varying strain, was put forward by John Herschel (son of astronomer William Herschel), as a minor aside in a paper devoted to volcanism and earthquakes (Herschel 1863). The earliest mathematical description of land tides in their dynamics was offered by George Darwin (son of naturalist Charles Darwin and great-grandson of poet and philosopher Erasmus Darwin). Following his predecessors Roche (1849) and Thompson (1863), who had calculated the figure of a static tide, Darwin (1879) assumed the Earth homogeneous and consisting of an incompressible fluid. To account for dynamics, he also assumed that the viscosity was the sole source of the tidal friction. Relying on this model, Darwin (1880, 1908) derived a tide-generated disturbing potential expanded into a Fourier series. Substitution thereof into the Lagrange-type planetary equations led him to expressions for the time derivatives of the orbital elements via partial derivatives of the disturbing potential with respect to the elements. An impressive generalisation of Darwin’s work by Kaula (1964), and the subsequent flow of new concepts and applications (MacDonald 1964; Goldreich 1966a,b; Goldreich & Peale 1966; Singer 1968; Mignard 1979, 1980; Touma & Wisdom 1994; Neron de Surgy & Laskar 1997; Krasinsky 2002, 2006; Getino, Escapa & García 2003; Ferraz Mello et al 2008; Efroimsky 2008) made bodily tides a rapidly developing area of the planetary astronomy. The vast and growing volume of the relevant material leaves one no chance to glean it all in one review. Therefore we shall concentrate on one special aspect of this research, the tidal torques emerging from the bodily tides. Moreover, we shall dwell solely on the techniques, not applications. Although our review will at times be very critical, it should from the beginning be agreed that our criticisms are intended in the spirit of the above quotation from Samuel Johnson. Along with reviewing the current state of the field, we shall provide some new results of our own. Specifically, we shall address the rheological models with the $Q$ factor scaling as the tidal frequency to a positive fractional exponential. We shall demonstrate that, contrary to the common opinion, such rheologies do _not_ cause infinities in the expression for the torque. We shall also derive an expression for the tidal torque decelerating a terrestrial planet obeying such a rheology. (That the realistic terrestrial bodies indeed obey this class of rheologies has been explained in Efroimsky & Lainey 2007.) ## 2 Trivia In this section, we shall briefly recall how a satellite-generated potential in a point on or inside the planet is expressed through the latitude, longitude, and the radial distance of the point. Let us begin from the first principles. The dynamics of point masses $\,m_{i}\,$ located at inertial-frame-related positions $\,{\mbox{{\boldmath$\vec{\rho}$}}}_{i}\;$, $m_{i}\;{\stackrel{{\scriptstyle\mbox{\bf{..}}}}{{\mbox{{\boldmath$\vec{\rho}$}}}}}_{i}\;=\;m_{i}\;\sum_{j\neq i}\,\;G\,m_{j}\;\frac{{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{ij}}{r_{ij}^{3}}\;\;\;,\;\;\;\;\;{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{ij}\,\equiv\,{\bf{\mbox{{\boldmath$\vec{\rho}$}}}}_{j}\,-\,{\bf{\mbox{{\boldmath$\vec{\rho}$}}}}_{i}\;\;,\;\;\;\;i,j=1,...,N\;\;,\;\;\;$ (1) may be conveniently reformulated in terms of the relative-to-the-primary locations $\mbox{{\boldmath$\vec{r}$}}_{i}\;\equiv\;\mbox{{\boldmath$\vec{r}$}}_{0i}\;\equiv\;{\bf{\mbox{{\boldmath$\vec{\rho}$}}}}_{i}\;-\;{\bf{\mbox{{\boldmath$\vec{\rho}$}}}}_{0}\;\;\;,$ (2) ${\bf{\mbox{{\boldmath$\vec{\rho}$}}}}_{0}\;$ standing for the position of the primary. The difference between ${\stackrel{{\scriptstyle\mbox{\bf{..}}}}{{\mbox{{\boldmath$\vec{\rho}$}}}}}_{i}\;=\;\sum_{j\neq i,0}\;\;G\;\frac{m_{j}\;{\mbox{{\boldmath$\vec{r}$}}}_{ij}}{r_{ij}^{3}}\;+\;G\;\frac{m_{0}\;{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{i0}}{r_{i0}^{3}}\;$ (3) and ${\stackrel{{\scriptstyle\mbox{\bf{..}}}}{{\mbox{{\boldmath$\vec{\rho}$}}}}}_{0}\;=\;\sum_{j\neq i,0}\;\;G\;\frac{m_{j}\;{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{0j}}{r_{0j}^{3}}\;+\;G\;\frac{m_{i}\;{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{0i}}{r_{0i}^{3}}\;$ (4) amounts to: ${\mbox{\boldmath$\ddot{\mbox{{\boldmath$\vec{r}$}}}$}}_{i}\;=\;\sum_{j\neq i,0}\;\;G\;\frac{m_{j}\;{\bf{\mbox{{\boldmath$\vec{r}$}}}}_{ij}}{r_{ij}^{3}}\;-\;\sum_{j\neq i,0}\;\;G\;\frac{m_{j}\;{\mbox{{\boldmath$\vec{r}$}}}_{j}}{r_{j}^{3}}\;-\;G\;\frac{\left(m_{i}\,+\,m_{0}\right)\;{\mbox{{\boldmath$\vec{r}$}}}_{i}}{r_{i}^{3}}\;=\;-\;\frac{\partial\,U_{i}}{\partial\,{\mbox{{\boldmath$\vec{r}$}}}_{i}}$ (5) $U_{i}\;$ being the potential: ${U}_{i}\;\equiv\;-\;\frac{G\;\left(m_{i}\;+\;m_{0}\right)}{r_{i}}\;+\;W_{i}\;\;,$ (6) with the disturbance $W_{i}\;\equiv\;-\;\sum_{j\neq i}\;\;G\;m_{j}\;\left\\{\frac{1}{r_{ij}}\;-\;\frac{{{\mbox{{\boldmath$\vec{r}$}}}}_{i}\,\cdot\,{{\mbox{{\boldmath$\vec{r}$}}}}_{j}}{r_{j}^{3}}\right\\}$ (7) singled out. This disturbing potential acting on mass $\,m_{i}\,$ is generated by the masses $\,m_{j}\,$ other than $\,m_{i}\,$ or the primary. It deviates from the Newtonian one by the amendment $\;G\,m_{j}\,{r_{j}^{-3}}\,{{\mbox{{\boldmath$\vec{r}$}}}_{i}\cdot{\mbox{{\boldmath$\vec{r}$}}}_{j}}\;$ emerging in the noninertial frame associated with the primary. In the simplest case of one secondary, a satellite of mass $\,m_{1}=M^{*}_{sat}\,$, located at a planetocentric position $\,\mbox{{\boldmath$\vec{r}$}}_{1}\,=\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,$, will be creating at some point $\,\mbox{{\boldmath$\vec{r}$}}_{2}\,=\,\mbox{{\boldmath$\vec{R}$}}\;$ a perturbing potential $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})\;=\;-\;G\;M^{*}_{sat}\;\left\\{\frac{1}{|\mbox{{\boldmath$\vec{R}$}}\;-\;\mbox{{\boldmath$\vec{r}$}}^{\;*}|}\;-\;\frac{{\mbox{{\boldmath$\vec{R}$}}}\,\cdot\,{\mbox{{\boldmath$\vec{r}$}}}^{\;*}}{|\mbox{{\boldmath$\vec{r}$}}^{\;*}|^{3}}\right\\}\;\;\;,$ (8) expandable over the Legendre polynomials (for $\,R\,<\,r^{*}\,$) by means of the formulae $\displaystyle\frac{1}{|\mbox{{\boldmath$\vec{R}$}}\;-\;\mbox{{\boldmath$\vec{r}$}}^{\;*}|}\;=\;\frac{1}{r^{*}}\;\sum_{{\it{l}}=0}^{\infty}\;\left(\,\frac{R}{r^{*}}\,\right)^{\it{l}}\;P_{\it{l}}(\cos\gamma)\;\;\;$ (9) and $\displaystyle\frac{{\mbox{{\boldmath$\vec{R}$}}}\,\cdot\,{\mbox{{\boldmath$\vec{r}$}}}^{\;*}}{|\mbox{{\boldmath$\vec{r}$}}^{\;*}|^{3}}\;=\;\frac{R\;r^{\;*}\;\cos\gamma}{{r^{\;*}}^{{\,{3}}}}\;=\;\frac{1}{r^{\;*}}\;\,\frac{R}{r^{\;*}}\;P_{1}(\cos\gamma)\;\;\;,$ (10) $\gamma\,$ being the angular separation between $\vec{R}$ and $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,$, subtended at the point of origin, which we shall naturally choose to coincide with the planet’s centre of mass. Together, the former and the latter formulae yield: $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})\;=\;-\;\frac{G\;M^{*}_{sat}}{r^{\,*}}\;\sum_{{\it{l}}=2}^{\infty}\,\left(\,\frac{R}{r^{\;*}}\,\right)^{\textstyle{{}^{\it{l}}}}\,P_{\it{l}}(\cos\gamma)~{}~{}~{},$ (11) where we have neglected the $\,{\it{l}}=0\,$ term $~{}-\,GM^{*}_{sat}/r^{\,*}~{}$, because it bears no dependence upon $\vec{R}$ , and in practical problems is attributed to the principal part of the potential, not to the one regarded as perturbation. The angle $\,\gamma\,$ can be expressed via spherical coordinates as: $\displaystyle\cos\gamma\;=\;\frac{\mbox{{\boldmath$\vec{R}$}}\cdot\mbox{{\boldmath$\vec{r}$}}^{\;*}}{R\;r^{\;*}}\;=\;\sin\phi\;\sin\phi^{*}\;+\;\cos\phi\;\cos\phi^{*}\;\cos(\lambda\,-\,\lambda^{*})\;\;\;,$ (12) $(R\,,\,\phi\,,\,\lambda)\;$ being the planetocentric distance, the latitude, and the longitude of the point where the disturbance is experienced; and $(r^{*}\,,\,\phi^{*}\,,\,\lambda^{*})\;$ being the spherical coordinates of the satellite. It is customary (though not at all obligatory) to reckon the longitudes from a planet-fixed meridian, in which case the subsequent formulae for the potential come out written in a reference frame co-rotating with the planet. A Legendre polynomial of $\,\cos\gamma\,$, too, can be expressed via the spherical coordinates: $\displaystyle P_{\it{l}}(\cos\gamma)\;=\;\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;(2\,-\,\delta_{0m})\;P_{{\it{l}}m}(\sin\phi)\;P_{{\it{l}}m}(\sin\phi^{*})\;\cos m(\lambda\,-\,\lambda^{*})~{}~{}~{},$ (13) substitution whereof into (11) results in $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\,\mbox{{\boldmath$\vec{r}$}}^{\;*})\,=\,-\,\frac{G\;M^{*}_{sat}}{r^{\,*}}\sum_{{\it{l}}=2}^{\infty}\left(\frac{R}{r^{\;*}}\right)^{\textstyle{{}^{\it{l}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(\sin\phi)P_{{\it{l}}m}(\sin\phi^{*})\;\cos m(\lambda-\lambda^{*})~{}~{}.~{}~{}~{}~{}~{}~{}$ (14) Evidently, this formalism will stay unaltered, if the role of the tide-raising satellite is played by the Sun, or by another satellite, or by another planet. (In this case, what we call $\,M_{sat}\,$ will, in fact, denote the mass of the Sun, or of the other satellite, or of the other planet.) Likewise, the formalism may in its entirety be applied to a satellite regarded as a tidally- disturbed primary, the planet being treated as a tide-raising body (and $\,M^{*}_{sat}\,$ now standing for the planetary mass). ## 3 The Kaula expansion for a tidal potential Kaula (1961) came up with a remarkable formula $\displaystyle\left(\,\frac{1}{r^{\,*}}\,\right)^{{\it l}+1}P_{\it{l}}(\sin\phi^{*})\;\left[\;\cos m\lambda^{*}\;+\;\sqrt{-1}\;\sin m\lambda^{*}\;\right]~{}=~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (15) $\displaystyle\left(\frac{1}{a^{\,*}}\right)^{{\it l}+1}\sum_{p=0}^{\infty}F_{{\it l}mp}({\it i}^{*})\sum_{q=\,-\,\infty}^{\infty}G_{{\it l}pq}(e^{*})\;\left\\{\begin{array}[]{c}\cos\left(\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,\right)\,+\,\sqrt{-1}\;\sin\left(\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,\right)\\\ \sin\left(\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,\right)\,-\,\sqrt{-1}\;\cos\left(\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,\right)\end{array}\right\\}^{{\it l}\,-\,m\;\;\mbox{\small even}}_{{\it l}\,-\,m\;\;\mbox{\small odd}~{}~{}~{}~{}~{}~{}~{}{\textstyle,}}~{}~{}~{}~{}$ (18) where $F_{{\it l}mp}({\it i})\,$ are the inclination functions (Gooding and Wagner 2008); $\,G_{{\it l}pq}(e)\,$ are the eccentricity polynomials identical to the Hansen coefficients $\,X_{({\it l}-2p+q)~{}}^{~{}(\,-\,{\it l}-1)\,,~{}({\it l}-2p)}\,$; the notation $\,\sqrt{-1}\,$ is used to avoid confusion with the inclination; and the auxiliary combinations $\,v_{{\it l}mpq}^{*}\,$ are defined as:111 This definition agrees with that by Kaula (1961, 1964, 1966), but differs from the one by Lambeck (1980) who incorporated $\;-\,m\,\theta^{*}\;$ into $\,v_{{\it l}mpq}^{*}\,$. $\displaystyle v_{{\it l}mpq}^{*}\;\equiv\;({\it l}-2p)\omega^{*}\,+\,({\it l}-2p+q){\cal M}^{*}\,+\,m\,\Omega^{*}~{}~{}~{}.$ (19) This development enabled Kaula (1961, 1964) to carry out a transformation from the tide-raising satellite’s spherical coordinates to its orbital elements and the sidereal time $\,\theta^{*}\,$. These elements (the semimajor axis $\,a^{*}\,$, eccentricity $\,e^{*}\,$, inclination $\,{\it i}^{*}\,$, periapse $\,\omega^{*}\,$, ascending node $\,\Omega^{*}\,$, mean anomaly $\,{\cal M}^{*}\,$) are introduced in a frame that is associated with the equator but is not co-rotating with it. In terms of these parameters, $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})\,=\,-\,\frac{G\,M^{*}_{sat}}{a^{*}}\sum_{{\it l}=2}^{\infty}\,\left(\frac{R}{a^{*}}\right)^{\textstyle{{}^{\it l}}}\,\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;(2\,-\,\delta_{0m})\,P_{{\it{l}}m}(\sin\phi)\,\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\,\sum_{q=\,-\,\infty}^{\infty}\,G_{{\it l}pq}(e^{*})~{}~{}~{}~{}~{}~{}~{}$ (20) $\displaystyle\left[\;\cos m\lambda\;\left\\{\begin{array}[]{c}\cos\\\ \sin\end{array}\right\\}^{{\it l}\,-\,m\;\;\mbox{\small even}}_{{\it l}\,-\,m\;\;\mbox{\small odd}}\;\left(\,v_{{\it l}mpq}^{*}\,-\;m\,\theta^{*}\,\right)+\;\sin m\lambda\;\left\\{\begin{array}[]{c}~{}~{}\sin\\\ -\,\cos\end{array}\right\\}^{{\it l}\,-\,m\;\;\mbox{\small even}}_{{\it l}\,-\,m\;\;\mbox{\small odd}}\;\left(\,v_{{\it l}mpq}^{*}\,-\;m\,\theta^{*}\,\right)\;\right]~{}~{}~{}~{}~{}~{}~{}$ (25) or, after carrying out the multiplication of the sine and cosine functions: $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})\;=\;-\;\frac{G\,M^{*}_{sat}}{a^{*}}\;\sum_{{\it l}=2}^{\infty}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\left.~{}~{}~{}-\;\delta_{0m}\,\right)\;P_{{\it{l}}m}(\sin\phi)\;\sum_{p=0}^{\it l}\;F_{{\it l}mp}({\it i}^{*})\;\sum_{q=\,-\,\infty}^{\infty}\;G_{{\it l}pq}(e^{*})\left\\{\begin{array}[]{c}\cos\\\ \sin\end{array}\right\\}^{{\it l}\,-\,m\;\;\mbox{\small even}}_{{\it l}\,-\,m\;\;\mbox{\small odd}}\;\left(v_{{\it l}mpq}^{*}-m(\lambda+\theta^{*})\right)~{}~{}~{}.~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (28) ## 4 Physical assumptions involved in Kaula’s theory If the primary is not a point mass, it becomes distorted by potential $\,W(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})\,$. The distortion of shape will, in its turn, generate some extra potential perturbation whose calculation is complicated by the tide-raising potential (28) evolving in time and having a rich spectrum of frequencies. The response of the primary’s shape to each of these is different and depends on the properties of the planet’s material. This is a situation where the linear approach becomes most helpful, when applicable.222 For most materials, departure from linearity becomes considerable when the strains approach $\,10^{-6}\,$. (Karato 2007) The linear theory of bodily tides comprises two independent assertions. One is that the energy attenuation rate $\langle\dot{E}\rangle$ at each harmonic depends solely on the frequency $\,\chi\,$ and on the amplitude $\,E_{peak}(\chi)\,$, and is not influenced by the rest of the spectrum. This is written down as $\langle\dot{E}(\chi)\rangle=-\chi E_{peak}(\chi)/Q(\chi)$, which is equivalent to $\Delta E_{cycle}(\chi)=-2\pi E_{peak}(\chi)/Q(\chi)$, where $\Delta E_{cycle}(\chi)$ is the one-cycle energy loss, and $Q(\chi)$ is the quality factor. The other assertion is that each _stationary_ tidal change of the potential, $\,W_{\it l}\,$, inflicts on the planet’s shape a linear deformation. Each of these deformations, in their turn, amend the potential of the primary with an addition proportional to the Love number $\,k_{\it l}\,$. As known from the potential theory, an addition proportional to $\,P_{\it l}(\cos\gamma)\,$ must be decreasing outside the spherical primary as $\,r^{-({\it l}+1)}\,$. Hence, were the external potential perturbation $\,W\,$ static (or, equivalently, were the response of the material instant), the tidal addition to the planetary potential would have assumed the form 333 Following MacDonald (1964) and Singer (1968), we denote the tide-raising potential with $\,W\,$ and the bodily-tide one with $\,U\,$. In his original paper, Kaula (1964) called these potentials $\,U\,$ and $\,T\,$, while in the book he switched to $\,U\,$ and $\,U_{\textstyle{{}_{T}}}\;$ (Kaula 1968). Be mindful that we are using a sign convention different from that of Kaula. As our forces are negative gradients of potentials, our potentials are negative to those of Kaula. $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{r}\,\right)^{{\it l}+1}\;W_{\it{l}}(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle=\;-\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{r}\,\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}_{sat}}{a^{*}}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\left.~{}~{}~{}-\;\delta_{0m}\,\right)\;P_{{\it{l}}m}(\sin\phi)\;\sum_{p=0}^{\it l}\;F_{{\it l}mp}({\it i}^{*})\;\sum_{q=\,-\,\infty}^{\infty}\;G_{{\it l}pq}(e^{*})\left\\{\begin{array}[]{c}\cos\\\ \sin\end{array}\right\\}^{{\it l}\,-\,m\;\;\mbox{\small even}}_{{\it l}\,-\,m\;\;\mbox{\small odd}}\;\left(\;v_{{\it l}mpq}^{*}-m(\lambda+\theta^{*})\;\right)~{}~{}~{}.~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (31) $R\,$ being the mean equatorial (equivolumetric) radius of the planet, $\,\mbox{{\boldmath$\vec{R}$}}\,=\,(R\,,\,\phi\,,\,\lambda)\,$ being a particular surface point, and $\,\mbox{{\boldmath$\vec{r}$}}\,=\,(r\,,\,\phi\,,\,\lambda)\,$ being an exterior point located right above the surface point $\vec{R}$ , at a planetocentric radius $\,r\,\geq\,R\,$. As we intend to study the effect of this potential on another external body, a similar transformation should be applied to the coordinates $\,(r\,,\,\phi\,,\,\lambda)\,$, to express $\,W\,$ through the orbital elements of this body. Employment of (15), this time not for $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,$ but for $\vec{r}$ , leads to: $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;-\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{a}\,\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}_{sat}}{a^{*}}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\;\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (32) $\displaystyle~{}~{}~{}~{}~{}\left.-\,\delta_{0m}\,\right)\,\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\cos\left[\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)\right]~{}~{}_{\textstyle{{}_{\textstyle,}}}$ a formula that generalises the tidal theory of Darwin (1908, p. 334) to $\,{\it l}\,$ and $\,|q|\,$ larger than $\,2\,$. Both Kaula (1964), who derived this milestone result, and Darwin, who had developed its simplified version, realised that this machinery would work only after the material’s delayed reaction to perturbation (28) is somehow taken into account. Until then (32) remains idealised, in that it corresponds to an unphysical case of instantaneous response. To account for damping, Kaula (1964) followed the path of Darwin (1880, 1908): he endowed each term of the Fourier series (32) with a real phase lag of its own, $\,\epsilon_{{\it l}mpq}\;$, whereafter the ultimate form of Kaula’s expansion became $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;-\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{a}\,\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}_{sat}}{a^{*}}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\;-\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (33) $\displaystyle\left.\delta_{0m}\,\right)\,\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\cos\left[\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)-\epsilon_{{\it l}mpq}\right]~{}~{}_{\textstyle{{}_{\textstyle.}}}$ This empirical method of including dissipation into the picture contains in itself an important omission, of which Sir George Darwin was aware, but which was overlooked by his successors. Briefly speaking, even in a linear system a dissipation process is _not_ fully accounted for by amending phases of the Fourier components. This observation happens to be of relevance in the theory of tidal torques. We shall return to this point in section 9. ## 5 The two sidereal angles Kaula’s construction contains a seemingly redundant fixture, which turns out to be an important and useful acquisition. This is Kaula’s introducing two sidereal angles instead of one. As these angles, $\,\theta\,$ and $\,\theta^{*}\,$, are not orbital elements of the tide-raising and tidally disturbed moons, but are parameters characterising the instantaneous attitude of the planet, it may look strange that Kaula (1964) assumed them to be different entities. To understand his point, let us trace the physical origin of the phase lag. The material of the primary is being deformed by a tidal stress whose spectrum contains an infinite number of frequencies, the reaction of the material to each of these being different. In a linear regime, the strain has the same spectrum, with each harmonic delayed by its own time lag $\Delta t_{\it{l}mpq}\,$. Singer (1968), and later Mignard (1979, 1980), assumed that all $\Delta t_{\it{l}mpq}$ are equal to one another: $\Delta t_{\it{l}mpq}=\Delta t$. If this were true, then in Kaula’s series each argument $\,v_{{\it l}mpq}^{*}-\,m\,\theta^{*}$ would have to be substituted with $\displaystyle v_{{\it l}mpq}^{*^{\;(delayed)}}-\,m\,\theta^{*^{\;(delayed)}}\equiv\,v_{{\it l}mpq}^{*}(t-\Delta t)\,-\,m\,\theta^{*}(t-\Delta t)=\,v_{{\it l}mpq}^{*}(t)-\,m\,\theta^{*}(t)\,-\,\left[\dot{v}_{{\it l}mpq}^{*}-\,m\,\dot{\theta}^{*}\right]\,\Delta t$ $\displaystyle=\;v_{{\it l}mpq}^{*}(t)\;-\,m\,\theta^{*}(t)\,-\;\left[\;({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;\right]\;\Delta t\;\;\;.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (34) In reality, however, the time lag is a function of frequency, for which reason the delays $\,\Delta t_{\it{l}mpq}\,$ will be different for each harmonic involved. This is why the arguments $\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,$ at the moment $\,t\,$ should rather be replaced with $\displaystyle v_{{\it l}mpq}^{*^{\;(delayed)}}\;-\,m\,\theta_{{\it l}mpq}^{*^{\;(delayed)}}\;=~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (35) $\displaystyle v_{{\it l}mpq}^{*}\;-\,m\,\theta^{*}\,-\;\left[\;({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;\right]\;\Delta t_{\it{l}mpq}\;\;\;.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ Specifically, $\displaystyle v_{{\it l}mpq}^{*^{\;(delayed)}}\;=\;v_{{\it l}mpq}^{*}\;-\;\left[\;({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;\dot{\Omega}^{*}\;\right]\;\Delta t_{\it{l}mpq}\;\;\;$ and $\displaystyle\theta_{{\it l}mpq}^{*^{\;(delayed)}}\;=\;\theta^{*}\;-\;\dot{\theta}^{*}\;\Delta t_{\it{l}mpq}\;\;\;,$ $\,\dot{\theta}^{*}\,$ being the planet spin rate. In brief, (35) can be rewritten as $\displaystyle v_{{\it l}mpq}^{*^{\;(delayed)}}\,-\,m\,\theta_{{\it l}mpq}^{*^{\;(delayed)}}\,=\,v_{{\it l}mpq}^{*}\,-\,m\,\theta^{*}\,-\,\omega_{\it{l}mpq}\;\Delta t_{\it{l}mpq}\;\;\;.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ We see that the total phase lags $\,\epsilon_{{\it l}mpq}\,$ introduced by Kaula are given by $\displaystyle\epsilon_{{\it l}mpq}=\left[\,({\it l}-2p)\,\dot{\omega}^{*}\,+\,({\it l}-2p+q)\,\dot{\cal{M}}^{*}\,+\,m\,(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\,\right]\,\Delta t_{\it{l}mpq}=\,\omega^{*}_{\it{l}mpq}\,\Delta t_{\it{l}mpq}=\,\pm\,\chi^{*}_{\it{l}mpq}\,\Delta t_{\it{l}mpq}~{}~{},~{}~{}~{}$ (36) the tidal harmonic $\,\omega^{*}_{\it{l}mpq}\,$ being introduced as $\displaystyle\omega^{*}_{{\it l}mpq}\;\equiv\;({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;~{}~{},~{}~{}~{}$ (37) the positively-defined physical frequency $\displaystyle\chi^{*}_{{\it l}mpq}\,\equiv\,|\,\omega^{*}_{{\it l}mpq}\,|\,=\,|\,({\it l}-2p)\,\dot{\omega}^{*}\,+\,({\it l}-2p+q)\,\dot{\cal{M}}^{*}\,+\,m\,(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;|~{}~{}~{}~{}~{}$ (38) being the actual physical $\,{{\it l}mpq}\,$ tidal frequency excited in the primary’s material. The appropriate positively-defined time delay $\,\Delta t_{\it{l}mpq}\,$ depends on this physical frequency, for which reason the delays $\,\Delta t_{\it{l}mpq}\,$ are, generally, different from one another. 444 When Kaula was developing his theory, the functional form of the dependence $\,\Delta t(\chi)\,$ was not yet known. Reliable data became available only in the final quarter of the past century. See formula (95) below. The sign on the right-hand side of (36) is simply the sign of $\,\omega^{*}_{{\it l}mpq}\,$. The sign evidently depends on whether $\,m\,\dot{\theta}\,$ falls short of or exceeds the linear combination $\,({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;\dot{\Omega}^{*}\,\approx\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,$. The origin and meaning of the phase lag $\,\epsilon_{\it{l}mpq}\,$ being now transparent, one may express the cosine functions in (33) either as $\displaystyle\cos\left[\,\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)-\epsilon_{{\it l}mpq}\,\right]\;$ (39) (where $\,\theta^{*}\,$ and $\,\theta\,$ are identical and cancel one another), or simply as $\displaystyle\cos\left[\,\left(v_{{\it l}mpq}^{*^{\;(delayed)}}\;-\,m\,\theta_{{\it l}mpq}^{*^{\;(delayed)}}\right)-\left(v_{{\it l}mhj}-m\theta\right)\;\right]\;\;\;.$ (40) In (40) we have the delayed siderial angle, $\,\theta_{{\it l}mpq}^{*^{\;(delayed)}}\,$, separated from the actual angle, $\,\theta\,$, by $\;-\,\dot{\theta}\,\Delta t_{{\it l}mpq}\,$, the time lag $\,\Delta t_{\it{l}mpq}\,$ being a function of $\,\chi_{{\it l}mpq}\,\equiv\,|\,\omega_{\it{l}mpq}\,|\,$. ## 6 The Darwin-Kaula-Goldreich expansion for the tidal torque Now we are prepared to calculate the planet-perturbing tidal torque. Since in what follows we shall dwell on the low-inclination case, it will be sufficient to derive the torque’s component orthogonal to the planetary equator: $\displaystyle{\tau}\;=\;-\;{M_{sat}}\;\frac{\partial U(\mbox{{\boldmath$\vec{r}$}})}{\partial\theta}\;\;\;,$ (41) $M_{sat}\,$ being the mass of the tide-disturbed satellite, and the “minus” sign emerging due to our choice not of the astronomical but of the physical sign convention. Adoption of the latter convention implies the emergence of a “minus” sign in the expression for the potential of a point mass: $\;-\,GM/r\,$. This “minus” sign then shows up on the right-hand sides of (6 \- 8) and, later, of (31 \- 33). It is then compensated by the “minus” sign standing in (41). The right way of calculating $\,{\partial U(\mbox{{\boldmath$\vec{r}$}})}/{\partial\theta}\,$ is to take the derivative of (40) with respect to $\,\theta\,$, then to insert (35) into the result, and finally to get rid of the sidereal angle completely, by imposing the constraint $\,\theta^{*}\,=\,\theta\,$. This will yield:555 Formally, one can as well differentiate (39) instead of (40), first ignoring the fact that $\,\theta^{*}\,$ and $\,\theta\,$ are identical and then, after differentiation, permitting them to cancel one another. Though this method produces the same result as the rigorous calculation, it nonetheless remains a formal procedure lacking physics in it. $\displaystyle{\tau}=-\,\sum_{{\it l}=2}^{\infty}k_{\it l}\left(\frac{R}{a}\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}_{sat}\,M_{sat}}{a^{*}}\left(\frac{R}{a^{*}}\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}2m\;\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\sin\left[\,v^{*}_{{\it l}mpq}\,-\;v_{{\it l}mhj}\,-\;\epsilon_{{\it l}mpq}\,\right]~{}~{}_{\textstyle{{}_{\textstyle,}}}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (42) In the case of the tide-raising satellite coinciding with the tide-perturbed one, $\,M_{sat}=M_{sat}^{*}\,$, and all the elements become identical to their counterparts with an asterisk. For a primary body not in a tidal lock with its satellite,666 This caveat is relevant, because in resonances expressions (45 \- 47) will require modifications. For example, the sidereal angle of a satellite tidally locked in a $\,1:1\,$ resonance will be: $\;\,\theta\,=\,\Omega\,+\,\omega\,+\,{\cal M}\,+\,180^{o}\,+\,\alpha\,+\,O(i^{2})\;\,$, letter $\alpha\,$ denoting the librating angle, which is subject to damping and therefore is normally small (less than $\,2"\,$ for the Moon). Inserting the said formula for $\,\theta\,$ into the expression (37) for the tidal harmonic, we obtain, in neglect of $\,\;-m\dot{\alpha}\,$: $\displaystyle\omega^{*}_{{\it{l}}mpq}\;\equiv\;({\it l}-2p-m)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q-m)\;\dot{\cal{M}}^{*}\;\;~{}~{}.~{}~{}~{}$ We now see that, since $\,\theta\,$ is a function of the other angles, different sets of the indices’s values will correspond to one value of the tidal frequency. We shall illustrate this by considering the so-called anomalistic modes $\,\pm\dot{{\cal{M}}}\,$ in the potential. These modes, corresponding to the physical frequency $\,|\dot{{\cal{M}}}|\,$, are given by $\,({\it{l}}mpq)\,=\,(201,\pm 1)~{}$ and also by $\,({\it{l}}mpq)\,=\,(220,\pm 1)\,$. Although the $\,m=0\,$ terms enter the potential, they will not be in the torque, as can be observed by differentiating (31) with respect to $\,\lambda\,$, or by differentiating equation (33) with respect to $\;-\theta\;$, or simply by noticing the presence of the factor $\,m\,$ on the right-hand side of (42). Nonetheless, we see that there exists a pair of $\,m=2\,$ terms, which provides an anomalistic input into the torque. This way, the case of libration deserves a separate consideration, as it is more involved than that of tidal despinning. Specifically, in the case of libration a value of the tidal frequency may correspond to different sets of the indices’ values. the torque (42) can be split into two parts. The first part is constituted by those terms of (42), in which indices $\,(p\,,\,q)\,$ coincide with $\,(h\,,\,j)\,$, and therefore all $\,v_{{\it{l}}mhj}\,$ cancel with $\,v_{{\it l}mpq}^{*}\,$, provided the tidally-perturbed satellite and the tide-raising one are the same body. This component of the torque is, therefore, constant. The rest of the total sum (42) will be denoted with $\,\tilde{\tau}\,$. It is comprised of the terms, in which the pairs $\,(p\,,\,q)\,$ differ from $\,(h\,,\,j)\,$. Accordingly, these terms contain the differences $\displaystyle v^{*}_{{\it{l}}mpq}\,-\;v_{{\it{l}}mhj}\,=~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle({\it{l}}\,-\,2\,p\,+\,q)\;{\cal{M}}^{*}\,-\;({\it{l}}\,-\,2\,h\,+\,j)\,{\cal{M}}\;+\;m\,({\Omega}^{*}\,-{\Omega})\;+\;{\it{l}}\;({\omega}^{*}\,-\,{\omega})\,-\,2\,p\,{\omega}^{*}\,+\,2\,h\,{\omega}~{}~{}~{}.~{}~{}~{}$ (43) When the tidally-perturbed and tide-raising moons are the same body, this becomes $\displaystyle v^{*}_{{\it{l}}mpq}\,-\;v_{{\it{l}}mhj}\,=~{}(2\,h\,-\,2\,p\,+\,q\,-\,j)\;{\cal{M}}\,+\,(2\,h\,-\,2\,p)\,{\omega}~{}~{}~{},~{}~{}~{}$ (44) whence we see that the oscillating component of the torque, $\,\tilde{\tau}\,$, consists of two parts. The part with $\;h\,-\,p\,=\,0\;$ and $\,q\,-\,j\neq\,0\;\,$ consists solely of short-period terms, and it averages out trivially. The mixed-period part of (42), with $\,h\,-\,p\,\neq\,0\;$, consists of both a short-period contribution dependent upon the mean anomaly, and a long-period contribution depending upon the argument of pericentre. All such terms contain multipliers like $\;F_{2mp}({\it i})\,F_{2mh}({\it i})\;$, where $\,h\,\neq\,p\;$, and $\;F_{220}\,=\,3\,+\,O({{\it i}}^{2})\;$, $\;\,F_{210}\,=\,3/2\,\sin{\it i}\,+\,O({{\it i}}^{3})\;$, $\;\,F_{211}\,=\,-\,3/2\,\sin{\it i}\,+\,O({\it i}^{3})\;$, $\;\,F_{221}\,=\,3/2\,\sin^{2}{\it i}\;$, the other relevant $\,F_{2mn}$’s being of higher order than $\,O({{\it i}}^{2})\,$. So the only long-period terms that we have to consider in (42) involve products: $\;F_{210}({\it i})\,F_{211}({\it i})\;$, $\,\;F_{211}({\it i})\,F_{210}({\it i})\;$, $\;F_{220}({\it i})\,F_{221}({\it i})\;$, and $\,\;F_{221}({\it i})\,F_{220}({\it i})\;$. However these products are of order $\,O({{\it i}}^{2})\,$. Thus, while the short-period terms in (42) average out over one rotation period of the moon about the planet, the long-period terms are of order $\,O(e^{2}{\it i}^{2})\,$, the $\,e^{2}\,$ coming from the $\,G_{2pq}\,$ functions. (Indeed, when $\,h\,$ and $\,p\,$ differ by $\,1\,$, then $\,q\,$ and $\,j\,$ must differ by $\,2\,$, to eliminate the mean anomaly, i.e., to make the term long-period and not short-period.) So both the short- and long- period contributions may be neglected in our approximation.777 Had we tried to expand our treatment to higher inclinations, our neglect of the short-period terms would remain legitimate, for they still would average out over one rotation period of the satellite about its primary. As for the long-period terms, it would be tempting to say that these average out over the apsidal- precession period. The latter is much shorter than the time scale of the planetary spin deceleration, a circumstance that may seem a safe justification for the neglect of the long-period terms also for higher inclinations. However, a word of warning would be appropriate here. As well known from Kozai (1959a), who took into account the primary’s nonsphericity, the pericentre of a satellite inclined by about $\,63^{o}\,$ or $\,117^{o}\,$ will neither advance nor retard, at least within the first-order (in $\,J_{2}\,$) perturbation theory. (For a critical review of Kozai’s theory see Taff 1985.) Kozai’s original attempt to introduce corrections owing to $\,J_{3}\,$ and $\,J_{4}\,$ was flawed because in the vicinity of the critical inclinations these terms should be considered not as higher-order but rather as leading. His later analysis demonstrated that at these inclinations the satellite’s perigee should librate about $\,90^{o}\,$ or $\,270^{o}\,$ (Kozai 1962). Under these circumstances, the long-period terms in our expression for the torque will not be averaged out. We however may neglect this possibility, because in the current work we consider only low-inclined moons. Another situation, which we exclude from our treatment, is libration of the satellite’s periapse about $\,90^{o}\,$ or $\,270^{o}\,$, caused by the pull of a third body (the star or some large neighbouring planet). The possibility of such librations may be derived from the presence of the $\,\cos 2\omega\,$ term on the right-hand side of the equation for $\,d\omega/dt\,$ in the theory of Kozai (1959b, 1962) – for an easy introduction into this theory see Innanen et al. (1997), and for its generalisation to finite obliquities see Gurfil et al. (2007). An important special case of the theory of the third-body-caused librations is the one of the satellite getting into a resonance with the third body. An indication that such resonances may cause the satellite’s $\,\omega\,$ librate comes from the mathematically similar theory of Pluto-Neptune resonances (Williams & Benson 1971): being in resonance with Neptune, Pluto has its periapse librating due to a high inclination. (To be exact, the behaviour of Pluto’s periapse is dictated not only by Neptune, but by the combined influence of all of the four gas giant planets. However, this does not change the main point: the outer body or bodies can cause apsidal libration.) In the Solar system, none of the large satellites is so highly perturbed as to have a periapse librating around $\,90^{o}\,$ or $\,270^{o}\,$ due to the above two mechanisms. In theory, though, this remains an option for exoplanets. Either librating mechanism might apply also to satellites of minor planets. In our current paper we do not consider such moons. Thus we arrive at: $\displaystyle{\tau}~{}=\sum_{{\it{l}}=2}^{\infty}2~{}k_{\it l}~{}G~{}M_{sat}^{\textstyle{{}^{2}}}~{}\frac{R^{\textstyle{{}^{2{\it{l}}\,+\,1}}}}{a^{\textstyle{{}^{2\,{\it{l}}\,+\,2}}}}\sum_{m=0}^{\it l}\frac{({\it{l}}\,-\,m)!}{({\it{l}}\,+\,m)!}\;m\;\sum^{\it l}_{p=0}\;F^{\textstyle{{}^{2}}}_{{\it{l}}mp}({\it i})\sum^{\it\infty}_{q=-\infty}G^{\textstyle{{}^{2}}}_{{\it{l}}pq}(e)\;\sin\epsilon_{{\it{l}}mpq}\;+\;\tilde{\tau}\;\;\;,~{}~{}~{}$ (45) the sum standing for the constant (${\cal{M}}$-independent) part of the torque, and $\tilde{\tau}$ denoting the oscillating part whose time-average is zero. As we pointed in the end of section 5, the sign of the phase lag $\,\epsilon_{{\it{l}}mpq}\,$ depends on whether $\,m\,\dot{\theta}\,$ falls short of or exceeds the linear combination $\,({\it{l}}-2p)\;\dot{\omega}^{*}\,+\,({\it{l}}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;\dot{\Omega}^{*}\,\approx\,({\it{l}}-2p+q)\;\dot{\cal{M}}^{*}\,$. Now we also understand that, outside resonances, the $\,{{\it{l}}mpq}\,$ component of the tidal torque experienced by the planet is decelerating if the values of $\,m\,\dot{\theta}\,$ exceed the given combination, and is accelerating otherwise. Expression (45) gets considerably simplified if we restrict ourselves to the case of $\,{\it l}\,=\,2\,$. Since $\,0\,\leq\,m\,\leq\,{\it l}\,$, and since $\,m\,$ enters the expansion as a multiplier, we see that only $\,m\,=\,1\,,\,2\,$ actually matter. As $\,0\,\leq\,p\,\leq\,{\it l}\,$, we are left with only six relevant $\,F$’s, those corresponding to $\;(\it{l}mp)\,=\,$ (210), (211), (212), (220), (221), and (222). By a direct inspection of the table of $\,F_{\it{l}mp}\,$ we find that five of these six functions happen to be $\,O({\it i})\,$ or $\,O(\,{\it i}^{2}\,)\,$, the sixth one being $\,F_{220}\,=\,\frac{\textstyle 3}{\textstyle 4}\,\left(\,1\,+\,\cos{\it i}\,\right)^{2}\,=\,3\,+\,O({\it i}^{2})\,$. Thus, in the leading order of $\,{\it i}\;$, the constant part of the torque reads: $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}~{}=~{}\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M_{sat}^{\textstyle{{}^{2}}}~{}~{}R^{\textstyle{{}^{5}}}\;a^{-6}\;G^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{20\mbox{\it{q}}}}}}}(e)\;k_{{2}}\;\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}\;+\;O({\it i}^{2}/Q)\;\;\;.$ (46) This is what is called Darwin-Kaula-Goldreich torque, or simply Darwin torque. The principal term of this series is $\displaystyle{\tau}_{\textstyle{{}_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}~{}=~{}\frac{3}{2}~{}G\,M_{sat}^{2}~{}k_{2}~{}R^{\textstyle{{}^{5}}}\;a^{{{-\,6}}}\sin\epsilon_{\textstyle{{}_{2200}}}\;\;\;.$ (47) Switching from the lags to quality factors via formula888 The phase lag $\,\epsilon_{\it{l}mpq}\,$ is introduced in (35 \- 36), while the tidal harmonic $\,\omega_{\it{l}mpq}\,$ is given by (37). The quality factor $\,Q_{\it{l}mpq}\,=\,|\,\cot\epsilon_{\it{l}mpq}\,|$ is, for physical reasons, positively defined. Hence the multiplier $\,\mbox{sgn}\,\omega_{\it{l}mpq}\,$ in (49). (As ever, the function $\,\mbox{sgn}(x)\,$ is defined to assume the values $\,+1\,$, $\,-1\,$, or $\,0\,$ for positive, negative, or vanishing $\,x\,$, correspondingly.) Mind that no factor of two appears in (48 \- 49), because $\epsilon$ is a phase lag, not a geometric angle. $\displaystyle Q_{\it{l}mpq}\,=\,|\,\cot\epsilon_{\it{l}mpq}\,|\;\;\;,$ (48) we obtain: $\displaystyle\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{{{{\it{l}}mpq}}}}}}}=\,\sin|\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\;\,\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}=\,\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}\cot^{2}\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=\;\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}Q^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=~{}\frac{~{}\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}~{}}{Q_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}+O(Q^{-3})~{}~{},~{}~{}~{}$ (49) whence $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}~{}=~{}\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M_{sat}^{2}~{}\;R^{\textstyle{{}^{5}}}\;a^{-6}\;G^{\textstyle{{}^{2}}}_{\textstyle{{}_{20\mbox{\it{q}}}}}(e)\;k_{\textstyle{{}_{2}}}\;\frac{~{}\mbox{sgn}\,\omega_{\textstyle{{}_{220\mbox{\it{q}}}}}\,}{Q_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}}\;\,\,+\,O({\it i}^{2}/Q)\,+\,O(Q^{-3})\;\;\;.$ Now, let us simplify the sign multiplier. If in expression (37) for $\omega_{\textstyle{{}_{{\it{l}}mpq}}}$ we get rid of the redundant asterisks, replace999 While in the undisturbed two-body setting $\,{\cal{M}}\,=\,{\cal{M}}_{0}+n\,(t-t_{0})\,$ and $\,\dot{\cal{M}}=n\,$, under perturbation these relations get altered. One possibility is to introduce (following Tisserand 1893) an _osculating mean motion_ $\,n(t)\,\equiv\,\sqrt{\mu/a(t)^{3}}\,$, and to stick to this definition under perturbation. Then the mean anomaly will evolve as $\,\;{\cal{M}}\;=\;{\cal{M}}_{\textstyle{{}_{0}}}(t)\,+\;\int_{\textstyle{{}_{t_{{}_{0}}}}}n(t)\;dt\;$, whence $\;\dot{\cal{M}}=\dot{\cal{M}}_{\textstyle{{}_{0}}}(t)+n(t)\,$. Other possibilities include introducing an _apparent_ mean motion, i.e., defining $\,n\,$ either as the mean-anomaly rate $\,d{\cal M}/dt\,$, or as the mean- longitude rate $\,dL/dt\,=\,d\Omega/dt\,+\,d\omega/dt\,+\,d{\cal{M}}/dt\,$ (as was done by Williams et al. 2001). It should be mentioned in this regard that, while the first-order perturbations in $a(t)$ and in the osculating mean motion $\sqrt{\mu/a(t)^{3}}$ do not have constant parts leading to secular rates, the epoch terms typically do have secular rates. These considerations explain why there exists a difference between the apparent mean motion defined as $dL/dt$ (or as $d{\cal M}/dt$) and the osculating mean motion $\sqrt{\mu/a(t)^{3}}$ . In many practical situations, the secular rate in $\,{\cal{M}}_{0}\,$ is of the order of the periapse rate, while the secular rate in $\,L_{0}\,$ turns out to be smaller. Hence the advantage of defining the apparent $\,n\,$ as the mean-longitude rate $\,dL/dt\,$, rather than as the mean-anomaly rate $\,d{\cal{M}}/dt\,$. (At the same time, for a satellite orbiting an oblate planet the secular rates of $\,M_{0}\,$, $\,L_{0}\,$, and periapse are of the same order.) Although the causes of orbit perturbations are beyond the scope of our paper, we would mention that in the expression (37) for $\omega_{\textstyle{{}_{{\it{l}}mpq}}}$ the notations $\,\dot{\cal{M}}\,$,$\,\dot{\omega}\,$, and $\,\dot{\Omega}\,$ generally imply the _secular_ rate. $\;\dot{\cal{M}}$ with $\dot{\cal{M}}_{0}+n\approx n$, and set ${\it{l}}=m=2$ and $p=0$, the outcome will be: $\displaystyle\mbox{sgn}\,\omega_{\textstyle{{}_{220\mbox{\it{q}}}}}\;=\;\mbox{sgn}\,\left[\,2\;\dot{\omega}\,+\,(2+q)\;n\,+\,2\,\dot{\Omega}-\,2\,\dot{\theta}\,\right]\;=\;\mbox{sgn}\,\left[\,\dot{\omega}\,+\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,+\,\dot{\Omega}-\,\dot{\theta}\,\right]~{}~{}~{}.$ As the node and periapse precessions are slow, the above expression may be simplified to $\displaystyle\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]~{}~{}~{}.$ All in all, the approximation for the constant part of the torque assumes the form: $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}\,=\,\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M_{sat}^{2}~{}\;R^{\textstyle{{}^{5}}}\;a^{-6}\,G^{\textstyle{{}^{2}}}_{\textstyle{{}_{20\mbox{\it{q}}}}}(e)\;k_{\textstyle{{}_{2}}}\;{Q^{\textstyle{{}^{-1}}}_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}}\;~{}\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]+O({\it i}^{2}/Q)+O(Q^{-3})\;\;.~{}~{}~{}$ (50) That the sign of the right-hand side in the above formula is correct can be checked through the following obvious observation: for a sufficiently high spin rate $\,\dot{\theta}\,$ of the planet, the multiplier $\,\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]\,$ becomes negative. Thereby the overall expression for $\,{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}\,$ acquires a “minus” sign, so that the torque points out in the direction of rotation opposite to the direction of increase of the sidereal angle $\,\theta\,$. This is exactly how it should be, because for a fixed $\,q\,$ and a sufficiently fast spin the $\,q$’s component of the tidal torque must be decelerating and driving the planet to synchronous rotation. Expansion (50) was written down for the first time, without proof, by Goldreich & Peale (1966). A schematic proof was later offered by Dobrovolskis (2007). ## 7 The MacDonald expression for the tidal torque The idea of representing the tidal pattern with one bulge is often mis- attributed to MacDonald (1964) who in fact borrowed it from Gerstenkorn (1955). This approach was furthered by Singer (1968) and greatly advanced by Mignard (1979, 1980). It is the latter two authors who realised that Gerstenkorn’s single-bulge simplification was acceptable only with a frequency-independent $\,\Delta t\,$, not with a frequency-independent $\,Q\,$ as in MacDonald (1964). Nevertheless we shall call this approach “the MacDonald torque”, to comply with the established convention. For the same reason, the afore-described Darwin-Kaula-Goldreich expansion will be referred to simply as “the Darwin torque”. In the preceding section, the Darwin torque’s component orthogonal to the equator was conveniently given by the fundamental formula (41). Within the MacDonald approach, it will be more practical to write the torque as a derivative taken with respect to the longitude. The torque acting on the tidally disturbed satellite of mass $\,M_{sat}\,$ is $\;-\,M_{sat}\;\partial U/\partial\lambda$, while the torque that this moon exerts on the planet is this expression’s negative: $\displaystyle\tau(\mbox{{\boldmath$\vec{r}$}})\;=\;M_{sat}\;\frac{\partial U(\mbox{{\boldmath$\vec{r}$}})}{\partial\lambda}~{}~{}~{}.~{}~{}~{}$ (51) Speaking rigorously, the formula furnishes the torque’s component perpendicular to the planetary equator. As can be seen from (56), formula (51) coincides with (41) for low inclinations. ### 7.1 Simplifications available for low _i_ In principle, we can as well insert into $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{r}\,\right)^{{\it l}+1}\;W_{\it{l}}({\mbox{{\boldmath$\vec{R}$}}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ the “raw” expression (14), the one as yet “unprocessed” by (15). This will give us $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;\,-\,{G\;M_{sat}^{*}}\sum_{{\it{l}}=2}^{\infty}k_{\it l}\;\frac{R^{\textstyle{{}^{2\it{l}+1}}}}{r^{\textstyle{{}^{\it{l}+1}}}{r^{\;*}}^{\textstyle{{}^{\it{l}+1}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(\sin\phi)P_{{\it{l}}m}(\sin\phi^{*})\;\cos m(\lambda-\lambda^{*})~{}~{}~{}~{}~{}~{}~{}~{}$ (52) or, for low inclinations of both the tidally-perturbed and tide-raising satellites: $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})=-{GM_{sat}^{*}}\sum_{{\it{l}}=2}^{\infty}k_{\it l}\frac{R^{\textstyle{{}^{2\it{l}+1}}}}{r^{\textstyle{{}^{\it{l}+1}}}{r^{\;*}}^{\textstyle{{}^{\it{l}+1}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(0)P_{{\it{l}}m}(0)\cos m(\lambda-\lambda^{*})~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\left.~{}\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}+O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{}.~{}~{}~{}$ (53) At this point we once again are faced with the question of how to bring damping into the picture, i.e., how to take care of the delayed reaction of the planet’s material to the tidal stress. It is tempting to substitute $m\lambda^{{}^{*}}$ with its delayed value. Then instead of $\cos m(\lambda-\lambda^{*})$ we get $\displaystyle\cos\left(~{}m~{}\lambda~{}-~{}m~{}\lambda^{{}^{*^{\,\textstyle{{}^{(delayed)}}}}}~{}\right)~{}=~{}\cos\left(\;m\;\lambda\;-\;\left[\,m\lambda^{{}^{*}}\,-\,m\stackrel{{\scriptstyle\centerdot}}{{\lambda}}{{}^{{}^{*}}}\Delta t\,\right]\;\right)\;\;\;,$ (54) This trick, suggested by Kaula (1968, page 201),101010 Mind the difference in notations. While in the original paper Kaula (1964) denoted the phase lags with $\,\epsilon_{\textstyle{{}_{\it{l}mpq}}}\,$, in his book Kaula (1968) called them $\,\varphi_{\textstyle{{}_{\it{l}mpq}}}\,$. For the longitudinal lag $\,2\stackrel{{\scriptstyle\centerdot}}{{\lambda}}{{}^{{}^{*}}}\Delta t\,$ emerging in our formula (58), Kaula (1968) used notation $\,2\delta\,$. This way, in the terms used by Kaula (1968) in his book, the geometric angle subtended at the primary’s centre between the directions to the bulge and the moon is called $\,2\,\delta\,$, not $\,\delta\,$ as in most literature. has a physical justification only if $\,\Delta t\,$ is the same for all frequencies, a model pioneered by Singer (1968) and furthered by Mignard (1979, 1980). It can be shown that this model is equivalent to the following rheological law:111111 Combining (36) with the relation $\,Q=1/\tan|\epsilon|\,$, we see that setting all $\,\Delta t_{\textstyle{{}_{\it{l}mpq}}}\,$ equal to the same $\,\Delta t\,$ is equivalent to saying that the quality factor scales as the inverse frequency: $\;Q\,=\,{1}/({\chi\;\Delta t})\;$, provided, of course, that the $Q$ factor is large. As can be seen from (49), a more exact relation will read: $\;\sin(\chi\,\Delta t)\;=\;{1}/{\sqrt{1\,+\,Q^{2}}}\;$, so that $\,\chi\,\Delta t\,=\,Q^{-1}\,+\,O(Q^{-3})\,$. Very special is the case when the values of the quality factor are very low (say, much less than 10). In this situation, the interconnection between the quality factor and the phase lag becomes quite different from the customary formula $\,Q\,=\,\cot\epsilon\,$ . See the Appendix for details. $\displaystyle Q_{\textstyle{{}_{\it{l}mpq}}}\;=\;\frac{1}{\chi_{\textstyle{{}_{\it{l}mpq}}}\,\;\Delta t}\;\;\;.$ (55) Even then, though, it remains unclear how to connect the longitude lag $\,m\stackrel{{\scriptstyle\centerdot}}{{\lambda}}{{}^{{}^{*}}}\Delta t\,\,$ with one or another $\,Q_{\textstyle{{}_{\it{l}mpq}}}\,$, in the spirit of (49). To see what can be done, write down the longitude (reckoned from a fixed meridian on the rotating planet) as $\displaystyle\lambda\;=\;-\;\theta\;+\;\Omega\;+\;\omega\;+\;\nu\;+\;O({\it i}^{2})\;=\;-\;\theta\;+\;\Omega\;+\;\omega\;+\;{\cal{M}}\;+\;2\;e\;\sin{\cal{M}}\;+\;O(e^{2})\;+\;O({\it i}^{2})\;\;\;,\;\;\;\;$ (56) $\nu\,$ being the true anomaly. Thence, in neglect of the nodal and apsidal precessions, the cosine becomes: $\displaystyle\cos\left(\;\left[\,m\;\lambda\;-\;m\lambda^{{}^{*}}\,\right]\;+\,m\stackrel{{\scriptstyle\centerdot}}{{\lambda}}{{}^{{}^{*}}}\Delta t\;\right)\;=\;\cos\left(\;\left[\,m\;\lambda\;-\;m\lambda^{{}^{*}}\,\right]\;+\,m\;\left[\dot{\nu}^{*}\,-\;\dot{\theta}^{*}\right]\;\Delta t\;\right)\;\;\;,\;\;\;\;\;$ (57) or, equivalently: $\displaystyle\cos\left(\left[m\lambda-m\lambda^{{}^{*}}\right]+m\stackrel{{\scriptstyle\centerdot}}{{\lambda}}{{}^{{}^{*}}}\Delta t\right)=\cos\left(\left[m\lambda-m\lambda^{{}^{*}}\right]+m\left[n^{*}-\dot{\theta}^{*}\right]\Delta t+2me^{*}n^{*}\Delta t\,\cos{\cal M}^{*}+O(e^{2})\right)\;\;\;.\;\;\;\;$ (58) Insertion of (57) into (53), along with substitution of $\,r^{*}(t)\,$ by $\,r^{*}(t-\Delta t)\,$, leads us to $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})=-{GM_{sat}^{*}}\sum_{{\it{l}}=2}^{\infty}k_{\it l}\frac{R^{\textstyle{{}^{2\it{l}+1}}}}{r(t)^{\textstyle{{}^{\it{l}+1}}}{r^{{}^{*}}}(t-\Delta t)^{\textstyle{{}^{\it{l}+1}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(0)P_{{\it{l}}m}(0)\,\cos\left(\;m\,\left[\,\lambda-\lambda^{{}^{*}}\,\right]\right.~{}~{}~{}~{}~{}~{}~{}$ (59) $\displaystyle\left.+\;m\,\left[\dot{\nu}^{*}-\dot{\theta}^{*}\right]\,\Delta t\;\right)+O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{}.~{}~{}~{}$ If we take into account only the $\;{\it l}\,=\,2\;$ contribution, expression (53) will simplify to $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})=\;-\;\frac{{G\;M_{sat}^{*}}\,k_{2}\;R^{\textstyle{{}^{5}}}}{r(t)^{\textstyle{{}^{3}}}{r^{{}^{*}}}(t)^{\textstyle{{}^{3}}}}\sum_{m=0}^{2}\frac{(2-m)!}{(2+m)!}(2-\delta_{0m})P_{2m}(0)P_{{2}m}(0)\,\cos m(\lambda-\lambda^{*})+O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{},~{}~{}$ (60) where only the $\,m\,=\,2\,$ term is important.121212 In (60), we may neglect the $\,\lambda$-independent term with $\,m\,=\,0\,$, because our eventual intention is to find the torque by differentiating $\,U(\mbox{{\boldmath$\vec{r}$}})\,$ with respect to $\,\lambda\,$. We may also omit the $\,m\,=\,1\,$ term, because $\;P_{21}(0)\,=\,0\;$. This omission brings up an error of order $\,O({\it i}{\it i}^{*})\,$ into equations (53), (59 \- 62), and (68) In the presence of dissipation, the appropriately simplified version of (60) will read: $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})=\,-\,\frac{3}{4}\;\frac{G\;M_{sat}^{*}\;k_{2}\;R^{\textstyle{{}^{5}}}}{r(t)^{\textstyle{{}^{3}}}{r^{{}^{*}}}(t-\Delta t)^{\textstyle{{}^{3}}}}\;\cos\left(\;\left[\,2\;\lambda\;-\;2\lambda^{{}^{*}}\,\right]\;+\,2\;\left[\dot{\nu}^{*}\,-\;\dot{\theta}^{*}\right]\;\Delta t\;\right)\;+\;O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{},~{}~{}~{}$ (61) while the corresponding expression for the torque exerted by the satellite on the planet will, in this approximation, be given by $\displaystyle\tau(\mbox{{\boldmath$\vec{r}$}})=M_{sat}\frac{\partial U(\mbox{{\boldmath$\vec{r}$}})}{\partial\lambda}=\frac{3\,GM_{sat}^{*}M_{sat}k_{2}R^{\textstyle{{}^{5}}}}{2\,r(t)^{\textstyle{{}^{3}}}{r^{{}^{*}}}(t-\Delta t)^{\textstyle{{}^{3}}}}\;\sin\left(\left[2\lambda-2\lambda^{{}^{*}}\right]+2\left[\dot{\nu}^{*}-\dot{\theta}^{*}\right]\Delta t\right)+O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{}.~{}~{}$ (62) In the case when the tidally disturbed satellite coincides with the tide- raising one, i.e., when $\,\lambda\,=\,\lambda^{*}\,$ and $\,M_{sat}\,=\,M_{sat}^{*}\,$, we obtain: $\displaystyle\tau$ $\displaystyle=$ $\displaystyle\frac{3}{2}\,{G\,M_{sat}^{2}}\;k_{2}\;\frac{R^{\textstyle{{}^{5}}}}{r(t)^{\textstyle{{}^{3}}}{r}(t-\Delta t)^{\textstyle{{}^{3}}}}\,\sin\left(2\left[\dot{\nu}-\,\dot{\theta}\right]\Delta t\right)+O({\it i}^{2}/Q)$ $\displaystyle=$ $\displaystyle\frac{3}{2}\;{G\,M_{sat}^{2}}\;k_{2}\;\frac{R^{\textstyle{{}^{5}}}}{r^{\textstyle{{}^{6}}}}\,\sin\left(2\left[\dot{\nu}-\dot{\theta}\right]\Delta t\right)+O({\it i}^{2}/Q)+O(en/Q^{2}\chi)\;\;,~{}~{}$ where the error $\,O(en/Q^{2}\chi)\,$ emerges when we identify the lagging distance $\,r(t-\Delta t)\,$ with $\,r\equiv r(t)\,$. Replacement of $\,r(t-\Delta t)\,$ with $\,r\,$ is convenient, though not necessary. In subsection 7.2 below, we shall explain that, after averaging over one revolution of the moon about the planet, the error caused by this replacement reduces to $\,O(e^{2}n^{2}/Q^{3}\chi^{2})\,$, which will be less than the largest error. The MacDonald torque (LABEL:43) is equivalent to the Darwin torque (46) with an important proviso that all time lags $\,\Delta t_{\textstyle{{}_{\it{l}mpq}}}\,$ are equal to one another or, equivalently, that the rheological model (55) is accepted. Physically, the special case of equal time lags is exactly the case when the tide may be rigorously interpreted as one double bulge of a variable rate and amplitude.131313 An attempt to generalise this simplified approach to arbitrary inclinations was undertaken by Efroimsky (2006). While for constant time lags that generalisation is likely to be acceptable, it remains to be explored whether it offers a practical approximation for actual rheologies (72). Mathematically, this model enables one to wrap up the infinite series (46) into the elegant finite form (LABEL:43). Formally, this wrapping can be described like this: expression (LABEL:43) mimics the principal term of the series (46), provided in this term the multiplier $\,G^{2}_{200}\,$ is replaced with unity, $\,a\,$ is replaced with $\,r\,$, and the principal phase lag $\displaystyle\epsilon_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\equiv\,2\,(n\,-\,\dot{\theta})\Delta t$ (64) is replaced with the longitudinal lag or, possibly better to say, with the quasi-phase $\epsilon\,\equiv\,2\,(\dot{\nu}\,-\,\dot{\theta})\,\Delta t\;\;\;.$ (65) Thus we see that within the MacDonald one-variable-bulge formalism the longitudinal lag (65) is acting as an _instantaneous_ phase lag associated with the _instantaneous_ tidal frequency $\,\chi\,\equiv\,2\,|\dot{\nu}\,-\,\dot{\theta}|\,$. This is why we may call it simply $\,\epsilon\,$, without a subscript. Evidently, $\,\epsilon\,$ is (up to a sign) twice the geometrical angle subtended at the primary’s centre between the directions to the moon and to the bulge.141414 As the subtended angle is $\;|\,(\dot{\nu}\,-\,\dot{\theta})\,\Delta t\,|\;$, its double is equal to the absolute value of $\,\epsilon\,$, and not to that of $\,\epsilon_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}\,=\,2\,(n-\dot{\theta})\,\Delta t\,$. The geometric meaning of the longitudinal lag being clear, let us consider its physical meaning, in the sense of this lag’s relation to the dissipation rate. For some fixed frequency $\,\chi_{{\it l}mpq}\,$, the corresponding phase lag $\,\epsilon_{{\it l}mpq}\,$ is related to the appropriate quality factor via $\,1/Q_{{\it l}mpq}\,=\,\tan|\epsilon_{{\it l}mpq}|\,$. To keep the analogy between the true lags and the instantaneous lag (65), one may conveniently _define_ a quantity $\,Q\,$ as the inverse of $\,\tan|\epsilon|\,$. This will enable one to express the MacDonald torque as $\displaystyle\tau~{}=~{}\frac{3}{2}~{}GM_{sat}^{2}\;k_{2}\frac{R^{\textstyle{{}^{5}}}}{r(t)^{\textstyle{{}^{3}}}{r}(t-\Delta t)^{\textstyle{{}^{3}}}}\sin\epsilon+O({\it i}^{2}/Q)~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (66) $\displaystyle=~{}\frac{3}{2}~{}{GM_{sat}^{2}}\;k_{2}\frac{R^{\textstyle{{}^{5}}}}{r^{\textstyle{{}^{6}}}}Q^{-1}\,\mbox{sgn}(\dot{\nu}-\dot{\theta})+O({\it i}^{2}/Q)+O(en/Q^{2}\chi)+O(Q^{-3})~{}~{}~{}.~{}~{}~{}~{}~{}~{}~{}~{}$ Since $\,Q\,$ was _defined_ as $\,1/\tan|\epsilon|\,$, it is not guaranteed to deserve the name of an overall quality factor. At each particular frequency $\,\chi_{\textstyle{{}_{{\it l}mpq}}}\,$, the corresponding quality factor $\,Q_{{\it l}mpq}\,\equiv\,1/\tan|\epsilon_{{\it l}mpq}|\,$ is related to the peak energy of this mode, $\,E_{peak}(\chi_{\textstyle{{}_{{\it l}mpq}}})\,$, and to the one-cycle energy loss at this frequency, $\,\Delta E_{cycle}(\chi_{\textstyle{{}_{{\it l}mpq}}})\,$, via $\displaystyle\Delta E_{cycle}(\chi_{\textstyle{{}_{{\it l}mpq}}})\;=\;-\;\frac{2\pi E_{peak}(\chi_{\textstyle{{}_{{\it l}mpq}}})}{Q_{\textstyle{{}_{{\it l}mpq}}}}\;\;\;.$ (67) However, it is not at all obvious if the quantity $\,Q\,$ defined through the longitudinal lag as $\,Q\,\equiv\,1/\tan|\epsilon|\,$ interconnects the overall tidal energy with the overall one-cycle loss, in a manner similar to (67). The literature hitherto has always taken for granted that it does. However, the proof (to be presented elsewhere) requires some effort. The proof is based on interpreting $\,\chi\,\equiv\,2\,|\dot{\nu}-\dot{\theta}|\,$ as an _instantaneous_ tidal frequency. The interconnection between $\,Q\,\equiv\,1/\tan|\epsilon|\,$ and the overall energy-damping rate mimics (67) only up to a relative error of order $\,O(en/Q\chi)\,=\,O(en\,\Delta t)\,$, i.e., up to an absolute error of order $\,O(eQ^{-1}n\,\Delta t)\,$. This is acceptable, because in realistic settings $\;n\,\Delta t\,\ll\,1\;$. ### 7.2 Further simplifications available in the zeroth order of _en/Q $\chi$_ Suppose we ignore the difference between $\,r\,$ and $\,r^{*}$, which are the two locations of the same satellite, separated by the time lag owing to the tidal response. We shall now demonstrate that, though the relative error of this approximations is $\,O(en/Q\chi)\,$, after averaging over a satellite period this approximation brings only a $\,O(e^{2}n^{2}/Q^{2}\chi^{2})\,$ relative error into the expression for the torque. From the well-known formulae $\,r\,=\,a\,(1\,-\,e^{2})/(1\,+\,e\,\cos\nu)\,$ and $\,\partial\nu/\partial M\,=\,(1\,+\,e\,\cos\nu)^{2}/(1\,-\,e^{2})^{3/2}\,$ we see that $\displaystyle\Delta r\equiv r(t)-r(t-\Delta t)=-\frac{a\,e\,(1\,-\,e^{2})}{(1\,+\,e\;\cos\nu)^{2}}\;\sin\nu\;\Delta\nu\,+\,O\left(e\,(\Delta\nu)^{2}\,\right)\,=\,-\,\frac{a\,e\;\sin\nu}{(1-e^{2})^{1/2}}\,n\,\Delta t\,+\,O\left(e\,(n\;\Delta t)^{2}\,\right)~{}~{}~{}.$ The time lag is interconnected with the phase shift and the quality factor via the relations $\displaystyle\chi\;\Delta t\;=\;\epsilon\;\approx\;Q^{-1}\;\;\;,$ $\,\chi\,=\,2\,|\dot{\theta}\,-\,\dot{\nu}\,|\,$ being the instantaneous tidal frequency. Hence $\displaystyle\Delta r\,\equiv\,r(t)\,-\,r(t-\Delta t)\,\approx\;-\;a\;\frac{e}{Q}\;\frac{n}{\chi}\;\sin\nu~{}~{}~{}.$ As $\,\Delta r\,$ is proportional to $\,\sin\nu$, only terms quadratic in $\,\Delta r\,$ survive averaging. Thus, while in $\displaystyle U\,=\,-\,\frac{3}{4}\,{G\,M_{sat}^{*}}\,k_{2}\,\frac{R^{\textstyle{{}^{5}}}}{r^{6}}\,\cos(2\lambda-2\lambda^{*}+\epsilon)+O(en/Q\chi)+O({\it i}^{2})+O({{\it i}^{*}}^{2})+O({\it i}{\it i}^{*})~{}~{},~{}~{}~{}$ (68) and $\displaystyle\tau~{}=~{}\frac{3}{2}\,{G\,M_{sat}\,M_{sat}^{*}}\,k_{2}\,\frac{R^{\textstyle{{}^{5}}}}{r^{6}}\,\sin(2\lambda-2\lambda^{*}+\epsilon)\;+\,O(en/Q\chi)\,+\,O({\it i}^{2})\,~{}~{},~{}~{}~{}$ (69) the _relative_ error is $\,O(en/Q\chi)+O({\it i}^{2})\,$, in the averaged expression151515 We recall that time averages over one revolution of the satellite about the primary are given by $\displaystyle\langle\;\,.\,.\,.\,\;\rangle\;\equiv\;\frac{1}{2\;\pi}\;\int_{0}^{2\pi}\;.\,.\,.\;\;\;d{\cal{M}}\;=\;\frac{\left(1\;-\;e^{2}\right)^{3/2}}{2\;\pi}\;\int_{0}^{2\pi}\;.\,.\,.\;\;\;\frac{d\nu}{\left(1\;+\;e\;\cos\nu\right)^{2}}\;\;\;\;,$ while the planetocentric distance is $\,r=a\left(1-e^{2}\right)/\left(1+e\,\cos\nu\right)\,$, with $\nu$ being the true anomaly. This way, $\displaystyle\langle\;\,\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\,\sin\epsilon\,\;\rangle\;=\;\frac{\left(1\;-\;e^{2}\right)^{3/2}}{2\;\pi}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\;\sin\epsilon\,\;\frac{d\nu}{\left(1\;+\;e\;\cos\nu\right)^{2}}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle=\;\frac{\left(1\;-\;e^{2}\right)^{3/2}}{2\;\pi}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\;\sin\epsilon\,\;\frac{r^{2}\;\;d\nu}{a^{2}\;\left(1\;-\;e^{2}\right)^{2}}\;=\;\frac{R^{\textstyle{{}^{2}}}}{a^{2}}\;\frac{1}{2\,\pi\,\;\left(1\;-\;e^{2}\right)^{1/2}\,}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{4}}}}{r^{4}}\;\;\sin\epsilon\,\;{d\nu}\;\;\;\;.$ $\displaystyle\langle\,\tau\,\rangle~{}=~{}-~{}\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\,k_{2}}{2\;R}\;\;\langle\;\,\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\,\sin\epsilon\,\;\rangle~{}\,+\,O(e^{2}n^{2}/Q^{3}\chi^{2})\,+\,O({\it i}^{2}/Q)~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (70a) $\displaystyle=\;-~{}\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\,k_{2}\,R}{4\;\pi\;a^{2}}\;\,\frac{1}{\,\left(1\;-\;e^{2}\right)^{1/2}\,}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{4}}}}{r^{4}}\;\,\sin\epsilon\;\;{d\nu}\,\,+\,O(e^{2}n^{2}/Q^{3}\chi^{2})\,+\,O({\it i}^{2}/Q)\;~{}~{}.~{}~{}~{}~{}~{}~{}~{}~{}$ (70b) it is only $\,O(e^{2}n^{2}/Q^{2}\chi^{2})+O({\it i}^{2})\,$. In the above expressions, we asserted after the differentiation that $\,M_{sat}^{*}\,=\,M_{sat}\,$ and $\,\lambda^{*}\,=\,\lambda\,$, implying that the tide-generating and tidally-perturbed moons are one and the same body. As soon as $\,\lambda\,$ is set to be equal to $\,\lambda^{*}\,$, the sine function in (70) becomes $\,\sin\epsilon\,\approx\,1/Q\,$. So, while the _relative_ error in (70) is $\;O(e^{2}n^{2}/Q^{2}\chi^{2})\,+\,O({\it i}^{2})\;$, the _absolute_ error becomes $\;O(e^{2}n^{2}/Q^{3}\chi^{2})\,+\,O({\it i}^{2}/Q)\;$. The error $\;O(e^{2}n^{2}/Q^{3}\chi^{2})\;$ becomes irrelevant for two reasons. First, our substitution of $\,\sin\epsilon\,$ with $\,\tan\epsilon\,=\,1/Q\,$ generates a relative error $\,O(Q^{-2})\,$, i.e., an absolute error $\,O(Q^{-3})\,$. Second, as explained in the end of subsection 7.1, the uncertainties inherent in our definition of the overall quality factor $\,Q\,$ entail an absolute error $\;O(en/Q^{2}\chi)\;$. Each of these two errors exceeds $\;O(e^{2}n^{2}/Q^{3}\chi^{2})\;$. We can then write: $\displaystyle\langle\,\tau\,\rangle~{}=~{}-~{}\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\,k_{2}}{2\;R}\;\;\langle\;\;\frac{\mbox{sgn}(\dot{\theta}\,-\;\dot{\nu})}{Q}\,\;\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\;\rangle~{}\,+\,O(Q^{-3})\,+\,O({\it i}^{2}/Q)\,+\,O(en/Q^{2}\chi)~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (71a) $\displaystyle=\,-\,\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\,k_{2}\,R}{4\;\pi\;a^{2}}\;\,\frac{1}{\,\left(1\;-\;e^{2}\right)^{1/2}\,}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{4}}}}{r^{4}}\;\,\frac{\mbox{sgn}(\dot{\theta}\,-\;\dot{\nu})}{Q}\,\;{d\nu}\,\,+\,O(Q^{-3})\,+\,O({\it i}^{2}/Q)\,+\,O(en/Q^{2}\chi)\;~{}.~{}~{}~{}~{}$ (71b) ## 8 Use and abuse of approximation (68 \- 71) Just as with the formula (68) for the potential, the elegant expression (69) for the torque remain correct only to the zeroth order in $\,e/Q\,$, while (71) is valid to the first order. This is the reason why the convenience of this approximation and of its corollaria is somewhat deceptive. Nevertheless, the (68) - (71) were employed by many an author. Goldreich & Peale (1966) used them to build a theory containing terms up to $\,e^{7}\,$. Now we see that some coefficients in their theory of capture into resonances must be reconsidered. The same pertains to some coefficients in the theory of Mercury’s rotation, recently offered by Peale (2005). Fortunately, the key conclusions of Peale (2005) stay unaltered, despite the corrections needed in the said coefficients.161616 Stan Peale, private communication, 2007. Interestingly, Kaula (1968) fell into this temptation, and so did Goldreich (1966b). Equation (4.5.29) in Kaula (1968), as well as equation (15) in Goldreich (1966b), is but the above formula (71) with the inverse quality factor taken out of the integral: $\displaystyle\tau^{Kaula}\;=\;-~{}\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\,k_{2}\,R}{4\;\pi\;a^{2}\;Q}\;\,\frac{1}{\,\left(1\;-\;e^{2}\right)^{1/2}\,}\;\int_{0}^{2\pi}\;\frac{R^{\textstyle{{}^{4}}}}{r^{4}}\;\,{\mbox{sgn}(\dot{\theta}\,-\,\,\dot{\nu})}\,\;{d\nu}\;~{}~{}.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}(\mbox{Kaula 1968, eqn$\,$4.5.29})$ Besides the afore-mentioned fact that this approach contains a relative error $\,O(en/Q\chi)\,+\,O(Q^{-2})+O({\it i}^{2})\,$, it suffers a greater defect. Taking $Q^{\textstyle{{}^{-1}}}$ out of the integral is illegitimate, because it implies frequency-independence of $\,Q\,$. This is then incompatible with Kaula’s implicit assumption of a constant $\,\Delta t\,$, an assumption tacitly present in (54).171717 This would be incompatible with the MacDonald (1964) treatment as well, because MacDonald’s formalism necessitates the rheology $\,Q\sim 1/\chi\,$. We shall return to this point in subsection 10.1. Goldreich (1966b) and Kaula (1968) used this oversimplified formula to investigate librations of a satellite trapped in a 1:1 resonance. Other authors used it to evaluate despinning rates of bodies outside this resonance. We shall dwell on the latter case in section 10. ## 9 Can the quality factor scale as a positive power of the tidal frequency? As of now, the functional form of the dependence $\,Q(\chi)\,$ for Jovian planets remains unknown. For terrestrial planets, the model $\,Q\,\sim\,1/\chi\,$ is definitely incompatible with the geophysical data. A convincing volume of measurements firmly witnesses that $\,Q\,$ of the mantle scales as the tidal frequency to a _positive_ fractional power: $\displaystyle Q\;=\;\left(\,{\cal E}\,\chi\,\right)^{\textstyle{{}^{\alpha}}}\;\;\;,~{}~{}~{}~{}~{}\mbox{where}\;\;\;\alpha\,=\,0.3\,\pm\,0.1\;\;\;,$ (72) ${\cal E}\,$ being an integral rheological parameter with dimensions of time. This rheology is incompatible with the postulate of frequency-independent time-delay. Therefore, insertion of the realistic model (72) into the formula presented in section 8 will remain insufficient. An honest calculation should be based on averaging the Darwin-Kaula-Goldreich formula (50), with the actual scaling law (72) inserted therein, and with the appropriate dependence $\,\Delta t_{\textstyle{{}_{lmpq}}}(\,\chi_{\textstyle{{}_{lmpq}}}\,)\,$ taken into account (see formula (95) below). ### 9.1 The “paradox” Although among geophysicists the scaling law (72) has long become common knowledge, in the astronomical community it is often met with prejudice. The prejudice stems from the fact that, in the expression for the torque, $\,Q\,$ stands in the denominator: $\displaystyle\tau\;\sim\;\frac{1}{Q}\;\;\;\,.\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (73) At the instant of crossing the synchronous orbit, the principal tidal frequency $\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,$ becomes nil, for which reason insertion of $\displaystyle Q\;\sim\;\chi^{\alpha}\;\;\;,~{}~{}~{}~{}~{}\alpha\,>\,0\;\;\;$ (74) into (73) seems to entail an infinitely large torque at the instant of crossing: $\displaystyle\tau\;\sim\;\frac{1}{Q}\;\sim\;\frac{1}{\chi^{\alpha}}\;\rightarrow\;\infty~{}~{},~{}~{}~{}~{}~{}{\mbox{for}}~{}~{}~{}\chi\;\rightarrow\;0~{}~{}~{},$ (75) a clearly unphysical result. Another, very similar objection to (72) originates from the fact that the quality factor is inversely proportional to the phase shift: $\,Q\,\sim\,1/\epsilon\,$. As the shift (36) vanishes on crossing the synchronous orbit, one may think that the value of the quality factor must, effectively, approach infinity. On the other hand, the principal tidal frequency vanishes on crossing the synchronous orbit, for which reason (72) makes the quality factor vanish. Thus we come to a contradiction. For these reasons, the long-entrenched opinion is that these models introduce discontinuities into the equations and can thus be considered as unrealistic approximations for rotating bodies. It is indeed true that, while law (72) works over scales shorter than the Maxwell time (about $\,10^{2}$ yr for most minerals), it remains subject to discussion in regard to longer timescales. Nonetheless, it should be clearly emphasised that the infinities emerging at the synchronous-orbit crossing can in no way disprove any kind of rheological model. They can only disprove the flawed mathematics whence they provene. ### 9.2 A case for reasonable doubt To evaluate the physical merit of the alleged infinite-torque “paradox”, recall the definition of the quality factor. As part and parcel of the linearity approximation, the overall damping inside a body is expanded in a sum of attenuation rates corresponding to each periodic disturbance: $\displaystyle\langle\,\dot{E}\;\rangle\;=\;\sum_{i}\;\langle\,\dot{E}(\chi_{\textstyle{{}_{i}}})\;\rangle$ (76) where, at each frequency $\,\chi_{i}\,$, $\displaystyle\langle\,\dot{E}(\chi_{\textstyle{{}_{i}}})~{}\rangle~{}=~{}-~{}2~{}\chi_{\textstyle{{}_{i}}}~{}\frac{\,\langle\,E(\chi_{\textstyle{{}_{i}}})~{}\rangle\,}{Q(\chi_{\textstyle{{}_{i}}})}~{}=\,\;-\;\chi_{\textstyle{{}_{i}}}\;\frac{\,E_{{}_{peak}}(\chi_{\textstyle{{}_{i}}})\,}{Q(\chi_{\textstyle{{}_{i}}})}\;\;\;,$ (77) $\langle\,.\,.\,.\,\rangle~{}$ designating an average over a flexure cycle, $\,E(\chi_{\textstyle{{}_{i}}})\,$ denoting the energy of deformation at the frequency $\,\chi_{\textstyle{{}_{i}}}\,$, and $Q(\chi_{\textstyle{{}_{i}}})\,$ being the quality factor of the medium at this frequency. This definition by itself leaves enough room for doubt in the above “paradox”. As can be seen from (77), the dissipation rate is proportional not to $\;1/Q(\chi)\;$ but to $\;\chi/Q(\chi)\;$. This way, for the dependence $\,Q\,\sim\,\chi^{\alpha}\,$, the dissipation rate $\,\langle\dot{E}\rangle\,$ will behave as $\,\chi^{1-\alpha}\,\;$. In the limit of $\,\chi\,\rightarrow\,0\,$, this scaling law portends no visible difficulties, at least for the values of $\,\alpha\,$ up to unity. While raising $\,\alpha\,$ above unity may indeed be problematic, there seem to be no fundamental obstacle to having materials with positive $\,\alpha\,$ taking values up to unity. So far, such values of $\,\alpha\,$ have caused no paradoxes, and there seems to be no reason for any infinities to show up. ### 9.3 The phase shift and the quality factor As another preparatory step, we recall that, rigorously speaking, the torque is proportional not to the phase shift $\,\epsilon\,$ itself but to $\,\sin\epsilon\,$. From (49) and (72) we obtain: $\displaystyle|\,\sin\epsilon\,|\;=\;\frac{1}{\textstyle\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}}}\;=\;\frac{1}{\sqrt{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}\;}}\;\;\;.$ (78) We see that only for large values of $\,Q\,$ one can approximate $\;\,|\,\sin\epsilon\,|\;\,$ with $\,1/Q\,$ (crossing of the synchronous orbit _not_ being the case). Generally, in any expression for the torque, the factor $\,1/Q\,$ must always be replaced with $\,1/\sqrt{1\,+\,Q^{2}}\,\;$. Thus instead of (73) we must write: $\displaystyle\tau\;\sim\;|\,\sin\epsilon\,|\;=\;\frac{1}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;=\;\frac{1}{\sqrt{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}\;}}\;\;\;\;\,,\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (79) ${\cal E}\,$ being a dimensional constant from (72). Although this immediately spares us from the fake infinities at $\chi\rightarrow 0$, we still are facing a strange situation: it follows from (78) that, for a positive $\alpha$ and vanishing $\chi$, the phase lag $\epsilon$ must be approaching $\pi/2$, thereby inflating the torque to its maximal value (while on physical grounds the torque should vanish for zero $\chi$). Evidently, some important details are still missing from the picture. ### 9.4 The stone rejected by the builders To find the missing link, recall that Kaula (1964) described tidal damping by employing the method suggested by Darwin (1880): he accounted for attenuation by merely adding a phase shift to every harmonic involved – an empirical approach intended to make up for the lack of a consistent hydrodynamical treatment with viscosity included. It should be said, however, that prior to the work of 1880 Darwin had published a less known article (Darwin 1879), in which he attempted to construct a self-consistent theory, one based on the viscosity factor of the mantle, and not on empirical phase shifts inserted by hand. Darwin’s conclusions of 1879 were summarised and explained in a more general mathematical setting by Alexander (1973). The pivotal result of the self-consistent hydrodynamical study is the following. When a variation of the potential of a tidally distorted planet, $\,U(\mbox{{\boldmath$\vec{r}$}})\,$, is expanded over the Legendre functions $\,P_{{\it{l}}m}(\sin\phi)\,$, each term of this expansion will acquire not only a phase lag but also a factor describing a change in amplitude. This forgotten factor, derived by Darwin (1879), is nothing else but $\;\,\cos\epsilon\;$. Its emergence should in no way be surprising if we recall that the damped, forced harmonic oscillator $\displaystyle\ddot{x}\;+\;2\;\gamma\;\dot{x}\;+\;\omega^{2}_{o}\,x\;=\;F\;e^{{\it i}\,\lambda\,t}$ (80) evolves as $\displaystyle x(t)~{}=~{}C_{1}~{}\,e^{\textstyle{{}^{(\,-\,\gamma\,+\,{\it i}\,\sqrt{\omega_{o}^{2}-\gamma^{2}\,}\,)\;t}}}\,+\;C_{2}~{}\,e^{\textstyle{{}^{(\,-\,\gamma\,-\,{\it i}\,\sqrt{\omega_{o}^{2}-\gamma^{2}\,}\,)~{}t}}}\,+~{}\frac{F~{}\cos\epsilon}{\omega_{o}^{2}\,-\,\lambda^{2}}~{}\,e^{\textstyle{{}^{{\it i}\,(\lambda\,t\,-\,\epsilon)}}}\;\;\;,$ (81) where the phase lag is $\displaystyle\tan\epsilon\;=\;\frac{2\;\gamma\;\lambda}{\left(\,\omega_{o}^{2}\,-\;\lambda^{2}\,\right)}\;\;\;,$ (82) and the first two terms in (81) are damped away in time.181818 As demonstrated by Alexander (1973), this example indeed has relevance to the hydrodynamical theory of Darwin, and is not a mere illustration. Alexander (1973) also explained that the emergence of the $\,\cos\epsilon\,$ factor is generic. (Darwin (1879) had obtained it in the simple case of $\,{\it l}\,=\,2\,$ and for a special value of the Love number: $\,k{\it{{}_{2}}}=\,1.5\,$.) A further investigation of this issue was undertaken in a comprehensive work by Churkin (1998), which unfortunately has never been published in English because of a tragic death of its Author. In this preprint, Churkin explored the frequency- dependence of both the Love number $\,k_{2}\,$ and the quality factor within a broad variety of rheological models, including those of Maxwell and Voight. It follows from Churkin’s formulae that within the Voigt model the dynamical $\,k_{2}\,$ relates to the static one as $\,\cos\epsilon\,$. In the Maxwell and other models, the ratio approaches $\,\cos\epsilon\,$ in the low-frequency limit. In the works by Darwin’s successors, the allegedly irrelevant factor of $\,\cos\epsilon\,$ fell through the cracks, because the lag was always asserted to be small. In reality, though, each term in the Fourier expansions (33), (42 \- 47), and (50) should be amended with $\,\cos\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,$. Likewise, the correct versions of (62 \- LABEL:43) and (66) should contain an extra factor of $\,\cos\epsilon_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,$. For the same reason, instead of (79), we should write down: $\displaystyle\tau\;\sim\;|\,\cos\epsilon\;\,\sin\epsilon\,|\;=\;\frac{Q}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;\frac{1}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;=\;\frac{{\cal E}^{\textstyle{{}^{\alpha}}}\;{\chi}^{\textstyle{{}^{\alpha}}}}{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}}\;\;\;\;\,.\,~{}$ (83) At this point, it would be tempting to conclude that, since (71) vanishes in the limit of $\chi\rightarrow 0\;$, _for any sign_ _of_ $\alpha\;$, then no paradoxes happen on the satellite’s crossing the synchronous orbit. Sadly, this straightforward logic would be too simplistic. In fact, prior to saying that $\,\cos\epsilon\,\sin\epsilon\rightarrow 0$, we must take into consideration one more subtlety missed so far. As demonstrated in the Appendix, taking the limit of $Q\rightarrow 0$ is a nontrivial procedure, because at small values of $\,Q\,$ the interconnection between the lag and the Q factor becomes very different from the conventional $Q=\cot|\epsilon|$. A laborious calculation shows that, for $\;Q\,<\,1-\pi/4\,$, the relation becomes: $\displaystyle|\,\sin\epsilon\,\cos\epsilon\,|\,=\;(3Q)^{1/3}\,\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;\;\;,$ which indeed vanishes for $Q\rightarrow 0$. Both $\,\epsilon_{\textstyle{{}_{2200}}}\,$ and the appropriate component of the torque change their sign on the satellite crossing the synchronous orbit. So the main conclusion remains in force: nothing wrong happens on crossing the synchronous orbit, Q.E.D. ## 10 Tidal despinning. The following formula for the average deceleration rate $\,\ddot{\theta}\,$ of a planet due to a tide-raising satellite has often appeared in the literature: $\displaystyle\langle\;\ddot{\theta}\;\,\rangle\;=\;-\;{\cal K}\;\left[\;\dot{\theta}\;\,{\cal A}(e)\;-\;n\;{\cal N}(e)\;\right]\;\;\;,$ (84) where $\displaystyle{\cal A}(e)\;=\;\left(\,1\;+\;3\;e^{2}\;+\;\frac{3}{8}\;e^{4}\,\right)\;\left(\,1\,-\,e^{2}\,\right)^{-9/2}~{}~{}~{},~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (85) and $\displaystyle{\cal N}(e)\;=\;\left(\,1\;+\;\frac{15}{2}\;e^{2}\;+\;\frac{45}{8}\;e^{4}\;+\;\frac{5}{16}\;e^{6}\,\right)\;\left(\,1\,-\,e^{2}\,\right)^{-6}~{}~{}~{},$ (86) $\theta\,$ being the sidereal angle, $\,\dot{\theta}\,$ being the primary’s spin rate, $\,{\cal K}\,$ being some constant, and the angular brackets designating an average over one revolution of the secondary about the primary. This expression was derived by different methods in Goldreich & Peale (1966) and Hut (1981), and was later employed by Dobrovolskis (1995, 2007) and Correia & Laskar (2004)191919 Our formula (78) differs from formula (4) in Correia and Laskar (2004) by a factor of $\,n/{\chi}\,$, because in Ibid. the quality factor had been introduced as $\,1/(n\,\Delta t)\,$ and not as $\,1/(\chi\,\Delta t)\,$.. ### 10.1 Derivation by means of the MacDonald torque The following proof of (84 \- 86) is implied in Goldreich & Peale (1966) and is presented in more detail in Dobrovolskis (2007). Their starting point was the MacDonald torque (71). Hut (1981), who approached the issue in the language of the Lagrange-type planetary equations, took into account, in the disturbing function, only the leading term of series (33), and substituted the principal tidal frequency $\,\chi_{\textstyle{{}_{2200}}}\,=\,2\,|\dot{\theta}\,-\,n|\,$ with the synodal frequency $\,\chi\,=\,2\,|\dot{\theta}\,-\,\dot{\nu}|\,$. Thereby, his approach was equivalent to that of Goldreich & Peale (1966) and Dobrovolskis (2007). Although not necessarily assumed by these authors,202020 It should be mentioned that the original treatment by MacDonald (1964) is inherently contradictory. On the one hand, MacDonald postulates (following Gerstenkorn 1958) that there exists one overall double bulge. As explained in subsection 7.1 above, this assertion unavoidably implies constancy of the time lag $\,\Delta t\,$, so that $\,Q\sim 1/\chi\,$ and $\,\epsilon\sim\chi\,$. However, MacDonald (1964) erroneously set $\,Q\,$ (and, thence, also $\,\epsilon\,$) frequency-independent, an assertion incompatible with his and Gerstenkorn’s postulate of existence of an overall double bulge. Whenever in the current paper we refer to MacDonald’s torque, we always imply his postulate that one double bulge exists. At the same time, to make the MacDonald-Gerstenkorn treatment consistent, we always adjust the MacDonald- Gerstenkorn treatment by letting $\,Q\,$ and $\,\epsilon\,$ scale as $\,1/\chi\,$ and $\,\chi\,$, correspondingly. their method, as we saw in the section above, inherently implied the following assertions: (I) The quantity $\,\chi\,=\,2\,|\dot{\theta}\,-\,\dot{\nu}|\,$ is treated as an instantaneous tidal frequency. Accordingly, the overall quality factor $\,Q\,$ is implied to be a function not of the principal frequency $\,\chi_{\textstyle{{}_{2200}}}\,$ but of the instantaneous frequency $\,\chi\,$. (II) The functional form of this dependence is chosen as $\;Q\,=\,{(\Delta t)^{-1}\;\chi^{-1}}\;$, where $\Delta t\,$ is the time lag. (III) The time lag $\,\Delta t\,$ is frequency-independent. This assertion is equivalent to (II), as can be demonstrated from (36). Beside this, those authors neglected the order-$en/Q\chi\,$ difference between $\,r\,$ and $\,r^{*}$ in (66), generating a relative error in $\,\tau\,$ of order $\,O(en/Q\chi)\,$ (which, luckily, reduced to $\,O(e^{2}n^{2}/Q^{2}\chi^{2})\,$ after orbital averaging). They also substituted $\,\sin\epsilon\,$ with $\,1/Q\,$, causing a relative error of order $\,O(1/Q^{2})\,$, because in reality $\,Q\,$ is the reciprocal of $\,\tan\epsilon\,$, not of $\,\sin\epsilon\,$. Assertion (II) can be written down in more generic notation: $\displaystyle Q\;=\;\left(\,{\cal E}\,\chi\,\right)^{\textstyle{{}^{\alpha}}}~{}~{}~{},~{}~{}~{}\mbox{with}~{}~{}~{}\alpha\;=\;-\;1~{}~{}~{}.$ (87) This form of the scaling law is more convenient, for it leaves one an opportunity to switch to different values of $\,\alpha\,$. For any value of $\,\alpha\,$ (not only for $\,-1\,$), the constant ${\cal E}\,$ is an integral rheological parameter (with the dimension of time), whose physical meaning is explained in Efroimsky & Lainey (2007). It can be shown that in the particular case of $\;\alpha\,=\,-\,1\;$ the parameter $\,{\cal E}\,$ coincides with $\,\Delta t\,$. In realistic situations, $\,\alpha\,$ differs from $\,-1\,$, while the parameter $\,{\cal E}\,$ is related to the time lag in a more sophisticated way (_Ibid._). To show how (84 \- 86) stem from the above Assertions, keep for the time being $\,\alpha=\,-\,1\,$. Also recall that the torque is despinning (so $\,\dot{\theta}\,>\,n\,$), and that for the averages over time $\displaystyle\langle\;\ddot{\theta}\;\,\rangle\;=\;\frac{\langle\,\tau\,\rangle}{C}\;\;\;,$ (88) $C\,=\,\xi\,M_{planet}\,R^{2}\,$ being the maximal moment of inertia of the planet. (For a homogeneous spherical planet, $\xi=2/5$.) Then plug (87) into (71) and average the torque:212121 As explained in the paragraph preceding formula (71), substitution of $\,\sin\epsilon\,$ with $\,1/Q\,$ in the expression for torque generates a relative error $\,O(Q^{-2})\,$, i.e., an absolute error $\,O(Q^{-3})\,$. Instead of inserting (87) into (71), one may directly use (65). Still, approximation of $\,\sin\epsilon\,$ with $\,\epsilon\,$ will entail, in (89) and its corollaria, a relative error $\,O(Q^{-2})\,$ and an absolute error $\,O(Q^{-3})\,$. The situation will become more complicated in the special case of low values of the quality factor. See the Appendix below $\displaystyle\langle\,\tau\,\rangle=\,-\;\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\cal E}}{R}\;\;\langle\;\,(\dot{\theta}\,-\,\dot{\nu})\,\;\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\;\rangle~{}~{}+O({\it i}^{2}/Q)+O(Q^{-3})+O(en/Q^{2}\chi)\;=~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle-\;\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\cal E}}{R}\;\dot{\theta}\;\;\langle\,\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\rangle\;+\;\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;k_{2}\;{\cal E}}{R}\;\langle\;\dot{\nu}\,\frac{R^{\textstyle{{}^{6}}}}{r^{6}}\;\rangle+O({\it i}^{2}/Q)+O(Q^{-3})+O(en/Q^{2}\chi)~{}~{}~{}~{}~{}$ (89a) $\displaystyle=$ $\displaystyle-$ $\displaystyle\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\cal E}}{R}\;\dot{\theta}\;\;\frac{R^{6}}{a^{6}}\left(1\,-\,e^{2}\right)^{-9/2}~{}\,\frac{1}{2\,\pi}\;\int_{0}^{2\pi}\left(1+e\;\cos\nu\right)^{4}\,d\nu\;~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (89b) $\displaystyle+$ $\displaystyle\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;k_{2}\,{\cal E}}{R}\,n\,\frac{R^{\textstyle{{}^{6}}}}{a^{6}}\left(1-e^{2}\right)^{-6}\frac{1}{2\pi}\int_{0}^{2\pi}\left(1+e\,\cos\nu\right)^{6}d{\nu}+O({\it i}^{2}/Q)+O(Q^{-3})+O(en/Q^{2}\chi)~{}~{},~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ where the absolute error $\,O(en/Q^{2}\chi)\,$ emerges due to an uncertainty in the definition of the overall quality factor $\,Q\,$ employed in MacDonald’s model. Evaluation of the above integrals is trivial and indeed leads to (84 \- 86), the constant being $\displaystyle{\cal K}\,=\,\frac{3\,G\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\cal E}}{C\;R}\;\frac{R^{\textstyle{{}^{6}}}}{a^{6}}\,=\,\frac{3\,n^{2}\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\Delta t}}{\xi\;M_{planet}\;(M_{planet}\,+\,M_{sat})}\;\frac{R^{\textstyle{{}^{3}}}}{a^{3}}\,=\,\frac{3\,n\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}}{\xi\;Q\;M_{planet}\;(M_{planet}\,+\,M_{sat})}\;\frac{R^{\textstyle{{}^{3}}}}{a^{3}}\;\,\frac{n}{\chi}\;\;\;,~{}~{}~{}$ (90) where we used the fact that for $\,\alpha=-1\,$ the rheological parameter ${\cal E}$ is simply the lag $\,\Delta t$. It should also be added that, since (89b) contains a relative error $\,O(Q^{-2})\,$, the usefulness of the $\,e^{4}\,$ and $\,e^{6}\,$ terms in (85 \- 86) depends on the values of the eccentricity and the quality factor. If, for example, $\,Q=70\,$, then the $\,e^{4}\,$ terms become unimportant for $\,e<0.12\,$, while the $\,e^{6}\,$ terms become unimportant for $\,e<0.24\,$. To draw to a close, we would mention that besides the above formula (84), in the literature hitherto we saw its sibling, an expression derived in a similar way, but with Assertion II rejected in favour of treating $\,Q\,$ as a frequency-independent constant. The result of this treatment suffers an incurable birth trauma – the incompatibility between the frequency- independence of $\,\Delta t\,$ and the frequency-independence of $\,Q\,$. ### 10.2 Calculation based on the Darwin torque The following alternative derivation is based on the same Assertions (I - III) and, naturally, leads to the same results. The idea is to calculate the despinning rate not in terms of the MacDonald torque, but in terms of the Darwin torque, keeping the eccentricity-caused relative error at the level of $\,O(e^{6})\,$. To keep the inclination-caused relative error at the level of $\,O({\it i}^{2})\,$, we still assume, in (46), that $\,{\emph{l}}\,=\,2\,$, $\;m\,=\,2\,$, $\;p\,=\,0\,$. As for the the values of $\,q\,$, we keep only the ones giving us terms of order up to $\,e^{4}\,$, inclusively. Besides, we assume the phase lags to be small, so that $\,\sin\epsilon_{\textstyle{{}_{lmpq}}}\,=\,\epsilon_{\textstyle{{}_{lmpq}}}\,+\,O(\epsilon^{3})\,=\,\epsilon_{\textstyle{{}_{lmpq}}}\,+\,O(Q^{-3})\,$. Under all these presumptions, the constant part of the tidal torque can be approximated with $\displaystyle\tau_{\textstyle{{}_{\textstyle{{}_{l=2}}}}}\;=\;\frac{3}{2}\;G\;M_{sat}^{2}\,R^{5}\,a^{-6}\,k_{2}\;\sum_{q=-2}^{2}\,G^{\textstyle{{}^{\,2}}}_{\textstyle{{}_{\textstyle{{}_{20\mbox{\it{q}}}}}}}\,\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}~{}+\,O(e^{6}/Q)\,+\,O({\it i}^{2}/Q)\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (91a) $\displaystyle=\;\frac{3}{2}\;G\;M_{sat}^{2}\,R^{5}\,a^{-6}\,k_{2}\;\sum_{q=-2}^{2}\,G^{\textstyle{{}^{\,2}}}_{\textstyle{{}_{\textstyle{{}_{20\mbox{\it{q}}}}}}}\,\epsilon_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}~{}+\,O(e^{6}/Q)\,+\,O({\it i}^{2}/Q)\,+\,O(Q^{-3})\;\;\;,$ (91b) where, according to the tables (Kaula 1966), $\displaystyle G^{2}_{\textstyle{{}_{\textstyle{{}_{20\;-2}}}}}\,=\,0\;\;\;,\;\;\;\;\;\;\;G^{2}_{\textstyle{{}_{\textstyle{{}_{20\;-1}}}}}\,=\,\frac{e^{2}}{4}\;-\;\frac{e^{4}}{16}\;+\;O(e^{6})\;\;\;\,,\;\;\;\;G^{2}_{\textstyle{{}_{\textstyle{{}_{200}}}}}\,=\,1\,-\,5\,e^{2}\,+\;\frac{63}{8}\;e^{4}\;+\;O(e^{6})\;\;\;\,,\;\;\;\;\;$ (92) $\displaystyle G^{2}_{\textstyle{{}_{\textstyle{{}_{20{{1}}}}}}}\,=\,\frac{49}{4}\;e^{2}\;-\;\frac{861}{16}\;e^{4}\;+\;O(e^{6})\;\;\;\;,\;\;\;\;\;\;\;G^{2}_{\textstyle{{}_{\textstyle{{}_{20{{2}}}}}}}\,=\,\frac{289}{4}\,e^{4}\,+\,O(e^{6})\;\;\;\;,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}\,$ and, according to formula (36), $\displaystyle\epsilon_{\textstyle{{}_{\textstyle{{}_{220\;-2}}}}}=\,(-\,2\,\dot{\theta}\,)\;\Delta t_{\textstyle{{}_{\textstyle{{}_{220\;-2}}}}}\;\;\;,\;\;\;\;\epsilon_{\textstyle{{}_{\textstyle{{}_{220\;-1}}}}}=\,(-\,2\,\dot{\theta}\,+\,n)\;\Delta t_{\textstyle{{}_{\textstyle{{}_{220\;-1}}}}}\;\;\;,\;\;\;\;\epsilon_{\textstyle{{}_{\textstyle{{}_{2200}}}}}=\,(-\,2\,\dot{\theta}\,+\,2\,n)\;\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\;\;\;,\;\;\;\;\;$ (93) $\displaystyle\epsilon_{\textstyle{{}_{\textstyle{{}_{220{{1}}}}}}}\,=\,(-\,2\,\dot{\theta}\,+\,3\,n)\;\Delta t_{\textstyle{{}_{\textstyle{{}_{220{{1}}}}}}}\;\;\;\;,\;\;\;\;\;\;\epsilon_{\textstyle{{}_{\textstyle{{}_{220{{2}}}}}}}\,=\,(-\,2\,\dot{\theta}\,+\,4\,n)\;\Delta t_{\textstyle{{}_{\textstyle{{}_{220{{2}}}}}}}\;\;\;\;.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ Provided the quality factor scales as inverse frequency, all the time lags are the same constant $\,\Delta t\,$, so the above formulae all together entail, in the case of nonresonant prograde spin: $\displaystyle\ddot{\theta}\;=\,\frac{\tau}{C}=\;{\cal K}\,\left[\;-\;\dot{\theta}\,\left(1\,+\,\frac{15}{2}\,e^{2}\,+\,\frac{105}{4}\,e^{4}\,+\,O(e^{6})\right)\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (94) $\displaystyle+\;\left.\,n\,\left(1\,+\,\frac{27}{2}\,e^{2}\,+\,\frac{573}{8}\,e^{4}\,+\,O(e^{6})\right)\,\right]~{}+\,O({\it i}^{2}/Q)\,+\,O(Q^{-3})\;\;\;,~{}~{}~{}~{}~{}~{}~{}~{}~{}$ which coincides with (84 \- 86) to the order $\,e^{4}\,$, inclusively, provided we substitute $\,\chi_{\textstyle{{}_{2200}}}\,$ instead of $\,\chi\,$ in the expression (90) for $\,{\cal{K}}\,$. ### 10.3 Rheologies different from $\,Q\,\sim\,1/\mbox{{\boldmath$\chi$}}\;\;$ A part and parcel of both afore-presented methods was the assertion of all the time lags $\,\Delta t_{\textstyle{{}_{lmpq}}}\,$ being equal. In reality, the time lags vary from one harmonic to another. Any particular functional form of the dependence $\,\Delta t(\chi)\,$ fixes the rheology: for example, the frequency-independence of $\,\Delta t\,$ constrains the value of the exponential $\,\alpha\,$ to $\,-1\,$ (while the parameter $\,{\cal{E}}\,$ becomes simply $\,\Delta t\,$). However, for an arbitrary $\,\alpha\,\neq\,-\,1\,$ the lags will read (Efroimsky & Lainey 2007): $\displaystyle\Delta t_{\textstyle{{}_{\textstyle{{}_{lmpq}}}}}\;=\;{\cal{E}}^{\textstyle{{}^{-\,\alpha}}}\;\chi_{\textstyle{{}_{\textstyle{{}_{lmpq}}}}}^{\textstyle{{}^{-\,(\alpha+1)}}}$ (95) While the MacDonald approach cannot be generalised to $\,\alpha\,\neq\,-\,1\,$, the Darwin-Kaula-Goldreich method can be well combined with (95). To this end, we shall insert (92 \- 93) and (95) into (91a), and shall also employ the evident formulae $\displaystyle\cos\epsilon_{\textstyle{{}_{\textstyle{{}_{{{{\it{l}}mpq}}}}}}}\,=\,\frac{~{}|\,\cot\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,|\;}{\sqrt{{\textstyle 1~{}+~{}\cot^{2}\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=\;\frac{\,Q_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}Q^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}~{}=~{}\frac{\;{\cal{E}}^{\textstyle{{}^{\alpha}}}\;\chi^{\textstyle{{}^{\alpha}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}{\cal{E}}^{\textstyle{{}^{2\alpha}}}\;\chi^{\textstyle{{}^{2\alpha}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}~{}~{}~{},~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (96) $\displaystyle\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{{{{\it{l}}mpq}}}}}}}=\,\sin|\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\;\,\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}=\,\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}\cot^{2}\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=\;\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}Q^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}~{}=~{}\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}{\cal{E}}^{\textstyle{{}^{2\alpha}}}\;\chi^{\textstyle{{}^{2\alpha}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}~{}~{}~{},~{}~{}~{}$ (97) with $\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,$ given by (37), and $\,|\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\,$ assumed (for reasons explained in the Appendix) not to approach too close to $\,\pi/2\;$. This will give us the following expression for (the constant part of) the deceleration rate of a non-resonant prograde spin: $\displaystyle\ddot{\theta}\,=\,\frac{\tau}{C}=\;-\;\frac{3}{2}\;\frac{G\;M^{2}_{sat}}{a^{3}}\,\frac{R^{5}}{a^{3}}\;\frac{k_{\textstyle{{}_{2}}}\;}{\xi\;M_{planet}\;R^{2}}\;\left[\;\frac{e^{2}}{4}\;\,\mbox{sgn}(2\,\dot{\theta}\,-\,n)\;\;\frac{{\cal{E}}^{\textstyle{{}^{\alpha}}}\,\;|2\,\dot{\theta}\,-\,n|^{\textstyle{{}^{\alpha}}}}{\;1\;+\;{\cal{E}}^{\textstyle{{}^{2\alpha}}}\,\;|2\,\dot{\theta}\,-\,n|^{\textstyle{{}^{2\alpha}}}\,}\;\;\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\left.+\;\left(1-5e^{2}+\frac{63}{8}e^{4}\right)\;\mbox{sgn}(2\,\dot{\theta}\,-\,2\,n)\;\;\frac{{\cal{E}}^{\textstyle{{}^{\alpha}}}\,\;|2\,\dot{\theta}\,-\,2\,n|^{\textstyle{{}^{\alpha}}}}{\,1\;+\;{\cal{E}}^{\textstyle{{}^{2\alpha}}}\,\;|2\,\dot{\theta}\,-\,2\,n|^{\textstyle{{}^{2\alpha}}}\,}\;\right.~{}~{}~{}~{}~{}~{}\,$ $\displaystyle+\,\left.\left(\frac{49}{4}e^{2}-\frac{861}{16}e^{4}\right)\;\mbox{sgn}(2\,\dot{\theta}\,-\,3\,n)\;\;\frac{{\cal{E}}^{\textstyle{{}^{\alpha}}}\,\;|2\,\dot{\theta}\,-\,3\,n|^{\textstyle{{}^{\alpha}}}}{\,1\;+\;{\cal{E}}^{\textstyle{{}^{2\alpha}}}\,\;|2\,\dot{\theta}\,-\,3\,n|^{\textstyle{{}^{2\alpha}}}\,}\right.~{}~{}~{}~{}~{}~{}$ $\displaystyle\left.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}+\;\frac{289}{4}\;e^{4}\;\mbox{sgn}(2\,\dot{\theta}\,-\,4\,n)\;\frac{{\cal{E}}^{\textstyle{{}^{\alpha}}}\,\;|2\,\dot{\theta}\,-\,4\,n|^{\textstyle{{}^{\alpha}}}}{\,1\;+\;{\cal{E}}^{\textstyle{{}^{2\alpha}}}\,\;|2\,\dot{\theta}\,-\,4\,n|^{\textstyle{{}^{2\alpha}}}\,}\;\right]\;+\;O({\it i}^{2}/Q)\;+\;O(e^{6}/Q)~{}~{}~{}.~{}~{}~{}~{}~{}~{}~{}$ (98) Be mindful, that a naive substitution of the formula (97) for $\,\sin\epsilon{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}$ into (91a) would result in an expression for the torque, attaining its maxima on approach to resonances (for a positive $\alpha$), an evidently unphysical behaviour. As explained in section 9, there exists a profound physical reason, for which the actual multiplier in (91a) must be not $\,\sin\epsilon{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;$ but: $\;\sin\epsilon{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,\cos\epsilon{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,$. Mathematically, the presence of the cosine is irrelevant unless $\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}$ and $Q_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}$ approach zero. If however $\,\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,$ becomes very small (i.e., if we approach a resonance), it is this long-omitted (though known yet to Darwin 1879) cosine multiplier that saves us from the unphysical maxima – see section 9 above. Under the extra assumptions222222 The smallness of $\;\,|\,\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\;\,$ enables one to employ (91b) instead of (91a). Then the multipliers ${\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ \frac{\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,\;\,\textstyle{{\cal E}^{\textstyle{{}^{\alpha}}}\,\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}^{\textstyle{{}^{\alpha}}}}}{\textstyle{1+{\cal E}^{\textstyle{{}^{2\alpha}}}\,\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}^{\textstyle{{}^{2\alpha}}}}}\,\;$ in (98) become $\;\,\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}=\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\frac{\textstyle{\Delta t_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}{\textstyle{\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}=\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\left(\frac{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}\right)^{\textstyle{{}^{\alpha+1}}}\\\ {\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ =\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\left(\frac{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}\right)^{\textstyle{{}^{\alpha}}}=\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\left(1\;+\;\frac{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}\;-\;\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}\right)^{\textstyle{{}^{-\,\alpha}}}.~{}~{}~{}$ Specifically, ${\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ {\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ \epsilon_{\textstyle{{}_{\textstyle{{}_{220{\textstyle{q}}}}}}}=\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{220{\textstyle{q}}}}}}}\;\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}~{}\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\left(1\,+\,\frac{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{{{2}}20{\textstyle{q}}}}}}}}-\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}\right)^{\textstyle{{}^{-\,\alpha}}}\approx\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{{2}}20{\textstyle{q}}}}}}}\;\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\;\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\left(1\,-\,\alpha\;\frac{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{{{2}}20{\textstyle{q}}}}}}}}-\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}{\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}\right)\\\ {\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ {\left.\;~{}\right.}^{\left.\;~{}\right.}\\\ =\mbox{sgn}\;\omega_{\textstyle{{}_{\textstyle{{}_{{{2}}20{\textstyle{q}}}}}}}\;\Delta t_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\;\left[\,(1\,+\,\alpha)\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,-\,\alpha\;\textstyle{\chi_{\textstyle{{}_{\textstyle{{}_{{{{2}}20{\textstyle{q}}}}}}}}}\right]\,\;$, the latter approximation being legitimate only under the ${\left.\;~{}\right.}^{\left.\;~{}\right.}$ condition of $\,\chi_{\textstyle{{}_{\textstyle{{}_{220{\textstyle{q}}}}}}}\,-\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,\ll\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,$, which turns out to be equivalent to $\,n\,\ll\,\dot{\theta}\,$. For example, ${\left.\;~{}\right.}^{\left.\;~{}\right.}$ $\;\frac{\chi_{\textstyle{{}_{\textstyle{{}_{220{\textstyle{q}}}}}}}\,-\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}\,=\,\frac{\textstyle(\,-\,2\,\stackrel{{\scriptstyle\centerdot}}{{\theta}}\,+\,4\,n)\,-\,(\,-\,2\,\stackrel{{\scriptstyle\centerdot}}{{\theta}}\,+\,2\,n)}{\textstyle{\,-\,2\,\stackrel{{\scriptstyle\centerdot}}{{\theta}}\,+\,2\,n}}\,=\,\frac{\textstyle 2\,n}{\textstyle\,-\,2\,\stackrel{{\scriptstyle\centerdot}}{{\theta}}\,+\,2\,n}\;\,$. So approximating $\,\epsilon_{\textstyle{{}_{\textstyle{{}_{220{\textstyle{q}}}}}}}\,$, for ${\left.\;~{}\right.}^{\left.\;~{}\right.}$ $\,q\,=\,-\,2\,,\,-\,1\,,\,0\,,\,1\,,\,2\,$, we arrive at formula (99). There exists one more reason to keep $\,n\,$ much smaller than $\,\dot{\theta}\,$ in (86 - 89). We derived (86 - 89) by inserting the customary relation (36) into (84 - 85). As explained in the Appendix below, (36) becomes invalid near spin- orbit commensurabilities. Indeed, at each commensurability a certain tidal harmonic becomes nil – see formula (26). According to (60), the appropriate Q, too, becomes nil. In this situation, one has to rely not on (36) but on a more general formula (105). The latter formula however entails vanishing of the appropriate component of the tidal torque on crossing the commensurability – see (113 - 114). This is why in (87 - 89) and even earlier, in (86), we should stay away from the commensurabilities $\,\dot{\theta}=n/2\,$, $\,\dot{\theta}=n\,$, $\,\dot{\theta}=3n/2\,$, or $\,\dot{\theta}=2n\,$. So we better keep $\,n\ll\dot{\theta}\,$. of $\,|\epsilon{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\,\ll\,1\,$ and $\,n\,\ll\,\dot{\theta}\,$, formula (98) simplifies to $\displaystyle\ddot{\theta}\,=\,\frac{\tau}{C}={\cal{K}}\left[\;\,-\;\,\dot{\theta}\,\left(1\,+\,\frac{15}{2}\,e^{2}\,+\,\frac{105}{4}\,e^{4}\,+\,O(e^{6})\right)\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle+\;\left.n\left(1+\left(\frac{15}{2}-6\alpha\right)e^{2}+\left(\frac{105}{4}-\frac{363}{8}\alpha\right)e^{4}+O(e^{6})\right)\right]+O({\it i}^{2}/Q)+O(Q^{-3})+O(n/\dot{\theta})\;\;\;~{}~{}~{}~{}~{}~{}$ (99) $\displaystyle\approx\;{\cal K}\;\left[\,-\,\dot{\theta}\,\left(1\,+\,\frac{15}{2}\,e^{2}\right)+\,n\,\left(1\,+\,\left(\frac{15}{2}\,-\,6\,\alpha\right)\,e^{2}\,\right)\,\right]\;\;\;,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (100) where the overall factor ${\cal K}$ is given by $\displaystyle{\cal K}\,=\,\frac{3\,n^{2}\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}\;{\Delta t_{\textstyle{{}_{2200}}}}}{\xi\;M_{planet}\;(M_{planet}\,+\,M_{sat})}\;\frac{R^{\textstyle{{}^{3}}}}{a^{3}}\,=\,\frac{3\,n\,M_{sat}^{\textstyle{{}^{\,2}}}\;\,k_{2}}{\xi\;Q_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\;M_{planet}\;(M_{planet}\,+\,M_{sat})}\;\frac{R^{\textstyle{{}^{3}}}}{a^{3}}\;\,\frac{n~{}~{}}{\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}\;\;\;,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (101) an expression identical to (90), except that it contains $\,\Delta t_{\textstyle{{}_{2200}}}\,$, $\,Q_{\textstyle{{}_{2200}}}\,$, and $\,\chi_{\textstyle{{}_{2200}}}\,$ instead of $\,\Delta t\,$, $\,Q\,$, and $\,\chi\,$, correspondingly. Were $\;\alpha\;$ equal to $\;\,-\,1\;$, sum (99) would coincide with (94), provided $\;\dot{\theta}\,>\,2\,n\;$ (but not otherwise!). For realistic mantles and crusts, though, the values of $\,\alpha\,$ will, as pointed above, reside within the interval $\,0.2-0.4\;$ (closer to $0.2$ for partial melts). ## 11 Conclusions In the article thus far we have provided a detailed review of a narrow range of topics. Our goal was to punctiliously spell out the assumptions that often remain implicit, and to bring to light those steps in calculations, which are often omitted as “self-evident”. This has helped us to demonstrate that MacDonald-style formula (69) for the tidal torque is valid only in the zeroth order of $\,en/Q\chi\,$, while its time-average is valid only in the first order. These restrictions mean that in the popular expressions for tidal despinning rate the terms with higher powers of $\,e\,$ become significant only for large eccentricities. Their significance is limited even further by the error $\,O(Q^{-3})\,$ emerging when the sine of the phase lag gets approximated by the inverse quality factor – see formula (71) and the paragraph preceding it. We have demonstrated that in the case, when the inclinations are small and the phase lags of the tidal harmonics are proportional to the frequency, the Darwin-Kaula expansion is equivalent to a corrected version of the MacDonald formalism. The latter method rests on the assumption of existence of one total double bulge. The necessary correction to MacDonald’s approach would be to assert (following Singer 1968) that the phase lag of this integral bulge is not constant, but is proportional to the instantaneous synodal frequency $\,2(\dot{\nu}-\dot{\theta})\,$, where $\nu$ and $\theta$ are the true anomaly and the sidereal angle. Any rheology different from this one will violate the equivalence of the Darwin-Kaula and MacDonald descriptions. It remains unexplored if their equivalence is violated also by setting the inclination high. We have demonstrated that no “paradoxes” ensue from the frequency-dependence $\,Q\sim\chi^{\alpha}\;$, with $\;\;\alpha\,=\,0.3\,\pm\,0.1\;$, found for the mantle. We have investigated the limitations of the popular formula interconnecting the quality factor $\,Q\,$ and the phase lag $\,\epsilon\,$. It turns out that for low quality factors (less than 10), the customary formula $\,Q\,=\,\cot|\epsilon|\,$ should be substituted with a far more complicated relation. Finally, we examined two derivations of the popular expressions (84 \- 86), and have pointed out that these expressions have limitations related to the frequency-dependence of the quality factor. First, dependent upon the values of $e$ and $Q$, the high-order terms in these expressions may become significant only for large eccentricities. Second, the expansion of the deceleration rate in even powers of $\,e\,$ will be different if $\,\Delta t\,$ is frequency-dependent (which is the case for solid materials). These two circumstances do not necessarily disprove any major result achieved in the bodily-tide theory. However, some coefficients may now have to be reconsidered. For the realistic rheology of terrestrial bodies, the despinning rate, in the absence of tidal locking, is given by our formulae (10.3 \- 101). Acknowledgments It is a pleasure for us to acknowledge the contribution to this work from Alessandra Celletti, whose incisive questions ignited a discussion and eventually compelled us to take pen to paper. Our profoundest gratitude goes also to Sylvio Ferraz Mello, who kindly offered a large number of valuable comments and important corrections. ME would also like deeply to thank Bruce Bills, Tony Dobrovolskis, Peter Goldreich, Shun-ichiro Karato, Valery Lainey, William Newman, Stan Peale, S. Fred Singer, Victor Slabinski, and Gabriel Tobie for numerous stimulating conversations on the theory of tides. Part of the research described in this paper was carried out at the Jet Propulsion Laboratory of the California Institute of Technology, under a contract with the National Aeronautics and Space Administration. ME is most grateful to John Bangert for his support of the project on all of its stages. Appendix. The lag and the quality factor: is the formula $\boldmath{Q=\cot|\mbox{{\boldmath$\epsilon$}}|}$ universal? The interrelation between the quality factor $\,Q\,$ and the phase lag $\,\epsilon\,$ is long-known to be $\displaystyle Q\,=\,\cot|\epsilon|\;\;\;,$ (102) and its derivation can be found in many books. In Appendix A2 of Efroimsky & Lainey(2007), that derivation is reproduced, with several details that are normally omitted in the literature. Among other things, we pointed out that the interrelation has exactly the form (102) only in the limit of small lags. For large phase lags, the form of this relation will change considerably. Since in section 9 of the current paper we address the case of large lags, it would be worth reconsidering the derivation presented in Efroimsky & Lainey (2007), and correcting a subtle omission made there. Before writing formulae, let us recall that, at each frequency $\,\chi\,$ in the spectrum of the deformation, the quality factor (divided by $\,2\,\pi\,$) is defined as the peak energy stored in the system divided by the energy damped over a cycle of flexure: $\displaystyle{Q}(\chi)\;\equiv\;-\;\frac{2\;\pi\;E_{peak}(\chi)}{\Delta E_{cycle}(\chi)}\;\;\;,$ (103) where $\,\Delta E_{cycle}(\chi)\,<\,0\,$ as we are talking about energy losses.232323 We are considering flexure in the linear approximation. Thus at each frequency $\,\chi\,$ the appropriate energy loss over a cycle, $\,\Delta E_{cycle}(\chi)\,$, depends solely on the maximal energy stored at that same frequency, $\,E_{peak}(\chi)\,$. An attempt to consider large lags (all the way up to $\,|\epsilon|\,=\,\pi/2\,$) sets the values of $\,Q/2\pi\,$ below unity. As the dissipated energy cannot exceed the energy stored in a free oscillator, the question becomes whether the values of $\,Q/2\pi\,$ can be that small. To understand that they can, recall that in this situation we are considering an oscillator, which is not free but is driven (and is overdamped). The quality factor being much less than unity simply implies that the eigenfrequencies get damped away during less than one oscillation. Nonetheless, motion goes on due to the driving force. Now let us switch to the specific context of tides. To begin with, let us recall that the dissipation rate in a tidally distorted primary is well approximated by the work that the secondary carries out to deform the primary: $\displaystyle\dot{E}\;=\;-\;\int\,\rho\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\;\cdot\;\nabla W\;d^{3}x$ (104) $\rho\,,\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,$, and $\,W\,$ denoting the density, velocity, and tidal potential inside the primary. The expression on the right-hand side can be transformed by means of the formula $\displaystyle\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla W\,=\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,W)\,-\,W\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho\,-\,W\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$})\,=\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,W)\,-\,W\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho\,+\,W\,\frac{\partial\rho}{\partial t}\;\;,\;\;\;\;$ (105) where the $W\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho$ and $\partial\rho/\partial t$ terms may be omitted under the assumption that the primary is homogeneous and incompressible. In this approximation, the attenuation rate becomes simply $\displaystyle\dot{E}\;=\;-\;\int\,\nabla\,\cdot\,(\rho\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\;W)\,d^{3}x\;=\;-\;\int\,\rho\;W\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,\cdot\,{\vec{\bf{n}}}\;\,dA\;\;\;,$ (106) ${\vec{\bf{n}}}\,$ being the outward normal to the surface of the primary, and $\,dA\,$ being an element of the surface area. It is now clear that, under the said assertions, it is sufficient to take into account only the radial elevation rate, not the horizontal distortion. This way, formula (104), in application to a unit mass, will get simplified to $\displaystyle\dot{E}\;=\;\left(-\,\frac{\partial W}{\partial r}\right)\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot{\vec{\bf{n}}}\;=\;\left(-\,\frac{\partial W}{\partial r}\right)\frac{d\zeta}{dt}\;\;\;,$ (107) $\zeta\,$ standing for the vertical displacement (which is, of course, delayed in time, compared to $\,W\,$). The amount of energy dissipated over a time interval $\,(t_{o}\,,\;t)\,$ is then $\displaystyle\Delta{E}\;=\;\int^{t}_{t_{o}}\;\left(-\,\frac{\partial W}{\partial r}\right)\;d\zeta\;\;\;.$ (108) We shall consider the simple case of an equatorial moon on a circular orbit. At each point of the planet, the variable part of the tidal potential produced by this moon will read $\displaystyle W\;=\;W_{o}\;\cos\chi t\;\;\;,$ (109) the tidal frequency being given by $\displaystyle\chi\,=\,2~{}|n\;-\;\omega_{p}|~{}~{}~{}.~{}~{}~{}$ (110) Let g denote the surface free-fall acceleration. An element of the planet’s surface lying beneath the satellite’s trajectory will then experience a vertical elevation of $\displaystyle\zeta\;=\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\cos(\chi t\;-\;|\epsilon|)\;\;\;,$ (111) $\,h_{2}\,$ being the corresponding Love number242424 For a homogeneous incompressible body, $\,k_{2}=(3/5)h_{2}\,$, for which reason (111) and the subsequent equations with $\,h_{2}\,$ can equally be written as proportional to $\,k_{2}\,$. The formulation employing $\,k_{2}\,$ is more fundamental, as it can, in principle, be generalised to a compressible body of a radially- changing density. Indeed, whatever the properties of the primary are, the dissipation rate in it is equal to the rate of change of the primary’s spin energy plus the rate of change of the orbital energy. Both the latter and the former are proportional to $\,k_{2}\,$., and $\,|\epsilon|\,$ being the _positive_ 252525 Were we not considering the simple case of a circular orbit, then, rigorously speaking, the expression for $\,W\,$ would read not as $\,W_{o}\,\cos\chi t\,$ but as $\,W_{o}\,\cos\omega_{\textstyle{{}_{tidal}}}t\,$, the tidal frequency $\,\omega_{\textstyle{{}_{tidal}}}\,$ taking both positive and negative values, and the physical frequency of flexure being $\,\chi\,\equiv\,|\omega_{\textstyle{{}_{tidal}}}|\,$. Accordingly, the expression for $\,\zeta\,$ would contain not $\,\cos(\chi t\,-\,|\epsilon|)\,$ but $\,\cos(\omega_{\textstyle{{}_{tidal}}}t\,-\,\epsilon)\,$. As we saw in equation (36), the sign of $\,\epsilon\,$ is always the same as that of $\,\omega_{\textstyle{{}_{tidal}}}\,$. For this reason, one may simply deal with the physical frequency $\,\chi\,\equiv\,|\omega_{\textstyle{{}_{tidal}}}|\,$ and with the absolute value of the phase lag, $\;|\epsilon|\;$. phase lag, which for the principal tidal frequency is simply the double geometric angle $\,\delta\,$ subtended at the primary’s centre between the directions to the secondary and to the main bulge: $\displaystyle|\epsilon|\;=\;2\;\delta\;\;\;.$ (112) Accordingly, the vertical velocity of this element of the planet’s surface will amount to $\displaystyle u\;=\;\dot{\zeta}\;=\;-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\sin(\chi t\;-\;|\epsilon|)\;=\;-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\left(\sin\chi t\;\cos|\epsilon|\;-\;\cos\chi t\;\sin|\epsilon|\right)\;\;.\;\;$ (113) The expression for the velocity has such a simple form because in this case the instantaneous frequency $\chi$ is constant. The satellite generates two bulges (on the facing and opposite sides of the planet) so each point of the surface is uplifted twice through a cycle. This entails the factor of two in the expression (110) for the frequency. The phase in (112), too, is doubled, though the necessity of this is less evident, – see footnote 4 in Appendix to Efroimsky & Lainey (2007). The energy dissipated over a time cycle $\,T\,=\,2\pi/\chi\,$, per unit mass, will, in neglect of horizontal displacements, be $\displaystyle\Delta E_{{}_{cycle}}$ $\displaystyle=$ $\displaystyle\int^{T}_{0}u\left(-\,\frac{\partial W}{\partial r}\right)dt=\,-\left(-\,h_{2}\;\chi\frac{W_{o}}{\mbox{g}}\right)\,\frac{\partial W_{o}}{\partial r}\int^{t=T}_{t=0}\cos\chi t\,\left(\sin\chi t\,\cos|\epsilon|\,-\,\cos\chi t\,\sin|\epsilon|\right)dt$ (114) $\displaystyle=$ $\displaystyle\,-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\sin|\epsilon|\,\;\frac{1}{\chi}\;\int^{\chi t\,=\,2\pi}_{\chi t\,=\,0}\;\cos^{2}\chi t\;\;d(\chi t)\;=\;-\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\pi\;\sin|\epsilon|\;\;,\;\;\;~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ while the peak energy stored in the system during the cycle will read: $\displaystyle E_{{}_{peak}}$ $\displaystyle=$ $\displaystyle\int^{T/4}_{|\epsilon|/\chi}u\left(-\,\frac{\partial W}{\partial r}\right)dt=\,-\left(-\,h_{2}\;\chi\,\frac{W_{o}}{\mbox{g}}\right)\frac{\partial W_{o}}{\partial r}\int^{t=T/4}_{t=|\epsilon|/\chi}\cos\chi t\,\left(\sin\chi t\,\cos|\epsilon|\,-\,\cos\chi t\,\sin|\epsilon|\right)dt$ (115) $\displaystyle=$ $\displaystyle\;\chi\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\left[\;\frac{\cos|\epsilon|}{\chi}\;\int^{\chi t\,=\,\pi/2}_{\chi t\,=\,|\epsilon|}\;\cos\chi t\;\sin\chi t\;\;d(\chi t)\;-\;\frac{\sin|\epsilon|}{\chi}\;\int^{\chi t\,=\,\pi/2}_{\chi t\,=\,|\epsilon|}\;\cos^{2}\chi t\;\;d(\chi t)\;\right]\;\;.~{}~{}~{}~{}~{}~{}~{}~{}\,$ In the appropriate expression in Appendix A1 to Efroimsky & Lainey (2007), the lower limit of integration was erroneously set to be zero. To understand that in reality integration over $\chi t$ should begin from $|\epsilon|$, one should superimpose the plots of the two functions involved, $\cos\chi t$ and $\sin(\chi t-|\epsilon|)$. The maximal energy gets stored in the system after integration through the entire interval over which both functions have the same sign. Hence $\chi t=|\epsilon|$ as the lower limit. Evaluation of the integrals entails: $\displaystyle E_{peak}\;=\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\left[\;\frac{1}{2}\;\cos|\epsilon|\;-\;\frac{1}{2}\;\left(\;\frac{\pi}{2}\;-\;|\epsilon|\;\right)\;\sin|\epsilon|\;\right]~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (116) whence $\displaystyle Q^{-1}\;=\;\frac{-\;\Delta E_{{}_{cycle}}}{2\,\pi\,E_{{}_{peak}}}\;=\;\frac{1}{2\,\pi}\;\,\frac{\pi\;\sin|\epsilon|}{~{}\frac{\textstyle 1}{\textstyle 2}\;\cos|\epsilon|\;-\;\frac{\textstyle 1}{\textstyle 2}\;\left(\;\frac{\textstyle\pi}{\textstyle 2}\;-\;|\epsilon|\;\right)\;\sin|\epsilon|}\;=\;\frac{\tan|\epsilon|}{1\;-\;\left(\;\frac{\textstyle\pi}{\textstyle 2}\;-\;|\epsilon|\;\right)\;\tan|\epsilon|}\;\;\;.~{}~{}~{}~{}~{}$ (117) As can be seen from (117), both the product $\,\sin\epsilon\,\cos\epsilon\,$ and the appropriate component of the torque attain their maxima when $\,Q\,=\,1\,-\,\pi/4\,$. Usually, $\,|\epsilon|\,$ is small, and we arrive at the customary expression $\displaystyle Q^{-1}\,=\,\tan|\epsilon|\;+\;O(\epsilon^{2})\;\;\;.$ (118) In the opposite case, when $Q\rightarrow 0$ and $|\epsilon|\rightarrow\pi/2$, it is convenient to employ the small difference $\displaystyle\xi\;\equiv\;\frac{\pi}{2}\;-\;|\epsilon|\;\;\;,$ (119) in terms whereof the inverse quality factor will read: $\displaystyle Q^{-1}\,=\,\frac{\cot\xi}{1\;-\;\xi\;\cot\xi}\;=\;\frac{1}{\tan\xi\;-\;\xi}\;=\;\frac{1}{z\;-\;\arctan z}\;=\;\frac{1}{\frac{\textstyle 1}{\textstyle 3}\;z^{3}\;\left[\,1\;-\;\frac{\textstyle 3}{\textstyle 5}\;z^{2}\,+\;O(z^{4})\,\right]}\;\;\;,\;\;\;$ (120) where $\;z\,\equiv\,\tan\xi\;$ and, accordingly, $\;\xi\;=\;\arctan z\;=\;z\,-\frac{\textstyle 1}{\textstyle 3}\,z^{3}\,+\,\frac{\textstyle 1}{\textstyle 5}\,z^{5}\,+\,O(z^{7})\;\,.\;$ Formula (120) may, of course, be rewritten as $\displaystyle z^{3}\;\left[\,1\;-\;\frac{3}{5}\;z^{2}\;+\;O(z^{4})\;\right]\;=\;3\;Q\;\;\;$ (121) or, the same, as $\displaystyle z\;=\;(3\,Q)^{1/3}\;\left[\,1\;+\;\frac{1}{5}\;z^{2}\;+\;O(z^{4})\,\right]\;\;\;.$ (122) While the zeroth approximation is simply $\;z\,=\,(3Q)^{1/3}\,+\,O(Q)\;$, the first iteration gives: $\displaystyle\tan\xi\;\equiv\;z\;=\;(3Q)^{1/3}\,\left[\,1\;+\;\frac{1}{5}\;(3Q)^{2/3}\;+\;O(Q^{4/3})\,\right]\;=\;q\;\left[\,1\;+\;\frac{1}{5}\;q^{2}\;+\;O(q^{4})\,\right]\;\;\;,~{}~{}~{}$ (123) with $\,q\,=\,(3Q)^{1/3}\,$ playing the role of a small parameter. We now see that the customary relation (118) should be substituted, for large lags, i.e., for small262626 The afore-employed expansion of $\,\arctan z\,$ is valid for $\,|z|\,<\,1\,$. This inequality, along with (120), entails: $\,Q\,=\,z\,-\,\arctan z\,<\,1\,-\,\pi/4\,$. values of $\,Q\,$, with: $\displaystyle\tan|\epsilon|\;=\;(3Q)^{-1/3}\,\left[\,1\;-\;\frac{1}{5}\;(3Q)^{2/3}\;+\;O(Q^{4/3})\,\right]$ (124) The formula for the tidal torque contains a multiplier $\,\sin\epsilon\,\cos\epsilon\,$, whose absolute value can, for our purposes, be written down as $\displaystyle\sin|\epsilon|\,\cos|\epsilon|=\cos\xi\,\sin\xi=\frac{\tan\xi}{1+\tan^{2}\xi}\,=\,\frac{q\,\left[1+\frac{\textstyle 1}{\textstyle 5}\,q^{2}+O(q^{4})\right]}{1+q^{2}\left[1+O(q^{2})\right]}=(3Q)^{1/3}\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;,~{}~{}~{}$ (125) whence $\displaystyle\sin\epsilon\;\cos\epsilon\;=\;\pm\;(3Q)^{1/3}\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;,~{}~{}~{}$ (126) an expression vanishing for $\,Q\,\rightarrow\,0\;$. Be mindful that both $\,\epsilon_{\textstyle{{}_{2200}}}\,$ and the appropriate component of the torque change their sign on the satellite crossing the synchronous orbit. ## References * [1] Alexander, M. E. 1973. “The weak-friction approximation and tidal evolution in close binary systems.” _Astrophysics and Space Sciences_ , Vol. 23, pp. 459 - 510 * [2] Bills, B. G., Neumann, G. A., Smith, D.E., & Zuber, M.T. 2005\. “Improved estimate of tidal dissipation within Mars from MOLA observations of the shadow of Phobos.” Journal of Geophysical Research, Vol. 110, pp. 2376 - 2406. doi:10.1029/2004JE002376, 2005 * [3] Churkin, V. A. 1998. “The Love numbers for the models of inelastic Earth.” Preprint No 121. Institute of Applied Astronomy. St.Petersburg, Russia. /in Russian/ * [4] Correia, A. C. M., & Laskar, J. 2004. “Mercury’s capture into the $\,3/2\,$ spin-orbit resonance as a result of its chaotic dynamics.” _Nature_ , Vol. 429, pp. 848 - 850 * [5] Darwin, G. H. 1879. “On the precession of a viscous spheroid and on the remote history of the Earth” Philosophical Transactions of the Royal Society of London, Vol.170, pp 447 -530 http://www.jstor.org/view/02610523/ap000081/00a00010/ * [6] Darwin, G. H. 1880. “On the secular change in the elements of the orbit of a satellite revolving about a tidally distorted planet.” Philosophical Transactions of the Royal Society of London, Vol. 171, pp. 713 - 891 http://www.jstor.org/view/02610523/ap000082/00a00200 * [7] Darwin, G. H. 1908. “Tidal friction and cosmogony.” In: Darwin, G. H., _Scientific Papers,_ Vol.2. Cambridge University Press, NY 1908. * [8] Dobrovolskis, A. 1995. “Chaotic rotation of Nereid?” _Icarus_ , Vol. 118, pp. 181 - 195 * [9] Dobrovolskis, A. 2007. “Spin states and climates of eccentric exoplanets.” _Icarus_ , Vol. 192, pp. 1 - 23 * [10] Efroimsky, M. 2006. The theory of bodily tides. The models and the physics. astro-ph/0605521 * [11] Efroimsky, M., & V. Lainey. 2007. “The Physics of Bodily Tides in Terrestrial Planets, and the Appropriate Scales of Dynamical Evolution.” _Journal of Geophysical Research – Planets_ , Vol. 112, p. E12003 doi:10.1029/2007JE002908 * [12] Efroimsky, M. 2008. “Can the tidal quality factors of terrestrial planets and moons scale as positive powers of the tidal frequency?” arXiv:0712.1056 * [13] Ferraz-Mello, S., Rodríguez, A., & Hussmann, H. 2008. “Tidal friction in close-in satellites and exoplanets: The Darwin theory re-visited.” _Celestial mechanics and Dynamical Astronomy,_ Vol. 101, pp. 171 - 201 * [14] Gerstenkorn, H. 1955. “Über Gezeitenreibung beim Zweikörperproblem.” Zeitschrift für Astrophysik, Vol. 36, pp. 245 - 274 * [15] Getino, J., Escapa, A., & García, A. 2003. “Spheroidal and Toroidal Modes for Tidal Kinetic Energy in Axisymmetric, Slightly Elliptical, Elastic Bodies.” _Romanian Astronomical Journal_ , Vol. , pp. 143 - 161 * [16] Goldreich, P. 1966a. “History of the Lunar Orbit.” _Reviews of Geophysics_. Vol. 4, pp. 411 - 439 * [17] Goldreich, P. 1966b. “Final spin states of planets and satellites.” _The Astronomical Journal_. Vol. 4, pp. 411 - 439 * [18] Goldreich, P., & Peale, S. 1966. “Spin-orbit coupling in the Solar System.” _The Astronomical Journal_. Vol. 71, pp. 425 - 438 * [19] Gooding, R. H., & Wagner, C. A. 2008. “On the inclination functions and a rapid stable procedure for their evaluation together with derivatives.” _Celestial Mechanics and Dynamical Astronomy_ , Vol. 101, pp. 247 - 272 * [20] Gurfil, P., Lainey, V., & Efroimsky, M. 2007. “Long-term evolution of orbits about a precessing oblate planet: 3. A semianalytical and a purely numerical approach.” _Celestial Mechanics and Dynamical Astronomy_ , Vol. 99, pp. 261 - 292 * [21] Herschel, J. F. W. 1863. “About Volcanoes and Earthquakes.” _Good Words_ , 4 February 1863, pp. 53 - 58. Reprinted in: Herschel, J. F. W. 1866. _Familiar Lectures on Scientific Subjects,_ pp. 1 - 46. Alexander Strahan Publishers, London & NY, 1866 * [22] Hut, P. 1981. “Tidal evolution in close binary systems.” _Astronomy & Astrophysics_, Vol. 99, pp. 126 - 140 * [23] Innanen, K. A., Zheng, J. Q., Mikkola, S., & Valtonen, M. J. 1997, “The Kozai Mechanism and the Stability of Planetary Orbits in Binary Star Systems”, The Astronomical Journal, Vol. 113, pp. 1915 - 1919. * [24] Johnson, Samuel. 1755. _A Dictionary of the English Language, in which the words are deduced from their originals, and illustrated with their different significations by examples from the best writers; to which are prefixed a history of the language, and an English grammar._ London. Printed by W. Strahan, for J. and P. Knapton, T. and T. Longman, C. Hitch, L. Hawes, A. Millar, R. and J. Dodsley. 1755. Folio. * [25] Kant, I. 1754. “Untersuchung der Frage, ob die Erde in ihrer Umdrehung um die Achse, wodurch sie die Abwechselung des Tages und der Nacht hervorbringt, einige Veränderung seit den ersten Zeiten ihres Ursprungs erlitten habe und woraus man sich ihrer versichern könne, welche von der Königl.” Akademie der Wissenschaften zu Berlin zum Preise für das jetztlaufende Jahr aufgegeben worden. In: _Kant’s gesammelte Schriften._ , Vol. I, pp. 183 - 191. Ed by the Royal Prussian Academy of Sciences, Georg Reimer Publishers, Berlin 1900 http://www.ikp.uni-bonn.de/Kant/aa01/Inhalt1.html English translations: Kant, I. 1754. _“Essay on the Retardation of the Rotation of the Earth.”_ Translation by William Hastie, in: Hastie, W. 1900. “Kant’s Cosmogony, as in his _Essay on the Retardation of the Rotation of the Earth_ and his _Natural History and Theory of the Heavens._ pp. 157 - 165. J Maclehose Publishers, Glasgow, 1900\. Reprinted: 1968, ed. Willy Ley (Greenwood Publishers, NY), and 1969, ed. Milton K. Munitz (University of Michigan Press). * [26] Karato, S.-i. 2007. _Deformation of Earth Materials. An Introduction to the Rheology of Solid Earth._ Cambridge University Press, UK. * [27] Kaula, W. M. 1961. “Analysis of gravitational and geometric aspects of geodetic utilisation of satellites.” The Geophysical Journal, Vol. 5, pp. 104 - 133 * [28] Kaula, W. M. 1964. “Tidal Dissipation by Solid Friction and the Resulting Orbital Evolution.” Reviews of Geophysics, Vol. 2, pp. 661 - 684 * [29] Kaula, W. M. 1966. _Theory of Satellite Geodesy: Applications of Satellites to Geodesy._ Blaisdell Publishing Co, Waltham MA. (Re-published in 2006 by Dover. ISBN: 0486414655.) * [30] Kaula, W. M. 1968. An Introduction to Planetary Physics. John Wiley & Sons, NY. * [31] Kozai, Y. 1959a. “The motion of a close earth satellite.” The Astronomical Journal, Vol. 64, pp. 367 - 377. * [32] Kozai, Y. 1959b. “On the effects of the Sun and the Moon upon the motion of a close Earth satellite.” SAO Special Report, Vol. 22, pp. 7 - 10. * [33] Kozai, Y. 1962. “Secular perturbations of asteroids with high inclination and eccentricity.” The Astronomical Journal, Vol. 67, pp. 591 - 598. * [34] Krasinsky, G. A. 2002. “Dynamical History of the Earth-Moon System.” _Celestial Mechanics & Dynamical Astronomy_, Vol. 84, pp. 27 - 55 * [35] Krasinsky, G. A. 2006. “Numerical theory of rotation of the deformable Earth with the two-layer fluid core. Part 1: Mathematical model.” _Celestial Mechanics & Dynamical Astronomy_, Vol. 96, pp. 169 - 217 * [36] Lambeck, K. 1980. The Earth’s Variable Rotation: Geophysical Causes and Consequences. Cambridge University Press, Cambridge UK * [37] MacDonald, G. J. F. 1964. “Tidal Friction.” Reviews of Geophysics. Vol. 2, pp. 467 - 541 * [38] Mignard, F. 1979. “The Evolution of the Lunar Orbit Revisited. I.” The Moon and the Planets. Vol. 20, pp. 301 - 315. * [39] Mignard, F. 1980. “The Evolution of the Lunar Orbit Revisited. II.” The Moon and the Planets. Vol. 23, pp. 185 - 201 * [40] Neron de Surgy, O., and Laskar, J. 1997. “On the long term evolution of the spin of the Earth.” _Astronomy & Astrophysics_, Vol. 318, pp. 975 - 989 * [41] Peale, S. 2005. “The free precession and libration of Mercury.” Icarus, Vol 178, pp 4 - 18 * [42] Roche, E. A. 1849. “Mémorie sur la figure d’une masse fluide, soumise a l’attraction d’un point éloingné.” _Académie des Sciences et Lettres de Montpellier. Mémories de la Section des Sciences._ , Tome 1, No 3, pp. 243 - 262 http://gallica.bnf.fr/ark:/12148/bpt6k209711r * [43] Singer, S. F. 1968. “The Origin of the Moon and Geophysical Consequences.” The Geophysical Journal of the Royal Astronomical Society, Vol. 15, pp. 205 - 226 * [44] Taff, L. G. 1985. _Celestial Mechanics. A Computational Guide for the Practitioner._ John Wiley & Sons, NY 1985, pp. 332 - 340 * [45] Thomson, W. 1863. “On the rigidity of the Earth.” _Philosophical Transactions of the Royal Society of London._ Vol. 153, pp. 573 - 582 http://www.jstor.org/view/02610523/ap000064/00a00270 * [46] Tisserand, F.-F. 1896. _Traité de Mécanique Céleste_. Tome I. _Perturbations des planètes d’après la méthode de la variation des constantes arbitraires._ Gauthier Villars, Paris 1896. Chapitre X. * [47] Touma, J., & Wisdom, J. 1994. “Evolution of the Earth-Moon system.” _The Astronomical Journal._ Vol. 108, pp. 1943 - 1961. * [48] Williams, J. G., & Benson, G. S. 1971. “Resonances in the Neptune-Pluto System.” _The Astronomical Journal,_ Vol. 71, pp. 167 - 176 * [49] Williams, J. G., Boggs, D. H., Yoder, C. F., Ratcliff, J. T. & Dickey, J. O. 2001. “Lunar rotational dissipation in solid body and molten core.” _The Journal of Geophysical Research - Planets_ , Vol. 106, No E11, pp. 27933 - 27968.

arxiv-papers

2024-09-04T02:48:54.482023

0803.3327

# Killing-Yano Forms of a Class of Spherically Symmetric Space-Times I: A Unified Generation of Killing Vector Fields Ö. Açık 1 ozacik@science.ankara.edu.tr Ü. Ertem 1 uertem@science.ankara.edu.tr M. Önder 2 onder@hacettepe.edu.tr A. Verçin 1 vercin@science.ankara.edu.tr 1 Department of Physics, Ankara University, Faculty of Sciences, 06100, Tandoğan-Ankara, Turkey 2 Department of Physics Engineering, Hacettepe University, 06800, Beytepe- Ankara, Turkey. ###### Abstract Killing-Yano one forms (duals of Killing vector fields) of a class of spherically symmetric space-times characterized by four functions are derived in a unified and exhaustive way. For well-known space-times such as those of Minkowski, Schwarzschild, Reissner-Nordstrøm, Robertson-Walker and several forms of de Sitter, these forms arise as special cases in a natural way. Besides its two well-known forms, four more forms of de Sitter space-time are also established with ten independent Killing vector fields for which four different time evolution regimes can explicitly be specified by the symmetry requirement. A family of space-times in which metric characterizing functions are of the general form and admitting six or seven independent Killing vector fields is presented. ###### pacs: 04.20.-q, 02.40.-k ## I Introduction Defining relations of Killing-Yano (KY) and conformal KY-forms are natural generalizations of Killing 1-forms and conformal Killing 1-forms. The latter are the dual of Killing and conformal Killing vector fields whose flows generate, respectively, local isometries and local conformal isometries of the metric in (pseudo)Riemannian geometry Benn-Tucker ; Thirring . Although they are not related to the isometries of the metric these higher rank generalizations have attracted increasing interest in various fields of physics and modern mathematics as well as in some related fields. Generally speaking, many interesting properties of a space-time are intimately connected with the existence of (conformal) KY-forms admitted by the corresponding metric. More specifically; the determination of KY-forms of a given metric, classification of space-times admitting KY-forms, analysis of the algebraic structures of these forms as well as specification of the symmetry algebra and related conserved quantities of the Dirac and related equations in a given curved background have gained increasing significance. Equally important objects (not considered in this study) are the totally symmetric Killing tensors and their conformal generalizations (see Dietz1 ; Dietz2 ; Benn-son and references therein). KY-forms play a prominent role in a unified description of null and non-null shear-free congruences Benn and in the search of force-free fields (divergence-free eigenvectors of the curl operator with position dependent eigenvalues) mostly encountered in astrophysics and fluid dynamics literatures Benn-Kress0 ; Benn-Kress3 ; Kress . A tangent vector which generates conformal transformations on its orthogonal complement is said to be the generator of shear-free congruence for its integral curve. Shear-free equation is a generalization of defining equation of conformal Killing equation. Together with Clifford calculus, KY-forms also provide efficient means in analyzing elliptic operators and the Dirac operator and in the further classification of (pseudo)Riemannian manifolds Benn-Tucker ; Semmelman1 . While conformal KY- forms take part in symmetry operators for the massless Dirac equation Benn- Charlton ; Benn-Kress3 , KY-forms are indispensable in constructing first order symmetries of the massless as well as massive Dirac equation in a curved space-time Benn-Kress1 . KY-forms are also necessary for the symmetries of the Kähler equation Benn-Tucker ; Benn-Kress2 . Longstanding interest in KY-forms largely stems from their constant use in general relativity and especially from their role in constructing conserved quantities in a number of ways. The studies of Penrose and his collaborators Walker ; Penrose , who have shown how the existence of a KY 2-form explains Carter’s result on the integrability of the geodesic equation in Kerr background, constitute a stepping-stone in this context Carter1 ; Carter2 . The fact that any KY p-form provides a quadratic first integral of the geodesic equation is by now a well-established particular result of the fact that the interior derivative of any KY p-form with respect to the tangent vector field of any geodesic remains parallel along the geodesic. More generally, any KY (p+1)-form is associated with a symmetric bilinear form, which is nothing more than the Killing tensor generalizing the so-called Stackel-Killing tensor that corresponds to a KY 2-form as first recognized by Penrose and Floyd. For more in this context, refer to Semmelman1 . On the other hand, as every KY form is co-closed, the Hodge dual of any KY $p$-form is directly associated with a conserved quantity. In the case of KY 1-forms, two types of conserved currents can be defined. For Ricci-flat space- times the Hodge dual of the exterior derivative of a KY 1-form $\omega$ is conserved, where ${}^{\ast}d\omega$ is known as the Komar form Benn-Tucker . Secondly, the current $j=i_{X^{a}}\omega\wedge^{\ast^{-1}}G_{a}$ defined in terms of the Einstein $3$-forms $G_{a}$, and KY 1-form $\omega$ is also conserved. Here, ∗ and ${}^{\ast^{-1}}$ represent the Hodge map and its inverse, $\wedge$ denotes exterior multiplication, $d$ and $i_{X}$ stand for exterior and interior derivatives (with respect to vector field $X$). It has recently been shown how the KY-forms of the flat space-time can be used to construct new, conserved gravitational charges for transverse asymptotically flat Kastor as well as for asymptotically anti de Sitter space-times Cebeci . These studies present another way of constructing a conserved current by taking a particular linear combination of wedge products of interior derivatives of the KY p-form and curvature characteristics of the underlying manifold. For KY 1-forms, this reduces to the usual current obtained from the Einstein $3$-form. The main purpose of this and the accompanying paper ozumav1 is to develop, by directly starting from the KY-equation, $\displaystyle\nabla_{X_{a}}\omega_{(p)}=\frac{1}{p+1}i_{X_{a}}d\omega_{(p)}\;,\quad p=1,2,3$ (1) a constructive method which makes it possible to generate all KY forms for a large class of spherically symmetric space-times in a unified and exhaustive way. Here $\nabla_{X}$ stands for the covariant derivative with respect to the vector field $X$. It should be noted that the KY 0-form $\omega_{(0)}$ can be any function, $\omega_{(1)}$ is the dual of a Killing vector field and $\omega_{(n)}$ is a constant (parallel), that is, it is a constant multiple $\omega_{(n)}=az$ of the volume form $z$ (for the orthonormal frame given below $z=e^{0123}$). The goal of our study is achieved by solving a coupled set of first order partial linear differential equations for the component functions of the KY forms $\omega_{(p)}$ in each case. The number of independent equations that have to be solved for $p=1,2,3$ are $10,\;18$ and $16$, respectively. To obtain the most general solutions in an exhaustive way, there are also $24,36$ and $24$ integrability conditions that must be carefully examined at the outset. We first solve a suitable set of the integrability conditions by which the set of solutions naturally branches into cases and subcases. We shall mainly use the notation of Benn-Tucker and adopt the following conventions and terminology. The underlying base manifold is supposed to be a $4$-dimensional ($4D$) pseudo-Riemannian manifold with the metric tensor $g$ having Lorentzian signature $(-+++)$ such that $g=-e^{0}\otimes e^{0}+e^{1}\otimes e^{1}+e^{2}\otimes e^{2}+e^{3}\otimes e^{3}$ in a local orthonormal co-frame $\\{e^{a}\\}$. By choosing the co-frame basis $\displaystyle e^{0}$ $\displaystyle=$ $\displaystyle H_{0}dt\;,\qquad e^{1}=TH_{1}dr\;,$ $\displaystyle e^{2}$ $\displaystyle=$ $\displaystyle TH_{2}d\theta\;,\quad e^{3}=TH_{2}\sin\theta d\varphi\;,$ a class of spherically symmetric metrics that will be considered can be parameterized by $\displaystyle T=\exp(\lambda(t))\;,\quad H_{j}=H_{j}(r)\;;\quad j=0,1,2\;$ which henceforth will be referred to as the (metric) coefficient functions. Here $(t,r,\theta,\varphi)$ specifies a local polar space-time chart with the usual range of variations. Whenever necessary, the range of $r$ can be bounded to keep the coefficient functions real. A generic property of this kind of metric is invariance under the transformation of the spatial rotation group $SO(3)$. If $g$ admits a time-like Killing vector field $K_{0}$, it is termed stationary and if, in addition, $K_{0}$ is orthogonal to a family of space- like hypersurfaces it is termed static. The dual tangent frame basis $\\{X_{a}\\}$ of the co-frame basis $\\{e^{a}\\}$ are, $\displaystyle X_{0}$ $\displaystyle=$ $\displaystyle\frac{1}{H_{0}}\partial_{t}\;,\qquad X_{1}=\frac{1}{TH_{1}}\partial_{r}\;,$ $\displaystyle X_{2}$ $\displaystyle=$ $\displaystyle\frac{1}{TH_{2}}\partial_{\theta}\;,\quad X_{3}=\frac{1}{TH_{2}\sin\theta}\partial_{\varphi}\;,$ where $e^{a}(X_{b})=\delta^{a}_{b}$ and $\partial_{x}=\partial/\partial x$. The metric dual of a vector field $X$ will be denoted by $\tilde{X}$ such that $\tilde{X}(Y)=g(X,Y)$ for any vector field $Y$. The torsion-free connection $1$-forms $\omega_{ab}$ for this class of metrics are well-known, and can be presented in the following antisymmetric matrix-valued $1-$form: $\displaystyle(\omega_{ab})=\frac{1}{T}\left(\begin{array}[]{cccc}0&&&\\\ \frac{h_{0}}{H_{1}}e^{0}+\frac{\dot{T}}{H_{0}}e^{1}&0&&\\\ \frac{\dot{T}}{H_{0}}e^{2}&\frac{h_{2}}{H_{1}}e^{2}&0&\\\ \frac{\dot{T}}{H_{0}}e^{3}&\frac{h_{2}}{H_{1}}e^{3}&\frac{\cot\theta}{H_{2}}e^{3}&0\end{array}\right)\;,$ (6) where we have used the abbreviations $\dot{T}=dT/dt,\;dH(r)/dr=H^{\prime}$ and $h_{j}=H^{\prime}_{j}/H_{j}$. We shall always use prime and over-dot to denote, respectively, the $r$-derivation and $t$-derivation of a function which depends only on $r$ and $t$. The partial derivative of a map $U$ of several variables with respect to $x$ will be denoted by $U_{x}$, and of a rational map or function $U/V$ by $\partial_{x}(U/V)$. The following matrix- valued $1$-form is helpful in carrying out the calculations $\displaystyle(\nabla_{X_{a}}e^{b})=-\frac{1}{T}\left(\begin{array}[]{cccc}\frac{h_{0}}{H_{1}}e^{1}&\frac{h_{0}}{H_{1}}e^{0}&0&0\\\ \frac{\dot{T}}{H_{0}}e^{1}&\frac{\dot{T}}{H_{0}}e^{0}&0&0\\\ \frac{\dot{T}}{H_{0}}e^{2}&-\frac{h_{2}}{H_{1}}e^{2}&\frac{\dot{T}}{H_{0}}e^{0}+\frac{h_{2}}{H_{1}}e^{1}&0\\\ \frac{\dot{T}}{H_{0}}e^{3}&-\frac{h_{2}}{H_{1}}e^{3}&-\frac{\cot\theta}{H_{2}}e^{3}&\frac{\dot{T}}{H_{0}}e^{0}+\frac{h_{2}}{H_{1}}e^{1}+\frac{\cot\theta}{H_{2}}e^{2}\end{array}\right)\;.$ (11) The corresponding curvature $2$-forms are presented in Appendix B. All the well-known spherically symmetric space-times such as the Minkowski, Schwarzschild, Reissner-Nordstrøm, Robertson-Walker and the six various forms of de Sitter models fall within the class of the considered metric as special cases and this provides the opportunity to give complete lists of their KY forms. As a particular result, we have found a completely solvable nonlinear ordinary differential equation $T^{2}\partial_{t}(\dot{T}/T)=\ell$, where $\ell$ is constant, characterizing five different time-dependent types of de Sitter space-time in a unified way. This fact also enables us to give explicit expressions of their KY-forms in a unified and exhaustive way. As the first part of our study, the present paper is entirely devoted to the unified generation of all Killing vector fields for the considered class of space- times. KY two and three forms are taken up in the next paper ozumav1 . Although KY-forms have been the subject of relatively recent active research, Killing vectors have been so intensely investigated that the following original points which hold for the KY-forms as well are worth emphasizing. (i) As has been mentioned above, there exist five well-known space-times that are covered by the considered class of metrics such that one of them (de Sitter) consists of six different types. Killing vector fields of these metrics are usually handled as case-by-case studies in scattered references. Our unified generation may remedy many inconveniences such as notational incompatibilities, proper range problems related to coordinates and relationships between cases. Moreover, one can explicitly observe the emergence of each case from the variations of the metric characterizing coefficients. (ii) Derivations of Killing vector fields for some types of de Sitter space-time from a five dimensional embedding flat manifold is geometrically very appealing and more inspiring physically (for a review of de Sitter spaces see Hawking ; Moschella and for recent interest see Randono and references therein). But the exact number of possible types naturally emerges from our study, and we identify an exactly solvable equation that determines this number by the number of its possible solutions. (iii) Our approach is exhaustive in the sense that all of the possible Killing vector fields of a given space-time can be completely determined from our approach so long as its metric belongs to the considered class. This fact makes it possible to reach decisive, or at least conclusive statements about a particular problem in which KY-forms are involved. We do not go into the detail of all possible cases but point out sufficiently symmetric cases, and have given some details of a particular case having six independent Killing vector fields that, as far as we know, does not appear in the literature. This example is also worth mentioning in light of the fact that its symmetry algebra changes drastically when a seemingly unimportant integration constant is changed. ## II KY 1-Forms: Defining Equations A $1$-form is a KY $1$-form if and only if it is the metric dual of a Killing vector field. This is equivalent to the fact that it satisfies equation (1) for $p=1$. For the components of $\displaystyle\omega_{(1)}=\alpha e^{0}+\beta e^{1}+\gamma e^{2}+\delta e^{3}\;,$ the KY equation gives, in view of (2) and (3), sixteen equations of which the following ten $\displaystyle\alpha_{t}$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}}{TH_{1}}\beta\;,\;\quad\quad\quad\beta_{r}=\dot{T}\frac{H_{1}}{H_{0}}\alpha\;,$ $\displaystyle T^{2}\partial_{t}\frac{\beta}{T}$ $\displaystyle=$ $\displaystyle-\frac{H^{2}_{0}}{H_{1}}\partial_{r}\frac{\alpha}{H_{0}}\;,\;\;\quad\beta_{\theta}=-\frac{H^{2}_{2}}{H_{1}}\partial_{r}\frac{\gamma}{H_{2}}\;,$ $\displaystyle T^{2}\partial_{t}\frac{\gamma}{T}$ $\displaystyle=$ $\displaystyle-\frac{H_{0}}{H_{2}}\alpha_{\theta}\;,\;\;\;\quad\quad\beta_{\varphi}=-\frac{H^{2}_{2}}{H_{1}}\sin\theta\partial_{r}\frac{\delta}{H_{2}}\;,$ (12) $\displaystyle T^{2}\partial_{t}\frac{\delta}{T}$ $\displaystyle=$ $\displaystyle-\frac{H_{0}}{H_{2}\sin\theta}\alpha_{\varphi}\;,\;\;\;\gamma_{\theta}=\dot{T}\frac{H_{2}}{H_{0}}\alpha-\frac{H^{\prime}_{2}}{H_{1}}\beta\;,$ $\displaystyle\delta_{\varphi}$ $\displaystyle=$ $\displaystyle\sin^{2}\theta\partial_{\theta}\frac{\gamma}{\sin\theta}\;,\quad\gamma_{\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{\delta}{\sin\theta}$ are independent. Although it appears difficult to solve this coupled set of first order partial differential equations directly, an exhaustive treatment of the problem with many classes of solutions is possible. At first to gain an initial insight into the above set of equations, let us look at some obvious solutions. We can immediately see that when all coefficient functions but $\alpha$ are zero, $T$ must be constant and $\alpha=H_{0}$. When $\beta$ alone is nonzero and proportional to $T$, the conditions $H^{\prime}_{0}=0=H^{\prime}_{2}$ must be fulfilled. The other two cases, in which only $\gamma$ or only $\delta$ is different from zero and proportional to $TH_{2}$, are forbidden by the last two equations of (4). But it is easy to verify that $\displaystyle\alpha=0=\beta\;,\quad\gamma=TH_{2}g\;,\quad\delta=TH_{2}(a_{3}\sin\theta+\cos\theta g_{\varphi})\;,$ is a solution, such that $g(\varphi)$ satisfies $g_{\varphi\varphi}+g=0$ and there is no additional constraint. We thus have, for $g=a_{1}\sin\varphi- a_{2}\cos\varphi$, the following three linearly independent KY 1-forms : $\displaystyle\tilde{K}_{1}$ $\displaystyle=$ $\displaystyle TH_{2}(\sin\varphi e^{2}+\cos\theta\cos\varphi e^{3})\;,$ $\displaystyle\tilde{K}_{2}$ $\displaystyle=$ $\displaystyle TH_{2}(-\cos\varphi e^{2}+\cos\theta\sin\varphi e^{3})\;,$ (13) $\displaystyle\tilde{K}_{3}$ $\displaystyle=$ $\displaystyle-TH_{2}\sin\theta e^{3}\;.$ The corresponding rotational Killing vector fields $\displaystyle K_{1}$ $\displaystyle=$ $\displaystyle\sin\varphi\partial_{\theta}+\cot\theta\cos\varphi\partial_{\varphi}\;,$ $\displaystyle K_{2}$ $\displaystyle=$ $\displaystyle-\cos\varphi\partial_{\theta}+\cot\theta\sin\varphi\partial_{\varphi}\;,$ (14) $\displaystyle K_{3}$ $\displaystyle=$ $\displaystyle-\partial_{\varphi}\;,$ are the well-known generators of the $so(3)$-algebra: $[K_{i},K_{j}]=\varepsilon_{ij}^{\;\;\;k}K_{k}$. As there is no constraint on these solutions, they must appear in every case independent of the specific form of the functions characterizing the metric. This is a typical characteristic of spherical symmetry. When $\dot{T}=0,\;\omega_{0}=H_{0}e^{0}$ is a KY $1$-form which corresponds to the time-like Killing vector field $K_{0}=\tilde{\omega}_{0}=-\partial_{t}$, and it can be combined with the above $so(3)$ solutions. In fact for $\alpha=H_{0},\;\beta=0$ and $\dot{T}=0$, the above $4D$ algebra is the dual of the most general solution of equations (4). Indeed in such a case, the first eight equations of (4) imply that $\gamma=TH_{2}G(\varphi)$ and $\delta=TH_{2}D(\theta,\varphi)$ and then the last two equations of (4) yield $\displaystyle D_{\varphi}=-\cos\theta G\;,\quad G_{\varphi}=-\sin\theta D_{\theta}+\cos\theta D\;.$ From the derivation of the second equation with respect to $\varphi$, we obtain $G_{\varphi\varphi}+G=0$ in view of the first equation. On the other hand, integration of the first equation gives $D=\cos\theta G_{\varphi}+f(\theta)$ which, upon substituting it into the second, gives $f_{\theta}=\cot\theta f$ whose integral is $f=a_{3}\sin\theta$. In the case of $H_{2}=r$, such space-times with $4D$ symmetry algebra include two physically important examples: the Reissner-Nordström (RN) and its special case Schwarzschild space-times, for which the other coefficient functions are given in Table I. It should be emphasized that the explicit forms of $H_{0}$ and $H_{1}$ are derived from the physical requirements, namely from the Einstein equations in the Schwarzschild case. For a general consideration, it turns out to be convenient to look for the solutions in the set of the solutions of some integrability conditions. This will also provide us with the necessary means to generate other sets of solutions. ## III Integrability Conditions and Their Solutions For $x=\alpha,\beta$ we have the integrability conditions $\displaystyle\partial_{\varphi}\partial_{\theta}(\frac{x}{\sin\theta})=0\;,\quad x_{\varphi\varphi}=\sin^{3}\theta\partial_{\theta}\frac{x_{\theta}}{\sin\theta}\;,$ (15) that follow from $\delta_{r\theta}=\delta_{\theta r},\delta_{rt}=\delta_{tr},\delta_{t\theta}=\delta_{\theta t},\delta_{t\varphi}=\delta_{\varphi t}$ and $\gamma_{r\varphi}=\gamma_{\varphi r},\gamma_{\theta\varphi}=\gamma_{\varphi\theta}$. There are two additional conditions for $\alpha$ and four conditions for $\gamma$ and $\delta$: $\displaystyle\partial_{r}\partial_{\theta}(\frac{\alpha}{H_{2}})$ $\displaystyle=$ $\displaystyle 0=\partial_{r}\partial_{\varphi}(\frac{\alpha}{H_{2}})\;,$ $\displaystyle\partial_{r}\partial_{\theta}(\frac{\delta}{H_{2}})$ $\displaystyle=$ $\displaystyle 0=\partial_{t}\partial_{\theta}(\frac{\delta}{T})\;,$ (16) $\displaystyle\partial_{t}\partial_{r}(\frac{\delta}{TH_{0}})$ $\displaystyle=$ $\displaystyle 0=\partial_{t}\partial_{r}(\frac{\gamma}{TH_{0}})\;.$ The first two conditions of (8) can be checked from $\alpha_{r\theta}=\alpha_{\theta r}$ and $\alpha_{r\varphi}=\alpha_{\varphi r}$. The second row of (8) follows from $\alpha_{\theta\varphi}=\alpha_{\varphi\theta},\beta_{\theta\varphi}=\beta_{\varphi\theta}$ and the last row can be seen from $\beta_{t\varphi}=\beta_{\varphi t}$ and $\beta_{r\theta}=\beta_{\theta r}$. The following two equations can be easily verified from the last row of (4) $\displaystyle y_{\varphi\varphi}+y=-\sin^{3}\theta\partial_{\theta}\frac{y_{\theta}}{\sin\theta}\;,$ (17) for $y=\gamma,\delta$. There are $24$ integrability conditions but only $20$ of them are independent, and the twelve shown above are sufficient for a unified and exhaustive investigation. #### III.0.1 The General Forms of $\alpha$ and $\beta$ In terms of the functions $f=f(t,r,\varphi),g=g(t,r,\theta)$ the first equation of (7) implies $\displaystyle x_{\varphi}=\sin\theta f\;,\quad x_{\theta}=\cot\theta x+g\;,$ and from the second equation of (7) we obtain $x=\sin^{2}\theta g_{\theta}-\sin\theta f_{\varphi}$. On substituting this solution into the above $x_{\varphi}$ and $x_{\theta}$ equations we arrive at $\displaystyle f_{\varphi\varphi}+f=0\;,\;g=\sin\theta\partial_{\theta}(\sin\theta g_{\theta})\;,$ whose general solutions can be written, with $f_{i}=f_{i}(t,r)$ and $g_{i}=g_{i}(t,r)$, as $\displaystyle f=-f_{1}\sin\varphi-f_{2}\cos\varphi\;,\quad g=-g_{1}\cot\theta-g_{2}\frac{1}{\sin\theta}\;.$ Here the minus signs are used for convenience. The general solution for $x$ is $\displaystyle x=g_{1}+g_{2}\cos\theta+\sin\theta(f_{1}\cos\varphi- f_{2}\sin\varphi)\;.$ In terms of the functions $A_{i}=A_{i}(t,r)$ and $B_{i}=B_{i}(t,r)$ let us define $\displaystyle\sigma^{A}$ $\displaystyle=$ $\displaystyle A_{1}\cos\varphi- A_{2}\sin\varphi\;,\quad\sigma^{B}=B_{1}\cos\varphi-B_{2}\sin\varphi\;,$ $\displaystyle A$ $\displaystyle=$ $\displaystyle\sin\theta\sigma^{A}+A_{3}\cos\theta\;,\quad\quad B=\sin\theta\sigma^{B}+B_{3}\cos\theta\;.$ Since the functions $f_{i}$ and $g_{i}$ are, in general, different for $\alpha$ and $\beta$, we can write their general forms very concisely as $\displaystyle\alpha=U+A\;,\quad\beta=V+B\;,$ (18) where $U$ and $V$ depend, like the $g_{1}$ term of $x$, on $t$ and $r$. Note that $A$ and $B$ depend on all of the coordinates and they satisfy the relations $\displaystyle A_{\theta\theta}+A=0=B_{\theta\theta}+B\;,\quad\sigma^{A}_{\varphi\varphi}+\sigma^{A}=0=\sigma^{B}_{\varphi\varphi}+\sigma^{B}\;.$ (19) In the case of $x=\alpha$, the first two conditions of (8) imply that the functions characterizing $A$ are proportional to $H_{2}$ and these enable us to write $A=H_{2}\xi$ such that $\displaystyle\xi=u\cos\theta+\sin\theta(v_{1}\cos\varphi- v_{2}\sin\varphi)\;,$ (20) where $u,v_{i}$ depend only on $t$. #### III.0.2 The General Forms of $\gamma$ and $\delta$ When $\alpha$ and $\beta$ given by (10) are substituted into the $\gamma_{\theta}$-equation of (4), we obtain $\displaystyle\gamma_{\theta}=(\dot{T}\frac{H_{2}}{H_{0}}U-\frac{H_{2}^{\prime}}{H_{1}}V)+\dot{T}\frac{H_{2}}{H_{0}}A-\frac{H_{2}^{\prime}}{H_{1}}B\;.$ For well-defined $\gamma$ solutions, the first term at the right hand side must be zero: $\displaystyle\dot{T}U=H_{0}PV\;,\quad(P=\frac{H_{2}^{\prime}}{H_{1}H_{2}})\;.$ (21) (Since the mentioned term is independent of $\theta$ it would lead, upon integration, to a $\gamma$ solution which linearly depends on $\theta$. In fact this condition results when the above $\gamma_{\theta}$ is used in eq. (9) of $\gamma$.) Then, in view of eq.(11), the $\gamma_{\theta}$-equation can be integrated to $\displaystyle\gamma=-(\dot{T}\frac{H_{2}}{H_{0}}A_{\theta}-\frac{H_{2}^{\prime}}{H_{1}}B_{\theta})+TH_{2}g\;,\;\;g_{\varphi\varphi}+g=0\;,$ (22) where the $\theta$-independent term $G=TH_{2}g(\varphi)$ comes from integration, and its form can be easily verified by substituting $\gamma$ in the $\alpha_{\theta}$ and $\beta_{\theta}$-equations of (4). Indeed, in the first case we get $\displaystyle T^{2}\partial_{t}(\frac{\dot{T}}{T}A_{\theta})-\frac{H_{0}^{2}}{H_{2}^{2}}A_{\theta}=T^{2}H_{0}P\partial_{t}\frac{B_{\theta}}{T}\;,$ (23) in addition to $\partial_{t}(G/T)=0$. In the second case, we obtain $\displaystyle\dot{T}\partial_{r}(\frac{A_{\theta}}{H_{0}})=\frac{H_{1}}{H_{2}^{2}}B_{\theta}+\partial_{r}(PB_{\theta})\;,$ (24) and $\partial_{r}(G/H_{2})=0$. These imply that $G$ is of the form $G=TH_{2}g(\varphi)$ and by the integrability condition (9) for $\gamma$, we see that $g$ must satisfy the equation given by (14). Having determined the general form of $\gamma$, we now use it in the last two equations of (4) to determine the form of $\delta$. The $\delta_{\varphi}$-equation of (4) directly gives $\displaystyle\delta_{\varphi}=\dot{T}\frac{H_{2}}{H_{0}}\sigma^{A}-\frac{H_{2}^{\prime}}{H_{1}}\sigma^{B}-TH_{2}\cos\theta g\;,$ which can be integrated to $\displaystyle\delta=-\dot{T}\frac{H_{2}}{H_{0}}\sigma^{A}_{\varphi}+\frac{H_{2}^{\prime}}{H_{1}}\sigma^{B}_{\varphi}+TH_{2}\cos\theta g_{\varphi}+aTH_{2}\sin\theta\;,$ (25) where $a$ is an integration constant and the $\varphi$-independent term $D=aTH_{2}\sin\theta$ again comes from integration, whose explicit form is determined from $\alpha_{\varphi},\;\beta_{\varphi}$ and the last two equations of (4). Indeed, the $\alpha_{\varphi}$-equation of (4) states that $D/T$ must be independent of $t$ and that the following condition must be satisfied : $\displaystyle T^{2}\partial_{t}(\frac{\dot{T}}{T}\sigma^{A}_{\varphi})-\frac{H_{0}^{2}}{H_{2}^{2}}\sigma^{A}_{\varphi}=T^{2}H_{0}P\partial_{t}\frac{\sigma^{B}_{\varphi}}{T}\;.$ (26) On the other hand, the $\beta_{\varphi}$-equation of (4) states that $D/H_{2}$ must be independent of $r$, and that the following condition must be satisfied : $\displaystyle\dot{T}\partial_{r}\frac{\sigma^{A}_{\varphi}}{H_{0}}=\frac{H_{1}}{H_{2}^{2}}\sigma^{B}_{\varphi}+\partial_{r}(P\sigma^{B}_{\varphi})\;.$ (27) As a particular result, we have $D=TH_{2}f(\theta)$ and when the found forms of $\gamma$ and $\delta$ are used in the last equation of (4), we obtain $f=a\sin\theta$. Note that the conditions (18) and (19) are contained in (15) and (16). As an intermediate result, all the metric coefficient functions have been specified in terms of seven functions $U,A_{i}$ and $B_{i}$, which can be determined from the first three equations of (4) and the conditions found in (13), (15) and (16). Therefore, seven equations of (4) have been analyzed, and the first three equations and conditions (15) and (16) remain to be solved. Note that the last terms of $\gamma$ and $\delta$ correspond to the $so(3)$ solutions. The rest of the investigation is entirely devoted to the specification of additional symmetries. #### III.0.3 The Clustering of Solutions We shall now substitute the solutions (10) and (12) into the first three equations of (4) to specify the unknown functions. The first three equations of (4) give the following three equations for $U$ $\displaystyle U_{t}$ $\displaystyle=$ $\displaystyle\frac{\dot{T}}{T}\frac{H_{0}^{\prime}H_{2}}{H_{0}H^{\prime}_{2}}U\;,$ $\displaystyle\partial_{r}(\frac{U}{H_{0}P})$ $\displaystyle=$ $\displaystyle H_{1}\frac{U}{H_{0}}\;,$ (28) $\displaystyle T^{2}\partial_{t}(\frac{\dot{T}}{T}U)$ $\displaystyle=$ $\displaystyle-\frac{H_{0}^{3}P}{H_{1}}\partial_{r}\frac{U}{H_{0}}\;,$ and the following three equations for $B$ $\displaystyle B=TL\xi_{t}\;,\quad B_{r}=\dot{T}\frac{H_{1}H_{2}}{H_{0}}\xi\;,\quad T^{2}\partial_{t}\frac{B}{T}=-\frac{H_{0}^{2}}{H_{1}}(\frac{H_{2}}{H_{0}})^{\prime}\xi\;,$ (29) where we have utilized relation (13) and the function $L=L(r)$ is defined by $\displaystyle L=\frac{H_{1}H_{2}}{H_{0}^{\prime}}\;.$ (30) As is obvious from the equations in this subsection, from here on the analysis critically depends on the derivatives of $T,H_{0}$ and $H_{2}$. Since $H_{2}^{\prime}$ is different from zero in all physically important space- times, we shall assume this to be the case throughout the paper. This means that the function $P$ is different from zero. The cases $H_{0}^{\prime}=0$ and $\dot{T}=0$ will be considered as particular cases in the last three sections. In the next section, we proceed to look for solutions for which both $H_{0}^{\prime}$ and $\dot{T}$ can be different from zero. ## IV Three classes of solutions Using the first equation of (21) in the other two equations of (21), we obtain two equations which accept separation of variables such that $\displaystyle m\xi_{t}=\frac{\dot{T}}{T}\xi\;,\quad T^{2}\xi_{tt}=-m_{0}\xi\;,$ (31) provided that $m$ and $m_{0}$ defined by $\displaystyle m=L^{\prime}\frac{H_{0}}{H_{1}H_{2}}\;,\quad m_{0}=\frac{H^{2}_{0}}{LH_{1}}(\frac{H_{2}}{H_{0}})^{\prime}\;,$ (32) are constants. When one side of a separable equation depends entirely on the metric coefficient functions, we shall use the letters $k,l$ and $m$ for separation constants. Other integration constants will be denoted by the letters $a,b$ and $c$. In view of the ansatz $U=H_{0}Y(r)f(t)$ the first two equations of (20) transform to $\displaystyle f_{t}=k\frac{\dot{T}}{T}f\;,\quad(\frac{Y}{P})^{\prime}=H_{1}Y\;,$ (33) where the metric constant $k$ is defined by $\displaystyle k=\frac{H^{\prime}_{0}H_{2}}{H_{0}H_{2}^{\prime}}\;.$ (34) On the other hand from the last equation of (20), with the same ansatz, we get $\displaystyle T^{2}\partial_{t}(\frac{\dot{T}}{T}f)=k_{0}f\;,\quad\frac{Y^{\prime}}{Y}=-k_{0}\frac{H_{1}}{H_{0}^{2}P}\;,$ (35) where $k_{0}$ is a separation constant. Two equations of (25) can be easily integrated to $\displaystyle f=c_{1}T^{k}\;,\quad Y=c_{2}H_{2}P\;.$ (36) Thus $U=cH_{0}H_{2}PT^{k}$ where $c_{i}$ are integration constants and $c=c_{1}c_{2}$. When these solutions are substituted into (27) we see that $T$ must satisfy $\displaystyle T\ddot{T}+(k-1)\dot{T}^{2}=k_{0}\;,$ (37) and $k_{0}$ must be the metric constant $\displaystyle k_{0}=-\frac{H_{0}^{2}}{H_{1}H_{2}}(\frac{H_{2}^{\prime}}{H_{1}})^{\prime}\;.$ (38) Noting that equations (20) and (21) are linear in $U$ and $B$, depending on the values of $m$ one can distinguish three classes of solutions for which both $H_{0}^{\prime}$ and $\dot{T}$ can be different from zero. These can be characterized as follows $\displaystyle(A)\;\;m=0\;,\quad(B)\;\;m\neq 0\;;\;\;U=0=V\;,\quad(C)\;\;m\neq 0\;.$ The cases (i) $\dot{T}=0,\;H_{0}^{\prime}\neq 0$, (ii) $\dot{T}\neq 0,\;H_{0}^{\prime}=0$ and (iii) $\dot{T}=0=H_{0}^{\prime}$ will be considered separately in the last three sections. ### IV.1 $m=0$ Solutions When $m$ is zero $L$ is a nonzero constant and we have ${H_{0}^{\prime}}L=H_{1}H_{2}$. In that case, equations given by (23) imply that $\xi=0=B$ and that $m_{0}$ need not be a constant. The conditions (15) and (16) are both trivially satisfied for this case and hence, $\gamma$ and $\delta$ solutions correspond only to the $so(3)$ solutions. Only the metric constants $k$ and $k_{0}$ are defined in that case and, provided that $T$ satisfies equation (29), we have the following solutions for $\alpha$ and $\beta$: $\displaystyle\alpha=U=cT^{k}H_{0}\frac{H_{2}^{\prime}}{H_{1}}\;,\quad\beta=V=c\dot{T}T^{k}H_{2}\;,$ (39) which determine the Killing vector field $\displaystyle X=-\frac{H_{2}^{\prime}}{H_{1}}T^{k}\partial_{t}+\dot{T}T^{k-1}\frac{H_{2}}{H_{1}}\partial_{r}\;,$ (40) commuting with the $so(3)$ vector fields. ### IV.2 $m\neq 0\;{\rm and}\;U=0=V$ Solutions When $m$ is different from zero, the first equation of (23) can be easily solved to be $\displaystyle\xi=T^{1/m}\zeta\;,\quad\zeta=a_{1}\cos\theta+\sin\theta(a_{2}\cos\varphi- a_{3}\sin\varphi)\;,$ (41) where $a_{i}$ are integration constants. The second equation of (23) then implies that $\displaystyle T^{(2m-1)/m}\partial_{t}(\dot{T}T^{(1-m)/m})=-mm_{0}\;.$ (42) This equation can equivalently be read as $\displaystyle(1-m)\dot{T}^{2}+mT\ddot{T}=-m^{2}m_{0}\;,$ (43) or more concisely, in terms of $K=T^{1/m}$, as $\displaystyle\ddot{K}=-m_{0}K^{1-2m}\;.$ (44) The corresponding $\alpha,\beta,\gamma$ and $\delta$ solutions are as follows: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle T^{1/m}H_{2}\zeta\;,\quad\beta=WL\zeta\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle\frac{1}{m_{0}}WH_{0}\zeta_{\theta}+TH_{2}g\;,\quad(g_{\varphi\varphi}+g=0)$ (45) $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\frac{1}{m_{0}}WH_{0}\sigma_{\varphi}+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;,$ where $W=\dot{T}T^{1/m}/m=\dot{K}T$ and $\displaystyle\sigma^{A}=T^{1/m}H_{2}\sigma\;,\quad\sigma^{B}=WL\sigma\;,\quad\sigma=a_{2}\cos\varphi- a_{3}\sin\varphi\;.$ In view of (24) and equation (35); it is easy to verify that while condition (15) yields, $\displaystyle m_{0}(\frac{H_{2}^{\prime}}{H_{0}^{\prime}}-m\frac{H_{2}}{H_{0}})=\frac{H_{0}}{H_{2}}\;,$ (46) for $A=T^{1/m}H_{2}\zeta$ and $B=WL\zeta$, condition (16) yields an equation that is just the derivative of (38). In writing $\gamma$ and $\delta$ of (37) the condition (38) has been used. ### IV.3 $m\neq 0$ Solutions In this case, solutions are just a combination of the above two classes: $\gamma$ and $\delta$ solutions are as in equations (37) but to the $\alpha$ and $\beta$ solutions of (37), one must add that given by (31). Equation (38) also hold in this case. However, there are four metric constants $m,m_{0},k$ and $k_{0}$, and $T$ must satisfy both the equations (29) and (34) (or equivalently (35) or (36)). A detailed investigation of the four metric constants presented in Appendix A shows that for $H^{\prime}_{2}$ to be nonzero, the following conditions must be satisfied: $\displaystyle km=1\;,\quad m=1+m_{0}\ell^{2}\;,$ where $\ell$ is another metric constant. In fact, from equations (29), (35) and the above relation, we also get $k_{0}=-mm_{0}$. To all these one must also add relation (38). In the case of (B) class solutions, we have six arbitrary real constants: $a$ and two constants determined by the function $g$ and three constants given by $\zeta$. These mean that we have six linearly independent Killing vector fields for class (B) which will be presented together with their Lie algebra in the next subsection. In the case of (C), which includes (A) and (B) solutions as special subcases, we have only one additional symmetry generator given by (32). Although some other subcases of (C) can be defined, this case will not be pursued any further as it has many additional conditions. ### IV.4 Killing Vector Fields and Lie Algebra for the case (B) In terms of the two nonzero constants $m$ and $m_{0}$ defined by (24), we have obtained, in addition to three $so(3)$ 1-forms given by (5), three additional linearly independent KY 1-forms: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle\cos\theta\omega-\frac{1}{m_{0}}WH_{0}\sin\theta e^{2}\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle\sin\theta\cos\varphi\omega+\frac{1}{m_{0}}WH_{0}(\cos\theta\cos\varphi e^{2}-\sin\varphi e^{3})\;,$ (47) $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle-\sin\theta\sin\varphi\omega-\frac{1}{m_{0}}WH_{0}(\cos\theta\sin\varphi e^{2}+\cos\varphi e^{3})\;,$ where the 1-form $\omega$ is defined by $\displaystyle\omega=T^{1/m}H_{2}e^{0}+WLe^{1}\;.$ (48) By noting that the metric dual of $e^{0}$ is $-H_{0}^{-1}\partial_{t}$, in terms of $K=T^{1/m}$ the vector field $X$ which is the metric dual of $\omega$ can be written as $\displaystyle X=\tilde{\omega}=-K\frac{H_{2}}{H_{0}}\partial_{t}+\dot{K}\frac{H_{2}}{H^{\prime}_{0}}\partial_{r}\;.$ (49) The Killing vector fields corresponding to (39) are then given by $\displaystyle X_{1}$ $\displaystyle=$ $\displaystyle\cos\theta X-MZ_{1}\;,$ $\displaystyle X_{2}$ $\displaystyle=$ $\displaystyle\sin\theta\cos\varphi X+MZ_{2}\;,$ (50) $\displaystyle X_{3}$ $\displaystyle=$ $\displaystyle-\sin\theta\sin\varphi X-MZ_{3}\;,$ where $M=\dot{K}H_{0}/m_{0}H_{2}$ and the vector fields $Z_{i}$ are defined as $\displaystyle Z_{1}=\sin\theta\partial_{\theta}\;,\quad Z_{2}=\cos\theta\cos\varphi\partial_{\theta}-\frac{\sin\varphi}{\sin\theta}\partial_{\varphi}\;,\quad Z_{3}=\cos\theta\sin\varphi\partial_{\theta}+\frac{\cos\varphi}{\sin\theta}\partial_{\varphi}\;.$ (51) By making use of $\displaystyle\;[Z_{1},Z_{2}]$ $\displaystyle=$ $\displaystyle K_{2}\;,\quad\quad[Z_{1},Z_{3}]=-K_{1}\;,\quad[Z_{2},Z_{3}]=-K_{3}\;,$ (52) $\displaystyle\;[Z_{1},K_{1}]$ $\displaystyle=$ $\displaystyle- Z_{3}\;,\quad[Z_{1},K_{2}]=Z_{2}\;,\quad\quad[Z_{1},K_{3}]=0\;,$ $\displaystyle\;[Z_{2},K_{1}]$ $\displaystyle=$ $\displaystyle 0\;,\quad\quad\;[Z_{2},K_{2}]=-Z_{1}\;,\;\quad[Z_{2},K_{3}]=-Z_{3}\;,$ (53) $\displaystyle\;[Z_{3},K_{1}]$ $\displaystyle=$ $\displaystyle Z_{1}\;,\quad\;\;[Z_{3},K_{2}]=0\;,\quad\;\;\quad[Z_{3},K_{3}]=Z_{2}\;,$ we obtain $\displaystyle\;[K_{1},X_{1}]$ $\displaystyle=$ $\displaystyle X_{3}\;,\quad\quad[K_{1},X_{2}]=0\;,\quad\quad\quad\quad[K_{1},X_{3}]=-X_{1}\;,$ $\displaystyle\;[K_{2},X_{1}]$ $\displaystyle=$ $\displaystyle X_{2}\;,\quad\quad[K_{2},X_{2}]=-X_{1}\;,\quad\quad\;\;[K_{2},X_{3}]=0\;,$ (54) $\displaystyle\;[K_{3},X_{1}]$ $\displaystyle=$ $\displaystyle 0\;,\quad\quad\;\;[K_{3},X_{2}]=-X_{3}\;,\quad\;\;\quad[K_{3},X_{3}]=X_{2}\;.$ On the other hand in terms of $s=X(M)+M^{2}$ we have $\displaystyle\;[X_{1},X_{2}]$ $\displaystyle=$ $\displaystyle sK_{2}\;,\quad[X_{1},X_{3}]=-sK_{1}\;,\quad[X_{2},X_{3}]=-sK_{3}\;.$ (55) For evaluation of $s$, we should recall that $K$ obeys the equation (36). For $m=1$ we have $K=T$ and $T\ddot{T}=-m_{0}$, which can be integrated to $\dot{T}^{2}+2m_{0}\ln T=\ell_{0}$. For $m\neq 1$, it can also be easily integrated once to obtain $\displaystyle\dot{K}^{2}=-\frac{m_{0}}{1-m}K^{2(1-m)}+m_{2}\;,$ (56) where $\ell_{0}$ and $m_{2}$ are integration constants determined by the metric. We can therefore write $\displaystyle s$ $\displaystyle=$ $\displaystyle-K\frac{H_{2}}{H_{0}}M_{t}+\dot{K}\frac{H_{2}}{H^{\prime}_{0}}M_{r}+M^{2}\;,$ $\displaystyle=$ $\displaystyle\frac{1}{m_{0}}[-K\ddot{K}+(1-m)\dot{K}^{2}]\;,$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{cc}\frac{1-m}{m_{0}}m_{2}\;,&\quad{\rm for}\quad m\neq 1,\\\ 1\;,&\quad{\rm for}\quad m=1\;,\end{array}\right.$ where we have made use of (38) in the second line. When $m\neq 1$ and $s\neq 0$ the $X_{i}$ generator can be normalized with the same constant such that at the right hand side of (47), there appear $\pm K_{j}$. Such a normalization does not affect relations (46). But when the integration constant $m_{2}$ is zero, the generator set $\\{X_{1},X_{2},X_{3}\\}$ form an abelian subalgebra. In that case the symmetry algebra is isomorphic to $3D$ Euclidean algebra $e(3)$. This shows how the symmetry of the metric may change for different values of an integration constant. ## V Maximal Symmetries : $\dot{T}=0,\;H_{0}^{\prime}\neq 0$ Solutions Let us begin this case by considering $\alpha$ and $\beta$ as given in equation (10), such that $A=(H_{0}^{\prime}/H_{1})\xi$ and the functions $V,\;V_{i},\;u$ and $v_{i}$ depend only on $\tau=t/T$ for $\beta$ is independent of $r$. Since $\dot{T}=0$ in this case, condition (13) requires that $V=0$ and we can then write, from the first equation of (4): $\displaystyle\alpha=U+\frac{H_{0}^{\prime}}{H_{1}}\xi\;,\quad\beta=\xi_{\tau}\;,$ (60) where $\xi$ is given by (12) and $U_{\tau}=0$. The third equation of (4) now provides us with $\displaystyle U=c_{0}H_{0}\;,\quad\xi_{\tau\tau}+k_{1}\xi=0\;,$ (61) where $c_{0}$ is an integration constant, and the metric constant $k_{1}$ is defined by $\displaystyle k_{1}=\frac{H_{0}^{2}}{H_{1}}(\frac{H_{0}^{\prime}}{H_{0}H_{1}})^{\prime}\;.$ (62) Note that the second equation of (51) is equivalent to three similar equations for $u,v_{1}$ and $v_{2}$, each of which gives two linearly independent solutions depending on the value of $k_{1}$. Having completely specified $\alpha$ and $\beta$ with $A=H_{0}^{\prime}\xi/H_{1}$ and $B=\xi_{\tau}$, we now turn to conditions (15) and (16), which in this case amount to $\displaystyle H_{0}H_{0}^{\prime}=k_{1}H_{2}H_{2}^{\prime}\;,\quad P^{\prime}=-\frac{H_{1}}{H_{2}^{2}}.$ (63) $k_{1}$ is a nonzero metric constant for $H_{0}^{\prime}\neq 0$. Both of these relations can be integrated to $\displaystyle H_{0}^{2}=k_{1}H_{2}^{2}+k_{2}\;,\quad P^{2}=k_{3}+\frac{1}{H_{2}^{2}}\;,$ (64) where $k_{2}$ and $k_{3}$ are integration constants. The first relation of (54) implies that for $H_{0}^{\prime}\neq 0,\;k_{1}$ must be different from zero. The conditions (52) and (53) also imply that $\displaystyle P^{2}=\frac{H_{0}^{2}}{k_{2}H_{2}^{2}}\;,\quad k_{1}=k_{2}k_{3}\;,$ (65) which means for $H_{2}^{\prime}\neq 0,\;k_{2}$ and $k_{3}$ must be different from zero as well. If $k_{1}$ is a negative constant such that $k_{1}=-\kappa^{2}$ where $\kappa$ is a nonzero real number, we can write, in terms of integration constants $c_{i}$, the $\xi$ solutions of (51) as $\displaystyle\xi$ $\displaystyle=$ $\displaystyle(c_{1}\cosh\kappa\tau+c_{2}\sinh\kappa\tau)\cos\theta+\sin\theta\sigma^{A}\;,$ $\displaystyle\sigma^{A}$ $\displaystyle=$ $\displaystyle(c_{3}\cosh\kappa\tau+c_{4}\sinh\kappa\tau)\cos\varphi-(c_{5}\cosh\kappa\tau+c_{6}\sinh\kappa\tau)\sin\varphi\;.$ Then $\gamma$ and $\delta$ can be determined from (14) and (17) as $\displaystyle\gamma=\frac{H_{2}^{\prime}}{H_{1}}\xi_{\tau\theta}+TH_{2}g\;,\quad\delta=\frac{H_{2}^{\prime}}{H_{1}}\sigma^{A}_{\tau\varphi}+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;.$ (66) The solutions (50) and (56) determine seven linearly independent KY $1$-forms in addition to $so(3)$ solutions. The first one is $\omega_{0}=H_{0}e^{0}$ which corresponds to $K_{0}=-\partial/\partial t$. For $k_{1}=-\kappa^{2}$, we can write these additional forms as follows: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle\cos\theta\psi_{1}-\kappa\frac{H_{2}^{\prime}}{H_{1}}\sinh\kappa\tau\sin\theta e^{2}\;,\quad\;\;\;\omega_{2}=\cos\theta\psi_{2}-\kappa\frac{H_{2}^{\prime}}{H_{1}}\cosh\kappa\tau\sin\theta e^{2}\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\sin\theta\cos\varphi\psi_{1}+\kappa\frac{H_{2}^{\prime}}{H_{1}}\sinh\kappa\tau\phi_{1}\;,\;\;\quad\omega_{4}=\sin\theta\cos\varphi\psi_{2}+\kappa\frac{H_{2}^{\prime}}{H_{1}}\cosh\kappa\tau\phi_{1}\;,$ (67) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle-\sin\theta\sin\varphi\psi_{1}-\kappa\frac{H_{2}^{\prime}}{H_{1}}\sinh\kappa\tau\phi_{2}\;,\quad\omega_{6}=-\sin\theta\sin\varphi\psi_{2}-\kappa\frac{H_{2}^{\prime}}{H_{1}}\cosh\kappa\tau\phi_{2}\;,$ where the $1$-forms $\psi_{i}$ and $\phi_{i},\;i=1,2$ are defined by $\displaystyle\psi_{1}$ $\displaystyle=$ $\displaystyle\cosh\kappa\tau\frac{H_{0}^{\prime}}{H_{1}}e^{0}+\kappa\sinh\kappa\tau e^{1}\;,\quad\phi_{1}=\cos\theta\cos\varphi e^{2}-\sin\varphi e^{3}\;,$ $\displaystyle\psi_{2}$ $\displaystyle=$ $\displaystyle\sinh\kappa\tau\frac{H_{0}^{\prime}}{H_{1}}e^{0}+\kappa\cosh\kappa\tau e^{1}\;,\quad\phi_{2}=\cos\theta\sin\varphi e^{2}+\cos\varphi e^{3}\;.$ The vector fields $W_{i}=\tilde{\psi}_{i}$ and $\bar{Z}_{i+1}=\tilde{\phi}_{i}$ are, in terms of $Z_{2}$ and $Z_{3}$ given by (43), as follows: $\displaystyle W_{1}$ $\displaystyle=$ $\displaystyle-\cosh\kappa\tau\frac{H_{0}^{\prime}}{TH_{0}H_{1}}\partial_{\tau}+\frac{\kappa}{TH_{1}}\sinh\kappa\tau\partial_{r}\;,\quad\bar{Z}_{2}=\frac{1}{TH_{2}}Z_{2}\;,$ (68) $\displaystyle W_{2}$ $\displaystyle=$ $\displaystyle-\sinh\kappa\tau\frac{H_{0}^{\prime}}{TH_{0}H_{1}}\partial_{\tau}+\frac{\kappa}{TH_{1}}\cosh\kappa\tau\partial_{r}\;,\quad\bar{Z}_{3}=\frac{1}{TH_{2}}Z_{3}\;.$ (69) It is also not difficult to verify that for $H_{2}=r$ we have, from (54) $\displaystyle H_{0}^{2}=k_{1}r^{2}+k_{2}\;,\quad H_{1}^{2}=\frac{1}{1+k_{3}r^{2}}\;,$ and $H_{0}H_{1}=1$ for $k_{2}=1$. In the case of $k_{1}=-1=k_{3}$ and $k_{2}=1$, we recover the static form of the de Sitter metric (see Thirring pp.492). KY 1-forms for five different forms of the de Sitter type space-times are obtained in the next section. ## VI $\rm{de}$ Sitter and Robertson-Walker Type Symmetries: $H_{0}^{\prime}=0,\dot{T}\neq 0$ Since $\alpha_{t}=0$ in this case, it is convenient to start by defining the constant $\displaystyle\ell=T^{2}\partial_{t}\frac{\dot{T}}{T}\;.$ (70) The nonzero and zero values of $\ell$ will then be considered separately. In these two cases, $T$ is restricted to be a special function of time by the symmetry requirement. Case A considered below leads us to a family of de Sitter type space-times with ten independent Killing vector fields and, depending on the values of $\ell$ and other integration constants, four different time evolution regimes can explicitly be specified by the symmetry requirement. The B case, specified by $\ell=0$, corresponds to the best known form of de Sitter space-time, again having ten independent Killing vector fields such that $T$ is an exponential function of time. However, there is an important special case specified by $\alpha=0$, and therefore $T$ is not restricted by any symmetry requirement. This corresponds to the Robertson-Walker space-time with six dimensional symmetry algebra in which they are the Einstein equations that give the time dependence as shown in Table I. ### VI.1 The case $\ell\neq 0$ We start with $\displaystyle\alpha=U(r)+H_{2}\zeta\;,\quad\beta=\dot{T}Y(r)+B\;,$ (71) where $\zeta$ is given by (33) and $B$ is defined as in Section III. Condition (13) implies that $U=H_{0}PY$ and the second and third equations of (4) then yield $\displaystyle Y^{\prime}$ $\displaystyle=$ $\displaystyle H_{1}PY\;,\quad\quad\quad\;Y=-\frac{H^{2}_{0}}{\ell H_{1}}(PY)^{\prime}\;,$ (72) $\displaystyle B_{r}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{1}H_{2}}{H_{0}}\zeta\;,\;\;T^{2}\partial_{t}\frac{B}{T}=-H_{0}H_{2}P\zeta\;.$ (73) These are the reduced forms of equations (20) and (21). The first equation of (62) gives $Y=c_{1}H_{2}$ and from the second equation we then obtain $\displaystyle P^{\prime}=-H_{1}(P^{2}+\frac{\ell}{H_{0}^{2}})\;.$ (74) The two equations of (63) imply that, in terms of $\displaystyle\eta_{2}=b_{1}\cos\theta+\sin\theta\sigma^{b}\;,\;\;\sigma^{b}=b_{2}\cos\varphi- b_{3}\sin\varphi\;,$ (75) the most general solution for $B$ is of the form $B=\dot{T}\eta_{1}(r,\theta,\varphi)+T\eta_{2}(\theta,\varphi)$. The second equation of (63) specifies $\eta_{1}$ in terms of $\zeta$: $\displaystyle\eta_{1}=-\frac{H_{0}}{\ell}H_{2}P\zeta\;,$ (76) and the first equation yields nothing but condition (64). Having completely specified $\alpha$ and $\beta$ with $A=H_{2}\zeta$ and $\displaystyle B=-\frac{H_{0}}{\ell}\dot{T}H_{2}P\zeta+T\eta_{2}=\sin\theta\sigma^{B}+\cos\theta(b_{1}T-\frac{H_{0}}{\ell}\dot{T}H_{2}P\sigma)\;,$ (77) such that $\displaystyle\sigma^{B}=T\sigma^{b}-\frac{1}{\ell}\dot{T}H_{0}H_{2}P\sigma\;,$ (78) the condition (14) amounts to $\displaystyle P^{2}=-\frac{\ell}{H_{0}^{2}}+\frac{1}{H_{2}^{2}}\;.$ (79) It is not difficult to verify that condition (64) is implied by (69), which also yields $P^{\prime}=-H_{1}/H_{2}^{2}$. (Note that the integration of (64) results in an additional constant multiplying the second term of (69)). In view of this relation, condition (16) is identically satisfied. That is, we only have condition (69). We can now turn to (14) and (17) to evaluate $\gamma$ and $\delta$, and the solutions can be collated as follows: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{1}H_{0}H_{2}P+H_{2}\zeta\;,$ $\displaystyle\beta$ $\displaystyle=$ $\displaystyle c_{1}\dot{T}H_{2}+\sin\theta\sigma^{B}+\cos\theta(b_{1}T-\frac{H_{0}}{\ell}\dot{T}H_{2}P\sigma)\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle-\dot{T}\frac{H_{2}^{2}}{H_{0}}\zeta_{\theta}+\dot{T}H_{2}P[\cos\theta\sigma^{B}-\sin\theta(b_{1}T-\frac{H_{0}}{\ell}\dot{T}H_{2}P\sigma)]+TH_{2}g\;,$ (80) $\displaystyle\delta$ $\displaystyle=$ $\displaystyle-\dot{T}\frac{H_{2}}{H_{0}}\sigma_{\varphi}+H_{2}P\sigma^{B}_{\varphi}+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;.$ Note that for $H_{0}=1$ and $H_{2}=r$, condition (69) yields $\displaystyle H^{2}_{1}=\frac{1}{1-\ell r^{2}}\;.$ (81) These are the characteristics of the de Sitter and Robertson-Walker space- times. It should be emphasized that the right hand side of condition (69) must be positive, which reflects the fact that $1/r^{2}>\ell$ is required to avoid any singularity in the corresponding space-times. We now turn to the explicit evolution of $T$. By multiplying both sides of equation (60) by $\dot{T}/T^{3}$, it can be integrated to $\displaystyle\dot{T}=\epsilon(\ell_{3}T^{2}-\ell)^{1/2}\;,$ (82) and then by integrating once more we get $\displaystyle T=\left\\{\begin{array}[]{ll}(\frac{\ell}{\ell_{3}})^{1/2}\cosh(\epsilon\ell_{3}^{1/2}t+a)\;;&\;\;{\rm for}\;\ell_{3}>0\;,\;\ell>0\;,\\\ \frac{\ell_{0}}{\ell_{3}^{1/2}}\sinh(\epsilon\ell_{3}^{1/2}t+a)\;;&\;\;{\rm for}\;\ell_{3}>0\;,\;\ell=-\ell_{0}^{2}<0\;,\\\ \frac{\ell_{0}}{k_{0}}\sin(\epsilon k_{0}t+a)\;;&\;\;{\rm for}\;\ell_{3}=-k_{0}^{2}<0\;,\;\ell=-\ell_{0}^{2}<0\;,\\\ \epsilon\ell_{0}t+a\;;&\;\;{\rm for}\;\ell_{3}=0,\;\ell=-\ell_{0}^{2}<0\;,\end{array}\right.$ (87) where $\ell_{3}$ and $a$ are integration constants and $\epsilon=\pm 1$. Note that the third solution can be inferred from the second one and that all the corresponding KY 1-forms can explicitly be read from (70). All of these de Sitter space-times of which the fourth one is a flat space-time, are spaces of constant curvature with the curvature scalar given by $\ell_{3}$ (see Appendix B). ### VI.2 de Sitter type Symmetries: $\ell=0=H_{0}^{\prime}$ The condition $\ell=0$ is equivalent to $\dot{T}=\lambda T$, that is to $\displaystyle T=T_{0}\exp(\lambda t)\;,$ where $\lambda$ is a metric constant and $T_{0}$ is an integration constant. In this case, to avoid excessive repetitions, we shall be content to present the solutions: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{0}H_{0}+\frac{1}{H_{0}P}\zeta\;,$ $\displaystyle\beta$ $\displaystyle=$ $\displaystyle\dot{T}[\frac{c_{0}}{P}+\frac{1}{2}(\frac{1}{\dot{T}^{2}}+\frac{1}{P^{2}})\zeta+b_{1}\cos\theta+\sin\theta\sigma^{b}]\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle\dot{T}H_{2}P\\{[-\frac{1}{H_{0}^{2}P^{2}}+\frac{1}{2}(\frac{1}{\dot{T}^{2}}+\frac{1}{P^{2}})]\zeta_{\theta}-b_{1}\sin\theta+\cos\theta\sigma^{b}\\}+TH_{2}g\;,$ (88) $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\dot{T}H_{2}P\\{[-\frac{1}{H_{0}^{2}P^{2}}+\frac{1}{2}(\frac{1}{\dot{T}^{2}}+\frac{1}{P^{2}})]\sigma_{\varphi}+\sigma_{\varphi}^{b}\\}+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;,$ which provide us with ten-dimensional symmetry algebra. Here, $\sigma^{b}$ is given by (65). The above solutions can be easily verified provided that the condition $H_{2}^{\prime}=\epsilon H_{1}$ holds, which implies that $H_{2}P=\epsilon$ and hence, $P^{\prime}=-H_{1}P^{2}$. ### VI.3 The Case $\alpha=0$: Robertson-Walker Symmetries When $\alpha=0$, the second and third equations of (4) imply that $\beta_{r}=0$ and $\displaystyle\beta=B=T(\sin\theta\sigma+a_{3}\cos\theta)\;,$ (89) with $\sigma=\sigma^{B}/T=a_{1}\cos\varphi-a_{2}\sin\varphi$. In that case, eq. (15) is identically satisfied and eq. (16) gives $P^{\prime}=-H_{1}/H^{2}_{2}$, which can be integrated to yield the second relation of eq. (54) and provide us with $\displaystyle H^{2}_{1}=\frac{1}{1+k_{3}r^{2}}\;,$ (90) for $H_{0}=1$ and $H_{2}=r$. Then the solutions can be easily read from (14) and (17) to be $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle T\frac{H^{\prime}_{2}}{H_{1}}(\cos\theta\sigma-a_{3}\sin\theta)+TH_{2}g\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle T\frac{H^{\prime}_{2}}{H_{1}}\sigma_{\varphi}+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;.$ With $\alpha=0$ and $\beta$ as in (75), the following Killing vector fields are obtained : $\displaystyle I_{1}$ $\displaystyle=$ $\displaystyle\frac{1}{H_{1}}\sin\theta\cos\varphi\partial_{r}+PZ_{2}\;,$ $\displaystyle I_{2}$ $\displaystyle=$ $\displaystyle-\frac{1}{H_{1}}\sin\theta\sin\varphi\partial_{r}-PZ_{3}\;,$ (91) $\displaystyle I_{3}$ $\displaystyle=$ $\displaystyle\frac{1}{H_{1}}\cos\theta\partial_{r}-PZ_{1}\;.$ These are the generalized translation generators which, in the usual cartesian coordinates read, respectively, as $H_{1}^{-1}\partial_{x}\;,-H_{1}^{-1}\partial_{y}$ and $H_{1}^{-1}\partial_{z}$. Together with the $so(3)$ generators, they close into the following Lie algebra structure: $\displaystyle\;[I_{1},I_{2}]$ $\displaystyle=$ $\displaystyle k_{3}K_{3}\;,\quad\;\quad[I_{2},I_{3}]=k_{3}K_{1}\;,\quad[I_{1},I_{3}]=-k_{3}K_{2}\;,$ $\displaystyle\;[K_{1},I_{2}]$ $\displaystyle=$ $\displaystyle- I_{3}=[K_{2},I_{1}]\;,\;\;[K_{1},I_{3}]=I_{2}=-[K_{3},I_{1}]\;,$ (92) $\displaystyle\;[K_{2},I_{3}]$ $\displaystyle=$ $\displaystyle I_{1}=[K_{3},I_{2}]\;,\quad\;\;[K_{i},I_{i}]=0\;,\quad i=1,2,3\;.$ ## VII Flat Space-Time Solutions: $\dot{T}=0,\;H_{0}^{\prime}=0$ In this case, $\alpha$ is independent of $\tau=t/T$, $\beta$ is independent of $r$, and condition (13) requires that $V=0$. Therefore it is convenient to start with $\alpha=c_{0}H_{0}+\eta_{1}$ and $\beta=\eta_{2}$, where $c_{0}$ is a constant and $\eta_{i}$ are as in (65) of the previous section, such that $\eta_{1}$ is independent from $\tau$ and $\eta_{2}$ is independent from $r$. The third equation of (4) gives $\eta_{2\tau}=-H_{0}\eta_{1r}/H_{1}$. Since the left hand side of this equality depends on $\tau$ and is independent of $r$, but the right hand side depends on $r$ and is independent of $\tau$, both sides must be equal $\displaystyle\eta_{2\tau}=\zeta=-\frac{H_{0}}{H_{1}}\eta_{1r}\;,$ (93) where $\zeta$ is given by (33). From the first equality, we obtain $\eta_{2}=\tau\zeta+\zeta^{(b)}$ where $\zeta^{(b)}$ is the same as $\zeta$ with $a=(a_{1},a_{2},a_{3})$ replaced by $b=(b_{1},b_{2},b_{3})$. On the other hand, for $A=\eta_{1}$ and $B=\tau\zeta+\zeta^{(b)}$, conditions (15) and (16) respectively yield $\displaystyle\eta_{1\theta}=-\frac{H_{2}^{2}P}{H_{0}}\zeta_{\theta}\;,\quad P^{\prime}=-\frac{H_{1}}{H_{2}^{2}}\;.$ (94) In view of the second equality of (79), we obtain $(H_{2}^{2}P)^{\prime}=H_{1}$ from the integrability condition $\eta_{1\theta r}=\eta_{1r\theta}$. From the second equation of (79) and $(H_{2}^{2}P)^{\prime}=H_{1}$, we get $P^{2}=1/H_{2}^{2}$, which can also be obtained from the integration of $P^{\prime}=-H_{1}/H_{2}^{2}$ with the zero integration constant (see the discussion following equations (53) and (54)). From (14) and (17), one can easily write out $\gamma$ and $\delta$ together of $\alpha$ and $\beta$ to be as : $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{0}H_{0}-\frac{1}{H_{0}P}\zeta\;,\quad\beta=\tau\zeta+\zeta^{(b)}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle H_{2}P(\tau\zeta+\zeta^{(b)})+TH_{2}g\;,$ (95) $\displaystyle\delta$ $\displaystyle=$ $\displaystyle H_{2}P(\tau\sigma_{\varphi}+\sigma^{(b)}_{\varphi})+TH_{2}(a\sin\theta+\cos\theta g_{\varphi})\;.$ We have ten linearly independent KY 1-forms and only one condition $P^{2}=1/H_{2}^{2}$ which is equivalent to $H_{2}^{\prime 2}=H_{1}^{2}$. For $H_{2}=r$, we have $H_{1}=\pm 1$, and the corresponding Killing vector fields are directly determined from (81) with $P=\pm 1/r$. ###### Acknowledgements. This work was supported in part by the Scientific and Technical Research Council of Turkey (TÜBİTAK). ## Appendix A Metric Constants In Section IV, the following four metric constants were defined : $\displaystyle m$ $\displaystyle=$ $\displaystyle L^{\prime}\frac{H_{0}}{H_{1}H_{2}}\;,\;\;m_{0}=\frac{H^{2}_{0}}{LH_{1}}(\frac{H_{2}}{H_{0}})^{\prime}\;,$ (96) $\displaystyle k$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}H_{2}}{H_{0}H_{2}^{\prime}}\;,\;\;\quad k_{0}=-\frac{H_{0}^{2}}{H_{1}H_{2}}(\frac{H_{2}^{\prime}}{H_{1}})^{\prime}\;.$ (97) The first relation of (A1) can be integrated to find $\displaystyle H^{\prime}_{0}=\ell_{1}H^{-m}_{0}H_{1}H_{2}\;,\quad L=\frac{1}{\ell_{1}}H_{0}^{m}\;,$ (98) which is also valid for $m=0$. The second relation of (A1) can be arranged as $\displaystyle\frac{H_{2}^{\prime}}{H_{2}}=\frac{H^{\prime}_{0}}{H_{0}}[m_{0}(\frac{H_{1}}{H^{\prime}_{0}})^{2}+1]\;.$ (99) The first relation of (A2) can also be integrated to find $H_{0}=\ell_{2}H_{2}^{k}$, where $\ell_{1}$ and $\ell_{2}$ are also non-zero metric constants. By combining (A4) with the first relation of (A2), we get $\displaystyle(1-k)H^{\prime\;2}_{0}=m_{0}kH^{2}_{1}\;,$ (100) and then, by (A3) and $H_{0}=\ell_{2}H_{2}^{k}$, we arrive at $\displaystyle(1-k)=m_{0}k(\frac{\ell_{2}^{m}}{\ell_{1}})^{2}H^{2(mk-1)}_{2}\;.$ (101) Thus for nonconstant $H_{2}$, we must have $\displaystyle km=1\;,\quad m=1+m_{0}(\frac{\ell_{2}^{m}}{\ell_{1}})^{2}\;.$ (102) Note that $k=1$ if and only if $m_{0}=0$ and if and only if $m=1$ for $H_{2}^{\prime}\neq 0$. ## Appendix B Curvature Forms The curvature 2-forms $R_{ab}=d\omega_{ab}+\omega_{ac}\wedge\omega^{c}_{\;b}$ for the considered class of spherically symmetric space-times are computed by using eqs. (2) and (3) to be as follows : $\displaystyle R_{01}$ $\displaystyle=$ $\displaystyle(S+\frac{h_{0}^{2}+h_{0}^{\prime}-h_{0}h_{1}}{T^{2}H_{1}^{2}})e^{01}\;,\;\;R_{12}=S_{2}e^{02}+S_{1}e^{12}\;,$ $\displaystyle R_{02}$ $\displaystyle=$ $\displaystyle(S+\frac{h_{0}h_{2}}{T^{2}H_{1}^{2}})e^{02}+S_{2}e^{12}\;,\quad R_{13}=S_{2}e^{03}+S_{1}e^{13}\;,$ (103) $\displaystyle R_{03}$ $\displaystyle=$ $\displaystyle(S+\frac{h_{0}h_{2}}{T^{2}H_{1}^{2}})e^{03}+S_{2}e^{13}\;,\quad R_{23}=(S_{1}+\frac{H_{1}P^{\prime}+(H_{1}/H_{2})^{2}}{T^{2}H_{1}^{2}})e^{23}\;,$ where $\displaystyle S=-\frac{\ddot{T}}{TH_{0}^{2}}\;,\quad S_{1}=\frac{\dot{T}^{2}}{T^{2}H_{0}^{2}}-\frac{P^{\prime}+H_{1}P^{2}}{T^{2}H_{1}}\;,\quad S_{2}=\frac{\dot{T}}{T^{2}}\frac{h_{0}}{H_{0}H_{1}}\;.$ (104) Note that $S=0=S_{2}$ when $\dot{T}=0$. The corresponding Ricci 1-forms and curvature scalar can be found from $P_{b}=i_{X^{a}}R_{ab}$ and $\Re=i_{X^{b}}P_{b}$. For $\dot{T}=0=H_{0}^{\prime}$ and $H_{2}^{\prime 2}=H_{1}^{2}$ we have $S_{2}=0$ in addition to $S=0=S_{1}$. It then follows that all the curvature components vanish, and we obtain the Minkowski space-time. In the case of the static form of de Sitter space-time, that is for $\dot{T}=0,\;H_{2}=r$ and $\displaystyle H_{0}^{2}=k_{1}r^{2}+k_{2}\;,\;H_{1}^{2}=\frac{1}{1+k_{3}r^{2}}\;,\;k_{1}=k_{2}k_{3}$ we have $S=0=S_{2}$ and $S_{1}=-k_{3}/T^{2}$ which gives the constant curvature solutions $\displaystyle R_{0j}=\frac{k_{3}}{T^{2}}e^{0j}\;,\quad R_{ij}=-\frac{k_{3}}{T^{2}}e^{ij}\;,$ (105) where $i,j=1,2,3$ and $i<j$. These two space-times are maximally symmetric with constant curvature. For $\dot{T}=\lambda T\;,H_{0}^{\prime}=0$ and $H_{2}^{\prime}=\epsilon H_{1}$ which indicate de Sitter space-time, we have $\displaystyle S=-\frac{\lambda^{2}}{H_{0}^{2}}=-S_{1}\;,\quad S_{2}=0\;.$ These lead us again to the curvature solution given by (B3) provided that $k_{3}/T^{2}$ is replaced by $-\lambda^{2}/H_{0}$. For the Robertson-Walker space-time, we have $\displaystyle R_{0j}=-\frac{\ddot{T}}{T}e^{0j}\;,\quad R_{ij}=[(\frac{\dot{T}}{T})^{2}+\frac{k}{T^{2}}]e^{ij}\;.$ (106) Finally, for the four forms of de Sitter space-time found in Section VI, we have $\displaystyle H_{0}=1\;,\quad H^{2}_{1}=\frac{1}{1-\ell r^{2}}\quad H_{2}=r\;,$ and $T$ expressions are explicitly given by equations (73). For these values we find $S_{2}=0$ and $S_{1}=\ell_{3}=-S$, where $\ell_{3}$ is a constant. Thus $\displaystyle R_{0j}=-\ell_{3}e^{0j}\;,\quad R_{ij}=\ell_{3}e^{ij}\;.$ (107) where we have used $\dot{T}=\epsilon(\ell_{3}T^{2}-\ell)^{1/2}$ found in equation (72). ## References * (1) I. M. Benn and R. W. Tucker, An Introduction to Spinors and Geometry with Applications in Physics, IOP Publishing Ltd, Bristol, 1987. * (2) W. Thirring, Classical Mathematical Physics: Dynamical Systems and Field Theory, Third edition, Springer, 1997. * (3) W. Dietz and R. Rüdiger, Proc. Roy. Soc. London Ser. A 375 (1981), 361. * (4) W. Dietz and R. Rüdiger, Proc. Roy. Soc. London Ser. A 381 (1982), 315. * (5) I. M. Benn, J. Math. Phys. 47 (2006), 022903. * (6) I. M. Benn, J. Math. Phys. 35 (1994), 1796. * (7) I. M. Benn, P. R. Charlton and J. Kress, J. Math. Phys. 38 (1997), 4504. * (8) J. Kress, Generalized Conformal Killing-Yano Tensors: Applications to Electrodynamics (1997), PhD Thesis (available at: http://web.maths.unsw.edu.au/ jonathan/thesis). * (9) I. M. Benn and J. Kress, J. Phys. A: Math. Gen. 29 (1996), 6295. * (10) U. Semmelmann, Math. Z. 243 (2003), 503. * (11) I. M. Benn and P. Charlton, Class. Quantum Grav. 14 (1997), 1037. * (12) I. M. Benn and J. Kress, Class. Quantum Grav. 21 (2004), 427. * (13) I. M. Benn and J. M. Kress, Symmetry operators for the Dirac and Hodge-deRham equations, Proceedings of the 9th DGA Conference, Prague, August 30 - September 3, 2004, pp. 421-430. * (14) M. Walker and R. Penrose, Commun. Math. Phys. 18 (1970), 265. * (15) R. Penrose, Ann. N. Y. Acad. Sci. 224 (1973), 125. * (16) B. Carter, Phys. Rev. D 174 (1968), 1559. * (17) B. Carter and R. G. McLenaghan, Phys. Rev. D 19 (1979), 1093. * (18) D. Kastor and J. Traschen, JHEP 0408 (2004), 045 [arXive:hep-th/0406052]. * (19) H. Cebeci, O. Sarioglu and B. Tekin, Phys.Rev. D. 74 (2006), 124021 [arXive:hep-th/0611011]. * (20) Ö. Açık, Ü. Ertem, M. Önder and A. Verçin, Killing-Yano Forms of a Class of Spherically Symmetric Space-Times II : A Unified Generation of Higher Forms. * (21) S. Hawking and G. Ellis, The large scale structure of space-time, Cambridge University Press, Cambridge, 1973. * (22) U. Moschella, The de Sitter and anti-de Sitter sightseeing tour, in Einstein 1905-2005, vol. 47 of Progress in Mathematical Physics, Birkhauser, Basel, 2006, pp. 120-133. * (23) A. C. Randono, In Search of Quantum de Sitter Space: Generalizing the Kodama State (2007), PhD dissertation (available at: arXiv:0709.2905v1 [gr-qc]). Table 1: Metric coefficient functions and the numbers of linearly independent KY-forms of some well-known spherically symmetric space-times. For all these cases $H_{2}$ is the radial coordinate $r$. The numbers of the fifth column represent the dimensions $d(4,1)$ of the corresponding symmetry algebras whose common part consists of three $so(3)$ generators given by the equation (6) of the main text. The last two columns denote the numbers $d(4,2)$ and $d(4,3)$ of the linearly independent KY 2-forms and 3-forms which are explicitly calculated in the accompanying paper. The fifth rows represent three different forms of the de Sitter space-time, of which the third consists of four cases with different time evolutions given in Section VI. The numbers $d(n,p)$ for the maximally symmetric space-times, such as those of Minkowski and de Sitter, represent the upper bounds for dimension $n=4$. Space-time | $T$ | $H_{0}$ | $H_{1}$ | $d(4,1)$ | $d(4,2)$ | $d(4,3)$ ---|---|---|---|---|---|--- Schwarzschild | $1$ | $\sqrt{1-\frac{2M}{r}}$ | $\frac{1}{\sqrt{1-\frac{2M}{r}}}$ | 4 | 1 | 0 Reissner-Nordstrøm | $1$ | $\sqrt{1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}}$ | $\frac{1}{\sqrt{1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}}}$ | 4 | 1 | 0 Robertson-Walker | $T(t)$ | $1$ | $\frac{1}{\sqrt{1-kr^{2}}}$ | 6 | 4 | 1 de Sitter | | | | | | static form | $1$ | $\sqrt{1-kr^{2}}$ | $\frac{1}{\sqrt{1-kr^{2}}}$ | 10 | 10 | 5 usual form | $e^{(\Lambda/3)^{1/2}t}$ | 1 | 1 | 10 | 10 | 5 four additional forms | $\dot{T}^{2}=\ell_{3}T^{2}-\ell$ | $1$ | $\frac{1}{\sqrt{1-\ell\;r^{2}}}$ | 10 | 10 | 5 Minkowski | $1$ | $1$ | $1$ | 10 | 10 | 5

arxiv-papers

2024-09-04T02:48:54.494661

0803.3328

# Killing-Yano Forms of a Class of Spherically Symmetric Space-Times II: A Unified Generation of Higher Forms Ö. Açık 1 ozacik@science.ankara.edu.tr Ü. Ertem 1 uertem@science.ankara.edu.tr M. Önder 2 onder@hacettepe.edu.tr A. Verçin 1 vercin@science.ankara.edu.tr 1 Department of Physics, Ankara University, Faculty of Sciences, 06100, Tandoğan-Ankara, Turkey 2 Department of Physics Engineering, Hacettepe University, 06800, Beytepe- Ankara, Turkey. ###### Abstract Killing-Yano (KY) two and three forms of a class of spherically symmetric space-times that includes the well-known Minkowski, Schwarzschild, Reissner- Nordstrøm, Robertson-Walker and six different forms of de Sitter space-times as special cases are derived in a unified and exhaustive manner. It is directly proved that while the Schwarzschild and Reissner-Nordstrøm space- times do not accept any KY 3-form and they accept only one 2-form, the Robertson-Walker space-time admits four KY 2-forms and only one KY 3-form. Maximal number of KY-forms are obtained for Minkowski and all known forms of de Sitter space-times. Complete lists comprising explicit expressions of KY- forms are given. ###### pacs: 04.20.-q, 02.40.-k ## I Introduction In the previous study ozumav1 (henceforth referred to as I), we developed a constructive method which provided a unified generation of all Killing vector fields for a class of four dimensional ($4D$) spherically symmetric space- times. As the sequel of I, in the present paper we shall investigate the explicit forms of KY two and three forms by solving the KY-equations $\displaystyle\nabla_{X_{a}}\omega_{(p)}=\frac{1}{p+1}i_{X_{a}}d\omega_{(p)}\;,$ (1) for this class of space-time metrics in the $p=2$ and $p=3$ cases. The underlying base manifold is supposed to be a $4D$ pseudo-Riemannian manifold endowed with the metric $g$ having the Lorentzian signature $(-+++)$ such that $\displaystyle g=-e^{0}\otimes e^{0}+e^{1}\otimes e^{1}+e^{2}\otimes e^{2}+e^{3}\otimes e^{3}\;.$ in a local orthonormal co-frame basis $\\{e^{a}\\}$. This metric can be parameterized by the (metric) characterizing functions $T=\exp(\lambda(t))$ and $H_{j}=H_{j}(r)$ by choosing $\displaystyle e^{0}$ $\displaystyle=$ $\displaystyle H_{0}dt\;,\qquad e^{1}=TH_{1}dr\;,$ $\displaystyle e^{2}$ $\displaystyle=$ $\displaystyle TH_{2}d\theta\;,\quad e^{3}=TH_{2}\sin\theta d\varphi\;.$ The corresponding orthonormal vector bases will be denoted by $X_{a}$. The torsion-free connection $1$-forms for this class of metrics and covariant derivatives of basis elements required for explicit calculations can be found in I. We shall mainly use the notation of I, which is in accordance with Benn- Tucker . As they span the kernel of the linear operator $\nabla_{X}-(p+1)^{-1}i_{X}d$, the space $Y_{p}$ of all KY p-forms constitute a linear space Semmelman1 ; Stepanov . Any function is a KY 0-form, $\omega_{(1)}$ is the dual of a Killing vector field and $\omega_{(n)}$ is a constant (parallel), that is, it is a constant multiple $\omega_{(n)}=az$ of the volume form, say $z=e^{0123}$ for $n=4$. Therefore while $Y_{0}$ is infinite dimensional, $Y_{n}$ is one dimensional. By a straightforward extension of the argument for determining the maximum number of Killing vectors, the upper bound for the dimension of $Y_{p}$’s can now be established for each $p$ Kastor ; Kastor1 . For this purpose, we should first note that equation (1) suggests an equivalent definition: a p-form is a KY p-form if and only if its symmetrized covariant derivatives vanish, that is, if and only if $\displaystyle i_{X}\nabla_{Y}\omega_{(p)}+i_{Y}\nabla_{X}\omega_{(p)}=0\;,$ is satisfied for all pair of the vector fields $X$ and $Y$. This is also equivalent to $i_{X}\nabla_{X}\omega_{(p)}=0$ and in particular every KY-form is divergent free, or equivalently co-closed, that is, $\delta\omega_{(p)}=-i_{X^{a}}\nabla_{X_{a}}\omega_{(p)}=0$. These statements imply that the covariant derivatives of KY forms are also totally anti- symmetric, with respect to the additional tensorial index. Hence the maximum numbers of linearly independent components and their first covariant derivatives are $C(n,p)$ and $C(n,p+1)$ respectively, where $C(n,p)$ stands for the binomial numbers. On the other hand “the second covariant derivatives”, that is, the Hessian of KY forms can be written, in terms of the curvature 2-forms $R^{c}_{\;a}$ as $\displaystyle(\nabla_{X_{a}}\nabla_{X_{b}}-\nabla_{\nabla_{X_{a}}X_{b}})\omega_{(p)}=\frac{1}{p}i_{X_{b}}(R^{c}_{\;a}\wedge i_{X_{c}}\omega_{(p)})\;.$ These imply that the value of any component of a KY-form at any point is entirely determined by the value of its first covariant derivative and by the value of the component itself at the same point. As a result the upper bounds for the numbers of linearly independent KY-forms is determined by the sum $C(n,p)+C(n,p+1)=C(n+1,p+1)$. In particular for $n=4$ the upper bounds are $10,10$ and $5$, respectively, for $p=1,p=2$ and $p=3$. These bounds are attained for the space-times of constant curvature. The rest of the paper is structured in two main parts. In the next section, the defining equations for KY 2-forms and their integrability conditions are obtained. By the integrability conditions, the set of solutions naturally breaks up into cases and subcases that are considered in the rest of first part consisting five sections. The second part is entirely devoted to determination of KY 3-forms. The final Section VIII presents a summary of the results, and we briefly point out how to calculate the associated Killing tensors and related linear and quadratic first integrals for the considered class of space-times. ## II KY 2-Forms: Defining Equations, Integrability Conditions For a 2-form $\displaystyle\omega_{(2)}=\alpha e^{01}+\beta e^{02}+\gamma e^{03}+\delta e^{12}+\epsilon e^{13}+\mu e^{23}\;,$ there are two types of equations that result from KY-equation (1) in the case of $p=2$. The first consists of twelve relatively simple equations resulting from the left hand side of the KY equations. Two of them are $\alpha_{t}=0=\alpha_{r}$, which imply that $\alpha=\alpha(\theta,\varphi)$, and the other ten equations are as follows: $\displaystyle\beta_{t}$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}}{TH_{1}}\delta\;,\;\qquad\qquad\beta_{\theta}=-\frac{H^{\prime}_{2}}{H_{1}}\alpha\;,$ $\displaystyle\delta_{r}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{1}}{H_{0}}\beta\;,\;\qquad\qquad\delta_{\theta}=-\dot{T}\frac{H_{2}}{H_{0}}\alpha\;,$ $\displaystyle\gamma_{t}$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}}{TH_{1}}\epsilon\;,\;\qquad\qquad\gamma_{\varphi}=-\frac{H^{\prime}_{2}}{H_{1}}\sin\theta\alpha-\cos\theta\beta\;,$ (2) $\displaystyle\epsilon_{r}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{1}}{H_{0}}\gamma\;,\;\qquad\qquad\epsilon_{\varphi}=-\dot{T}\frac{H_{2}}{H_{0}}\sin\theta\alpha-\cos\theta\delta\;,$ $\displaystyle\mu_{\theta}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{2}}{H_{0}}\gamma-\frac{H^{\prime}_{2}}{H_{1}}\epsilon\;,\quad\mu_{\varphi}=-\dot{T}\frac{H_{2}}{H_{0}}\sin\theta\beta+\frac{H^{\prime}_{2}}{H_{1}}\sin\theta\delta\;.$ The second type also consists of twelve equations, but four of them are just a sum of two other equations of the same type. Hence, there are eight independent additional equations, six of which can be put in the following forms: $\displaystyle\alpha_{\theta}$ $\displaystyle=$ $\displaystyle-\frac{H^{2}_{2}}{H_{1}}\partial_{r}\frac{\beta}{H_{2}}\;,\quad\alpha_{\varphi}=-\frac{H_{2}^{2}}{H_{1}}\sin\theta\partial_{r}\frac{\gamma}{H_{2}}\;,$ $\displaystyle\partial_{r}\frac{\beta}{H_{0}}$ $\displaystyle=$ $\displaystyle-T^{2}\frac{H_{1}}{H^{2}_{0}}\partial_{t}\frac{\delta}{T}\;,\quad\gamma_{\theta}=-T^{2}\frac{H_{2}}{H_{0}}\partial_{t}\frac{\mu}{T}\;,$ (3) $\displaystyle\partial_{r}\frac{\gamma}{H_{0}}$ $\displaystyle=$ $\displaystyle-T^{2}\frac{H_{1}}{H^{2}_{0}}\partial_{t}\frac{\epsilon}{T}\;,\quad\epsilon_{\theta}=-\frac{H^{2}_{2}}{H_{1}}\partial_{r}\frac{\mu}{H_{2}}\;.$ The last two equations (presented below) do not depend on the metric coefficient functions, and therefore much of the essence of spherical symmetry is encoded in them: $\displaystyle\beta_{\varphi}$ $\displaystyle=$ $\displaystyle-\sin^{2}\theta\partial_{\theta}\frac{\gamma}{\sin\theta}\;,\quad\quad\delta_{\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{\epsilon}{\sin\theta}\;.$ (4) Before solving these eighteen equations, there are some integrability conditions that must be met. Two of them, which enable us to find the solutions in a systematic way, are $\displaystyle\dot{T}H^{\prime}_{0}\alpha$ $\displaystyle=$ $\displaystyle 0\;,$ (5) $\displaystyle\;[\rho(r)+\varrho(t)\;]\alpha$ $\displaystyle=$ $\displaystyle 0\;,$ (6) where $\displaystyle\rho=\frac{H_{0}^{3}}{H_{1}H_{2}}(\frac{H_{2}^{\prime}}{H_{0}H_{1}})^{\prime}\;,\quad\varrho=T\ddot{T}-\dot{T}^{2}=T^{2}\partial_{t}(\frac{\dot{T}}{T})\;.$ (7) Condition (5) separately follows from each of $\displaystyle\beta_{t\theta}=\beta_{\theta t},\;\gamma_{t\varphi}=\gamma_{\varphi t},\;\delta_{r\theta}=\delta_{\theta r}\;,\;\epsilon_{r\varphi}=\epsilon_{\varphi r}\;,$ and condition (6) can be checked from $\delta_{t\theta}=\delta_{\theta t}$ or $\epsilon_{\varphi t}=\epsilon_{t\varphi}$. The integrability conditions imply that solutions must be investigated in two classes: (A) $\alpha\neq 0$ and (B) $\alpha=0$, such that the first consists of three important subclasses characterized by $\displaystyle(i)\;H_{0}^{\prime}=0\;,\;\;\dot{T}\neq 0\;,\quad(ii)\;H_{0}^{\prime}\neq 0\;,\;\;\dot{T}=0\;,\quad(iii)\;H_{0}^{\prime}=0=\dot{T}\;.$ In fact there are $6\times 6$ integrability conditions, some of which are trivially satisfied and the remaining ones will be considered in the following classes. We should also note that whenever $T$ or $H_{k}$’s are constant, they must be considered to be nonzero to keep the metric non-degenerate. In the next three sections, we shall consider the first (A) case in which $\alpha$ is different from zero. At the outset of these sections, we should note the following two equations for $\alpha$: $\displaystyle\alpha_{\theta\theta}+\ell_{1}\alpha$ $\displaystyle=$ $\displaystyle 0\;,$ (8) $\displaystyle\alpha_{\varphi\varphi}+\ell_{1}\sin^{2}\theta\alpha+\sin\theta\cos\theta\alpha_{\theta}$ $\displaystyle=$ $\displaystyle 0\;,$ (9) which do not change in the following subcases. These are obtained by differentiating the first two equations of (3) and by making use of the equation for $\beta_{\theta}$ and $\gamma_{\varphi}$ from (2). Here the constant $\ell_{1}$ is defined, in terms of $P=H_{2}^{\prime}/H_{1}H_{2}$, as follows: $\displaystyle\ell_{1}=-\frac{H_{2}^{2}}{H_{1}}P^{\prime}\;,$ (10) We should finally note that, as there are no equations involving the first power of $\mu$, $\displaystyle\omega_{0}=TH_{2}e^{23}\;$ is a solution of the KY-equation without any constraint. This observation means that all the space-times within the considered class of metrics admit of at least one KY 2-form. This fact was also observed for the static space-times in Howarth ; Collinson 1 . In fact, it turns out that $\omega_{0}$ is the only KY 2-form admitted by two important cases, the Schwarzschild and the Reissner- Nordstrøm space-times, which do not accept any KY 3-forms. ## III KY 2-Forms for $H_{0}^{\prime}=0,\;\dot{T}\neq 0\;,\alpha\neq 0$ In the case of nonzero $\alpha$ and $H_{0}^{\prime}=0$, two equations of (2) show that $\beta$ and $\gamma$ are time independent and the integrability conditions (5-7) require that $\ell$, defined in terms of $P$ by $\displaystyle\frac{H_{0}^{2}}{H_{1}H_{2}}(H_{2}P)^{\prime}=-\ell=T^{2}\partial_{t}(\frac{\dot{T}}{T})\;,$ (11) must be a constant. By multiplying both sides of the first equality by $H_{2}H_{2}^{\prime}$ it can easily be integrated to write $\displaystyle P^{2}=\frac{\ell_{1}}{H^{2}_{2}}-\frac{\ell}{H^{2}_{0}}\;,$ (12) where $\ell_{1}$ is taken as an integration constant, since the defining relation (10) of $\ell_{1}$ is implied by the condition (12). Then, in terms of $\displaystyle{\cal B}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{2}}{H_{0}}\beta\;,\quad{\cal D}=H_{2}P\delta\;,$ $\displaystyle{\cal G}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{2}}{H_{0}}\gamma\;,\quad{\cal E}=H_{2}P\epsilon\;,$ we define $\displaystyle x={\cal G}-{\cal E}\;,\quad y={\cal B}-{\cal D}\;.$ (13) The first four equations appearing in the second column of (2) and two equations of (4) give the following equations for $x$ and $y$: $\displaystyle y_{\theta}=0\;,\quad x_{\varphi}=-\cos\theta y\;,\quad y_{\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{x}{\sin\theta}\;.$ (14) These immediately imply $y_{\varphi\varphi}+y=0$ and hence $\displaystyle y=y_{1}\cos\varphi+y_{2}\sin\varphi\;,\quad x=\cos\theta y_{\varphi}+y_{3}\sin\theta\;,$ (15) where three $y_{i}$ functions depend on both $t$ and $r$, and the expression of $x$ is obtained by integration from (14). #### III.0.1 The general forms of $\gamma,\epsilon$ and $\mu$ In terms of $x$ and $y$, the last two equations of (2) are simply $\mu_{\theta}=x,\;\mu_{\varphi}=-y\sin\theta$, and therefore they specify the general form of $\mu$ as $\displaystyle\mu=y_{\varphi}\sin\theta-y_{3}\cos\theta+c_{0}TH_{2}\;,$ (16) where $c_{0}$ is an integration constant. In specifying the last term of (16), we make use of the $\gamma_{\theta}$ and $\epsilon_{\theta}$-equations of (3), which transform to $\displaystyle\gamma_{\theta}=T^{2}\frac{H_{2}}{H_{0}}\partial_{t}\frac{x_{\theta}}{T}\;,\quad\epsilon_{\theta}=\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{x_{\theta}}{H_{2}}\;.$ (17) Integration of these equations with respect to $\theta$ yield $\displaystyle\gamma=T^{2}\frac{H_{2}}{H_{0}}\partial_{t}\frac{x}{T}+g(r,\varphi)\;,\quad\epsilon=\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{x}{H_{2}}+\varepsilon(t,r,\varphi)\;.$ (18) By substituting these expression into $x={\cal G}-{\cal E}$, we obtain $\displaystyle\varepsilon=\frac{\dot{T}}{H_{0}}\frac{g}{P}\;,$ (19) and a set of three equations for $y_{i}$, which in terms of $Y_{i}=y_{i}/TH_{2}$ can be written as $\displaystyle Y_{i}=T\dot{T}(\frac{H_{2}}{H_{0}})^{2}Y_{it}-\frac{H_{2}^{2}P}{H_{1}}Y_{ir}\;,\quad i=1,2,3\;.$ (20) In writing equations (20), we have equated the coefficient functions of different trigonometric functions forming a basis, and relation (19) results from the fact that $x$ does not contain a term independent from $\theta$. We now concentrate on specifying $g$ and $y_{i}$ functions. For this purpose we consider the coupled $\epsilon_{r}$-equation of (2) and $\gamma_{r}$-equation of (3) which, in terms of $P$ and $\ell$, give $\displaystyle\partial_{r}(\frac{g}{P})=H_{1}g\;,\quad g_{r}=-\frac{\ell}{H_{0}^{2}}\frac{H_{1}}{P}g\;,$ (21) in addition to two sets of equations for $y_{i}$ : $\displaystyle\partial_{r}(\frac{H^{2}_{2}}{H_{1}}Y_{ir})$ $\displaystyle=$ $\displaystyle T\dot{T}\frac{H_{1}H^{2}_{2}}{H^{2}_{0}}Y_{it}\;,$ (22) $\displaystyle\partial_{r}(H^{2}_{2}Y_{it})$ $\displaystyle=$ $\displaystyle-H^{2}_{2}Y_{itr}\;.$ (23) By dividing the first equation of (21) with $P$, it can be easily integrated to yield $\displaystyle g=H_{2}Pu_{\varphi}\;,$ (24) where for convenience the last factor of $g$ has been written as the derivative of $u=u(\varphi)$. From the second equation of (21) we then obtain just one of the conditions of (11), when $u_{\varphi}\neq 0$. After a slight rearrangement, the equation (23) can be integrated to find $\displaystyle Y_{i}=\frac{q_{i}(t)}{H_{2}}+\frac{z_{i}(r)}{H_{2}}\;$ (25) The substitution of this relation into (20) and (22) yield $\displaystyle T\dot{T}\dot{q}_{i}-[\ell+(1-\ell_{1})\frac{H^{2}_{0}}{H_{2}^{2}}]q_{i}$ $\displaystyle=$ $\displaystyle[\ell+(1-\ell_{1})\frac{H_{0}^{2}}{H_{1}}]z_{i}+H_{0}^{2}\frac{P}{H_{1}}z_{i}^{\prime}\;,$ (26) $\displaystyle T\dot{T}\dot{q}_{i}-\ell q_{i}$ $\displaystyle=$ $\displaystyle\ell z_{i}-\frac{H_{0}^{2}H_{1}^{\prime}}{H_{1}^{3}}z_{i}^{\prime}+\frac{H_{0}^{2}}{H^{2}_{1}}z_{i}^{\prime\prime}\;,$ (27) for $i=1,2,3$. As we are about to see at the beginning of next section, the constant $\ell_{1}$ must be $1$. Anticipating this result here, we see that the above equations are separable, and therefore each side of them must be a constant such that $\displaystyle T\dot{T}\dot{q}_{i}-\ell q_{i}$ $\displaystyle=$ $\displaystyle m_{i}\;,$ $\displaystyle\ell z_{i}-\frac{H_{0}^{2}H_{1}^{\prime}}{H_{1}^{3}}z_{i}^{\prime}+\frac{H_{0}^{2}}{H^{2}_{1}}z_{i}^{\prime\prime}$ $\displaystyle=$ $\displaystyle m_{i}\;,$ (28) $\displaystyle\ell z_{i}+\frac{H_{0}^{2}}{H_{1}}Pz_{i}^{\prime}$ $\displaystyle=$ $\displaystyle m_{i}\;.$ It is not hard to see that the last two sets of (28) can be integrated to obtain $\displaystyle z_{i}=c_{i}H_{2}P-\tilde{c}_{i}\;,\quad i=1,2,3$ (29) with $m_{i}=-\ell\tilde{c}_{i}$. The equations in the first line of (28) need not be integrated at this point, since they will be directly solved during the investigation of next subsection. #### III.0.2 The general forms of $\alpha,\beta$ and $\delta$ Relations found by (25) specify $y_{i}$ for $i=1,2,3$ as follows : $\displaystyle y_{i}=Tq_{i}(t)+Tz_{i}(r)\;.$ (30) By noting that $(y_{i}/T)_{tr}=0$, the substitution of the expression of $\gamma$ given by (18) into the second equation of (3) yield $\alpha_{\varphi}=\ell_{1}\sin\theta u_{\varphi}$ in view of (24). This can easily be integrated to write $\alpha=\ell_{1}\sin\theta u+f(\theta)$, and then by using this in equations (8) and (9) we obtain $\displaystyle\ell_{1}(\ell_{1}-1)\sin\theta u+f_{\theta\theta}+\ell_{1}f$ $\displaystyle=$ $\displaystyle 0\;,$ $\displaystyle\ell_{1}[u_{\varphi\varphi}+(\cos^{2}\theta+\ell_{1}\sin^{2}\theta)u]+\ell_{1}\sin\theta f+\cos\theta f_{\theta}$ $\displaystyle=$ $\displaystyle 0\;.$ Since $u$ is a functions of $\varphi$, these relations imply that $\ell_{1}$ must be either $0$ or $1$. In fact, the former value is not compatible with the very definition of spherical symmetry, since it would lead to metric coefficient functions depending on some finite powers of angular coordinates. Indeed, equations (8) and (9) show that when $\ell_{1}=0$, $\alpha$ can be assumed to be a nonzero constant. But the equations in the second column of (2) then show that there is no way to get rid of the above mentioned angular dependence. Therefore from now on, we take $\ell_{1}=1$ by which the above two equations are reduced to the following forms : $\displaystyle f_{\theta\theta}+f=0\;,\quad u_{\varphi\varphi}+u+\sin\theta f+\cos\theta f_{\theta}=0\;.$ (31) Hence, in terms of integration constants $c_{4},c_{5},c_{6},\tilde{c}_{4}$ and $\displaystyle v=v(\varphi)=c_{5}\cos\varphi+c_{6}\sin\varphi\;,$ (32) we have $f=c_{4}\cos\theta+\tilde{c}_{4}\sin\theta$ and $u=v-\tilde{c}_{4}$ which implies $u_{\varphi}=v_{\varphi}$. These completely specify $\alpha$ as $\displaystyle\alpha=c_{4}\cos\theta+\sin\theta v\;.$ (33) Using (33) in the second equation of (2) and the first equation of (3) leads us to $\displaystyle\beta=H_{2}P\alpha_{\theta}+H_{2}\bar{f}(\varphi)\;,\quad\delta=\dot{T}\frac{H_{2}}{H_{0}}(\alpha_{\theta}+\frac{\bar{f}}{P})-\frac{y}{H_{2}P}\;,$ (34) where $\delta$ is obtained from the definition $y={\cal B}-{\cal D}$. Substitutions of the solutions (33) and (34) into the $\gamma_{\varphi}$-equation of (2) yield, in terms of constants $c_{7}$ and $c_{8}$ $\displaystyle\bar{f}=\frac{v_{1}}{H_{0}}\;,\quad v_{1}=v_{1}(\varphi)=c_{7}\cos\varphi+c_{8}\sin\varphi\;,$ (35) and $\dot{q}_{1}=c_{7}/T^{2},\;\dot{q}_{2}=c_{8}/T^{2}$ since $\gamma$ is independent of time. #### III.0.3 KY 2-forms for time-dependent de Sitter space-time: $\ell\neq 0$ solutions From the first set of equations of (28), $q_{1}$ and $q_{2}$ can be completely specified as $\displaystyle q_{1}=c_{7}\frac{\dot{T}}{\ell T}+\tilde{c}_{1}\;,\quad q_{2}=c_{8}\frac{\dot{T}}{\ell T}+\tilde{c}_{2}\;,$ (36) by recalling the relations $m_{1}=-\ell\tilde{c}_{1}$ and $m_{2}=-\ell\tilde{c}_{2}$. In view of (29), (30) and (36), $y$ is also completely specified as $\displaystyle y=\frac{\dot{T}}{\ell}v_{1}+TH_{2}Pv_{2}\;,$ (37) where we have defined $\displaystyle v_{2}=v_{2}(\varphi)=c_{1}\cos\varphi+c_{2}\sin\varphi\;.$ (38) In view of (35) and (37), while $\beta$ and $\delta$ have been completely determined, there remains to determine $q_{3}(t)$ of the $y_{3}$-function $\displaystyle y_{3}=Tq_{3}(t)+T(c_{3}H_{2}P-\tilde{c}_{3})\;$ (39) for complete specification of $\gamma,\epsilon$ and $\mu$. The resulting $\beta,\;\delta$ solutions identically satisfy the $\beta_{r}$-equation of (3) and $\delta_{r}$-equation of (2), and therefore they give nothing new. On the other hand, if (37) and (38) are used in the $\gamma$-expression of (18), we see that like $q_{1}$ and $q_{2}$, $q_{3}$ must be equal to $c_{9}(\dot{T}/\ell T)+\tilde{c}_{3}$ since $\gamma$ does not depend on $t$. The resulting $\gamma$ and $\epsilon$ also satisfy the $\epsilon_{r}$-equation of (2), as well as the definition $x={\cal G}-{\cal E}$. But, as they do not take part in any metric coefficient functions, the constants $\tilde{c}_{1},\tilde{c}_{2}$ and $\tilde{c}_{3}$ become redundant. We are now ready to present all the coefficient functions together : $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{4}\cos\theta+\sin\theta v\;,\quad\beta=H_{2}P\alpha_{\theta}+\frac{H_{2}}{H_{0}}v_{1}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle\frac{H_{2}}{H_{0}}(c_{9}\sin\theta+\cos\theta v_{1\varphi})+H_{2}Pv_{\varphi}\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{2}}{H_{0}}\alpha_{\theta}-\frac{1}{\ell}\dot{T}H_{2}Pv_{1}-Tv_{2}\;,$ (40) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle-\sin\theta(c_{9}\frac{1}{\ell}\dot{T}H_{2}P+c_{3}T)-\cos\theta(\frac{1}{\ell}\dot{T}H_{2}Pv_{1\varphi}+Tv_{2\varphi})+\dot{T}\frac{H_{2}}{H_{0}}v_{\varphi}\;,$ $\displaystyle\mu$ $\displaystyle=$ $\displaystyle\sin\theta(\frac{1}{\ell}\dot{T}v_{1\varphi}+TH_{2}Pv_{2\varphi})-\cos\theta(\frac{c_{9}}{\ell}\dot{T}+c_{3}TH_{2}P)+c_{0}TH_{2}\;,$ which determine ten linearly independent KY $2$-forms, one for each $c_{i},\;i=0,1,\ldots,9$: $\displaystyle\omega_{0}$ $\displaystyle=$ $\displaystyle TH_{2}e^{23}\;,$ $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle-T\cos\varphi e^{12}+T\sin\varphi(\cos\theta e^{13}+H_{2}P\sin\theta e^{23})\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle-T\sin\varphi e^{12}-T\cos\varphi(\cos\theta e^{13}-H_{2}P\sin\theta e^{23})\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle-T(\sin\theta e^{13}+H_{2}P\cos\theta e^{23})\;,$ $\displaystyle\omega_{4}$ $\displaystyle=$ $\displaystyle\cos\theta e^{01}-H_{2}\sin\theta\Omega_{1}\wedge e^{2}\;,$ (41) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle\cos\varphi(\sin\theta e^{01}+H_{2}\cos\theta\Omega_{1}\wedge e^{2})-H_{2}\sin\varphi\Omega_{1}\wedge e^{3}\;,$ $\displaystyle\omega_{6}$ $\displaystyle=$ $\displaystyle\sin\varphi(\sin\theta e^{01}+H_{2}\cos\theta\Omega_{1}\wedge e^{2})+H_{2}\cos\varphi\Omega_{1}\wedge e^{3}\;,$ $\displaystyle\omega_{7}$ $\displaystyle=$ $\displaystyle H_{2}\cos\varphi\Omega_{2}\wedge e^{2}-H_{2}\sin\varphi(\cos\theta\Omega_{2}\wedge e^{3}+\frac{\dot{T}}{\ell}\sin\theta e^{23})\;,$ $\displaystyle\omega_{8}$ $\displaystyle=$ $\displaystyle H_{2}\sin\varphi\Omega_{2}\wedge e^{2}+H_{2}\cos\varphi(\cos\theta\Omega_{2}\wedge e^{3}+\frac{\dot{T}}{\ell}\sin\theta e^{23})\;,$ $\displaystyle\omega_{9}$ $\displaystyle=$ $\displaystyle H_{2}\sin\theta\Omega_{2}\wedge e^{3}-\frac{\dot{T}}{\ell}\cos\theta e^{23}\;.$ where the 1-forms $\Omega_{1}$ and $\Omega_{2}$ are defined, for the sake of simplicity, as $\displaystyle\Omega_{1}=Pe^{0}+\frac{\dot{T}}{H_{0}}e^{1}\;,\quad\Omega_{2}=\frac{1}{H_{0}}e^{0}-\frac{\dot{T}}{\ell}Pe^{1}\;.$ (42) #### III.0.4 KY 2-forms for the usual form of de Sitter space-time: $\ell=0$ solutions When $\ell$ is zero, the integrability conditions (11) and condition (10) for $\ell_{1}=1$ imply that $\displaystyle H_{2}P=\varepsilon\;,\quad P^{\prime}=-\frac{H_{1}}{H^{2}_{2}}\;,\quad\dot{T}=\lambda T\;,$ (43) where $\varepsilon$ and $\lambda$ are some nonzero constants. One can easily verify that the first two equations of (43) yield $\varepsilon^{2}=1$, therefore $\varepsilon=\pm 1$, and that the first relation implies the second one. The nine equations of (28) are then as follows for $i=1,2,3$ : $\displaystyle\dot{q}_{i}=\frac{m_{i}}{\lambda T^{2}}\;,\quad z^{\prime}_{i}=\frac{m_{i}}{H^{2}_{0}}\frac{H_{1}}{P}\;,\quad z_{i}^{\prime\prime}-\frac{H_{1}^{\prime}}{H_{1}}z_{i}^{\prime}=\frac{m_{i}}{H^{2}_{0}}H^{2}_{1}\;.$ (44) The first two sets can be easily solved, such that $m_{i}=2c_{i}H^{2}_{0}$ and $\displaystyle q_{i}=-c_{i}\frac{H^{2}_{0}}{\lambda^{2}T^{2}}+a_{i}\;,\quad z_{i}=c_{i}H^{2}_{2}+b_{i}\;,$ (45) where $a_{i}$ and $b_{i}$ are new integration constants. These $z_{i}$ solutions identically satisfy the last relations of (44) without any extra condition. By defining $d_{i}=a_{i}+b_{i}$ and $\displaystyle v_{3}=v_{3}(\varphi)=d_{1}\cos\varphi+d_{2}\sin\varphi\;,$ (46) $y_{i}$ and hence $x,\;y$ functions can be explicitly written from (15) and (30) as $\displaystyle y_{i}$ $\displaystyle=$ $\displaystyle T(c_{i}S_{-}+d_{i})\;,\quad y=T(S_{-}v_{2}+v_{3})\;,$ (47) $\displaystyle x$ $\displaystyle=$ $\displaystyle T(S_{-}v_{2\varphi}+v_{3\varphi})\cos\theta+T(c_{3}S_{-}+d_{3})\sin\theta\;,$ (48) where $v_{2}$ is given by the relation (38) and $\displaystyle S_{\pm}=H^{2}_{2}\pm\frac{H^{2}_{0}}{\lambda^{2}T^{2}}\;.$ (49) The substitution of $\alpha$ and $\beta$ given by (33) and (34) into the $\gamma_{\varphi}$-equation of (2) yields $\displaystyle\bar{f}(\varphi)=\frac{T^{2}}{H_{0}}\partial_{t}\frac{y}{T}=2\frac{H_{0}}{\lambda}v_{2}\;.$ (50) Using (38),(47) and (48) in (18) and (34) give the following : $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{4}\cos\theta+\sin\theta v\;,\quad\beta=H_{2}P\alpha_{\theta}+2\frac{H_{0}}{\lambda}H_{2}v_{2}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle 2\frac{H_{0}}{\lambda}H_{2}(c_{3}\sin\theta+\cos\theta v_{2\varphi})+H_{2}Pv_{\varphi}\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\frac{\lambda}{H_{0}}TH_{2}\alpha_{\theta}+T\frac{1}{H_{2}P}(S_{+}v_{2}-v_{3})\;,$ (51) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\varepsilon TS_{+}(c_{3}\sin\theta+\cos\theta v_{2\varphi})-\varepsilon T(d_{3}\sin\theta+\cos\theta v_{3\varphi})+\frac{\lambda}{H_{0}}TH_{2}v_{\varphi}\;,$ $\displaystyle\mu$ $\displaystyle=$ $\displaystyle T(S_{-}v_{2\varphi}+v_{3\varphi})\sin\theta-T(d_{3}+c_{3}S_{-})\cos\theta+c_{0}TH_{2}\;,$ which determine ten linearly independent KY $2$-forms, one for each $c_{i},\;i=0,1,\ldots,6$ and $d_{j},\;j=1,2,3$. The corresponding KY 2-forms can be written as $\displaystyle\omega_{0}$ $\displaystyle=$ $\displaystyle TH_{2}e^{23}\;,$ $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle 2\frac{H_{0}}{\lambda}H_{2}e^{0}\wedge\Phi_{\varphi}+\varepsilon TS_{+}e^{1}\wedge\Phi-TS_{-}\sin\theta\sin\varphi e^{23}\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle 2\frac{H_{0}}{\lambda}H_{2}e^{0}\wedge\Phi-\varepsilon TS_{+}e^{1}\wedge\Phi_{\varphi}+TS_{-}\sin\theta\cos\varphi e^{23}\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle 2\frac{H_{0}}{\lambda}H_{2}\sin\theta e^{03}+\varepsilon TS_{+}\sin\theta e^{13}-TS_{-}\cos\theta e^{23}\;,$ $\displaystyle\omega_{4}$ $\displaystyle=$ $\displaystyle\cos\theta e^{01}-\sin\theta\Psi\wedge e^{2}\;,$ (52) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle\cos\varphi(\sin\theta e^{01}+\cos\theta\Psi\wedge e^{2})-\sin\varphi\Psi\wedge e^{3}\;,$ $\displaystyle\omega_{6}$ $\displaystyle=$ $\displaystyle\sin\varphi(\sin\theta e^{01}+\cos\theta\Psi\wedge e^{2})+\cos\varphi\Psi\wedge e^{3}\;,$ $\displaystyle\omega_{7}$ $\displaystyle=$ $\displaystyle-\varepsilon T\cos\varphi e^{12}+T\sin\varphi(\varepsilon\cos\theta e^{13}-\sin\theta e^{23})\;,$ $\displaystyle\omega_{8}$ $\displaystyle=$ $\displaystyle-\varepsilon T\sin\varphi e^{12}-T\cos\varphi(\varepsilon\cos\theta e^{13}-\sin\theta e^{23})\;,$ $\displaystyle\omega_{9}$ $\displaystyle=$ $\displaystyle-\varepsilon T\sin\theta e^{13}-T\cos\theta e^{23}\;,$ where the 1-forms $\Phi$ and $\Psi$ are defined, for the sake of simplicity, as $\displaystyle\Phi=\cos\varphi e^{2}-\cos\theta\sin\varphi e^{3}\;,\quad\Psi=\varepsilon e^{0}+\frac{\lambda}{H_{0}}TH_{2}e^{1}\;,$ (53) and $\Phi_{\varphi}=-(\sin\varphi e^{2}+\cos\theta\cos\varphi e^{3})$. ## IV KY 2-Forms for $\dot{T}=0\;,\;H^{\prime}_{0}\neq 0\;,\alpha\neq 0$ In this case, condition (7) requires $\displaystyle H^{\prime}_{2}=kH_{0}H_{1}\;,$ (54) such that $k$ is a nonzero constant and, in addition to $\alpha_{\tau}=0=\alpha_{r}$, we immediately have $\delta_{r}=0=\delta_{\theta}$ and $\epsilon_{r}=0$ from equations (2). Moreover, the $\epsilon_{\varphi}$-equation of (2) becomes independent of the metric characterizing functions, and leads to $\delta_{\varphi\varphi}+\delta=0$ when substituted into the $\varphi$-derivative of the second equation of (4). Therefore, $\delta$ is a harmonic function of $\varphi$ and when this fact is used in the integration of $\epsilon_{\varphi}=-\cos\theta\delta$, we obtain $\displaystyle\delta$ $\displaystyle=$ $\displaystyle D_{1}(\tau)\cos\varphi+D_{2}(\tau)\sin\varphi\;,$ $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\cos\theta\delta_{\varphi}+E(\tau)\sin\theta\;,$ (55) $\displaystyle\mu$ $\displaystyle=$ $\displaystyle- H_{2}P\sin\theta\delta_{\varphi}+E(\tau)H_{2}P\cos\theta+U(\tau,r)\;,$ where we have also made use of the $\mu_{\theta}$ and $\mu_{\varphi}$-equations of (2). The $\epsilon_{\theta}$-equation then yields $P^{\prime}=-H_{1}/H^{2}_{2}$, which means that $\ell_{1}=1$ and $U=H_{2}u(\tau)$. Therefore, $\alpha$ is again given by (35) and $\beta$ can be written, from the second equation of (2) and the first equation of (3), as $\beta=H_{2}P\alpha_{\theta}+H_{2}B(\tau,\varphi)$. It remains to determine only the $\tau$ dependence of $\delta,\epsilon$ and $\mu$. The $\beta_{\tau}$-equation of (2) and the third equation of (3) give $\displaystyle B_{\tau}=m\delta\;,\quad m_{1}B=-\delta_{\tau}\;,$ (56) provided that $m$ and $m_{1}$ defined by $\displaystyle m=\frac{H_{0}^{\prime}}{H_{1}H_{2}}\;,\quad m_{1}=(\frac{H_{2}}{H_{0}})^{\prime}\frac{H^{2}_{0}}{H_{1}}=kH_{0}^{2}-mH_{2}^{2}\;,$ (57) are new constants, such that $m$ is supposed to be nonzero in this section. Thus, $\delta$ must satisfy $\delta_{\tau\tau}+mm_{1}\delta=0$, and therefore its coefficient functions can be obtained, depending on the value of $mm_{1}$, from $\displaystyle D_{i\tau\tau}+mm_{1}D_{i}=0\;,\quad i=1,2\;.$ (58) From here on, the discussion proceeds in to ways: (A) $m_{1}\neq 0$ and (B) $m_{1}=0$. ### IV.1 KY 2-Forms of the static de Sitter space-time: $m_{1}\neq 0$ Solutions In this case, the $\gamma_{\varphi}$-equation of (2), the second equation of (3) and the first equation of (4) specify $\gamma$ as $\displaystyle\gamma=H_{2}Pv_{\varphi}-\frac{H_{2}}{m_{1}}\cos\theta\delta_{\tau\varphi}+H_{2}\sin\theta g(\tau)\;.$ (59) Only three equations remain unused so far; the $\gamma_{\tau},\gamma_{\theta}$-equations and the equation $\partial_{r}(\gamma/H_{0})=-H_{1}\epsilon_{\tau}/H_{0}^{2}$ of (3). One can easily check that the first two of these equations yield $\displaystyle g_{\tau}=mE\;,\quad g=-kE_{\tau}\;,\quad km_{1}=1\;,\quad u_{\tau}=0\;,$ (60) and the last equation gives nothing new. In accordance with the previous solutions, we take the constant $u$ as $u=c_{0}T$. The first two equations of (60) also give $E_{\tau\tau}+mm_{1}E=0$. We can now write the the complete solutions of the case: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{4}\cos\theta+\sin\theta v\;,\quad\beta=H_{2}P\alpha_{\theta}-\frac{H_{2}}{m_{1}}\delta_{\tau}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle H_{2}Pv_{\varphi}-\frac{H_{2}}{m_{1}}(\cos\theta\delta_{\tau\varphi}+\sin\theta E_{\tau})\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle D_{1}\cos\varphi+D_{2}\sin\varphi\;,$ (61) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\cos\theta\delta_{\varphi}+E\sin\theta\;,$ $\displaystyle\mu$ $\displaystyle=$ $\displaystyle- H_{2}P\sin\theta\delta_{\varphi}+EH_{2}P\cos\theta+c_{0}TH_{2}\;,$ where for, say $mm_{1}=-w_{0}^{2}<0$, we have $\displaystyle D_{1}$ $\displaystyle=$ $\displaystyle a_{1}\cosh w_{0}\tau+a_{2}\sinh w_{0}\tau\;,$ $\displaystyle D_{2}$ $\displaystyle=$ $\displaystyle a_{3}\cosh w_{0}\tau+a_{7}\sinh w_{0}\tau\;,$ (62) $\displaystyle E$ $\displaystyle=$ $\displaystyle a_{8}\cosh w_{0}\tau+a_{9}\sinh w_{0}\tau\;.$ When $mm_{1}=w^{2}_{0}>0$, it is enough to replace the hypergeometric functions of (62) by the corresponding trigonometric functions and the minus sign by a plus in the expression of $H_{0}^{2}$. The above solutions define ten linearly independent KY 2-forms $\displaystyle\omega_{0}$ $\displaystyle=$ $\displaystyle TH_{2}e^{23}\;,$ $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle-\frac{w_{0}}{m_{1}}H_{2}\sinh w_{0}\tau e^{0}\wedge\Phi+\cosh w_{0}\tau(e^{1}\wedge\Phi+H_{2}P\sin\theta\sin\varphi e^{23})\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle-\frac{w_{0}}{m_{1}}H_{2}\cosh w_{0}\tau e^{0}\wedge\Phi+\sinh w_{0}\tau(e^{1}\wedge\Phi+H_{2}P\sin\theta\sin\varphi e^{23})\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\frac{w_{0}}{m_{1}}H_{2}\sinh w_{0}\tau e^{0}\wedge\Phi_{\varphi}-\cosh w_{0}\tau(e^{1}\wedge\Phi_{\varphi}+H_{2}P\sin\theta\cos\varphi e^{23})\;,$ $\displaystyle\omega_{4}$ $\displaystyle=$ $\displaystyle\cos\theta e^{01}-H_{2}P\sin\theta e^{02}\;,$ (63) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle\sin\theta\cos\varphi e^{01}+H_{2}P(\cos\theta\cos\varphi e^{02}-\sin\varphi e^{03})\;,$ $\displaystyle\omega_{6}$ $\displaystyle=$ $\displaystyle\sin\theta\sin\varphi e^{01}+H_{2}P(\cos\theta\sin\varphi e^{02}+\cos\varphi e^{03})\;,$ $\displaystyle\omega_{7}$ $\displaystyle=$ $\displaystyle\frac{w_{0}}{m_{1}}H_{2}\cosh w_{0}\tau e^{0}\wedge\Phi_{\varphi}-\sinh w_{0}\tau(e^{1}\wedge\Phi_{\varphi}+H_{2}P\sin\theta\cos\varphi e^{23})\;,$ $\displaystyle\omega_{8}$ $\displaystyle=$ $\displaystyle-\frac{w_{0}}{m_{1}}H_{2}\sinh w_{0}\tau\sin\theta e^{03}+\cosh w_{0}\tau(\sin\theta e^{1}+H_{2}P\cos\theta e^{2})\wedge e^{3}\;,$ $\displaystyle\omega_{9}$ $\displaystyle=$ $\displaystyle-\frac{w_{0}}{m_{1}}H_{2}\cosh w_{0}\tau\sin\theta e^{03}+\sinh w_{0}\tau(\sin\theta e^{1}+H_{2}P\cos\theta e^{2})\wedge e^{3}\;,$ corresponding, respectively, to $c_{i},\;i=0,4,5,6$ and $a_{j},\;j=1,2,3,7,8,9$. Here, $\Phi$ is given by (53) and the metric coefficient functions are as follows: $\displaystyle H_{0}^{2}=m_{1}^{2}-w^{2}_{0}H_{2}^{2}\;,\quad H_{1}=m_{1}\frac{H^{\prime}_{2}}{H_{0}}\;.$ (64) ### IV.2 $m_{1}=0$ Solutions When $m_{1}$ is zero, we have $H_{2}=m_{0}H_{0}$ where $m_{0}$ is a nonzero constant and, by virtue of (54) and (57), $k=mm^{2}_{0}$. Therefore $k$ and $m$ have the same sign, and (54) and (57) amount to the same relation. In that case, the equations of (56) are of the forms $\delta_{\tau}=0$ and $B_{\tau}=m\delta$, which imply that $B=m\tau\delta+C_{\varphi}(\varphi)$ and $\displaystyle\beta=H_{2}(P\alpha_{\theta}+m\tau\delta+C_{\varphi})\;,\quad\delta=b_{1}\cos\varphi+b_{2}\sin\varphi\;.$ (65) where $b_{i}$’s are constants. The $\gamma_{\varphi}$-equation of (2) and the second equation of (3) provide us with $\displaystyle\gamma=H_{2}[Pv_{\varphi}+\cos\theta(m\tau\delta_{\varphi}-C(\varphi))+G(\tau,\theta)]\;,$ (66) such that $\ell_{1}=1$. Thus, $\alpha$ is still given by (33) and $\epsilon,\mu$ are as in equations (55). For the above solutions, we have $\partial_{r}(\gamma/H_{0})=0$ and the fifth equation of (3) implies that $\epsilon_{\tau}=0$, that is, $E$ is independent of time. On the other hand, the $\gamma_{\tau}$-equation of (2) and the first equation of (4) yield $\displaystyle G_{\tau}=mE\sin\theta\;,\quad C_{\varphi\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{G}{\sin\theta}\;.$ We are finally left with the $\gamma_{\theta}$-equation, which gives $\displaystyle G_{\theta}=\sin\theta(m\tau\delta_{\varphi}-C)-\frac{H_{2}}{H_{0}}u_{\tau}\;.$ As $G$ is independent of $\varphi$, the only possible solutions of the last three equations are $G_{\theta}=-c\sin\theta$ and $\delta_{\varphi}=0=u_{\tau}$ such that $C=c=$constant. Since $G_{\tau\theta}=G_{\theta\tau}$ implies $E=0$, we can write $\epsilon=0=\delta$ and $u=c_{0}T$. Although $G=c\cos\theta$ is a solution, $c$ does not take part in any component functions. The results can therefore be written as follows: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{4}\cos\theta+\sin\theta v\;,\quad\beta=H_{2}P\alpha_{\theta}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle H_{2}v_{\varphi}\;,\quad\delta=0=\epsilon\;,\quad\mu=c_{0}TH_{2}\;,$ which define four linearly independent KY 2-forms; $\omega_{0}$ and $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle\cos\theta e^{01}-mm_{0}H_{2}\sin\theta e^{02}\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle-\cos\varphi\omega_{1\theta}-H_{2}\sin\varphi e^{03}\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle-\sin\varphi\omega_{1\theta}+H_{2}\cos\varphi e^{03}\;.$ (The corresponding 1-forms can be found from Section IV-A or V with $k_{1}=0$ of the first paper). ## V KY 2-Forms of The Flat Space-Time In this section we take both $\dot{T}$ and $H_{0}^{\prime}$ to be zero, which imply $\beta_{\tau}=0=\gamma_{\tau},\;\delta_{r}=0=\delta_{\theta}$ and $\epsilon_{r}=0$. Thus, as in the previous section, we have $\displaystyle\beta$ $\displaystyle=$ $\displaystyle H_{2}P\alpha_{\theta}+H_{2}B(\varphi)\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle D_{1}(\tau)\cos\varphi+D_{2}(\tau)\sin\varphi\;,$ (67) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\cos\theta\delta_{\varphi}+E(\tau)\sin\theta\;,\quad\mu=H_{2}(P\epsilon_{\theta}+u(\tau))\;,$ provided that $P^{\prime}=-H_{1}/H^{2}_{2}$ and hence $\alpha$ is given by (33). On the other hand, the $\beta_{r}$-equation of (3) gives $\displaystyle(H_{2}P)^{\prime}=0\;,\quad B=-\frac{H_{1}}{H_{0}H^{\prime}_{2}}\delta_{\tau}\;.$ (68) The first relation is equivalent to the first integrability condition of (7), and when it is combined with $P^{\prime}=-H_{1}/H^{2}_{2}$, we obtain $P=\varepsilon/H_{2}$, that is $\varepsilon H_{2}^{\prime}=H_{1}$ with $\varepsilon=\pm 1$. Five equations remain untouched so far; the $\gamma_{\varphi}$-equation of (2), three equations of (3) involving $\gamma$, and the first equation of (4). We first consider the $\gamma_{\theta}$-equation of (3) by inserting the $\mu$-solution of (67) into it. The resulting equation implies that $u$ must be a constant, which we take to be $u=c_{0}T$, and we are left with $\displaystyle\gamma_{\theta}=\frac{H^{2}_{2}P}{H_{0}}(\sin\theta\delta_{\tau\varphi}-E_{\tau}\cos\theta)\;.$ (69) As $\gamma$ is independent of $\tau$, $E_{\tau}$ and $D_{j\tau}$ must be constants, and therefore $E=a_{0}\tau+b_{0},\;D_{1}=b_{1}\tau+b_{3}$ and $D_{2}=b_{2}\tau+b_{4}$, where $a_{i}$ and $b_{j}$ are constants. Thus, in terms of $\displaystyle z_{1}(\varphi)=b_{1}\cos\varphi+b_{2}\sin\varphi\;,\quad z_{2}(\varphi)=b_{3}\cos\varphi+b_{4}\sin\varphi\;,$ (70) we can write $\delta=\tau z_{1}+z_{2}$, which implies that $B=-z_{1}/H_{0}H_{2}P$ by virtue of the second relation of (68). Therefore, (69) can be integrated to obtain the explicit expression of $\gamma$, presented together with the other solutions below: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle c_{4}\cos\theta+\sin\theta v\;,\quad\beta=H_{2}P\alpha_{\theta}-\frac{1}{H_{0}P}z_{1}\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle-\frac{H^{2}_{2}P}{H_{0}}(\cos\theta z_{1\varphi}+a_{0}\sin\theta)+H_{2}Pv_{\varphi}\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\tau z_{1}+z_{2}\;,$ (71) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\tau(\cos\theta z_{1\varphi}+a_{0}\sin\theta)+\cos\theta z_{2\varphi}+b_{0}\sin\theta\;,$ $\displaystyle\mu$ $\displaystyle=$ $\displaystyle H_{2}(P\epsilon_{\theta}+c_{0}T)\;.$ The last term of $\gamma$ which arises from the $\theta$-integration mentioned above is specified by the $\beta_{\varphi}$-equation of (4). It is straightforward to verify that the remaining three equations of $\gamma$ are identically satisfied. The corresponding KY 2-forms are, in addition to $\omega_{0}$, as follows: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle-\varepsilon\frac{H_{2}}{H_{0}}(\cos\varphi e^{02}-\cos\theta\sin\varphi e^{03})+\tau(\cos\varphi e^{12}-\sin\varphi A\wedge e^{3})\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle-\varepsilon\frac{H_{2}}{H_{0}}(\sin\varphi e^{02}+\cos\theta\cos\varphi e^{03})+\tau(\sin\varphi e^{12}+\cos\varphi A\wedge e^{3})\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\cos\varphi e^{12}-\sin\varphi A\wedge e^{3}\;,\;\;\;\qquad\omega_{4}=e^{0}\wedge A\;,$ (72) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle-\cos\varphi e^{0}\wedge A_{\theta}-\varepsilon\sin\varphi e^{03}\;,$ $\displaystyle\omega_{6}$ $\displaystyle=$ $\displaystyle-\sin\varphi e^{0}\wedge A_{\theta}+\varepsilon\cos\varphi e^{03}\;,\quad\omega_{7}=\sin\varphi e^{12}+\cos\varphi A\wedge e^{3}\;,$ $\displaystyle\omega_{8}$ $\displaystyle=$ $\displaystyle-\varepsilon\frac{H_{2}}{H_{0}}\sin\theta e^{03}-\tau A_{\theta}\wedge e^{3}\;,\;\;\quad\omega_{9}=-A_{\theta}\wedge e^{3}\;,$ where $A=\cos\theta e^{1}-\varepsilon\sin\theta e^{2}$. ## VI Solutions For $\alpha=0$ In this case the second and fourth equations of (2) give $\beta_{\theta}=0=\delta_{\theta}$ and $\displaystyle\beta=H_{2}B(t,\varphi)\;,\quad\gamma=H_{2}G(t,\theta,\varphi)\;,$ (73) are implied by the first two equations of (3). Fourteen equations remain to be solved. But when $\alpha$ is set to zero, the $\gamma_{\varphi}$ and $\epsilon_{\varphi}$-equations of (2) are freed from metric coefficient functions, such that $\gamma_{\varphi}=-\cos\theta\beta$ and $\epsilon_{\varphi}=-\cos\theta\delta$. When these are combined with two equations of (4), they considerably ease the investigation by providing general forms of the solutions. Indeed, differentiating both equations of (4) with respect to $\varphi$ gives $\beta_{\varphi\varphi}+\beta=0=\delta_{\varphi\varphi}+\delta$. Therefore, the general forms of $\beta,\gamma,\delta,\epsilon$ and $\mu$ can be written, in view of (73), as follows: $\displaystyle\beta$ $\displaystyle=$ $\displaystyle H_{2}(B_{1}(t)\cos\varphi+B_{2}(t)\sin\varphi)\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle\cos\theta\beta_{\varphi}+G(t)H_{2}\sin\theta\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle D_{1}(t,r)\cos\varphi+D_{2}(t,r)\sin\varphi\;,$ (74) $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\cos\theta\delta_{\varphi}+E(t,r)\sin\theta\;,$ $\displaystyle\mu$ $\displaystyle=$ $\displaystyle(\dot{T}\frac{H_{2}}{H_{0}}\beta_{\varphi}-\frac{H^{\prime}_{2}}{H_{1}}\delta_{\varphi})\sin\theta+U(t,r,\theta)\;,$ where functions $B_{i},D_{i},G,E$ and $U$ are to be determined from the remaining nine equations: the five equations in the first column of (2) and the last four equations of (3). The second and fourth relations are obtained by integrations and then by using the results in equations (4). The last relation of (74) is obtained by first using $\beta$ and $\delta$ obtained in the last relation of (2) and then by integrating the resulting equation with respect to $\varphi$. When the solutions (74) are substituted into the five equations appearing in the first column of equations (2), we obtain $\displaystyle\dot{B_{1}}$ $\displaystyle=$ $\displaystyle\frac{M}{T}D_{1}\;,\;\;\quad\dot{B_{2}}=\frac{M}{T}D_{2}\;,$ $\displaystyle D_{1r}$ $\displaystyle=$ $\displaystyle\dot{T}LB_{1}\;,\quad D_{2r}=\dot{T}LB_{2}\;,$ (75) $\displaystyle\dot{G}$ $\displaystyle=$ $\displaystyle\frac{M}{T}E\;,\qquad E_{r}=\dot{T}LG\;,$ $\displaystyle U_{\theta}$ $\displaystyle=$ $\displaystyle(\dot{T}\frac{H_{2}^{2}}{H_{0}}G-\frac{H_{2}^{\prime}}{H_{1}}E)\sin\theta\;,$ where we have defined the functions $\displaystyle M=\frac{H_{0}^{\prime}}{H_{1}H_{2}}\;,\quad L=\frac{H_{1}H_{2}}{H_{0}}\;.$ (76) ### VI.1 A General Case To be as general as possible, we shall first seek solutions for which both $\dot{T}$ and $H_{0}^{\prime}$ can be different from zero. In such a case, provided that $m$ defined by $\displaystyle m=\frac{M^{\prime}}{M^{2}L}\;$ (77) is a constant, from the first four equations of (75) one can easily obtain $\displaystyle B_{i}=b_{i}K\;,\quad D_{i}=b_{i}\frac{\dot{K}}{K^{m}M}\;;\quad i=1,2\;,$ (78) where $b_{1}$ and $b_{2}$ are integration constants, $K=T^{-1/m}$ and $m$ is supposed to be different from zero. In a similar way, the fifth and sixth equations of (75) yield $\displaystyle G=b_{3}K\;,\quad E=b_{3}\frac{\dot{K}}{K^{m}M}\;,\quad U_{\theta}=-b_{3}\frac{\dot{K}R}{K^{m}M}\sin\theta\;,$ (79) where $b_{3}$ is a constant, and we have defined $\displaystyle R=\frac{H_{2}}{H_{1}}(m\frac{H_{0}^{\prime}}{H_{0}}+\frac{H_{2}^{\prime}}{H_{2}})\;.$ (80) In terms of $g=b_{1}\cos\varphi+b_{2}\sin\varphi$, the solutions can now be rewritten as $\displaystyle\beta$ $\displaystyle=$ $\displaystyle KH_{2}g\;,\quad\;\;\gamma=KH_{2}(g_{\varphi}\cos\theta+b_{3}\sin\theta)\;,$ $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\frac{\dot{K}}{K^{m}M}g\;,\quad\epsilon=\frac{\dot{K}}{K^{m}M}(g_{\varphi}\cos\theta+b_{3}\sin\theta)\;,$ (81) $\displaystyle\mu$ $\displaystyle=$ $\displaystyle-\frac{\dot{K}R}{K^{m}M}(g_{\varphi}\sin\theta- b_{3}\cos\theta)+u(t,r)\;,$ where the integration of $U_{\theta}$ done with respect to $\theta$. There remain the last four equations of (3) that have not been used so far. Substitution of these solutions into the last four equations of (3) yield $\displaystyle\ddot{K}$ $\displaystyle=$ $\displaystyle- m_{1}K^{2m+1}=-m_{2}K^{2m+1}\;,$ $\displaystyle\partial_{t}(\frac{u}{T})$ $\displaystyle=$ $\displaystyle 0=\partial_{r}(\frac{u}{H_{2}})\;,$ (82) $\displaystyle\frac{H_{1}}{H_{2}^{2}}$ $\displaystyle=$ $\displaystyle-M(\frac{R}{H_{2}M})^{\prime}\;,$ provided that $m_{1}$ and $m_{2}$ defined by $\displaystyle m_{1}=M\frac{H_{0}^{2}}{H_{1}}(\frac{H_{2}}{H_{0}})^{\prime}\;,\quad m_{2}=M\frac{H_{0}}{R}\;,$ (83) are constant. In fact, the first equality of (82) implies that $m_{1}=m_{2}$, and $\displaystyle\ddot{K}$ $\displaystyle=$ $\displaystyle-m_{1}K^{2m+1}\;,\quad R\frac{H_{0}}{H_{1}}(\frac{H_{2}}{H_{0}})^{\prime}=1\;.$ (84) The equations appearing in the second line of (82) give $u=TH_{2}$, and hence $\mu$ has been completely specified as $\displaystyle\mu=-\frac{\dot{K}R}{K^{m}M}(g_{\varphi}\sin\theta- b_{3}\cos\theta)+c_{0}TH_{2}\;.$ (85) We have obtained five linearly independent KY 2-forms for a family of space- times characterized by two constants $m$ and $m_{1}$. Having determined the coefficient functions of $\omega_{2}$, we now turn to the conditions which restrict the functions determining the metric tensor. In addition to two conditions given by (84), we have two more conditions defining the constants $m$ and $m_{1}$. The first, given by (77), can be integrated to yield $\displaystyle H_{0}^{\prime}=kH_{0}^{-1/m}H_{1}H_{2}\;$ (86) and hence, $M=kH_{0}^{m}$. From the first equation of (83) and the second of (84), we find $\displaystyle R=\frac{k}{m_{1}}H_{0}^{m+1}\;,\quad H_{2}^{\prime}=kH_{0}^{m-1}H_{2}^{2}+\frac{m_{1}}{k}H_{1}H_{0}^{-m-1}\;.$ Substitutions of these into (73) finally yields $\displaystyle m_{1}H_{2}(m+\frac{1}{H_{1}})=H_{0}^{2}-(\frac{m_{1}}{k})^{2}H_{0}^{-2m}\;.$ (87) Although several subclasses of space-times can be identified for particular values of $m$ and $m_{1}$, this general consideration will not be pursued any further. It will suffice to exhibit the physically important final case. ### VI.2 KY $2$-forms of the Robertson-Walker space-time The Robertson-Walker space-time is characterized by $\displaystyle H_{0}=1\;,\quad H_{2}=r\;,\quad H_{1}^{2}=\frac{1}{1+k_{3}r^{2}}\;,$ (88) such that $T$ is specified by special cosmological models. Two such specifications are $T=t^{1/2}$ and $T=t^{2/3}$ which correspond, respectively, to radiation-dominated and matter-dominated universes. In this final subsection, we shall present KY 2-forms of this space-time such that there is no constraint on $T$. It turns out that such a case is possible only if we take $\alpha=0=\beta=\gamma$ and $H^{\prime}_{0}=0$. The solutions then are $\displaystyle\delta$ $\displaystyle=$ $\displaystyle T(c_{1}\cos\varphi+c_{2}\sin\varphi)\;,$ $\displaystyle\epsilon$ $\displaystyle=$ $\displaystyle\cos\theta\delta_{\varphi}+c_{3}T\sin\theta\;,$ (89) $\displaystyle\mu$ $\displaystyle=$ $\displaystyle- H_{2}P\sin\theta\delta_{\varphi}+T(c_{3}H_{2}P\cos\theta+c_{0}H_{2})\;,$ provided that $P^{\prime}=-H_{1}/H_{2}^{2}$, which leads to $H_{1}$ of (94). The above solutions provide us, in addition to $\omega_{0}$, with three linearly independent KY 2-forms: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle T(\cos\varphi e^{12}-\cos\theta\sin\varphi e^{13})+H_{2}P\sin\theta\sin\varphi e^{23}\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle T(\sin\varphi e^{12}+\cos\theta\cos\varphi e^{13})-H_{2}P\sin\theta\cos\varphi e^{23}\;,$ (90) $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle T(\sin\theta e^{13}+H_{2}P\cos\theta e^{23})\;.$ ## VII KY 3-Forms For a 3-form $\displaystyle\omega_{(3)}=\alpha e^{012}+\beta e^{013}+\gamma e^{023}+\delta e^{123}\;,$ the KY-equation gives sixteen equations, five of which have the following simple forms: $\displaystyle\alpha_{t}=0=\beta_{t}\;,\quad\alpha_{r}=0=\beta_{r}\;,\quad\alpha_{\theta}=0\;.$ These imply that $\alpha$ depends only on $\varphi$, and $\beta$ is a function of $\theta$ and $\varphi$. Seven of the remaining equations have the following two-term forms: $\displaystyle\beta_{\varphi}$ $\displaystyle=$ $\displaystyle-\cos\theta\alpha\;,$ $\displaystyle\gamma_{t}$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}}{TH_{1}}\delta\;,\quad\gamma_{\theta}=-\frac{H^{\prime}_{2}}{H_{1}}\beta\;,\;\quad\gamma_{\varphi}=\frac{H^{\prime}_{2}}{H_{1}}\sin\theta\alpha\;,$ (91) $\displaystyle\delta_{r}$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{1}}{H_{0}}\gamma\;,\quad\delta_{\theta}=-\dot{T}\frac{H_{2}}{H_{0}}\beta\;,\quad\delta_{\varphi}=\dot{T}\frac{H_{2}}{H_{0}}\sin\theta\alpha\;.$ The last four equations give three independent equations, which can be written as: $\displaystyle\alpha_{\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{\beta}{\sin\theta}\;,\quad\beta_{\theta}=-\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{\gamma}{H_{2}}\;,\quad\ \partial_{r}\frac{\gamma}{H_{0}}=-T^{2}\frac{H_{1}}{H^{2}_{0}}\partial_{t}\frac{\delta}{T}\;.$ (92) Two of the most important integrability conditions for these equations are as follows: $\displaystyle\dot{T}H^{\prime}_{0}\beta=0\;,\quad[\rho(r)-\varrho(t)]\beta=0\;,$ (93) where $\rho$ and $\varrho$ are given by (7). The first follows from $\gamma_{t\theta}=\gamma_{\theta t}$ and $\delta_{r\theta}=\delta_{\theta r}$, and the second from $\delta_{t\theta}=\delta_{\theta t}$. There are also identical conditions with $\beta$ replaced by $\alpha$ that can be checked from $\gamma_{t\varphi}=\gamma_{\varphi t}\;,\delta_{r\varphi}=\delta_{\varphi r}$ and $\delta_{t\varphi}=\delta_{\varphi t}$. But noting that $\beta=0$ implies $\alpha=0$, we see that the above conditions include the second (see also relations (94)). There are also some other conditions which should be considered in investigating the cases implied by the above conditions. Therefore, we shall present the general solutions in two classes: (A) $\beta\neq 0$ and (B) $\beta=0$. The essence of spherical symmetry seems to be encoded in the first equations of (91) and (92), for they do not depend on the metric coefficient functions. We are thus able to start with their general solutions $\displaystyle\alpha=a_{1}\cos\varphi+a_{2}\sin\varphi\;,\quad\beta=\cos\theta\alpha_{\varphi}+a_{3}\sin\theta\;,$ (94) which can easily be verified. Here, $a_{i}$ are integration constants. ### VII.1 Solutions for $\beta\neq 0$ For the fulfilment of conditions (93) in the case of nonzero $\beta$, two sets of conditions must be distinguished: $\displaystyle{\bf(i)}:\;\;\quad\dot{T}$ $\displaystyle=$ $\displaystyle 0\;,\quad H^{\prime}_{2}=kH_{0}H_{1}\;,$ $\displaystyle{\bf(ii)}:\quad H_{0}^{\prime}$ $\displaystyle=$ $\displaystyle 0\;,\quad\dot{T}^{2}-T\ddot{T}=-\ell=\frac{H_{0}^{2}}{H_{1}H_{2}}(H_{2}P)^{\prime}\;,$ Here $k$ and $\ell$ are constants such that $k\neq 0$. The well-known maximal symmetric Minkowski and the static form of de Sitter space-times, each having five independent KY $3$-forms are obtained among the ${\bf(i)}$ solutions as special cases. On the other hand, four time dependent forms plus the most well-known form of de Sitter and Robertson-Walker space-times, emerge in the second case. We should emphasize the fact that the former space-time is obtained without any restriction on $T$, which is obtained by taking $H_{0}^{\prime}$ zero and by starting in such a way that the last two conditions of (ii) are avoided. #### VII.1.1 KY 3-Forms for the Minkowski and Static Form of de Sitter Space- time In the (i) case we have, in addition to $\delta=f(\tau)$, the following five equations: $\displaystyle\gamma_{\tau}$ $\displaystyle=$ $\displaystyle\frac{H^{\prime}_{0}}{H_{1}}f\;,\quad\qquad\gamma_{\theta}=-kH_{0}\beta\;,\;\quad\gamma_{\varphi}=kH_{0}\sin\theta\alpha\;,$ $\displaystyle f_{\tau}$ $\displaystyle=$ $\displaystyle-\frac{H^{2}_{0}}{H_{1}}\partial_{r}\frac{\gamma}{H_{0}}\;,\;\;\beta_{\theta}=-\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{\gamma}{H_{2}}\;.$ where $\tau=t/T$. Provided that $m$ is a nonzero constant such that $\displaystyle H_{2}H_{0}^{\prime}-H_{0}H_{2}^{\prime}=mH_{1}\;,$ (96) the last two equations of (95) imply that $m\gamma$ must be equal to $H_{2}f_{\tau}-H_{0}\beta_{\theta}$. The first and second equations of (95) in this case require $km=-1$ and $f_{\tau\tau}+m_{1}f=0$, where the constant $m_{1}$ is defined by $\displaystyle H_{0}^{\prime}$ $\displaystyle=$ $\displaystyle km_{1}H_{1}H_{2}\;.$ (97) The third equation of (95) is then identically satisfied. As an alternative approach one can first integrate the $\gamma_{\theta}$-equation with respect to $\theta$, and then use it in the $\gamma_{\varphi}$-equation to find $\gamma=kH_{0}\beta_{\theta}+G(\tau,r)$. The remaining three equations then give the same solution. As a result, under the ${\bf(i)}$ conditions the general forms of the coefficient functions for KY $3$-form are, in addition to that given by (94), as follows: $\displaystyle\gamma=-k(f_{\tau}H_{2}-H_{0}\beta_{\theta})\;,\quad\delta=f(\tau)\;.$ (98) Depending on the value of $m_{1}$, one can easily write the explicit form of $f$. For $m_{1}=0$ we have, in terms of integration constants $a,b$, $\displaystyle f=a+b\tau\;,\quad H_{2}^{\prime}=kH_{0}H_{1}\;.$ (99) and $H^{2}_{0}=k^{-2}$ from equations (96) and (97). For nonzero $m_{1}$, we have $\displaystyle H^{2}_{0}=k_{0}+m_{1}H_{2}^{2}\;,\quad H_{0}H_{1}=k_{0}kH_{2}^{\prime}\;,\quad k_{0}k^{2}=1\;$ (100) and $f$ is as follows ($k_{0},b_{1}$ and $b_{2}$ are integration constants): $\displaystyle f=\left\\{\begin{array}[]{cc}b_{1}\cos\omega_{0}\tau+b_{2}\sin\omega_{0}\tau\;;&\quad m_{1}=\omega_{0}^{2}>0\;,\\\ b_{1}\cosh\omega_{0}\tau+b_{2}\sinh\omega_{0}\tau\;,&\quad m_{1}=-\omega_{0}^{2}<0\;,\end{array}\right.$ (103) In any case, we have five linearly independent $3$-forms. For $T=1,\;H_{0}=1=H_{1}$ and $H_{2}=r$ we get, from the equations (94), (97) and (99), 3-forms of Minkowski space-time: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle w_{1}+\sin\theta\sin\varphi e^{023}\;,\quad\omega_{2}=w_{2}-\sin\theta\cos\varphi e^{023}\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\sin\theta e^{013}+\cos\theta e^{023}\;,$ (104) $\displaystyle\omega_{4}$ $\displaystyle=$ $\displaystyle e^{123}\;,\quad\omega_{5}=-re^{023}+\tau e^{123}\;,$ where the $3$-forms $w_{1}$ and $w_{2}$ are defined, for brevity, as $\displaystyle w_{1}=\cos\varphi e^{012}-\cos\theta\sin\varphi e^{013}\;,\quad w_{2}=\sin\varphi e^{012}+\cos\theta\cos\varphi e^{013}\;.$ (105) Note that $k=1$ (hence $m=-1$) and $m_{1}=0$ for this case. On the other hand, for the values $\displaystyle T=1\;,\quad H_{2}=r\;,\quad k_{0}=1=k\;$ (106) we obtain, by virtue of equations (94), (97) and (100), the KY $3$-forms $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle w_{1}+H_{0}\sin\theta\sin\varphi e^{023}\;,\quad\omega_{2}=w_{2}-H_{0}\sin\theta\cos\varphi e^{023}\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\sin\theta e^{013}+H_{0}\cos\theta e^{023}\;,\quad\omega_{4}=r\sinh\tau e^{023}+\cosh\tau e^{123}\;,$ (107) $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle-r\cosh\tau e^{023}+\sinh\tau e^{123}\;$ for the static form of de Sitter space-time, specified also by $H_{1}^{2}=1+m_{1}r^{2}$ and $H_{0}H_{1}=1$. #### VII.1.2 Solutions for Four Time-Dependent Forms of de Sitter Space-time From here on, we consider the ${\bf(ii)}$ conditions. The condition $H_{0}^{\prime}=0$ gives $\gamma_{t}=0$ and leaves us with seven equations to be solved. It is easy to integrate the $\gamma_{\theta}$-equation of (91) and then use it in the $\gamma_{\varphi}$-equation to find $\gamma=H_{2}P\beta_{\theta}+G(r)$. By substituting this solution into the $\beta_{\theta}$-equation, we find $G=cH_{2}$ and hence, $\gamma=H_{2}P\beta_{\theta}+cH_{2}$, provided that $P^{\prime}=-H_{1}/H_{2}^{2}$. Here, $c$ is an integration constant. The $\delta_{\theta}$-equation can also be integrated with respect to $\theta$, and then by substituting the solution into $\delta_{\varphi}$-equation, we find $\displaystyle\delta=\dot{T}\frac{H_{2}}{H_{0}}\beta_{\theta}+D(t,r)\;.$ (108) The following two equations of (91) and (92) remain to be solved: $\displaystyle\delta_{r}=\dot{T}\frac{H_{1}}{H_{0}}\gamma\;,\;\quad\frac{H_{0}}{H_{1}}\gamma_{r}=-T^{2}\partial_{t}\frac{\delta}{T}\;.$ By substituting the above $\gamma$ solution and (106) into these equations, we obtain $\displaystyle D_{r}=c\dot{T}\frac{H_{1}H_{2}}{H_{0}}\;,\quad T^{2}\partial_{t}\frac{D}{T}=-cH_{0}H_{2}P\;,$ (109) provided that $\displaystyle(H_{2}P)^{\prime}=-\ell\frac{H_{1}H_{2}}{H_{0}^{2}}\;,$ (110) which is just one of the integrability conditions of (ii). Now it is not difficult to see that for nonzero values of $\ell$, the general solution for $D$ is $D=-(c/\ell)\dot{T}H_{0}H_{2}P+c_{0}T$, where $c_{0}$ is another integration constant. We can now collate the solutions of the case as follows: $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle a_{1}\cos\varphi+a_{2}\sin\varphi\;,\quad\beta=\cos\theta\alpha_{\varphi}+a_{3}\sin\theta\;,$ $\displaystyle\gamma$ $\displaystyle=$ $\displaystyle H_{2}P\beta_{\theta}+cH_{2}\;,$ (111) $\displaystyle\delta$ $\displaystyle=$ $\displaystyle\dot{T}\frac{H_{2}}{H_{0}}\beta_{\theta}-\frac{c}{\ell}\dot{T}H_{0}H_{2}P+c_{0}T\;.$ Depending on the values of $\ell$ and other integration constants arising when integrating the equation $T^{2}\partial_{t}(\dot{T}/T)=\ell$, four different time regimes were presented in Part I (see relations (72) in I). The above solutions provide five independent KY $3$-forms: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle\cos\varphi e^{012}-\cos\theta\sin\varphi e^{013}+H_{2}\sin\theta\sin\varphi B\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle\sin\varphi e^{012}+\cos\theta\cos\varphi e^{013}-H_{2}\sin\theta\cos\varphi B\;,$ $\displaystyle\omega_{3}$ $\displaystyle=$ $\displaystyle\sin\theta e^{013}+H_{2}\cos\theta B\;,$ (112) $\displaystyle\omega_{4}$ $\displaystyle=$ $\displaystyle H_{2}(e^{023}-\frac{H_{0}}{\ell}\dot{T}Pe^{123})\;,$ $\displaystyle\omega_{5}$ $\displaystyle=$ $\displaystyle Te^{123}\;,$ where we have defined the 3-form $B=Pe^{023}+H^{-1}_{0}\dot{T}e^{123}$. #### VII.1.3 KY 3-Forms of de Sitter Space-Time with Exponential Time Dependence When $\ell=0$, we have $\dot{T}=\lambda T$ and $H_{2}^{\prime}=m_{0}H_{1}$ from (108). In such a case, the most general solution of (107) turns out to be $\displaystyle D=c\frac{\lambda T}{2m_{0}H_{0}}[H_{2}^{2}+(\frac{m_{0}H_{0}}{\lambda T})^{2}]+c_{1}\frac{1}{T}\;,$ where $c_{1}$ is an integration constant. The solutions are then given by (109), with $\delta$ replaced by $\delta=\dot{T}(H_{2}/H_{0})\beta_{\theta}+D$. From $P^{\prime}=-H_{1}/H_{2}$, it follows that $m_{0}^{2}=1$, that is, $m_{0}=\varepsilon=\pm 1$ and we again obtain five linearly independent $3$-forms for de Sitter space-time. The first three forms are the same as those given in (110), and the last two are as follows: $\displaystyle\omega_{4}=H_{2}e^{023}+\frac{\lambda T}{2m_{0}H_{0}}[H_{2}^{2}+(\frac{m_{0}H_{0}}{\lambda T})^{2}]e^{123}\;,\quad\omega_{5}=\frac{1}{T}e^{123}\;.$ (113) #### VII.1.4 KY 3-form of the Robertson-Walker Space-Time For the solutions obtained so far under ${\bf(ii)}$ conditions, $T$ is restricted as a special function of time. It turns out (see also the next section) that the only possible solution in which there are no constraints on $T$ is $\displaystyle\omega=Te^{123}\;,$ (114) provided that $H_{0}^{\prime}=0$. In particular, there is only one KY 3-form for the Robertson-Walker space-time. ### VII.2 Solutions for $\beta=0$ For $\beta=0$, the equations (91) and (92) imply that $\alpha=0,\;\gamma=H_{2}f(t)$ and $\delta=D(t,r)$, and the following three equations $\displaystyle T\dot{f}$ $\displaystyle=$ $\displaystyle\frac{H_{0}^{\prime}}{H_{1}H_{2}}D\;,$ $\displaystyle\dot{T}f$ $\displaystyle=$ $\displaystyle\frac{H_{0}}{H_{1}H_{2}}D_{r}\;,$ (115) $\displaystyle\frac{T^{2}}{f}\partial_{t}\frac{D}{T}$ $\displaystyle=$ $\displaystyle-\frac{H_{0}^{2}}{H_{1}}(\frac{H_{2}}{H_{0}})^{\prime}\;,$ determine $f$ and $D$. If $H_{0}^{\prime}$ is zero, then $f$ is a constant, say $c$, and we get $\gamma=cH_{2}$ and $\delta=-(c/\ell)\dot{T}H_{0}H_{2}P+c_{0}T$, which are special cases of solutions (109). Moreover, the special case $c=0$ produces the solution (112) for the Robertson-Walker space-time. For nonzero $H_{0}^{\prime}$, taking $D$ from the first equation of (113) and substituting it into the other two equations lead us to two separate equations. Each side of these equations must be constant such that $\displaystyle\frac{H_{0}}{H_{1}H_{2}}(\frac{H_{1}H_{2}}{H_{0}^{\prime}})^{\prime}=m_{1}\;,\quad\frac{H_{0}^{2}H_{0}^{\prime}}{H^{2}_{1}H_{2}}(\frac{H_{2}}{H_{0}})^{\prime}=m_{2}\;.$ (116) This leaves us with two simple equations for $f$: $\displaystyle\frac{\dot{T}f}{T\dot{f}}=m_{1}\;,\quad T^{2}\frac{\ddot{f}}{f}=-m_{2}\;.$ (117) For $m_{1}=0$, we have $\dot{T}=0$ and $\displaystyle f_{\tau\tau}+m_{2}f=0\;,\quad H_{0}^{\prime}=m_{3}H_{1}H_{2}\;,\quad D=m_{3}^{-1}f_{\tau}.$ (118) where $m_{3}$ is a nonzero constant. These provide two independent $3$-forms: $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle\tau H_{2}e^{023}+\frac{1}{m_{3}}e^{123}\;,\quad\omega_{2}=H_{2}e^{023}\;,$ for $m_{2}=0$ and $\displaystyle\omega_{1}$ $\displaystyle=$ $\displaystyle H_{2}\cosh w_{0}\tau e^{023}+\frac{w_{0}}{m_{3}}\sinh w_{0}\tau e^{123}\;,$ $\displaystyle\omega_{2}$ $\displaystyle=$ $\displaystyle H_{2}\sinh w_{0}\tau e^{023}+\frac{w_{0}}{m_{3}}\cosh w_{0}\tau e^{123}\;,$ for $m_{2}=w_{0}^{2}>0$. For nonzero values of $m_{1}$, the first equation of (115) can be easily integrated to yield $f=c_{1}T^{1/m_{1}}$, and by substituting this into the second equation, we get $\ddot{K}=-m_{2}K^{1-2m_{1}}$, where $K=T^{1/m_{1}}$. As long as this last condition and that given by (114) are satisfied, the general solutions which define only one KY 3-form are as follows: $\displaystyle\gamma=c_{1}H_{2}T^{1/m_{1}}\;,\quad\delta=\frac{c_{1}}{m_{1}}\dot{T}T^{1/m_{1}}\frac{H_{1}H_{2}}{H_{0}^{\prime}}\;.$ ## VIII Conclusion By directly starting from the KY-equation, we have developed a constructive method which makes it possible to generate all KY forms for a large class of spherically symmetric space-times in a unified and exhaustive way. Our results for the well-known spherically symmetric space-times are quantitatively summarized in Table I of the first paper and their KY two and three forms are computed in this second paper. In particular, we have found an exactly solvable nonlinear time equation for de Sitter type space-times which enables us to generate all of their KY forms in a unified manner. We have also reported solutions in some detail for sufficiently symmetric new cases which fall within the considered class of metrics. Our results can be used to reach decisive, or at least conclusive statements in analyzing the algebraic structures of KY-forms Gibbons ; Kastor1 ; Cariglia , in specifying of the symmetry algebra and related conserved quantities of the Dirac as well as other equations in spherically symmetric curved backgrounds Benn-Charlton ; Benn-Kress1 ; Benn-Kress2 ; Benn-Kress3 ; Cebeci . Finally, as an application, we indicate an approach for calculating Killing tensors and associated first integrals for the considered class of space-times Benn4 . As has been mentioned before, to each KY (p+1)-form $\omega$, there corresponds an associated Killing tensor $K$ that can be defined by $K(X,Y)=g_{p}(i_{X}\omega,i_{Y}\omega)$, where $g_{p}$ is the compatible metric in the space of $p$-forms induced by the space-time metric $g$. Then $i_{\dot{\gamma}}\omega$ is parallel-transported along the affine- parameterized geodesic $\gamma$ with tangent field $\dot{\gamma}$, and $K(\dot{\gamma},\dot{\gamma})$ is the associated quadratic first integral. The first statement follows from the fact that the covariant and interior derivatives with respect to the same geodesic tangent field commute and the second statement follows from the fact that the cyclicly permuted sum of $\nabla_{X}K(Y,Z)$ vanishes. In particular, Killing tensor fields and associated first integrals for the space-times given in the Table I of I can be computed and used in investigating some integrability problems. Our study on the symmetries of the Dirac equation and related matter, is in progress, and soon will be reported elsewhere ozumav2 . ###### Acknowledgements. This work was supported in part by the Scientific and Technical Research Council of Turkey (TÜBİTAK). ## References * (1) Ö. Açık, Ü. Ertem, M. Önder and A. Verçin, Killing-Yano Forms of a Class of Spherically Symmetric Space-Times I : A Unified Generation of Killing Vector Fields. * (2) I. M. Benn and R. W. Tucker, An Introduction to Spinors and Geometry with Applications in Physics, IOP Publishing Ltd, Bristol, 1987. * (3) U. Semmelmann, Math. Z. 243 (2003), 503. * (4) S. E. Stepanov, Theor. Math. Phys. 134 (2003), 333. * (5) D. Kastor and J. Traschen, JHEP 0408 (2004), 045 [arXive:hep-th/0406052]. * (6) D. Kastor, S. Ray and J. Traschen, Class. Quantum Grav. 24 (2007), 3759. * (7) L. Howarth and C. D. Collinson, Gen. Rel. Grav. 32 (2000), 1845. * (8) C. D. Collinson and L. Howarth, Gen. Rel. Grav. 32 (2000), 1767. * (9) G. W. Gibbons, R. H. Rietdijk and J. W. von Holten , Nucl. Phy. B 404 (1993), 42. * (10) M. Cariglia, Class. Quantum. Grav. 21 (2004), 1051. * (11) I. M. Benn and P. Charlton, Class. Quantum Grav. 14 (1997), 1037. * (12) I. M. Benn and J. Kress, Class. Quantum Grav. 21 (2004), 427. * (13) I. M. Benn and J. M. Kress, Symmetry operators for the Dirac and Hodge-deRham equations, Proceedings of the 9th DGA Conference, Prague, August 30 - September 3, 2004, pp. 421-430. * (14) I. M. Benn, P. R. Charlton and J. Kress, J. Math. Phys. 38 (1997), 4504. * (15) H. Cebeci, O. Sarioglu and B. Tekin, Phys.Rev. D. 74 (2006), 124021 [arXive:hep-th/0611011]. * (16) I. M. Benn, J. Math. Phys. 47 (2006), 022903. * (17) Ö. Açık, Ü. Ertem, M. Önder and A. Verçin, Symmetries of Dirac equation in curved space-times (in preparation).

arxiv-papers

2024-09-04T02:48:54.501352

0803.3331

# Variable range hopping in thin film with large dielectric constant B. I. Shklovskii Theoretical Physics Institute, University of Minnesota, Minneapolis, Minnesota 55455 ###### Abstract In a film with large dielectric constant $\kappa$ the electric field of an electron spreads inside the film before exiting the film at large distances of order of $\kappa d$ ($d$ is the film width). This leads to the logarithmic Coulomb repulsion between electrons and modifies the shape of the Coulomb gap in the density of localized states in a doped film. As a result the variable range hopping conductivity in such a film has a peculiar temperature dependence, where the domain of the $\ln\sigma(T)\propto(T_{0}/T)^{p}$ dependence, with the index $p\simeq 0.7$, is sandwiched between the two domains with $p=1/2$. Variable range hopping (VRH) is the generic mechanism of the low temperature transport in systems with localized electron states. When electrons repel each other via the Coulomb potential energy $V(r)=\frac{e^{2}}{\kappa r},$ (1) where $\kappa$ is the dielectric constant of the solid the density of localized states $g(E)$ has the soft Coulomb gap. As a result the VRH conductivity $\sigma$ obeys the Efros-Shklovskii (ES) law ES ; SE84 $\sigma(T)=\sigma_{0}\exp\left[-\left(\frac{T_{0}}{T}\right)^{1/2}\right]$ (2) both in two-dimensional (2D) and three-dimensional (3D) cases. Here the characteristic temperature $T_{0}=\frac{Ce^{2}}{\kappa a},$ (3) (we use the energy units for the temperature $T$), $a$ is the localization length of electrons and $C$ is the numerical coefficient SE84 . This paper deals with the situation when the Coulomb interaction between electrons has a more complicated form. We consider a thin film with the thickness $d\gg a$ and the large dielectric constant $\kappa\gg 1$. The film is surrounded by the media with much smaller dielectric constant, for example just by the air with $\kappa_{ext}=1$. The energy of the Coulomb repulsion of two electrons in the film was calculated exactly in Ref. Keldysh . Here we present only asymptotic results with their physics interpretation. Let us assume that the film is defined by the surfaces $z=\pm d/2$ and one electron is at $z=x=y=0$. Then at distances $r\ll d$ its electric field (induction) spreads isotropically. At larger distances the electric field lines are forced by the large $\kappa$ to stay inside the film, so that the field spreads along the radius $\rho=\sqrt{x^{2}+y^{2}}$ of the cylindrical coordinate system with the same $z$-axis. At $\rho\sim\kappa d$ electric field lines exit from the film and eventually spread uniformly over the whole $4\pi$ body angle again. Let us discuss the potential energy of repulsion of two electrons in the film. At the distance $\rho\gg\kappa d$ two electrons interact via the ”external” Coulomb interaction $V(\rho)=\frac{e^{2}}{\rho}.$ (4) On the other hand, in intermediate range of distances $d\ll\rho\ll\kappa d$ $V(\rho)=2\epsilon_{d}\ln\left(\frac{\kappa d}{\rho}\right),$ (5) where $\epsilon_{d}=e^{2}/\kappa d$. Finally at even smaller distances $\rho\ll d$ we arrive at the the ”internal” Coulomb interaction $V(r)=\frac{e^{2}}{\kappa r}+2\epsilon_{d}\ln\kappa,$ (6) which differs from Eq. (1) by the logarithmically large energy accumulated when the second electron is moved from infinity to $\rho=d$. The resulting potential energy $V(\rho)$ is plotted as a function of $\rho$ in Fig. 1 in all three ranges. Figure 1: A schematic plot of the potential energy of repulsion of two charges located inside the film with a large dielectric constant as a function of the distance $\rho$ between electrons. Equation numbers describing different segments of this plot are shown next to each segment. Let us assume that the film is uniformly doped by a concentration of donors $N_{D}$ and compensated by smaller concentration of acceptors so that low temperature transport is due to VRH on donors. How does the two-dimensional electrostatics affect ES law? This question was first addressed in Ref. Larkin for another object, a weakly disordered two-dimensional gas, for example, in a silicon inversion layer. In this case, according to Ref. Abrahams localized states have an exponentially large localization length. Large localized states play the dual role in this theory. First, VRH conductivity is related to long distance hopping between those of them which are close to the Fermi level. Second, all other large localized states according to Ref. Larkin contribute to the large effective dielectric constant which keeps the electric lines of a charge in the plane of the two-dimensional gas. In order to calculate VRH conductivity the authors of Ref. Larkin used an intuitive shortcut avoiding discussion of the Coulomb gap. The authors found two ranges of the temperature dependence of the VRH conductivity $\sigma(T)$. At very low temperatures where the characteristic length of the hop $r_{h}$ is much larger than $\kappa d$ they arrived at ES law with $\kappa=1$ and and at higher temperature range where $d\ll r_{h}\ll\kappa d$ they obtained activated conductivity. In this paper, we calculate the VRH conductivity in the framework of our simpler model, where large $\kappa$ is of the lattice origin or is a result of close three-dimensional metal-insulator transition. In this way we avoid the controversial subject of the 2D metal-insulator transition Kravchenko . We follow the ”orthodox” ES logic ES ; SE84 deriving first the Coulomb gap and then the conductivity. Let us first formulate our results moving from high to lower temperatures. At high temperatures $T\gg\epsilon_{d}a/d$ the VRH hopping conductivity is determined by the ”internal” ES law $\sigma(T)=\sigma_{0}\exp\left[-\left(\frac{T_{0}}{T}\right)^{1/2}-\frac{\epsilon_{d}\ln\kappa}{T}\right],$ (7) which at high enough temperatures coincides with Eq. (2). It is followed by the range of intermediate temperatures $\frac{\epsilon_{d}a}{\kappa d}\ll T\ll\frac{\epsilon_{d}a}{d}$ (8) with ”activated” VRH conductivity $\sigma(T)=\sigma_{0}\exp\left[-\frac{T_{1}(T)}{T}\right],$ (9) where the ”activation energy” $T_{1}(T)=\epsilon_{d}\ln\left(\frac{T}{\epsilon_{d}}\frac{\kappa d}{a}\right).$ (10) of the intermediate regime is weakly temperature dependent. If one approximates Eq. (9) as $\sigma(T)=\sigma_{0}\exp\left[-\left(\frac{T_{p}}{T}\right)^{p}\right],$ (11) the power $p=1-\frac{1}{\ln(T\kappa d/\epsilon_{d}a)}.$ (12) Close to $T=\epsilon_{d}a/d$ we get $p=1-1/\ln\kappa$. For example, at $\kappa=40$ one gets $p=0.7$. At even smaller temperatures $T\ll\epsilon_{d}a/\kappa d$ we arrive at the ”external” ES law: $\sigma(T)=\sigma_{0}\exp\left[-\left(\frac{T_{0}^{ext}}{T}\right)^{1/2}\right],$ (13) where $T_{0}^{ext}=Ce^{2}/a$. Thus, the ”activated” VRH conductivity Eq. (9) is sandwiched between the two different ES regimes of $\sigma(T)$, the ”internal” one, Eq. (7), at the high temperature side, and the ”external” one, Eq. (13), on the low temperature side. The two low temperature regimes Eq. (9) and Eq. (13) are in agreement with Ref. Larkin . The new high temperature ”internal” ES regime exists only if $d\gg a$. The experimental literature on the VRH conductivity in thin films is controversial (see Refs. Goldman ; Baturina and references therein.) How large is the film dielectric constant and how important is contribution of large localized states Larkin in most of cases is not clear. On the insulating side of the superconductor-insulator transition in ultrathin quench-condensed Ag, Bi, Pb and Pd films Goldman VRH data agree with Eq. (11) with $p=2/3$. On the other hand, for relatively thick TiN films Baturina the crossover from $p\simeq 1/2$ to $p\simeq 1$ is observed with the decreasing temperature. Finding an explanation for this crossover is challenging Baturina because one would normally expect that the ES law emerges at the low temperature limit. Our theory shows that in relatively thick film with the decreasing temperature one can see the crossover from the ”internal” ES law to the activated transport. This may explain results of Ref. Baturina . In order to simulate a film with a large dielectric constant one can make a two-dimensional array of isolated metallic islands overhanging each other Delsing ; Fistul . Although these arrays were originally designed as arrays of Josephson junctions they perfectly simulate a large dielectric constant in the normal state. Indeed, such array keeps electric lines in its plane if the capacitance between two islands is larger than the capacitance of each island to the ground. As a result such a normal array in presence of some disorder should show activated VRH discussed above. Before switching to the derivation of our results we would like to dwell on the related theoretical paper Fisher which deals with the VRH transport of point like vortexes responsible for the low temperature resistance of a superconductor film in the external magnetic film. Two vortexes interact via the logarithmic potential at small distances, while again at large distances their interaction follows the ”external” Coulomb potential Pearl , so that one could expect to see the two low temperature ranges discussed above, the ”activated” regime and the ”external” ES law. However, Ref. Fisher argues that ES approach is not valid in this case because ”logarithmic interaction grows without bound with particle separation” and, therefore, ”single-particle energies can not be defined” so that ”multi-vortex hopping dominates the above single-particle effects”. The multi-particle estimate Fisher leads to Eq. (11) with $2/3<p<4/5$. This is close to what we got for $p$ at $\kappa=40$. In contrary to the above statements of Ref. Fisher Fig. 1 clearly shows that the repulsion energy vanishes at infinity similarly to the standard Coulomb potential Eq. (4). Thus, there is no problem to introduce a single particle energy for a system of localized electrons interacting with the pairwise potential $V(\rho)$. We can proceed with the ES argument and study the new shape of the Coulomb gap in the density of states (DOS) of single particle excitations and eventually the VRH conductivity. We return to the discussion of the role of multi-particle processes in the end of this paper. Below we calculate the zero temperature DOS $g(\epsilon)$ following Refs. ES ; SE84 ; ES85 . In the ground state of the system we define the single electron energy $\epsilon_{i}$ of an occupied donor $i$ as the energy necessary to extract an electron from this state to the state with the energy right at the Fermi level at infinity. The single electron energy $\epsilon_{j}$ of an empty in the ground state donor $j$ is defined as the energy necessary to bring to it an electron from a state at infinity with the energy right at the Fermi level $\epsilon_{F}$. By the definition of the ground state $\epsilon_{j}>\epsilon_{F}>\epsilon_{i}$. Another stronger stability condition can be formulated for each pair of occupied and empty states as follows $\epsilon_{j}-\epsilon_{i}-V(r_{ij})>0.$ (14) Here $V(r_{ij})$ is the repulsion energy of two electrons on sites $i$ and $j$ and $-V(r_{ij})$ is the Coulomb energy of attraction of the electron which moved to the site $j$ by the hole it has left at the site $i$. In other words, this term describes the exciton effect. Eq. (14) requires that any two states close in energy to $\epsilon_{F}$ should be far enough in the space. This limits the density of states (DOS) close to the Fermi level. For the Coulomb potential the result for the DOS is known and we do not repeat derivations from ES ; SE84 ; ES85 , but list the results. The ”external” Coulomb potential Eq. (4) limits two-dimensional density of states at the level $g(\epsilon)=\frac{2}{\pi}\frac{\epsilon}{e^{4}},$ (15) For the ”internal” Coulomb potential Eq. (6) we get $g(\epsilon)=\frac{3}{\pi}\frac{\kappa^{3}\epsilon^{2}d}{e^{6}},$ (16) where the factor $d$ converts the three-dimensional DOS to the two-dimensional one. Apparently the asymptote Eq. (15) is valid at large distances, i. e. at small energies $\epsilon$ and the asymptote Eq. (15) describes small lengths or large energies. Let us now derive the DOS for the intermediate range of distances and energies. To this end we have to calculate the number of states in the band of the width $2\epsilon$ around $\epsilon_{F}$. Using Eq. (14) and Eq. (5) we get $\epsilon_{d}\ln(\kappa d/\rho)<\epsilon$. This means that in this energy band there is no more than one state in the disc with radius $\rho(\epsilon)=\kappa d\exp\left[-\frac{\epsilon}{\epsilon_{d}}\right].$ (17) This leads to the following estimate of the DOS for $\epsilon_{d}\ll\epsilon\ll\epsilon_{d}\ln\kappa$: $g(\epsilon)\sim\frac{1}{\rho(\epsilon)^{2}\epsilon}=\frac{1}{(d\kappa)^{2}\epsilon}\exp\left[\frac{2\epsilon}{\epsilon_{d}}\right].$ (18) DOS Eq. (18) matches DOS Eq. (15) at $\epsilon=\epsilon_{d}$ and Eq. (16) at $\epsilon=\epsilon_{d}\ln\kappa$ (see Fig. 2)). At large energies the parabolic range of the Coulomb gap, Eq. (16), is limited by the total width of the impurity band $\epsilon_{max}=e^{2}N_{D}^{1/3}/\kappa+\epsilon_{d}\ln\kappa$. At $\epsilon\gg\epsilon_{max}$ the DOS decreases (see the dashed line in Fig. 2) similarly to the three-dimensional DOS of the classical impurity band (see Ch.14 of Ref. SE84 ). Figure 2: Schematic plot of the DOS of localized electrons limited by Coulomb interaction of electrons as a function of energy. Equation numbers describing different segments of this plot are shown next to each segment. Dashed line shows beginning of the DOS decline at large energies Now we can apply the calculated above DOS in order to estimate the exponential term of the VRH conductivity of the film at different temperatures. This estimate closely follows the original Mott’s approach Mott . We define an energy band around the Fermi level and estimate contribution of this band to the VRH conductivity. Then we can optimize result with respect of the band width. This calculations are quite straightforward and we will not go through them here. Results are already formulated in the beginning of our paper. Let us return to the multi-electron effects on ES law. For the standard Coulomb potential Eq. (1) they were studied in Refs. Efros ; Baran ; SE84 ; ES85 . It was shown that they may change only the coefficient $C$ in Eq. (2). Here we only briefly remind what was done. In 3D case a small energy single- electron excitation strongly interacts with dipole moments of surrounding compact electron-hole excitations forming together with them a composite charged multi-electron excitation, the electronic polaron. Polarons being charged particles obey stability criterion Eq. (14), have the Coulomb gap and lead to ES law. The only difference is that coefficient $C$ can be somewhat larger because every hop of a single-electron excitation in 3D is accompanied by the tunnelling depolarization of many polaron pairs. A simple estimate showed that the total length of these small hops is of the order of the length of the main long ES VRH hop. This gives no more than the factor 2 in the expression for $C$. In 2D number of dipole excitations in the polaron atmosphere is of order one and polaron effects provide only small corrections to $C$. We estimated similar effects in the framework of the complex potential of this paper (Eqs. (4), (5), (6)). It turns out that in the low temperature range polaron effects provide only small corrections to $C$ in Eq. (13). On the other hand, for the two higher temperature ranges (Eqs. (9) and (2) the situation is similar to the 3D case studied in Refs. Baran ; SE84 ; ES85 . Namely, at these temperatures multi-electron effects may add a numerical coefficient in the exponential. For Eq. (9) this means that $T_{1}$ may become twice larger. Thus, multi-electron effects do not change the power $p$ in Eq. (11). This conclusion is in disagreement with Ref. Fisher which assumes that even in 2D the total length of small hops is much larger than the ES hop and, therefore, overestimates importance of the multi-electron effects. Above we discussed only the ohmic transport in a weak electric field. If the electric field is so strong that $eEa/2\gg T$ one can replace $T$ by the effective VRH temperature $eEa/2$ in Eqs. (2), (9), (11) and (13) to obtain non-ohmic current-voltage characteristics Shklovskii (see also Larkin ; Marianer ). For the intermediate activated regime we then arrive at the current-voltage characteristics $J\propto J_{0}\exp(-E_{0}/E)^{p}$, where $p$ is close to unity. This result is in agreement with the earlier theory Nelson of the VRH transport of pinned vortexes in superconductors under the influence of a strong current. Until now we dealt with a film. Now let us briefly discuss the similar physics in a long cylindrical nano-wire or nano-rod with radius $d$ made from a semiconductor with a large dielectric constant $\kappa\gg 1$, for simplicity, in the air environment. Here again the electric field of an electron at distances $r\ll d$ spreads isotropically, then stays inside nano-rod for a distance $x<\xi=d\kappa^{1/2}$ along the cylinder axis, and then leaks from the cylinder and eventually spreads isotropically in the air at large enough $r$. Thus, the potential energy of repulsion of two electrons again changes from the very short range ”internal” Coulomb interaction Eq. (4) to the very long range ”external” Coulomb interaction Eq. (1) with the large intermediate range of $x$, where interaction has the one-dimensional character. The interaction in this range was studied Kamenev ; Zhang for an ion channel in a lipid membrane, where the cylindrical pore with radius $d$ is filled by water with $\kappa=81$ and is surrounded by lipids with $\kappa=2$. Translated to our problem the potential energy of two electrons located at the nano-rod axis at the intermediate distance $x$ from each other is well approximated by $V(x)=eE_{0}\xi[\exp(-x/\xi)-1],$ (19) where $E_{0}=2e/\kappa d^{2}$. (see Sec. VIII of Ref. Zhang ). Using this potential for repulsion of two electrons in the nano-rod and following the Coulomb gap based derivation similar to one used above for a film (or the shortcut approach of Ref. Larkin ) one can calculate the temperature dependence of the VRH conductivity of the nano-rod in the intermediate temperature range. The result is the strict activation regime with the temperature independent activation energy $T_{a}=eE_{0}\xi$. It is, of course, sandwiched between the two ES laws, the ”internal” one on the high temperature side and the ”external” one Eq. (13) on the low temperature side. I am grateful to T. Baturina, A. M. Goldman, A. Kamenev, D. E. Khmelnitskii and A. I. Larkin for useful discussions. ## References * (1) A. L. Efros and B. I. Shklovskii, J. Phys. C 8, L49 (1975). * (2) B. I. Shklovskii and A. L. Efros, Electronic Properties of Doped Semiconductors, (Springer, New York, 1984). This book is available from http://www.tpi.umn.edu/shklovskii/ * (3) L. V. Keldysh, Sov. Phys. JETP Lett. 30, 245 (1979). * (4) A. I. Larkin, D. E. Khmelnitskii, Sov. Phys. JETP 56, 647 (1982). * (5) E. Abrahams, P. W. Anderson, D. C. Liccardello, T. V. Ramakrishnan, Phys. Rev. Lett. 42, 673 (1979). * (6) E. Abrahams, S. V. Kravchenko, M. P. Sarachik, Rev. Mod. Phys. 73, 251 (2001) * (7) N. Markovic, C. Christiansen, D. E. Grupp, A. M. Mack, G. Martinez-Arizala, A. M. Goldman, Phys. Rev. B 62, 2195 (2000). * (8) T. I. Baturina, A. Yu. Mironov, V. M. Vinokur, M. R. Baklanov, C. Strunk, Phys. Rev. Lett. 99, 257003 (2007). * (9) E. Chow, P. Delsing, D. B. Haviland, Phys. Rev. Lett. 81, 204 (1998). * (10) M. V. Fistul, V. M. Vinokur, T. I. Baturina, Phys. Rev. Lett. 100, 086805 (2008). * (11) M. P. A. Fisher, T. A. Tokuasu, A. P. Young, Phys. Rev. Lett. 66, 2931 (1991). * (12) J. Pearl, Appl. Phys. Lett. 5, 65 (1964), P. G. De Gennes, Superconductivity of Metalls and alloys. (W. A. Benjamin, New York - Amterdam, 1966). * (13) A. L. Efros, B. I. Shklovskii, in Electron-electron interaction in disordered systems , ed. by A.L. Efros, M. Pollak. (North-Holland, Amsterdam, 1985). * (14) N. F. Mott, J. Non-Cryst. Solids 1, 1 (1968). * (15) A. L. Efros, J. Phys. C 9, 2021 (1976) * (16) S. D. Baranovskii, B. I. Shklovskii, A. L. Efros, Sov. Phys. JETP 51, 199 (1980). * (17) B. I. Shklovskii, Sov. Phys.-Semicond. 6, 1964 (1973). * (18) S. Marianer, B. I. Shkiovskii, Phys. Rev. B 46, 13100 (1992). * (19) D. R. Nelson, V. M. Vinokur, Phys. Rev. B 48, 13060 (1993). * (20) A. Kamenev, J. Zhang, A. I. Larkin, B. I. Shklovskii, Physica A 359, 129 (2006). * (21) J. Zhang, A. Kamenev, B. I. Shklovskii, Phys. Rev. E 73, 051205 (2006).

arxiv-papers

2024-09-04T02:48:54.508552

0803.3460

# Density-density functionals and effective potentials in many-body electronic structure calculations F. A. Reboredo Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831, USA P. R. C. Kent Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37831, USA ###### Abstract We demonstrate the existence of different density-density functionals designed to retain selected properties of the many-body ground state in a non- interacting solution starting from the standard density functional theory ground state. We focus on diffusion quantum Monte Carlo applications that require trial wave functions with optimal Fermion nodes. The theory is extensible and can be used to understand current practices in several electronic structure methods within a generalized density functional framework. The theory justifies and stimulates the search of optimal empirical density functionals and effective potentials for accurate calculations of the properties of real materials, but also cautions on the limits of their applicability. The concepts are tested and validated with a near-analytic model. ## I Introduction Density Functional Theory (DFT) hohenberg ; kohn is based on the Hohenberg- Kohn proof of a functional correspondence between the ground state energy and the ground state density $E[\rho(r)]$. In the formulation of Kohn and Sham (K-S),kohn the interacting electron gas is replaced by non-interacting electrons moving in an effective potential. In this construction, the non- interacting density $\bar{\rho}(r)$ is equal to the interacting one $\rho(r)$ but no other property of the interacting ground state is in principle retained in the non-interacting wave function. Initially DFT was formulated to describe the total ground state energy of an interacting system and $\rho(r)$.hohenberg ; kohn ; practice Although progress towards more accurate density functionals is ongoing, current approximations such as the local density approximation (LDA)kohn ; perdew81 and more recent gradient-based extensionsRMMartinBook are already successful in predicting many electronic properties of real materials. This success has led to the use of DFT beyond its formal scope and unfortunately tempted some to believe that if we had the exact ground state density functional, we would only need to solve non-interacting problems even for properties not related to the ground state energy and density. While the virtues and limitations of the Kohn-Sham eigenvalues are discussed in textbooks,RMMartinBook ; parr the possible reasons for the success or failure of Kohn-Sham wave-functions in many-body problems are little understood and widely dispersed throughout the literature. It is often assumed, without a formal proof, that the Kohn-Sham non- interacting ground state wave-function forms a good description of the interacting ground state wave-function to be used as the foundation of theories that go beyond DFT such as GW-Bethe Salpeter Equation (GW-BSE), Quantum Monte Carlo (QMC), or even configuration interaction (CI). This leads to an apparent contradiction in the literature since density functionals that provide wave-functions that are a good starting points in one field (as judged by comparison with experiment) are found inadequate in others. Broadly summarizing: for structural properties gradient corrected density functionalsPBE are nowadays preferred over LDA.kohn ; perdew81 Structural properties depend essentially of atomic forces which in turn are related to the density. However, for GW-BSE calculations of optical properties, an LDA- based ground state is preferredaulbur00 . In this approach a good initial approximation for the Green function is required. In QMC calculations a (non- interacting) Hartree-Fock (HF) ground state might be preferred over LDA, but the subject is still under debate. In CI calculations, instead, it is empirically claimed that the convergence with natural orbitalslowdin55 is more rapid than HF orbitals. In Diffusion Quantum Monte Carlo (DMC) a trial wave function enforces the antisymmetry of the electronic many-body wave functionanderson79 and the nodal structure of the solution. The accuracy of the trial wave-function is critical and determines the success or failure of the method to accurately predict properties of real materials. The trial wave-function is usually a product of a Slater determinant $\Phi_{T}({\bf R})$ and a Jastrow factor $e^{J}({\bf R})$. $\Phi_{T}({\bf R})$ is often constructed with single particle Kohn-Sham orbitals or from other mean field approaches such as HF. The Jastrow, in turn, is a symmetric factor which does not change the nodes, but accelerates convergence and improves the algorithm’s numerical stability. The DMC algorithm finds the lowest energy of the set of all wave-functions that share the nodes of $\Psi_{T}({\bf R})$. The exact ground state energy is obtained only if the exact nodes are provided. Since any change to an antisymmetric wave-function must result in a higher energy than the antisymmetric ground state, the energy obtained with arbitrary nodes is an upper bound to the exact ground state energy.anderson79 Only in small systems is it possible to improve the nodes bajdich05 ; filippi00 ; umrigar07 ; rios06 ; luchow07 or even avoid the trial wave-function approach altogetherkalos00 ; zhang91 . Consequently, a general formalism that could alleviate the nodal error in large systemsalfe04 ; reboredo05 is highly desired. Quite recently it has been shown that within the single Slater determinant approach the computational cost of the DMC algorithm can have an almost linear scaling with the number of electronswilliamson ; alfe04 ; reboredo05 ; kolorenc . It is claimed, if not formally proved, that the nodes of the many-body wave-function are not too far from those of a wave function obtained via a mean field approach. However, this might not continue to hold as electron-electron interactions become more important. To improve the accuracy of these approaches and increase the range of materials to which they can be applied it is important to examine the advantages of different mean-field wave-functions. In this paper, we demonstrate that density-density functionals can be obtained by finding the minimum of different cost functions relating the set of non- interacting $v$-representable ground state with an interacting many-body state. The minimum of these cost functions establishes a correspondence between the non-interacting and the interacting wave-functions and their associated densities and potentials. The cost function can be designed to retain selected properties of the many-body wave function in the non- interacting one. Crucially, for DMC applications the nodes can be optimized. Under certain conditions density-density functionals exist that can lead to standard scalar-density functionals. As in the case of standard DFT, this proof does not mean that we know the expression of each functional or associated potential but it will certainly stimulate the search of methods to find or approximate them. For DMC applications it is enough to prove that an optimal mean field potential for nodes exists. In order to test the theory, we find the ground state wave function of a model interacting system. Then we obtain (i) the exact DFT effective exchange correlation potential associated with the ground state density, (ii) the potential that maximizes the projection of $\Phi_{T}({\bf R})$ with the ground state. Finally, (iii) we optimize a potential to match the nodes and find that surprisingly, for this model, the non-interacting solution in the same potential as the interacting problem is a very good approximation for the nodes while the exact non- interacting Kohn-Sham ground state is particularly poor. This paper is organized as follows: In Section II we demonstrate the existence of an different density-density correspondences associated with cost functions. We prove the existence of this functional correspondence for the case of optimal nodes required in DMC. In Section III we solve an interacting problem up to numerical precision and find its many-body ground state wave- function. Subsequently we optimize different cost functions to retain specific properties of the ground state. Finally in Section IV we discuss the relevance of our results for many-body electronic structure and give our conclusions. ## II Generalized density-density functionals Given an interaction in a many-body system, the Hohenberg-Kohn theoremhohenberg establishes a functional correspondence between densities $\rho({\bf r})$, external potentials $V(r)[\rho({\bf r})]$ and ground state wave-functions $\Psi({\bf R})[\rho({\bf r})]$; where $[\rho({\bf r})]$ denotes a functional dependence on the ground state density, and ${\bf R}$ denotes a point in the many-body $3N$ space. Since the density changes according to the strength and functional form of the interaction, this correspondence is different for different interactions. For a fixed interaction, the subset of densities $\rho({\bf r})$ corresponding to a ground state of an interacting system under an external potential $V(r)$ are denoted as pure state $v$-representable.parr A non-interacting pure state $v$-representable density is given instead by $\bar{\rho}({\bf r})=\sum_{\nu}|\phi_{\nu}\left({\bf r}\right)|^{2}$ where $\phi_{\nu}\left({\bf r}\right)$ are the Kohn-Sham-like single particle orbitals, or eigenvectors, of the Hamiltonian: $\left[-\frac{1}{2}{\bf\nabla}^{2}+\bar{V}\left({\bf r}\right)\right]\phi_{\nu}\left({\bf r}\right)=\varepsilon_{\nu}\phi_{\nu}\left({\bf r}\right),$ (1) where $\bar{V}\left({\bf r}\right)$ is an effective single particle potential. For simplicity we denote here a density to be $v$-representable if it is both pure state non-interacting and pure state v-representable. In the following we also imply pure state when we write only $v$-representable. Each point in the sets of $v$-representable densities is associated with two different points in the wave-functions Hilbert space. In figure 1 we schematize the subset of $v$-representable densities and the functional correspondence with the subsets of the interacting and non-interacting ground state wave-functions. Note that, in principle, the two subsets of ground state wave-functions do not necessarily overlap. In the non interacting case the wave-function is given by a Slater determinant of Kohn-Sham-like orbitals but for interacting problems this simplification is not longer possible. The Kohn-Sham scheme for density functional theory establishes a correspondence between interacting and non-interacting wave-functions represented as line (1) in Fig 1. This Kohn-Sham correspondence between wave- functions is implicit in the Khon-Sham construction for the external effective potential.kohn Figure 1 emphasizes that the wave-functions joined by line (1), while different, give the same electronic density. In more technical terms they both belong to the same Percus-Levy partition of the Hilbert spacepercus ; levy but they are the minimum energy wave-function for different interactions. The exchange-correlation potential is by construction the difference that one has to add to the external potential in a non- interacting problem so that its ground state density is the same as the interacting one. If the energy-density functional $E[\rho({\bf r})]$ is known, the effective non-interacting potential can be obtained following the standard Kohn-Sham approach. If the ground state density $\rho({\bf r})$ is known, the same correspondence between interacting and non-interacting densities can be achieved by minimizing the following function $K_{\rho}=\frac{1}{2}\int{\bf dr}\left[\bar{\rho}({\bf r})-\rho({\bf r})\right]^{2}.$ (2) within the subset of non-interacting $v$-representable densities. Formally, this could be done by exploring all values $\bar{V}\left({\bf r}\right)$ in Eq (1). In practice, if the density of the interacting ground state is known, the potential $\bar{V}_{K_{\rho}}({\bf r})$ that minimizes Eq. (2) can be obtained numerically with a procedure similar in spirit to the optimized effective potential (OEP) method. The change in the density required to minimize Eq. (2) is $\Delta_{\rho}=-[\bar{\rho}({\bf r})-\rho({\bf r})].$ (3) Within linear response, the change in the potential required to produce $\Delta_{\rho}$ is $\Delta\bar{V}_{K_{\rho}}({\bf r})=\int{\bf dr^{\prime}}\left[\rho({\bf r^{\prime}})-\bar{\rho}({\bf r^{\prime}})\right]\frac{\delta V\left({\bf r^{\prime}}\right)}{\delta\rho\left({\bf r}\right)}$ (4) Adding recursively $\Delta\bar{V}_{K_{\rho}}({\bf r})$ we can find the potential $\bar{V}_{K_{\rho}}({\bf r})$ associated with $K_{\rho}=0$ (see an example below). Figure 1: (Color online) a) Representation of the sets of pure state $v$-representable interacting densities. b) Sets of interacting and non- interacting ground state wave-functions. The Kohn-Sham formulation of DFT relates a $v-representable$ density with a pair interacting and a non- interacting wave-function. The same functional correspondence can be obtained minimizing Eq. (2) (line 1 in the figure). Different cost functions relate an interacting $v$-representable density with a different non-interacting-$v$ representable density (see lines 2 and 3). ### II.1 Other density-density correspondences It is often desirable o preserve properties in addition to the density of the many-body ground state $\Psi({\bf R})$ in a non-interacting wave-function $\Phi_{T}({\bf R})$ to be used as a starting point for theories that go beyond DFT. This task involves exploring all the non-interacting $v$-representable set in order to find a wave-function that best describes a given property. This is a typical optimization problem. One of the most common strategies in optimization is the design of a cost function. One example is Eq (2), a measure of the difference in two densities. Another example of a cost function is $K_{Det}=-\left|\left<\Psi\right.\left|\Phi_{T}\right>\right|^{2}$ (5) which involves a projection of the interacting ground state $\Psi$ in the set of non-interacting $v$-representable wave-functions $\\{\Phi_{T}\\}$. The minimum of Eq. (5) is the non-interacting ground state Slater determinant with maximum projection in the interacting ground state. We have claimed above that the interacting wave-functions might be in general very different from a single non-interacting Slater determinant. Accordingly, we expect $K_{Det}>-1$. We expect to find a different minimum in the non-interacting ground state set, if we change the functional form of the cost function from Eq. (2) to Eq. (5) for the following reasons: 1) We can visualize the cost function as a scalar potential defined in the full Hilbert-space. Although different cost functions can share the same minimum in the complete Hilbert space, in the restricted subset of non- interacting ground state wave-functions, different cost functions can have a different minimum: the optimal point found depends on the functional form of the cost function. Accordingly, while all the cost functions we propose here would be minimized if we could reach the interacting many body state $\Psi$ ( where, of course, every property is retained exactly), because our search is constrained to non-interacting $v$-representable subset the minimum we would find will depend on the properties we wish to retain. 2) The Hohenberg Kohn theorem, when applied to the non-interacting $v$-representable case implies that, in the absence of degeneracy, there is at most one wave-function that has the same density as the interacting case. Therefore, once the minimum of an arbitrary cost function is found, its associated non-interacting density can no longer be equal to the interacting density unless the property enforced by the cost function can be related back to density. Enforcing the non-interacting density to remain equal to the interacting ground state density prevents all other properties of the non- interacting wave-function from being further improved. If we intend to optimize other properties, we have to relax the density constraint finding a different wave-function associated with a different density. The minimization of different cost-functions, relating the interacting ground state $\Psi({\bf R})$ with the non-interacting $v$-representable set, provide in-principle different correspondences between interacting and non-interacting wave-functions represented as different lines in figure 1. Each cost function $K$ defines a correspondence different than the identity between pure state $v$-representable densities and non-interacting pure-state $v-$representable densities. As a consequence, the idealized optimization processes outlined here defines an operator $U_{K}$ that turns each $\rho({\bf r})$ into a non- interacting density corresponding to the wave-functions $\Phi_{T}({\bf R})$ which is the minimum of a cost function $K$. $\bar{\rho}_{K}({\bf r})=U_{K}\left[\rho({\bf r})\right].$ (6) Note that if the minimum of a given cost function $K$ is a single $\Phi_{T}({\bf R})$ for every $v$-representable density, then $U_{K}$ defines a density-density functional. When more than one non-interacting $v$-representable wave-function give the same optimal value for $K$, the degeneracy can be broken by additional requirements in the cost function [e.g. also minimizing Eq. (2), the difference between the current and pure state densities ]. Since we only need one optimal wave-function, any from a degenerate minimum can be chosen to construct the density-density functional $U_{K}$. When minimization of a cost function defines a one to one correspondence with an inverse, a more usual energy-density functional of the form $E\\{U_{K}^{-1}[\bar{\rho}_{K}({\bf r})]\\}$ can be constructed. Only a restricted class of cost functions lead to density transformations with an inverse. Minimization of the cost functions among all pure-state-non- interacting $v$-representable densities defines the optimal effective potential which is a function of this density. Given a cost function, $K$, finding an approximation for the density transformation operator $U_{K}$ could certainly be as demanding as finding an approximation for the energy-density functional $E\left[\rho({\bf r})\right]$ required by DFT based methods. This task is beyond the goal of this paper. However, we will show that we can expect the operator associated to the best nodes for DMC ($U_{DMC}$) to be non-local and very different from the identity. Accordingly we can expect non-interacting wave-functions with good nodes to be a poor source of densities. Moreover, for the example considered below, we find, that the direction we might have to explore to optimize the potential might be surprisingly different than the attempts considered so farwagner03 ; kolorenc . ### II.2 The Diffusion Monte Carlo case We next show that optimization of the nodes for DMC among the set of $v-$representable wave-functions leads to a correspondence between pure state $v$-representable densities and pure state non-interacting $v$-representable densities of the class described above. These in turn demonstrate the existence of an optimal effective non-interacting nodal potential. Since, the ground state density $\rho({\bf r})$ determines the ground state wave-function $\Psi({\bf R})[\rho({\bf r})]$,hohenberg $\rho({\bf r})$ defines also the points ${\bf R}$ of the nodal surface $S_{0}({\bf R})[\rho({\bf r})]$ where $\Psi({\bf R})[\rho({\bf r})]=0$. We can also classify the nodal surfaces in pure state v-representable and pure-state-non- interacting $v$-representable. The DMC algorithm in the fixed node approximation finds the lowest energy of the set of all wave-functions that share the nodes or the trial wave-function. For Slater determinant Jastrow wave-functions, the nodes of the trial wave- function are by construction those of $\Phi_{T}({\bf R})$; that is they are pure-state non-interacting $v$-representable. The DMC energy, $E_{DMC}$ is also a function of the external potential which in turn is a function of the interacting ground state density $V(r)[\rho({\bf r})]$. Thus minimization of $E_{DMC}[\Phi_{T}({\bf R}),\rho({\bf r})]$ in the set of non-interacting $v$-representable wave-functions $\Phi_{T}({\bf R})$ determines one $\Phi_{T}({\bf R})$ with the best nodes. Every optimal $\Phi_{T}({\bf R})$ defines an optimal auxiliary density $\bar{\rho}_{DMC}({\bf r})$. As a consequence optimizing the nodes of the trial wave-function by perturbing the nodes of pure state non-interacting wave-functions implies finding another correspondence between interacting and non-interacting densities (another line in figure 1). The best cost function for optimal nodes is ultimately the DMC energy itself. Since we restrict the search to pure-state non-interacting $v$-representable nodes, the minimum energy $E_{DMC}[\rho({\bf r}]$ will be larger than the true ground state energy $E[\rho({\bf r}]$, because of the upper bound theorem, unless $S_{0}({\bf R})$ is non-interacting $v$-representable. Note that for an arbitrary interaction $S_{0}({\bf R})$ is not expected to be, in general, pure-state-not-interacting v-representable. However, if $S_{0}({\bf R})$ were non-interacting v-representable, the best Slater Determinant $\Phi_{T}({\bf R})$ for DMC could be formally found by finding the minimum of the cost function $K_{S_{0}}=\int_{S_{0}}{\bf dS}\left|\Phi_{T}({\bf R})\right|^{2}.$ (7) where $\int_{S_{0}}$ denotes a surface integral over the interacting nodal surface. ## III Cost function minimization To demonstrate the theoretical concepts above we solve a simple non-trivial interacting model as a function of the interacting potential strength and shape. We then optimize the wave-functions to minimize the cost functions in Eqs. (2), (5) and (7) so as to find the exact DFT wave-function, the wave- functions that maximize the projection on the interacting ground state and minimize the projection on the nodes. Subsequently, we estimate the volume of the Hilbert space enclosed between the nodes of the interacting wave-functions and the optimized non-interacting ones. ### III.1 A model interacting ground state For illustrative purposes we choose the interacting problem to be as simple as possible and yet not trivial. We solve the ground state of two spin-less electrons moving in a two dimensional square of side length 1 with a repulsive interaction potential of the form $V({\bf r},{\bf r^{\prime}})=8\gamma\cos{[\alpha\pi(x-x^{\prime})]}\cos{[\alpha\pi(y-y^{\prime})]}$. units While this potential is different than the Coulomb interaction, it shares some of its properties. For positive $\gamma$ and $|\alpha|<1$ the interaction is repulsive with a repulsion that increases monotonically when for shorter distances. The amplitude of the repulsion as compared to the kinetic energy can be changed by adjusting $\gamma$. Since the Coulomb interaction is self similar, changing $\gamma$ mimics what happens in a real system when we change the size of the system. The shape of the potential within the confined region can be altered by changing $\alpha$. In the limit of $\alpha\rightarrow 0$ the interaction potential is separable which allows several limits to be tested (such as the nodes). The functional form facilitates an analytical treatment of the problem by removing the singularity of the Coulomb interaction at short distances. We expanded the many-body wave-function in a full CI on non-interacting Slater determinants with the same symmetry as the ground state. The ground state is degenerate because there are only two electrons. We chose one of the ground state wave-functions according to the $D_{2}$ subgroup of the $D_{4}$ symmetry of the Hamiltonian. With this choice, $\rho({\bf r})$ has $D_{2}$ symmetry ($x$ is not equivalent to $y$). The basis of Slater determinants was constructed with functions of the form $16\mathrm{sin}(n\pi x)\mathrm{sin}(m\pi y)$ with $n$ and $m<8$. Since parity is preserved by the interaction and Slater determinants of identical functions are zero, the size of the basis set is reduced to only 300 in our calculations. Most of the calculations reported here were done analytically with the help of the Mathematica package, including all of the electron-electron interaction integrals.notebook The only source of errors are numerical truncation and the size of the basis, which was tested for convergence. In Figure 2 we show the quadrants of densities corresponding to wave-functions that are even for reflections in the $y$ direction and odd in the $x$ direction for $\gamma=2$ and $\alpha=1$. Figure 2(a) shows the upper left quadrant of the density of the interacting ground state of two spin-less electrons obtained with full CI. Figure 2(b) shows the upper right quadrant of the non-interacting density corresponding to the effective potential obtained by adding recursively $\Delta V_{K_{\rho}}({\bf r})$ [Eq. (4)], which is exactly the reflection of $\rho({\bf r^{\prime}})$ Fig. 2(a) up to numerical precision [$K_{\rho}=0$ in Eq. (2)]. Because of the Hohenberg-Kohn theorem,hohenberg $V_{K_{\rho}}$ and the Kohn-Sham potential $V_{KS}(\rho)$ can only differ by a constant and thus the wave-functions coming from this potential are the exact DFT wave-functions for our interaction. The properties of the wave-functions will be discussed later in the text. The densities in Figs. 2(d) and 2(d) correspond to the minimum of the cost functions given in Eqs (5) and (7) obtained as described below. Figure 2: Ground state densities, in particles per unit area, for two interacting spin-less electrons in a square box. The complete density can be obtained by reflections (see also Fig 3). (a) Full CI ground state interacting density. (b) Exact DFT solution obtained minimizing Eq. (2). (c) Slater determinant with maximum projection with CI the ground state [see Eq. (5)]. (d) Slater determinant with minimum amplitude on the nodes of the CI ground state [see Eq. (7)]. ### III.2 Effective potential optimization We now consider more difficult cost functions than density differences, Eq. (2). For non-interacting $v$-representable densities there are also functional correspondences between ground state wave-functions, potentials and densities. This concept has been exploited in the optimized effective potential (OEP) for exact exchange. xxtheory ; xxsemiconductors ; xx2deg ; us The exchange potential can be calculated in OEP xxtheory ; xxsemiconductors ; xx2deg ; us as: $\displaystyle V_{x}({\bf r})=\frac{\delta E_{x}}{\delta\rho\left({\bf r}\right)}$ (8) $\displaystyle=\sum_{\nu}^{occ}\int\\!\\!\\!\\!\int\\!\\!{\bf dr^{\prime}}{\bf dr^{\prime\prime}}\left[\frac{\delta E_{x}}{\delta\phi_{\nu}\left({\bf r^{\prime\prime}}\right)}\frac{\delta\phi_{\nu}\left({\bf r^{\prime\prime}}\right)}{\delta V_{KS}\left({\bf r^{\prime}}\right)}+c.c.\right]\\!\\!\frac{\delta V_{KS}\left({\bf r^{\prime}}\right)}{\delta\rho\left({\bf r}\right)}.$ In Eq. (8), the functional derivative $\delta E_{x}/\delta\phi_{\nu}\left({\bf r}\right)$ is evaluated directly from the explicit expression for the exchange energy $E_{x}$ in terms of $\phi_{\nu}\left({\bf r}\right)$. Next $\delta\phi_{\nu}\left({\bf r}\right)/\delta V_{KS}\left({\bf r^{\prime}}\right)$ is evaluated using first-order perturbation theory from Eq. (1). Finally ${\delta V_{KS}\left({\bf r^{\prime}}\right)}/{\delta\rho\left({\bf r}\right)}$ is the inverse of the linear susceptibility operator. If there are fixed boundary conditions such as the number of particles, the susceptibility operator is singular xxsemiconductors ; us . Excluding these null spaces it can be inverted numerically.xxsemiconductors We use earlier this susceptibility operator in Eq. (4). Equation (8) is by construction the gradient of the exchange energy in the set of pure-state-non-interacting $v$-representable densities. While we are not going to attempt an exact exchange approach in this paper, the ability to calculate gradients allow us to minimize cost functions as long as the cost function $K$ can be expressed in terms of non-interacting ground state wave wave-functions or eigenvalues. The potential that minimizes $K$ can be obtained by recursively applying the formula $\delta V_{K}({\bf r})=\epsilon\sum_{\nu}^{occ}\int\\!\\!{\bf dr^{\prime}}\\!\\!\frac{\delta K}{\delta\phi_{\nu}\left({\bf r^{\prime}}\right)}\frac{\delta\phi_{\nu}\left({\bf r^{\prime}}\right)}{\delta V_{KS}\left({\bf r}\right)}+c.c.$ (9) Equation (9) gives the direction we need to change the potential to minimize the cost function. The magnitude of the change is controlled by $\epsilon$, which can be adjusted as one reaches the minimum. Replacing $K$ by $K_{Det}$ in Eq. (9) and using Eq (5) and first order perturbation theory we find $\displaystyle\delta V_{K_{Det}}({\bf r})$ $\displaystyle=$ $\displaystyle\epsilon\left<\Psi\right.\left|\Phi_{T}\right>\sum_{\nu}^{o}\sum_{n}^{u}\left<\Psi\right|c^{{\dagger}}_{n}c_{\nu}\left|\Phi_{T}\right>\frac{\phi_{n}({\bf r})\phi_{nu}({\bf r})}{\varepsilon_{\nu}-\varepsilon_{n}}$ (10) $\displaystyle+$ $\displaystyle c.c.$ In equation (10) $\sum_{n}^{o}$ ( $\sum_{n}^{u}$ ) means sum over occupied (unoccupied) states, while $c^{{\dagger}}_{n}$ and $c_{\nu}$ are creation and destruction operators on the non-interacting ground state $\left|\Phi_{T}\right>$. One can understand also the state $c^{{\dagger}}_{n}c_{\nu}\left|\Phi_{T}\right>$ as the many body wave-function $\Phi_{T}^{n,\nu}(R)$ resulting from replacing the occupied state $\phi_{\nu}$ by the $\phi_{n}$. This is equivalent to creating an electron hole pair excitation in a non-interacting ground state. In Eq. (10) a term in the potential is added every time an electron hole pair excitation has no zero projection to the interacting ground state. Since the basis of products of wave-functions $\phi_{n}({\bf r})\phi_{\nu}({\bf r})$ is over-complete, there are linear combinations with non-zero coefficients that add up to zero. A minimum is found when the gradient of the cost function with respect to variations of the effective potential is zero. If the absolute minimum is found, the wave function can only be improved further by a multi-determinant expansion, that is, outside the set of pure-state non-interacting densities. Since we choose a basis expansion for the single particle orbitals to be sine functions, the products $\phi_{n}({\bf r})\phi_{nu}({\bf r})$ are linear combinations of sine products. These sine products can be transformed analytically to cosines. The change in the potential is thus written in a cosine basis which is complete. All coefficients must vanish in the cosine basis when a minimum is found. This allows us to verify that the gradient in the potential can be minimized up to numerical precision. The integrated effective potential is thus naturally expressed as a linear combinations of cosines, which allows the analytical calculation of the coefficients of the effective potential matrix in a basis of sines, where the kinetic energy is diagonal. The Slater determinant $|\Phi_{T}>$ is written in the same basis as the interacting ground state $|\Phi>$. The projections involved in Eq. (10) are then reduced to a scalar product of the vectors of coefficients. In figure 2(c) we show the ground state density associated to minimization of Eq. (5) for the same parameters as the interacting ground state density in Fig 2(a). We see that while optimizing the cost function (2) allows matching the interacting density exactly, optimizing the wave-function projection requires a significant change in the resulting density. Similarly, replacing $K$ by $K_{S_{0}}$ in Eq. (9) and using Eq (7) we get $\displaystyle\delta V_{K_{S_{0}}}({\bf r})$ $\displaystyle=$ $\displaystyle\epsilon\sum_{\nu}^{o}\sum_{n}^{u}\int_{S_{0}}\;\;{\bf dS}\Phi_{T}^{n,\nu}({\bf R})\Phi_{T}({\bf R})\frac{\phi_{n}({\bf r})\phi_{nu}({\bf r})}{\varepsilon_{\nu}-\varepsilon_{n}}$ (11) $\displaystyle+$ $\displaystyle c.c.$ Unlike Eqs.4 and 10, a complication appears when evaluating the integral over the nodal surface $\int_{S_{0}}{\bf dS}$. This integral involves finding the points where the many-body wave-function is zero. The problem is simplified because the derivatives of the many body wave-functions can be obtained analytically. Consequently, starting from an arbitrary point ${\bf R}$ we can find a zero recursively with the Newton-Raphson method, ${\bf R}_{n+1}={\bf R}_{n}+\Psi({\bf R}){\bf\nabla}\Psi({\bf R})/\left|{\bf\nabla}\Psi({\bf R})\right|^{2}$. Next we make a random displacement ${\bf\Delta R}$ in the hyper-plane perpendicular to ${\bf\nabla}\Psi({\bf R})$ within a circle of radius 0.05 and find a node again. We repeat this process $50$ times and select an element of $\\{{\bf R}\\}_{S}$. With this parameters, the random position ${\bf R}_{n}$ can travel across the full size of the system so that the distribution is homogeneous. By repeating this process $N=500$ times, excluding points at the boundaries which are zero by construction, we generate an homogeneous distribution of points at the nodal surface $\\{{\bf R}\\}_{S}$. We approximate the integral in Eq. (7) as a sum on the values on the set $\\{{\bf R}\\}_{S}$. Note that while the total area of the surface would be in general involved as a factor, the value of this area is not relevant since we are interested in finding a minimum of the cost function and the position of the minimum of any function is not altered by a positive multiplicative constant. The sum over random points introduces a relative error of order $1/\sqrt{500}$ . Replacing the integral with a summation creates also many local minima in the landscape of Eq. (11). Accordingly, we tested different initial conditions; the best results are obtained starting from $V_{K_{S_{0}}}=0$. The density resulting from minimization of Eq. (7) is plotted in Fig 2(d). We see here again a significant change as compared with the fully interacting CI ground state [see Fig 2(a)] and the the exact DFT non-interacting solution Fig 2(b). Figure 2 is a clear example that corroborates our claim in Section II.1 that enforcing different properties on the non-interacting wave-function implies a density-density correspondence different than the identity between the interacting and non interacting systems. Similar results are observed as function of the strength and shape of the interaction [controlled by $\gamma$ and $\alpha$] A comparison between Eqs. (10) and (11) clearly shows that the relative values of the coefficient multiplying $\phi_{n}({\bf r})\phi_{nu}({\bf r})$ depends fundamentally on the cost function. Therefore, even starting from the same effective potential and $\Phi_{T}({\bf R})$ the coefficient affecting each individual product $\phi_{n}({\bf r})\phi_{nu}({\bf r})$ depends on the functional form of the cost function. This change in the potential remains present when the potential is written in the complete cosine basis. Thus, the effective potential must change in accord with the property of the interacting ground state that one aims to enforce in the non-interacting ground state with a cost function. Figure 3 shows the effective potentials used for the calculations shown in Fig 2. We show in Fig 3(a) a constant, since in the interacting problem solved with full CI no effective external potential was added. Figures 3(b) [minimum of Eq. (2)], 3(c) [minimum of Eq. (5)] and 3(d) [minimum of Eq. (7)] show a clear change in the effective potential depending on the cost functions. As argued earlier Fig. 3(b) shows the exact Kohn-Sham DFT potential for this interaction which implies that a different density functional must be used to obtain non-interacting wave-functions preserving properties other than the density. Note that Eq. (7) could be zero only for a pure-state-non-interacting $v$-representable nodal surface. However, if the nodes are not, replacing in $K_{S_{0}}$ could result in a potential that simply prevents the non- interacting wave-function to reach regions of space where the nodes are more troublesome. The potential shown in Fig. 3(d) presents a maximum in regions where instead Figs. 3(b) and 3(c) develop a minimum. These are the regions where the electrons in the many-body wave-functions tend to localize because of correlation effects. The maximum in Fig 3(d) suggest the possibility of non-$v$ representability by a non-interacting wave-function in this model. Figure 3: Optimized effective potentials corresponding to the densities in Fig. 2. The complete potentials can be obtained by reflection on the black lines. Gray level valuesunits are given on the right. The optimal effective potentials are strongly dependent on the property we target to retain in the wave-function. ### III.3 Wave-function internal structure In order to quantitatively test the quality of the nodes of the wave-functions found by minimization of Eqs. (2), (5) and (7) and to test the convergence of the nodes of the full CI calculations, we take advantage of the homogeneous distribution of points $\\{{\bf R}\\}_{S}$ at the nodal surface $S_{0}({\bf R})$ described earlier. For each point ${\bf R}$ in $\\{{\bf R}\\}_{S}$ we can find the distance $\ell_{i}$ to the node of another wave-function $\Phi_{T}({\bf R})$ in the direction of ${\bf\nabla}\Psi({\bf R})$. Thus $\Delta V=\frac{1}{N}\sum_{i}\ell_{i}$ (12) is an approximated measure of the fraction of the Hilbert space volume between the nodal surface of $\Phi_{T}({\bf R})$ and $\Psi({\bf R})$ and $\delta\rho=\frac{1}{3N}\sum_{i}\ell_{i}\left|\Phi_{T}\left({\bf R}_{i}\right)\right|^{2}$ (13) measures the probability density inside $\Delta V$. We can use Eqs. (12) and (13) to test the convergence of the CI ground state nodes as a function of the size of the basis set. While the ground state energy requires $40$ basis functions, the nodes are more difficult to converge requiring four times as many. The size of the basis required to converge the nodes was determined for $\gamma=2$ plotting $\delta\rho$ between the CI ground state with 300 wave-functions and the CI ground state obtained using a reduced basis. The quantities in Eqs. (12) and (13) can be used also to characterize the nodes of different wave-functions as compared with the exact node. In figure 4 we show the volume enclosed between the nodes $\Delta V$ of different optimized $\Phi_{T}({\bf R})$ and the interacting ground state $\Psi({\bf R})$ as a function of the strength of the interaction potential $\gamma$. Note that the Kohn-Sham DFT solution gives a significantly larger volume than other optimized wave-functions. The difference increases as the interaction strength increases. The wave-function that results from minimizing Eq. (5) which targets wave-function projection fares very well over the range explored. In turn, minimization of $K_{S_{0}}$ [see Eq. (7)] results in nodes that are only sometimes marginally better. Surprisingly, the non-interacting solution, that is the non-interacting ground state in the absence of any effective potential, is remarkably good. Similar results are found by altering the shape of the potential with $\alpha$. Figure 4: (Color online) Fraction of the Hilbert space $\Delta V$ between the full CI node and the nodes of different optimized wave-functions. Triangles correspond to the exact DFT wave-function [Eq. (2)], squares to maximum projection [Eq. (5)], rhombi to the minimum amplitude at the nodes [Eq. (7)] and circles to the non-interacting ground state. The inset shows the method used to estimate $\Delta V$ In figure 5 we plot the values $\delta\rho$ for different optimized wave- functions as a function of $\gamma$. We see again that the exact Kohn-Sham DFT solution is not the best. The quantity $\delta\rho$ is a measure on how much the error in the nodes would affect the probability density and thus it can be understood as the as a measure of the nodal error in the ground state energy. Again in this case the non-interacting ground state without any effective potential is the best approximation. Figure 5: (Color online) Probability density inside the volume between the nodes of the full CI wave-function and optimized wave-functions. Same conventions and symbols as in Fig. 2 ## IV Discussion Although the numerical investigation of the different density-density functionals described above required numerical representation of the many body ground state wave-function, the conclusions that we draw have general value. The model we explore is simplified but has the advantages that the results can be converged and are free of significant approximations. The simplified interaction used in the model retains essential features of the Coulomb interaction. We have shown (Fig. 2 and 3) that the effective densities and potentials are explicit functions of a cost function. Potentials and densities very different to the exact DFT solutions are obtained if we enforce properties beyond $\rho({\bf r})$ in the cost function. The exact DFT wave-function matches $\rho({\bf r})$ with complete disregard to other elements of the many body wave-function structure. Since the Hohenberg-Kohn theoremhohenberg is valid, optimizing other properties of the non-interacting ground state in general requires changing the potential with a resulting impact on the density. We find that mean field methods, while giving an accurate description of the density can mislead us in other aspects of the wave-function structure such as the nodal surface. In this paper we argue that, among the pure-state-non- interacting $v$-representable densities, there is at least one that more accurately describes the interacting ground state nodes. We can optimize the wave-function associated with this density with a cost function. The cost function form depends on the property we target to retain and also the optimal density we find. For the nodes, the optimal cost function is clearly the DMC energy. A fixed cost function establishes a density-density correspondence which can be described as an operator $U$ that transforms interacting densities into non-interacting ones. While finding the functional form of $U_{DMC}$ is a task beyond the scope of this paper, we argue that we can expect this operator to be highly non-local and very different from the identity, in particular for strong electron- electron correlations. We find that we can improve the nodes with some simple cost functions, but the best nodes we found were obtained solving the non- interacting problem without the addition of any effective potential. We cannot exclude the possibility that this result might well be an accident of the model. Our result, however, shows that the popular expectation that the DFT solution is a good starting point for nodes is not valid in general. Optimal wave-functions can be found by altering an external potential. This idea is not new. In practice wave-functions are optimized with the trial wave- function only minimizing the ground state energy $K_{VMC}=\left<\Phi_{T}\right|e^{-J}He^{-J}\left|\Phi_{T}\right>/(\left<\Phi_{T}\right|e^{-2J}\left|\Phi_{T}\right>)$, or the variance of the ground state energy. Replacing $K_{VMC}$ into Eq. (9) leads to a procedure similar to the optimization of Filippi and Fahy filippi00 providing additional support to that method. In the case of Refs. wagner03 ; kolorenc the nodes are selected by adjusting the mix of density functionals that gives the lowest DMC energy for a small system. The same mix is then used in a larger system. This procedure is in fact equivalent to optimizing the shape of the effective potential with the restriction of remaining a linear combination of two of more exchange-correlation potentials. We find that the change in the effective potential required to optimize the nodes could be of the order of the Hartree potential, since the wave-function with the best nodes is the non-interacting ground state, without any effective potential, for all the range of interaction strengths and shapes explored. Our results suggest that counter intuitive directions for potential optimizations should be explored to improve the nodes. Potential optimization has also been applied for the prediction of electronic excitations. Since $\rho({\bf r})$ determines $V({\bf r})$ (but from a constant), the excitation spectra $\\{E_{\nu,n}\\}$ is a function of $\rho({\bf r})$. This allows defining cost functions $K_{ex}$ to match the spectra of a non-interacting system. In order to minimize $K_{ex}$ one should do the derivatives $\delta\varepsilon_{\nu}/\delta V_{KS}({\bf r})$ as in Ref us . When $\\{E_{\nu}\\}$ is taken from experiment, the search of a potential giving a non-interacting density that minimizes $K_{ex}$ is equivalent to the empirical potential method.wang95 . Unfortunately, in this case the electronic density can no longer be used to obtain the forces on the atoms. The existence of a single density functional that can be used to obtain the excitation spectra of any system is then a subject of debate. In summary, although the popular languages of electronic structure theory all share the same quantum mechanical underpinnings, when applied by experts to physical systems we often reach different conclusions. Many experts in QMC prefer HF wave-functions, while in contrast calculations done within the GW- BSE approach often rely on LDA derived wave-functions and energiesaulbur00 , while some hybrid density functionals obtain single particle excitations in direct agreement with excitation spectrabarone05 . We argue that as different theories need to retain different properties of the same ground state wave- function to minimize errors, different functionals should also be used. In some cases these functionals correspond to the minimization of cost functions designed to retain properties of the many body ground state in the non- interacting wave-function. Since some properties are favored at the expense of others, it is unlikely that we can use the same functional universally: we find that a function designed for optimal nodes is a bad source of densities and vise versa. Here we give a qualitative picture of the size of the differences that one can expect as correlations start to dominate. With increasing interaction strength, the exact Kohn-Sham non-interacting wave- function becomes a much poorer description of several properties of the many- body ground state. Methods that go beyond DFT are limited to the nearly non- interacting limit if they depend strongly on DFT derived wave-functions. Research performed at the Materials Science and Technology Division and the Center of Nanophase Material Sciences at Oak Ridge National Laboratory sponsored the Division of Materials Sciences and the Division of Scientific User Facilities U.S. Department of Energy. The authors would like thank R. Q. Hood, M. Kalos and M. L. Tiago for discussions. ## References * (1) P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964). * (2) W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965). * (3) In practice, an approximate exchange correlation potential is used. Due to this approximation $\bar{\rho}(r)$ and $\rho(r)$ are in practice different. * (4) J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981). * (5) R. M. Martin, Electronic Structure, Basic Theory and Practical Methods Cambridge University Press, (2004). * (6) R. G. Parr and W. Yang, Density-Functional Theory of Atoms and Molecules Oxford Science Publications (1989). * (7) J.P.Perdew, et al., Phys. Rev. Lett. 77, 3865 (1996). * (8) W.G. Aulbur, L. Jonsson, and J.W. Wilkins, Solid State Physics, eds. F. Seitz, D. Turnbull, and H. Ehrenreich, 54, 1 (2000). * (9) P.-O. Löwdin, Phys. Rev. 97, 1474 (1955). * (10) J. B. Anderson, Int. J. of Quantum Chem. 15, 109 (1979). * (11) M. Bajdich, L. Mitas, G. Drobny, L.K. Wagner, Phys. Rev. B 72, 075131 (2005); L. Mitas, Phys. Rev. Lett. 96, 240402 (2006). * (12) C. Filippi and S. Fahy in J. Chem. Phys. 112, 3523 (2000). * (13) C. J. Umrigar, et al. Phys. Rev. Lett. 98, 110201 (2007). * (14) P. LópezRios, et. al, Phys. Rev. E 74 066701 (2006). * (15) A. Lüchow, etl al, J. Chem. Phys. 126 144110 (2007). * (16) M.H. Kalos and F. Pederiva, Phys Rev. Lett. 85, 3547 (2000). * (17) S. W. Zhang, M. H. Kalos, Phys. Rev. Lett. 67, 3074 (1991). * (18) D. Alfe, M.J. Gillan, Phys. Rev. B 70, 161101(R) (2004). * (19) F.A. Reboredo, A.J. Williamson, Phys. Rev. B 71, 121105(R) (2005). * (20) A. J. Williamson, R. Q. Hood, J. C. Grossman, Phys. Rev. Lett. 87 246406 (2001). * (21) J. Kolorenc̆. and L. Mitas http://arxiv.org/pdf/0712.3610. * (22) J. K. Percus Int. J. Quantum Chem, 13, 89 (1978). * (23) M. Levy, Proc Nat Acad. Sci. USA, 76, 6062 (1979). * (24) L. Wagner and L. Mitas, Chem. Phys. Lett. 370, 412 (2003). * (25) We define the energy unit to be $\hbar^{2}/(2m\pi^{2})$. * (26) The Mathematica notebook is available upon request. * (27) A. Görling and M. Levy, Phys. Rev. B 47, 13105 (1993); A. Görling and M. Levy, Phys. Rev. A 50, 196 (1994); J. B. Krieger, Y. Li, and G. J. Iafrate, Phys. Rev. A 45, 101 (1992); S. Ivanov, S. Hirata, and R. J. Bartlett, Phys. Rev. Lett. 83, 5455 (1999); A. Görling, Phys. Rev. Lett. 83, 5459 (1999); F. Della Sala et. al, Phys. Rev. Lett. 89, 033003 (2002). * (28) M. Städele, et. al. , Phys. Rev. Lett. 79, 2089 (1997); M. Städele, et. al Phys. Rev. B 59, 10031 (1999). * (29) A. R. Goñi, et. al., Phys. Rev. B 65, 121313(R) (2004). * (30) F. A. Reboredo and C. R. Proetto, Phys. Rev. B 67, 115325 (2003). * (31) L.-W. Wang and A. Zunger, Phys. Rev. B 51, 17398 (1995). * (32) V. Barone, et. al., Nano Lett 5, 1621 (2005).

arxiv-papers

2024-09-04T02:48:54.514331

0803.3472

# Tracking Vector Magnetograms with the Magnetic Induction Equation P. W. Schuck††affiliation: schuck@ppdmail.nrl.navy.mil Plasma Physics Division, United States Naval Research Laboratory 4555 Overlook Ave., SW, Washington, DC 20375-5346 ###### Abstract The differential affine velocity estimator (DAVE) developed in Schuck (2006) for estimating velocities from line-of-sight magnetograms is modified to directly incorporate horizontal magnetic fields to produce a differential affine velocity estimator for vector magnetograms (DAVE4VM). The DAVE4VM’s performance is demonstrated on the synthetic data from the anelastic pseudospectral ANMHD simulations that were used in the recent comparison of velocity inversion techniques by Welsch et al. (2007). The DAVE4VM predicts roughly 95% of the helicity rate and 75% of the power transmitted through the simulation slice. Inter-comparison between DAVE4VM and DAVE and further analysis of the DAVE method demonstrates that line-of-sight tracking methods capture the shearing motion of magnetic footpoints but are insensitive to flux emergence — the velocities determined from line-of-sight methods are more consistent with horizontal plasma velocities than with flux transport velocities. These results suggest that previous studies that rely on velocities determined from line-of-sight methods such as the DAVE or local correlation tracking may substantially misrepresent the total helicity rates and power through the photosphere. magnetic fields — Sun: atmospheric motions — methods: data analysis ††slugcomment: Submitted to Ap. J. ## 1 Introduction Coronal mass ejections (CMEs) are now recognized as the primary solar driver of geomagnetic storms Gosling (1993). Several theoretical mechanisms have been proposed as drivers of CMEs, including large scale coronal reconnection Sweet (1958); Parker (1957); Antiochos et al. (1999), emerging flux cancellation of the overlying coronal field Linker et al. (2001), flux injection Chen (1989, 1996), the kink instability of filaments Rust & Kumar (1996); Török et al. (2004); Kliem et al. (2004), and photospheric footpoint shearing Amari et al. (2000, 2003a, 2003b); Schrijver et al. (2005). All of these CME mechanisms are driven by magnetic forces. The main differences depend on whether the magnetic helicity and energy are first stored in the corona and later released by reconnection and instability or whether the helicity and Poynting fluxes are roughly concomitant with the eruption. The timing and magnitude of the transport of magnetic helicity and energy through the photosphere provides an important discriminator between the mechanisms. In addition, eruption precursors in the photospheric magnetic field might provide reliable forecasting for space weather events. However, reliable, repeatable photospheric precursors of CMEs have so far eluded detection Leka & Barnes (2003a, b, 2007). The magnetic helicity and Poynting flux may be estimated from photospheric velocities inferred from a sequence of magnetograms Berger & Ruzmaikin (2000); Démoulin & Berger (2003). However, accurately estimating velocities from a sequence of images is extremely challenging because image motion is ambiguous. The “aperture problem” occurs when different velocities produce image dynamics that are indistinguishable Stumpf (1911); Marr & Ullman (1981); Hildreth (1983, 1984). Optical flow methods solve these under-determined or ill-posed problems that have no unique velocity field solution by applying additional assumptions about flow structure or flow properties. Both Schuck (2006) and Welsch et al. (2007) provide an overview of optical flow methods for recovering estimates of photospheric velocities from a sequence of magnetograms Kusano et al. (2002, 2004); Welsch et al. (2004); Longcope (2004); Schuck (2005, 2006); Georgoulis & LaBonte (2006). Currently, most methods for estimating photospheric velocities implement some form of the normal component of the induction equation $\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$V$}}_{h}-V_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=0,$ (1) where the plasma velocity $V$ and the magnetic fields $B$ are decomposed into a local right-handed Cartesian coordinate system with vertical direction along the $z$-axis and the horizontal plane, denoted generically by the subscript “h,” containing the $x$\- and $y$-axes. Démoulin & Berger (2003) observed that the geometry of magnetic fields embedded in the photosphere implied that $\mbox{\boldmath{$F$}}=\mbox{\boldmath{$U$}}\,B_{z}\equiv{B}_{z}\,\mbox{\boldmath{$V$}}_{h}-V_{z}\,\mbox{\boldmath{$B$}}_{h}=\widehat{\mbox{\boldmath{$z$}}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$V$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}\right)=\widehat{\mbox{\boldmath{$z$}}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$V$}}_{\perp}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}\right),$ (2a) where $F$ denotes the flux transport vector, $U$ is the horizontal footpoint velocity or flux transport velocity $\left(\mbox{\boldmath{$U$}}\cdot\widehat{\mbox{\boldmath{$z$}}}=0\right)$ and $\mbox{\boldmath{$V$}}_{\perp}$ is the plasma velocity perpendicular to the magnetic field $\mbox{\boldmath{$V$}}_{\perp}\cdot\mbox{\boldmath{$B$}}=0$. The flux transport vectors are composed of two terms ${B}_{z}\,\mbox{\boldmath{$V$}}_{h}$ and $V_{z}\,\mbox{\boldmath{$B$}}_{h}$ representing shearing due to horizontal motion and flux emergence due to vertical motion respectively. Equation (2a) may be used to transform (1) into a continuity equation for the vertical magnetic field $\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$U$}}\,B_{z}\right)=0,$ (2b) where plasma velocity may be written generally in terms of the flux transport velocity as $V$ $\displaystyle=$ $\displaystyle\mbox{\boldmath{$U$}}-\frac{\left(\mbox{\boldmath{$U$}}\cdot\mbox{\boldmath{$B$}}_{h}\right)\,\mbox{\boldmath{$B$}}}{|\mbox{\boldmath{$B$}}|^{2}}+V_{\parallel}\,\frac{\mbox{\boldmath{$B$}}}{\left|\mbox{\boldmath{$B$}}\right|},$ (3a) $\displaystyle\mbox{\boldmath{$V$}}_{\perp{h}}$ $\displaystyle=$ $\displaystyle\mbox{\boldmath{$U$}}-\frac{\left(\mbox{\boldmath{$U$}}\cdot\mbox{\boldmath{$B$}}_{h}\right)\,\mbox{\boldmath{$B$}}_{h}}{|\mbox{\boldmath{$B$}}|^{2}},$ (3b) $\displaystyle\mbox{\boldmath{$V$}}_{\perp{z}}$ $\displaystyle=$ $\displaystyle-\frac{\left(\mbox{\boldmath{$U$}}\cdot\mbox{\boldmath{$B$}}_{h}\right)\,B_{z}}{|\mbox{\boldmath{$B$}}|^{2}},$ (3c) and the subscripts “$\parallel$” and “$\perp$” denote plasma velocities parallel and perpendicular to the magnetic field respectively. Equations (3a-c) are the algebraic decomposition Welsch et al. (2004) generalized for arbitrary parallel velocity $V_{\parallel}$, but the value of $V_{\parallel}$ does not affect the perpendicular plasma velocity (3b)-(3c) or the perpendicular electric field $c\,\mbox{\boldmath{$E$}}_{\perp}=-\mbox{\boldmath{$V$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}=-\overbrace{\mbox{\boldmath{$U$}}\mbox{\boldmath{$\times$}}\widehat{\mbox{\boldmath{$z$}}}\,B_{z}}^{\mbox{$\mbox{\boldmath{$E$}}_{\perp{h}}$}}-\overbrace{\mbox{\boldmath{$U$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}_{h}}^{\mbox{$E_{\perp{z}}$}},$ (4) which both depend only on the flux transport velocity $U$. Equations (1)-(3) should be formally distinguished from the inverse problem for determining an estimate of the plasma velocity $v$ from vector magnetograms using the normal component of the magnetic induction equation $\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=0,$ (5a) where $\mbox{\boldmath{$f$}}=\mbox{\boldmath{$u$}}\,B_{z}=B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}=\widehat{\mbox{\boldmath{$z$}}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$v$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}\right)=\widehat{\mbox{\boldmath{$z$}}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$v$}}_{\perp}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}\right),$ (5b) and the inverse problem for determining flux transport velocity $u$ from the evolution of the vertical magnetic field or line-of-sight component $\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)=0.$ (6) The notation $\vartheta$, denoting an optical flow estimate, emphasizes that $\vartheta$ determined from (6) is not necessarily immediately identified with the flux transport velocity $u$. Equations (5)-(6) are ill-posed inverse problems because of two ambiguities: 1. 1. The Helmholtz decomposition of the flux transport vectors Welsch et al. (2004); Longcope (2004) $\mbox{\boldmath{$f$}}=\mbox{\boldmath{$u$}}\,B_{z}=B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}=-\left(\mbox{\boldmath{$\nabla$}}_{h}\phi+\mbox{\boldmath{$\nabla$}}_{h}\psi\mbox{\boldmath{$\times$}}\widehat{\mbox{\boldmath{$z$}}}\right),$ (7) where $\phi$ is the inductive potential and $\psi$ is the electrostatic potential manifestly demonstrates that only inductive potential $\psi$ may be unambiguously determined from the local evolution of $B_{z}$ in (5). The electrostatic potential $\phi$ must be constrained by additional assumptions. By analogy, (6) is also ill-posed for the same reason; $\mbox{\boldmath{$\nabla$}}_{h}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ is not constrained by the local evolution of $\partial_{t}B_{z}$. 2. 2. For (5a) and (5b) $V_{\parallel}$ is not constrained by the local evolution of $\partial_{t}B_{z}$. For (6), there is no a priori relationship between $\vartheta$ and $u$ or $\vartheta$ and $v$ for the inverse problem. However, if $\vartheta$ is identified with the flux transport velocity $u$ then $u$ and $v$ will satisfy the same relationships as $U$ and $V$ in (3). The first ambiguity may be resolved for (5a) by the induction method (IM) Kusano et al. (2002, 2004), minimum energy fit (MEF) Longcope (2004), or the differential affine velocity estimator for vector magnetograms (DAVE4VM) presented in § 2. These methods produce a unique solution for $v$, but not necessarily the unique solution that corresponds to $V$. The first ambiguity may be resolved for (6) by local optical flow methods such as the differential affine velocity estimator (DAVE) Schuck (2006), its nonlinear generalization Schuck (2005), global methods Wildes et al. (2000), the minimum structure reconstruction (MSR) Georgoulis & LaBonte (2006) which imposes $\mbox{\boldmath{$v$}}_{\perp{z}}=0$ as an assumption, or hybrid local-global methods such as inductive local correlation tracking (ILCT) Welsch et al. (2004). These methods produce a unique solution for $\vartheta$, but not necessarily the unique solution that corresponds to $u$. Several assumptions have been used either explicitly or implicitly to resolve the second ambiguity. Chae et al. (2001) conjecture that local correlation tracking (LCT) Leese et al. (1970, 1971); November & Simon (1988) provides a direct estimate of the horizontal photospheric plasma velocity: $\mbox{\boldmath{$\vartheta$}}^{\left(\mathrm{LCT}\right)}=\mbox{\boldmath{$V$}}_{h}$. Démoulin & Berger (2003) conjecture that line-of-sight tracking methods, and in particular LCT, estimate the total flux transport velocity $\mbox{\boldmath{$\vartheta$}}^{\left(\mathrm{LCT}\right)}=\mbox{\boldmath{$U$}}$. Schuck (2005) formally demonstrated that LCT is consistent with the advection equation $\partial_{t}B_{z}+\mbox{\boldmath{$\vartheta$}}^{\left(\mathrm{LCT}\right)}\cdot\mbox{\boldmath{$\nabla$}}_{h}{B_{z}}=0,$ (8) not the continuity equation in (6), but that LCT could be modified to be consistent with (6) by direct integration along Lagrangian trajectories in an affine velocity profile. Nonetheless, both conjectures may be considered in the context of (6). Under Chae et al.’s (2001) assumption, the flux transport velocity would be derived from line-of-sight optical flow methods via $\mbox{\boldmath{$u$}}\,B_{z}\equiv\mbox{\boldmath{$\vartheta$}}\,B_{z}-v_{z}\,\mbox{\boldmath{$B$}}_{h},$ (9) where in principle, $v_{z}$ might be approximately determined from Doppler velocities near disk center. Under Démoulin & Berger’s (2003) assumption, the total flux transport velocity would be derived from line-of-sight optical flow methods via $\mbox{\boldmath{$u$}}\equiv\mbox{\boldmath{$\vartheta$}}\mbox{ for }B_{z}\neq 0.$ (10) The Ansatz $\mbox{\boldmath{$u$}}=\mbox{\boldmath{$\vartheta$}}$ has important implications for solar observations. This conjecture implies that the total helicity and Poynting flux may be estimated by tracking the vertical magnetic field or by tracking the line-of-sight component near disk center as a proxy for the vertical magnetic field. Démoulin & Berger’s (2003) Ansatz has largely been accepted by the solar community Welsch et al. (2004, 2007); Kusano et al. (2004); Schuck (2005, 2006); LaBonte et al. (2007); Santos & Büchner (2007); Tian & Alexander (2008); Zhang et al. (2008); Wang et al. (2008). However, equivalence between $\vartheta$ and $u$ for line-of-sight methods has never been practically established. These two different hypotheses (9) and (10) for the interpretation of $\vartheta$ inferred by DAVE will be considered in § 4. The second ambiguity usually is not resolved using only information about the magnetic fields. The velocity field inferred by the IM Kusano et al. (2002, 2004) does produce a component of the plasma velocity along the magnetic field, but this was simply subtracted off in Welsch et al. (2007). In the absence of a reference flow, possibly derived from Doppler measurements or LCT, the MEF imposes $v_{\parallel}=0$ Longcope (2004). ILCT and the original algebraic decomposition both assume $v_{\parallel}=0$ Welsch et al. (2004). Georgoulis & LaBonte (2006) describe a method for inferring $v_{\parallel}$ from Doppler measurements for MSR. For DAVE4VM the second ambiguity is resolved simultaneously with the first. The DAVE4VM method estimates a field aligned plasma velocity from only magnetic field observations! Using established computer vision techniques Lucas & Kanade (1981); Lucas (1984); Baker & Matthews (2004), Schuck (2006) developed the DAVE from a short time-expansion of the modified LCT method discussed in Schuck (2005) for estimating velocities from line-of-sight magnetograms. The DAVE locally minimizes the square of the continuity equation (2b) subject to an affine velocity profile. Using “moving paint” experiments, Schuck (2006) demonstrated that this technique was faster and more accurate than existing LCT algorithms for data satisfying (2b). The DAVE method has been used to study the apparent motion of active regions Schuck (2006), flux pile up in the photosphere Litvinenko et al. (2007), and helicity flux in the photosphere Chae (2007). However, nagging questions remain about its performance. Welsch et al. (2007) set an important new standard for evaluating scientific optical flow methods used for studying the Sun. For the first time many existing methods for estimating photospheric velocities from magnetograms were tested on a reasonable approximation to synthetic photospheric data from anelastic pseudospectral ANMHD simulations Fan et al. (1999); Abbett et al. (2000, 2004). The methods tested were Lockheed Martin’s Solar and Astrophysical Laboratory’s (LMSAL) LCT code DeRosa (2001), Fourier LCT (FLCT) Welsch et al. (2004), the DAVE Schuck (2006), the IM Kusano et al. (2002, 2004), ILCT Welsch et al. (2004), the MEF Longcope (2004), and MSR Georgoulis & LaBonte (2006). Unfortunately the results were not entirely encouraging. Welsch et al. (2007) treated the velocities estimated from line-of-sight methods as the flux transport velocities consistent with the hypothesis of Démoulin & Berger (2003) in (10). Evaluation of the DAVE’s performance on the ANMHD data under this assumption revealed that the DAVE method did not estimate the helicity flux or Poynting flux reliably. In fact none of the pure line-of-sight methods: LMSAL’s LCT, FLCT, or the DAVE—estimated these fluxes reliably, reproducing (at best) respectively 11%, 9%, and 23% of the helicity rate, and reproducing respectively 6%, 11%, and 22% of the power injected through the surface. Of course the ANMHD data have limitations. The simulation models the rise of a buoyant magnetic flux rope in the convection zone and represents the magnetic structure of granulation or super-granulation rather than the dynamics of an active region (See § 2 in Welsch et al., 2007, for a complete discussion). In addition, Welsch et al. (2007) noted that tracking methods performed better on real magnetograms than on the synthetic ANMHD data using “moving paint” experiments where images were simply shifted relative to one another. These results provoked them to comment “that the ANMHD data set either lacks some characteristic present in real solar magnetograms or contains artifacts not present in solar data.” Consequently, the poor performance of tracking methods on ANMHD data might be attributed to the de-aliasing method for nonlinear terms in ANMHD (truncating the spatial Fourier spectrum effectively smoothes small-scale structures) or perhaps to the Fourier ringing near strong fields in the ANMHD data set. While these issues are important to resolve, they fail to fully explain the poor performance of the tracking methods to accurately reproduce the quantity they were designed to estimate, namely the helicity flux! This paper has two primary goals: 1. 1. Develop a modified DAVE Schuck (2006) that incorporates horizontal magnetic fields, termed the “differential affine velocity estimator for vector magnetograms” (DAVE4VM), and demonstrate its performance on the ANMHD simulation data. DAVE4VM performs much better than the original DAVE technique and roughly on par with the minimum energy fit (MEF) method developed by Longcope (2004) which was deemed to have performed the best overall in Welsch et al.’s (2007) comparison of velocity-inversion techniques. 2. 2. Identify the reasons for the poor performance of DAVE in Welsch et al. (2007). The paper attempts to follow, as closely as possible, the presentation of the DAVE in Schuck (2006) and the analysis of velocity inversion techniques by Welsch et al. (2007). For the remainder of this paper, lower case variables are used to represent the flux transport vector, flux transport velocity, plasma velocity, electric field, Poynting flux, and helicity flux estimates from the DAVE4VM and DAVE: $f$, $u$, $\mbox{\boldmath{$v$}}_{\perp}$, $\mbox{\boldmath{$e$}}_{\perp}$, $s_{z}$ and $h$ and the corresponding uppercase variables are used to represent the “ground truth” from ANMHD: $F$, $U$, $\mbox{\boldmath{$V$}}_{\perp}$, $\mbox{\boldmath{$E$}}_{\perp}$, $S_{z}$, and $H$. The one deviation from this notation involves $\vartheta$ which denotes an optical flow estimate based on (6). Section (2) describes the DAVE4VM model and § 3 describes its application to the ANMHD data. For the most part, the plots and quantitative analysis presented in Welsch et al. (2007) are produced for the DAVE4VM and DAVE to facilitate inter-comparison and comparison to the other methods considered in Welsch et al. (2007). For the DAVE this analysis involves the explicit assumption that $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$. In § 4 the assumption $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ for the DAVE is relaxed and compared with an alternative hypotheses that $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$ — that the DAVE produces a biased estimate of the total horizontal plasma velocity. ## 2 The DAVE4VM Model The extension of the DAVE for horizontal magnetic fields is straight-forward. The plasma velocity is modeled with a three-dimensional affine velocity profile: $\mbox{\boldmath{$v$}}\left(\mbox{\boldmath{$P$}};\mbox{\boldmath{$x$}}\right)=\left(\begin{array}[]{c}\widehat{u}_{0}\\\ \widehat{v}_{0}\\\ \widehat{w}_{0}\end{array}\right)+\left(\begin{array}[]{cc}\widehat{u}_{x}&\widehat{u}_{y}\\\ \widehat{v}_{x}&\widehat{v}_{y}\\\ \widehat{w}_{x}&\widehat{w}_{y}\end{array}\right)\,\left(\begin{array}[]{c}x\\\ y\end{array}\right),$ (11) where the hatted variables model the local plasma velocity profile. The coordinate system for the affine velocity profile is not aligned with the magnetic field. Therefore, the velocities are not guaranteed to be orthogonal to $B$. However, the parallel $\left(\parallel\right)$ and perpendicular $\left(\perp\right)$ components of the plasma velocity may be determined from $\displaystyle v_{\parallel}$ $\displaystyle=$ $\displaystyle\frac{\left(\mbox{\boldmath{$v$}}\cdot\mbox{\boldmath{$B$}}\right)\,\mbox{\boldmath{$B$}}}{B^{2}},$ (12a) $\displaystyle\mbox{\boldmath{$v$}}_{\perp}$ $\displaystyle=$ $\displaystyle\mbox{\boldmath{$v$}}-\frac{\left(\mbox{\boldmath{$v$}}\cdot\mbox{\boldmath{$B$}}\right)\,\mbox{\boldmath{$B$}}}{B^{2}},$ (12b) $\displaystyle\mbox{\boldmath{$v$}}_{\perp{h}}$ $\displaystyle=$ $\displaystyle\mbox{\boldmath{$v$}}_{h}-\frac{\left(\mbox{\boldmath{$v$}}\cdot\mbox{\boldmath{$B$}}\right)\,\mbox{\boldmath{$B$}}_{h}}{B^{2}},$ (12c) $\displaystyle v_{\perp{z}}$ $\displaystyle=$ $\displaystyle v_{z}-\frac{\left(\mbox{\boldmath{$v$}}\cdot\mbox{\boldmath{$B$}}\right)\,B_{z}}{B^{2}}.$ (12d) The error metric $\displaystyle\mathcal{C}_{\mbox{SSD}}$ $\displaystyle=$ $\displaystyle\int{dt}{dx^{2}}\,w\left(\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}},t-\tau\right)\left\\{\partial_{t}B_{z}\left(\mbox{\boldmath{$x$}},t\right)+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left[B_{z}\left(\mbox{\boldmath{$x$}},t\right)\,\mbox{\boldmath{$v$}}_{h}\left(\mbox{\boldmath{$P$}},\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}}\right)\right.\right.,$ $\displaystyle\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\left.\left.-v_{z}\left(\mbox{\boldmath{$P$}},\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}}\right)\,\mbox{\boldmath{$B$}}_{h}\left(\mbox{\boldmath{$x$}},t\right)\right]\right\\}^{2},$ $\displaystyle=$ $\displaystyle\mbox{\boldmath{$\eta$}}\cdot\left\langle\SS\right\rangle\cdot\mbox{\boldmath{$\eta$}}$ (13b) characterizes how well the local velocity profile satisfies the magnetic induction equation over a subregion of the magnetogram sequence defined by the window function $w\left(\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}},t-\tau\right)$ where $\mbox{\boldmath{$P$}}=\left(\widehat{u}_{0},\widehat{v}_{0},\widehat{u}_{x},\widehat{v}_{y},\widehat{u}_{y},\widehat{v}_{x},\widehat{w}_{0},\widehat{w}_{x},\widehat{w}_{y}\right)$ is a vector of parameters and $\mbox{\boldmath{$\eta$}}\equiv\left(\mbox{\boldmath{$P$}},1\right)$. The plasma velocity $\mbox{\boldmath{$v$}}\left(\mbox{\boldmath{$P$}},\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}}\right)$ in (13) is referenced from the center of the window at $\mbox{\boldmath{$x$}}=\mbox{\boldmath{$\chi$}}$ so that $\widehat{u}_{0}$, $\widehat{v}_{0}$, and $\widehat{w}_{0}$ represent the plasma velocities at the center of the window and the subscripted parameters represent the best fit local shears in the plasma flows, i.e. $\widehat{u}_{x}=\partial_{x}\,\left(\widehat{\mbox{\boldmath{$x$}}}\cdot\mbox{\boldmath{$v$}}\right)$. The matrix elements of $\left\langle\SS\right\rangle$ are defined by $\left\langle\SS\right\rangle=\int{dt}\,{dx^{2}}\,w\left(\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}},t-\tau\right)\,\SS\left(\mbox{\boldmath{$\chi$}};x,t\right),$ (13c) where $\SS\left(\mbox{\boldmath{$\chi$}};x,t\right)\equiv\left[\begin{array}[]{cc}\mbox{\boldmath{$\mathsf{A}$}}&\mbox{\boldmath{$b$}}\\\ \mbox{\boldmath{$b$}}&\mathcal{G}_{99}\end{array}\right]=\left[\begin{array}[]{cccccccccc}\mathcal{G}_{00}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\\ \mathcal{G}_{10}&\mathcal{G}_{11}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\\ \mathcal{G}_{20}&\mathcal{G}_{21}&\mathcal{G}_{22}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\\ \mathcal{G}_{30}&\mathcal{G}_{31}&\mathcal{G}_{32}&\mathcal{G}_{33}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\\ \mathcal{G}_{40}&\mathcal{G}_{41}&\mathcal{G}_{42}&\mathcal{G}_{43}&\mathcal{G}_{44}&\cdot&\cdot&\cdot&\cdot&\cdot\\\ \mathcal{G}_{50}&\mathcal{G}_{51}&\mathcal{G}_{52}&\mathcal{G}_{53}&\mathcal{G}_{54}&\mathcal{G}_{55}&\cdot&\cdot&\cdot&\cdot\\\ \mathsf{s}_{60}&\mathsf{s}_{61}&\mathsf{s}_{62}&\mathsf{s}_{63}&\mathsf{s}_{64}&\mathsf{s}_{65}&\mathsf{s}_{66}&\cdot&\cdot&\cdot\\\ \mathsf{s}_{70}&\mathsf{s}_{71}&\mathsf{s}_{72}&\mathsf{s}_{73}&\mathsf{s}_{74}&\mathsf{s}_{75}&\mathsf{s}_{76}&\mathsf{s}_{77}&\cdot&\cdot\\\ \mathsf{s}_{80}&\mathsf{s}_{81}&\mathsf{s}_{82}&\mathsf{s}_{83}&\mathsf{s}_{84}&\mathsf{s}_{85}&\mathsf{s}_{86}&\mathsf{s}_{87}&\mathsf{s}_{88}&\cdot\\\ \mathcal{G}_{90}&\mathcal{G}_{91}&\mathcal{G}_{92}&\mathcal{G}_{93}&\mathcal{G}_{94}&\mathcal{G}_{95}&\mathsf{s}_{96}&\mathsf{s}_{97}&\mathsf{s}_{98}&\mathcal{G}_{99}\\\ \end{array}\right],$ (14) is a real symmetric $\SS=\SS^{*}$ positive semidefinite structure tensor where a superscript “*” indicates the matrix transpose. The matrix elements of $\SS$ are provided in Appendix A. The elements $\mathcal{G}_{ij}$ correspond to the original DAVE method Schuck (2006) and the remainder $\mathsf{s}_{ij}$ represent corrections due to the horizontal components of the magnetic field and flows normal to the surface. The least-squares solution is $\mbox{\boldmath{$P$}}=-\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle^{-1}\cdot\left\langle\mbox{\boldmath{$b$}}\right\rangle,$ (15) when the aperture problem is completely resolved $\det\left(\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle\right)\neq 0$ and the velocity field is unambiguous. However, there are important new terms in the structure tensor $\left\langle\SS\right\rangle$ involving $\mbox{\boldmath{$B$}}_{h}$. Situations where $\det\left(\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle\right)=0$ because $\mbox{\boldmath{$B$}}_{h}=0$ or $B_{z}=0$ over the region contained within the window must be considered. In general, the Moore-Penrose pseudo- inverse $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle^{\dagger}$ provides a numerically stable estimate of the optical flow parameters even when $\det\left(\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle\right)=0$ $\mbox{\boldmath{$P$}}=-\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle^{{\dagger}}\cdot\left\langle\mbox{\boldmath{$b$}}\right\rangle,$ (16a) where $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle^{\dagger}\equiv\mbox{\boldmath{$\mathsf{V}$}}\,\mbox{\boldmath{$\Sigma$}}^{\dagger}\,\mbox{\boldmath{$\mathsf{U}$}}^{*},$ (16b) is defined in terms of the singular value decomposition Golub & Van Loan (1980) $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle=\mbox{\boldmath{$\mathsf{U}$}}\,\mbox{\boldmath{$\Sigma$}}\,\mbox{\boldmath{$\mathsf{V}$}}^{*}.$ (16c) Here $\mathsf{U}$ and $\mathsf{V}$ are orthonormal matrices corresponding to the nine principle directions, $\Sigma$ is a diagonal matrix containing the nine singular values, and $\mbox{\boldmath{$\Sigma$}}^{\dagger}$ is computed by replacing every nonzero element of $\Sigma$ by its reciprocal. If $\mbox{\boldmath{$B$}}_{h}=0$, the singular values along the vertical direction are zero and $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle$ is rank deficient. In this case, the method implemented produces the minimum norm least squares solution resulting in no vertical flows: $w_{0}=w_{x}=w_{y}=0$. ## 3 Application to ANMHD Simulations This paper considers the pair of vector magnetograms $B$ separated by the shortest time interval $\Delta t\approx 250$ s between data dumps of the ANMHD simulation slice archived by Welsch et al. (2007). The “ground truth” data are derived from the time-averaged velocity and magnetic fields from ANMHD over the shortest time interval. The region of interest in the ANMHD simulations corresponds to a $101\times 101$ pixel region centered on a convection cell. Welsch et al. (2007) thresholded on the vertical magnetic field and considered only pixels with $|B_{z}|>370$ G for all plots and quantities except for the total helicity where a different masking of results was used Welsch et al. (2008). In a departure from the original presentation of Welsch et al. (2007) this paper considers pixels with $B=\sqrt{B_{x}^{2}+B_{y}^{2}+B_{z}^{2}}>370$ G. This corresponds to 7013 pixels or roughly 70% of the region of interest. This difference in thresholding is important because most of the flux emergence and helicity flux in the simulation occurs along the neutral line in regions of weak vertical field that are missed with vertical field thresholding used in the original study. Note that the modified thresholding mask contains weak vertical field regions and contains substantially more points than the roughly 3800 used in Welsch et al. (2007). The difference between the helicity flux in this comparison region and the total simulation is 0.023%. Since the DAVE4VM is a local optical flow method that determines the plasma velocities within a windowed subregion by constraining the local velocity profile, the choice of window size is a crucial issue for estimating velocities accurately. The window must be large enough to contain enough structure to uniquely determine the coefficients of the flow profile and resolve the aperture problem, but not so large as to violate the affine velocity profile (11) which is only valid locally (See Schuck, 2006, for discussion of the aperture problem in the context of the DAVE). In Welsch et al. (2007), the optimal window size was chosen for the DAVE by examining the Pearson correlation and slope between $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ and $\Delta B_{z}/\Delta t$ from ANMHD. If the method satisfies the induction equation exactly everywhere, the Pearson correlation and slope would both be equal to $-1$. However both the DAVE and DAVE4VM were conceived with the recognition that real magnetograms contain noise and should not satisfy the magnetic induction equation exactly; these methods satisfy the induction equation statistically within the window by minimizing the mean squared deviations from the ideal induction equations. Consequently, how well these methods satisfy the magnetic induction equation globally or over a subset of pixels can be used to assess overall performance. In Welsch et al. (2007), the DAVE was “optimized” over a subset of pixels with $\left|B_{z}\right|>370$ G that did not include the weak vertical field regions discussed in this study. Using the Pearson correlation and slope, an asymmetric window of $21\times 39$ performed the best on the ANMHD data. For the present study, the DAVE was “re- optimized” over the new criteria $\left|\mbox{\boldmath{$B$}}\right|>370$ G to provide an “apples-to-apples” comparison. However, I emphasize that generally this optimization cannot be carried out for the DAVE because this method was proposed for deriving flux-transport velocities from line-of-sight magnetograms. In this situation, the threshold mask can only be applied on the line-of-sight component as a proxy for the vertical magnetic field. Nonetheless, as a practical matter, understanding the accuracy limitation of the line-of-sight method in comparison to the vector method when synthetic vector magnetograms are available will reveal the relative reliability of helicity flux studies that used optical flow methods that rely exclusively on the line-of-sight magnetic field under the “best case scenario” for the tracking methods: the true vertical magnetic field is tracked and regions that contain interesting physics are known a-priori. By using the “ground truth” vertical magnetic field, the evaluation is biased to favor the performance of the DAVE method over what would be possible under realistic conditions where only the line-of-sight magnetic field is available. Figure 1: Optimization curves for the DAVE (left) and DAVE4VM (right). (top) The Spearman rank order (dashed black curve) and Pearson (solid black curve) correlations and slope (red curve) between $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ and $\Delta B_{z}/\Delta t$ for DAVE and $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ and $\Delta B_{z}/\Delta t$ for DAVE4VM as a function of window size. (bottom) The power (black) and helicity (red) as a function of window size; this assumes $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ for DAVE. The horizontal black and red dashed lines correspond to the ground truth power and helicity rate through the simulation slice from ANMHD. The vertical dot-dashed lines indicate the “optimal” window size of 23 pixels chosen for both DAVE and DAVE4VM. Table 1: Comparison between the DAVE and DAVE4VM over the 7013 pixels that satisfy $\left|\mbox{\boldmath{$B$}}\right|>370$ G in Figure 2. This mask contain regions of weak vertical field not considered in Welsch et al. (2007). | | DAVE (Assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$) | DAVE4VM ---|---|---|--- Quantities | Spearman | Pearson | Slope | Spearman | Pearson | Slope $u_{x}\,B_{z}$ | $U_{x}\,B_{z}$ | 0.34 | 0.57 | 0.15 | 0.88 | 0.89 | 0.80 $u_{y}\,B_{z}$ | $U_{y}\,B_{z}$ | 0.70 | 0.76 | 0.71 | 0.94 | 0.90 | 0.89 $v_{\perp{x}}$ | $V_{\perp{x}}$ | 0.87 | 0.85 | 0.91 | 0.89 | 0.88 | 0.94 $v_{\perp{y}}$ | $V_{\perp{y}}$ | 0.93 | 0.92 | 1.20 | 0.94 | 0.94 | 1.00 $v_{\perp{z}}$ | $V_{\perp{z}}$ | 0.17 | 0.28 | 0.07 | 0.80 | 0.80 | 0.79 $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ | $\Delta B_{z}/\Delta t$ | -0.85 | -0.95 | -0.97 | -0.79 | -0.94 | -0.97 $e_{\perp{x}}$ | $E_{\perp{x}}$ | 0.70 | 0.76 | 0.71 | 0.94 | 0.90 | 0.89 $e_{\perp{y}}$ | $E_{\perp{y}}$ | 0.34 | 0.57 | 0.15 | 0.88 | 0.89 | 0.80 $e_{\perp{z}}$ | $E_{\perp{z}}$ | 0.96 | 0.96 | 1.20 | 0.94 | 0.97 | 1.00 $s_{z}$ | $S_{z}$ | 0.20 | 0.12 | 0.04 | 0.88 | 0.83 | 0.71 Five different criteria were used to optimize window selection. Only symmetric windows were considered and some improvement in the results can be achieved by implementing asymmetric windows as in Welsch et al. (2007). Figure 1 shows the optimization curves for the DAVE (left) and DAVE4VM (right). The top plots show the Spearman rank order ($\rho$; dashed black) and Pearson ($C$; solid black curve) correlations and slope ($S$; red curve) between $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ and $\Delta B_{z}/\Delta t$ for DAVE and $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ and $\Delta B_{z}/\Delta t$ for DAVE4VM using pixels that satisfy $\left|\mbox{\boldmath{$B$}}\right|>370$ G. The gradients were computed with 5-point least-squares optimized derivatives Jähne (2004). Window sizes between 15 and 30 pixels provide the best balance for achieving performance approaching $C=\rho=S=-1$. Increasing the window size beyond 30 pixels continues to improve the Spearman and Pearson correlations but with diminishing returns while the slopes degrade significantly. The bottom plots show the power (black curve) and helicity rate (red curve) as a function of window size for the DAVE and DAVE4VM. For DAVE these calculations require the assumption that $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$. The horizontal black and red dashed lines correspond to the ground truth Poynting and helicity flux from ANMHD. The magnitude of these fluxes are maximum near 20 pixels with roughly uniform performance between 15 and 30 pixels. A window size111Only windows with odd numbers of pixels on each side of the window are possible in these implementations of the DAVE and DAVE4VM. of 23 pixels was chosen for both the DAVE and DAVE4VM as indicated by the vertical dot-dashed lines in Figure 1. These objective metrics for evaluation of global performance can be implemented without knowledge of ground truth. Only future tests with more realizations of synthetic data can reveal whether they are robust metrics for optimizing window choice. Table 3 presents a summary of the correlation coefficients on the mask $\left|\mbox{\boldmath{$B$}}\right|>370$ G characterizing the accuracy of the DAVE and DAVE4VM for the quantities discussed in this section. Table 2: Comparison of accuracy of the velocity estimates between the DAVE and DAVE4VM over the 7013 pixels that satisfy $\left|\mbox{\boldmath{$B$}}\right|>370$ G in Figure 2. This mask contains regions of weak vertical field not considered in Welsch et al. (2007). Here $\left\langle\left|\delta\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$ is the average fractional error, $\left\langle\delta\left|\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$ is the average error in magnitude, $C_{\mathrm{vec}}$ is the vector correlation, $C_{\mathrm{CS}}$ is the direction correlation, and $\left\langle\cos\theta\right\rangle_{W}$ is weighted direction cosine. ### 3.1 Flux Transport Vectors and Plasma Velocities Figure 2: Top left: Grey scale image of the vertical magnetic field $B_{z}$ overlaid with the horizontal magnetic field vectors $\mbox{\boldmath{$B$}}_{h}$ in aqua. Blue contours indicate smoothed neutral lines. Top right: Distribution of angles for $\mbox{\boldmath{$B$}}_{h}$ in the horizontal plane. Bottom: (red arrows) $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ for the DAVE (left) and $\mbox{\boldmath{$u$}}\,B_{z}$ for the DAVE4VM (right) plotted over ANMHD’s flux transport vectors $\mbox{\boldmath{$U$}}\,B_{z}$ (green arrows). Vectors are shown only for pixels in which $|B|>370$ G, and for clarity, only every third vector is displayed. Determining the flux transport vectors $\mbox{\boldmath{$f$}}=\mbox{\boldmath{$u$}}\,B_{z}$ is equivalent to determining the perpendicular plasma velocities $\mbox{\boldmath{$v$}}_{\perp}$. The accuracy of the flux transport vectors is critical for estimating other MHD quantities: perpendicular plasma velocities, electric field, helicity flux, Poynting flux, etc, since all of these quantities may be derived directly from flux transport vectors. #### 3.1.1 Flux Transport Velocities and Perpendicular Plasma Velocities The top left of Figure 2 shows the region of interest from the ANMHD simulations with grey scale image of vertical magnetic field overlaid with the horizontal magnetic field vectors $\mbox{\boldmath{$B$}}_{h}$ in aqua. The blue contours indicate smoothed neutral lines. The top right shows the distribution of angles for $\mbox{\boldmath{$B$}}_{h}$ in the horizontal plane. The horizontal magnetic field is largely aligned with the $\widehat{\mbox{\boldmath{$x$}}}$-axis as indicated by the aqua vectors in the left panel and the strong peak in the histogram near $\arctan\left(B_{y},B_{x}\right)\approx 0^{\circ}$ in the right panel.222$\arctan\left(y,x\right)\equiv\arctan\left(y/x\right)$. There is also significant alignment of the magnetic field with $\pm 60^{\circ}$ and alignment of weak fields with $-140^{\circ}$. The bottom panels show $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ from DAVE (left) and $\mbox{\boldmath{$u$}}\,B_{z}$ from DAVE4VM in red (right) and the flux transport vectors $\mbox{\boldmath{$F$}}=\mbox{\boldmath{$U$}}\,B_{z}$ from ANMHD in green. The improvement between the DAVE and DAVE4VM is manifest — finding a region where the DAVE4VM performs qualitatively worse than the DAVE is difficult. The DAVE4VM performs the worst in the region $140\\--160\times 150\\--170$ where the flux transport vectors run roughly anti-parallel to the horizontal magnetic field and there is little structure in the vertical component. Figure 3: Scatter plots of $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ for the DAVE (left) and $\mbox{\boldmath{$u$}}\,B_{z}$ for the DAVE4VM (right) versus ANMHD’s flux transport vectors, $\mbox{\boldmath{$F$}}=\mbox{\boldmath{$U$}}\,B_{z}$. Red and blue are used to distinguish $x$\- and $y$-components, respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown for both components of the flux transport vectors. Figure 3 shows scatter plots of the $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ from DAVE (left) and $\mbox{\boldmath{$u$}}\,B_{z}$ from DAVE4VM (right) versus the flux transport vectors $\mbox{\boldmath{$U$}}\,B_{z}$ from ANMHD. Red and blue are used to distinguish $x$\- and $y$-components, respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown for both components of the flux transport vectors. Both visually and quantitatively the DAVE4VM’s correlation with ANMHD is much higher than the DAVE’s. The correlation coefficients even match or exceed the correlation coefficients for the flux transport vectors from the DAVE and MEF reported for the restricted mask $\left|B_{z}\right|>370$ G in Welsch et al. (2007). In particular, DAVE does not accurately estimate the flux transport vectors in the $\widehat{\mbox{\boldmath{$x$}}}$-direction. The correlation coefficients for this $\widehat{\mbox{\boldmath{$x$}}}$-component of the flux transport vectors are $\rho_{x}=0.34$ and $C_{x}=0.57$ with a slope of $S_{x}=0.15$. Since that the $\widehat{\mbox{\boldmath{$x$}}}$-direction is the predominant direction of the horizontal magnetic for the ANMHD data, the low correlation coefficients suggest that DAVE is insensitive to flux emergence which is proportional to $v_{z}\,\mbox{\boldmath{$B$}}_{h}$. This will be discussed further in § 4. Figure 4: Scatter plots of the estimated perpendicular plasma velocities $\mbox{\boldmath{$v$}}_{\perp}$ from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus the perpendicular plasma velocities $\mbox{\boldmath{$V$}}_{\perp}$ from ANMHD. Red, blue, and black correspond to the $x$-, $y$\- and $z$-components respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) are shown for each component of the perpendicular plasma velocities. Figure 4 shows scatter plots of the estimated perpendicular plasma velocities $\mbox{\boldmath{$v$}}_{\perp}$ from the DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus ANMHD’s perpendicular plasma velocities $\mbox{\boldmath{$V$}}_{\perp}$. Red, blue, and black correspond to the $x$-, $y$-, and $z$-components respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown for each component of the perpendicular plasma velocities. The DAVE4VM’s correlation coefficients match or exceed the correlation coefficients for the perpendicular plasma velocities from the DAVE. Particularly striking is the DAVE4VM’s relatively higher correlation for the perpendicular vertical plasma velocity $\mbox{\boldmath{$v$}}_{\perp{z}}$ which exceeds the correlation for the DAVE by roughly $0.5\\--0.6$. The improvement in the DAVE4VM’s estimate is due to the explicit inclusion of horizontal magnetic fields and vertical flows. The flux transport and perpendicular plasma velocity estimates are further quantified by considering the metrics used by Schrijver et al. (2006), Welsch et al. (2007), and Metcalf et al. (2008). The fractional error between the estimated vector $f$ and the true vector $F$ at the $i$th pixel is $\left|\delta\smash{\widetilde{\mbox{\boldmath{$f$}}}}_{i}\right|\equiv\frac{\left|\mbox{\boldmath{$f$}}_{i}-\mbox{\boldmath{$F$}}_{i}\right|}{\left|\mbox{\boldmath{$F$}}_{i}\right|},$ (17a) whereas the fractional error in magnitude is $\delta\left|\smash{\widetilde{\mbox{\boldmath{$f$}}}}_{i}\right|\equiv\frac{\left|\mbox{\boldmath{$f$}}_{i}\right|-\left|\mbox{\boldmath{$F$}}_{i}\right|}{\left|\mbox{\boldmath{$F$}}_{i}\right|}.$ (17b) The moments of these error metrics or any quantity $q$ may be accumulated over the $N$ pixels within the masks (either $\left|\mbox{\boldmath{$B$}}\right|>370$ G or $\left|B_{z}\right|>370$ G) producing the average $\left\langle{q}\right\rangle\equiv\frac{1}{N}\sum_{i=1}^{N}q_{i},$ (18a) and the variance $\sigma^{2}_{q}\equiv\frac{1}{N-1}\sum_{i=1}^{N}\left(q_{i}-\left\langle q\right\rangle\right)^{2}.$ (18b) For perfect agreement between the estimates and the “ground truth” from ANMHD, $\left\langle\left|\delta\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$, $\left\langle\delta\left|\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$, and their associated variances would be zero. Two measures of directional error are considered, the vector correlation $C_{\mathrm{vec}}=\frac{\left\langle\mbox{\boldmath{$f$}}\cdot\mbox{\boldmath{$F$}}\right\rangle}{\sqrt{\left\langle\mbox{\boldmath{$f$}}^{2}\right\rangle\,\left\langle\mbox{\boldmath{$F$}}^{2}\right\rangle}},$ (19a) and the direction correlation $C_{\mathrm{CS}}=\left\langle\frac{\mbox{\boldmath{$f$}}\cdot\mbox{\boldmath{$F$}}}{\sqrt{\mbox{\boldmath{$f$}}^{2}\,\mbox{\boldmath{$F$}}^{2}}}\right\rangle\equiv\left\langle\cos\theta\right\rangle.$ (19b) Both metrics range from $-1$ for antiparallel vector fields, to $0$ for orthogonal vector fields, and to $1$ for parallel vector fields (perfect agreement). Table 3 shows these metrics for the DAVE and DAVE4VM over the mask $\left|\mbox{\boldmath{$B$}}\right|>370$ G. The DAVE4VM has fractional errors less than or equal to $0.4$ whereas the fractional errors for the DAVE exceed $0.7$ for both the flux transport vectors and the perpendicular plasma velocities. The average bias error in the magnitude is improved for the DAVE4VM over the DAVE. For the flux transport vectors the bias error in magnitude is $-0.09$ and $-0.47$ for DAVE4VM and DAVE respectively which corresponds to a factor of $5$ improvement. For the plasma velocity, the bias error in magnitude is 0.01 and 0.09 for DAVE4VM and DAVE respectively which corresponds to a factor of $9$ improvement. The vector correlation is larger for DAVE4VM than for DAVE. For DAVE4VM $C_{\mathrm{vec}}\gtrsim 0.9$ for both the flux transport velocity and the perpendicular plasma velocities. In contrast, for the DAVE there is a substantial difference in the accuracy of the flux transport vectors with $C_{\mathrm{vec}}=0.61$ and the perpendicular plasma velocities with $C_{\mathrm{vec}}=0.81$. Finally the directional errors are smaller for the DAVE4VM than for the DAVE. For the DAVE4VM $C_{\mathrm{CS}}\gtrsim 0.9$ for both the flux transport vectors and the perpendicular plasma velocities. Again, for the DAVE there is a substantial difference in the accuracy of the flux transport vectors with $C_{\mathrm{CS}}=0.52$ and the perpendicular plasma velocities with $C_{\mathrm{CS}}=0.77$. Figure 5: Histograms of the angles between $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ and $\mbox{\boldmath{$U$}}\,B_{z}$ for the DAVE (left) and between $\mbox{\boldmath{$u$}}\,B_{z}$ and $\mbox{\boldmath{$U$}}\,B_{z}$ for the DAVE4VM (right). The mean (bias) and standard deviation are reported in the upper left-hand corners. The direction correlation $C_{\mathrm{CS}}$ is difficult to translate into average angular error because it is a nonlinear function of $\theta$ and does not indicate whether the estimated vectors “lead” or “lag” the “ground truth” on average. For the 2D flux transport vectors the moments of the distribution of angular errors $\theta=\arctan\left[\left(\mbox{\boldmath{$u$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$U$}}\right)_{z},\mbox{\boldmath{$u$}}\cdot\mbox{\boldmath{$U$}}\right],$ (20) can be more informative. Figure 5 shows histograms of the angles between $\mbox{\boldmath{$u$}}\,B_{z}$ and $\mbox{\boldmath{$U$}}\,B_{z}$ for the DAVE (left) and DAVE4VM (right). This is a quantitative estimate of the errors in directions of the flux transport vectors. The DAVE4VM represents a dramatic improvement over the DAVE. The DAVE4VM produces a nearly unimodal distribution peaked near $0^{\circ}$, whereas the DAVE produces a multi-peaked distribution with the largest peak at $80^{\circ}$ and a variance that is more than twice as large as DAVE4VM. Metrics such as (17) and (19) weight all estimates equally. To address this Metcalf et al. (2008) suggested weighting the errors. For example, the weighted direction cosine between an inferred vector and the ground truth vector may be defined as $\left\langle\cos\theta\right\rangle_{W}\equiv\frac{\left\langle{W}\,\cos\theta\right\rangle}{\left\langle{W}\right\rangle}$ (21) where $W$ represents weights. For the flux transport velocity, the errors in the orientation of $u$ are more important where $\left|\mbox{\boldmath{$U$}}\,B_{z}\right|$ is large and less important where $\left|\mbox{\boldmath{$U$}}\,B_{z}\right|$ is small which suggests a weighting factor $W_{i}=\left|\left(\mbox{\boldmath{$U$}}\,B_{z}\right)\right|$. For perfect agreement $\left\langle\cos\theta\right\rangle_{W}=1$. The weighted direction cosines for the flux transport vectors and the plasma velocities are reported in Table 3. Comparing the values of $C_{\mathrm{CS}}$ and $\left\langle\cos\theta\right\rangle_{W}$ demonstrates that weighting the direction cosine improves the apparent performance of DAVE but the results for DAVE4VM are essentially unchanged. This suggests that DAVE4VM estimates velocities better than DAVE in regions of weak flux transport. #### 3.1.2 Parallel Velocity Under ideal conditions, the magnetic field is only affected by $\mbox{\boldmath{$\nabla$}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$v$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$B$}}\right)$. Consequently, only the inductive potential $\phi$ in (7) may be uniquely determined from the evolution of $B$ alone. The electrostatic potential $\psi$ must and can be estimated with additional judicious assumptions. These additional assumptions correspond to the minimum photospheric velocity consistent with (7) for the global method MEF and to the prescribed affine form of the local plasma velocity for the local method DAVE4VM. The constraint of the affine velocity profile permits DAVE4VM to determine the electrostatic potential $\psi$ from the nonlocal structure of the inductive potential $\phi$. DAVE4VM uses unambiguous “pieces” of the plasma velocity within the window aperture to reconstruct the total plasma velocity at the center of the aperture. Within the notation of the Helmholtz decomposition, DAVE4VM estimates the local electrostatic field from the structure of the nonlocal inductive field within the aperture window by imposing a smoothness constraint on the velocity (the affine velocity profile). The accuracy of this estimate for the electrostatic field depends on the validity of the local affine velocity profile and the amount of structure in the local magnetic field; local methods cannot detect motion in regions of uniform magnetic field. Figure 6: Schematic diagram of a uniform plasma flow across a diverging magnetic field above the photosphere at $h=0$. The black arrows indicate the strength and direction of the magnetic field, the red arrows indicate the direction of the spatially uniform total plasma velocity $v$, and the blue arrows indicate the magnitude and direction of the perpendicular plasma velocity $\mbox{\boldmath{$v$}}_{\perp}$. The aperture in the photosphere is indicated by the gray box. While researchers have widely recognized that estimating the electrostatic potential $\psi$ from (5a) requires additional assumptions, they have not generally recognized that the parallel velocity $v_{\parallel}$ may be estimated by the analogous arguments. In the absences of a reference flow, the MEF constrains the velocity to be perpendicular to the magnetic field: $v_{\parallel}=0$. In contrast to the velocity estimated by DAVE4VM is not constrained to be perpendicular to the local magnetic field. Instead, DAVE4VM fits an affine velocity model to the magnetic induction equation in an aperture window. This affine velocity model couples the dynamics across pixels within the window aperture. If there is sufficient structuring in the direction of the magnetic field within the aperture, i.e., the perpendicular plasma velocity points in different directions at different pixels within the aperture, then DAVE4VM can resolve the ambiguity in the field-aligned component of the plasma velocity at the center of the aperture. Consider the simplified two-dimensional situation illustrated by the schematic diagram in Figure 6 of spatially uniform plasma flow across a diverging magnetic field above the photosphere at $h=0$. The black arrows indicate the magnitude and direction of the magnetic field, the red arrows indicate the direction of the spatially uniform total plasma velocity $v$, and the blue arrows indicate the magnitude and direction of the perpendicular plasma velocity $\mbox{\boldmath{$v$}}_{\perp}$. The aperture in the photosphere is indicated by the gray box. Within the aperture, the perpendicular plasma velocity captures a different component of the total plasma velocity at different locations; this is a consequence of the structuring of the magnetic field. Under the smoothness assumption of a uniform velocity profile, the velocity along the magnetic field in the $\widehat{\mbox{\boldmath{$z$}}}$-direction at $x=0$ may be determined from the components of the perpendicular plasma velocity in the $\widehat{\mbox{\boldmath{$z$}}}$ direction at other locations within the aperture. Using the $N$ pixels in the window aperture results in an overdetermined system for the total plasma velocity: $\underbrace{\left[\begin{array}[]{cc}\frac{B_{z}^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}&-\frac{B_{x}\left(\mbox{\boldmath{$x$}}_{1}\right)\,B_{z}\left(\mbox{\boldmath{$x$}}_{1}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}\\\ -\frac{B_{x}\left(\mbox{\boldmath{$x$}}_{1}\right)\,B_{z}\left(\mbox{\boldmath{$x$}}_{1}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}&\frac{B_{x}^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{1}\right)}\\\ \vdots&\vdots\\\ \frac{B_{z}^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}&-\frac{B_{x}\left(\mbox{\boldmath{$x$}}_{N}\right)\,B_{z}\left(\mbox{\boldmath{$x$}}_{N}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}\\\ -\frac{B_{x}\left(\mbox{\boldmath{$x$}}_{N}\right)\,B_{z}\left(\mbox{\boldmath{$x$}}_{N}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}&\frac{B_{x}^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}{B^{2}\left(\mbox{\boldmath{$x$}}_{N}\right)}\end{array}\right]}_{\mbox{\boldmath{$\mathsf{D}$}}}\,\left(\begin{array}[]{cc}v_{x}\\\ v_{z}\end{array}\right)=\underbrace{\left[\begin{array}[]{cc}v_{\perp{x}}\left(\mbox{\boldmath{$x$}}_{1}\right)\\\ v_{\perp{z}}\left(\mbox{\boldmath{$x$}}_{1}\right)\\\ \vdots\\\ v_{\perp{x}}\left(\mbox{\boldmath{$x$}}_{N}\right)\\\ v_{\perp{z}}\left(\mbox{\boldmath{$x$}}_{N}\right)\end{array}\right]}_{{\mbox{\boldmath{$\mathsf{d}$}}}},$ (22a) which has the solution Golub & Van Loan (1980) $\left(\begin{array}[]{cc}\widehat{v}_{x}\\\ \widehat{v}_{z}\end{array}\right)=\left({\mbox{\boldmath{$\mathsf{D}$}}}^{*}\,{\mbox{\boldmath{$\mathsf{D}$}}}\right)^{-1}\,{\mbox{\boldmath{$\mathsf{D}$}}}^{*}{\mbox{\boldmath{$\mathsf{d}$}}}.$ (22b) Note that ${\mbox{\boldmath{$\mathsf{D}$}}}^{*}\,{\mbox{\boldmath{$\mathsf{D}$}}}$ is analogous to $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle$ and ${\mbox{\boldmath{$\mathsf{D}$}}}^{*}{\mbox{\boldmath{$\mathsf{d}$}}}$ is analogous to $\left\langle\mbox{\boldmath{$b$}}\right\rangle$ in (15). This pedagogical example illustrates how DAVE4VM may analogously estimate the field-aligned plasma velocity for the more general case of a spatially variable plasma flow in an inhomogeneous magnetic field for (15). The accuracy of the estimate of the parallel velocity will be limited by the structuring in direction of the magnetic field within the aperture; if the magnetic field has a uniform orientation in the aperture window, no useful estimate of the field- aligned plasma velocity can be made from the magnetic measurements alone. The quality of the estimate may be assessed with the conditioning of ${\mbox{\boldmath{$\mathsf{D}$}}}^{*}\,{\mbox{\boldmath{$\mathsf{D}$}}}$ in (22b) for the pedagogical example or $\left\langle\mbox{\boldmath{$\mathsf{A}$}}\right\rangle$ in (15) for the full system. Figure 7: Scatter plots of (left) the estimated parallel plasma velocities $v_{\parallel}$ from DAVE4VM versus the parallel plasma velocities $V_{\parallel}$ from ANMHD and (right) the estimated total plasma velocity $v$ from DAVE4VM versus the total plasma velocities $V$ from ANMHD. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown. The pairs of numbers represent correlations between $v$ and $V$ and $\mbox{\boldmath{$v$}}_{\perp}$ and $V$. Figure (7) shows scatter plots of (left) the estimated parallel plasma velocities $v_{\parallel}$ from DAVE4VM versus the parallel plasma velocities $V_{\parallel}$ from ANMHD and (right) the estimated total plasma velocity $v$ from DAVE4VM versus the total plasma velocities $V$ from ANMHD. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown. The comma-separated pairs of numbers in the right plot, corresponding to correlations between $v$ and $V$ and $\mbox{\boldmath{$v$}}_{\perp}$ and $V$ respectively, represent the relative improvement in total velocity estimate over the simple null hypothesis $H_{0}:\mbox{\boldmath{$v$}}=\mbox{\boldmath{$v$}}_{\perp}$ that the total plasma velocity is the perpendicular plasma velocity. The correlation of the $\widehat{\mbox{\boldmath{$x$}}}$-component of total velocity is significantly improved over the null hypothesis $H_{0}$. This improvement id interesting since the horizontal magnetic field is predominantly aligned with the $x$-axis (See Figure 2). The correlation of the $\widehat{\mbox{\boldmath{$y$}}}$-component of total velocity is slightly worse than the null hypothesis. Finally, the correlation of the $\widehat{\mbox{\boldmath{$z$}}}$-component of total velocity is mixed with the Spearman correlation $\rho$ slightly worse than the null hypothesis and the Pearson correlation $C$ slightly better than the null hypothesis. However, the slopes of all three components are improved over the null hypothesis. The significance of the correlations in the left plot may be tested against the null hypothesis $H_{0}:\rho=0$ by the Fisher permutation test. Fieller et al. (1957) have demonstrated with analysis backed Monte-Carlo simulation that Fisher’s $z$-transform of the correlation coefficient $z_{\mathrm{S}}\left(\rho\right)=\frac{1}{2}\,\log\left|\frac{1+\rho}{1-\rho}\right|,$ (23) produces approximately normally distributed values. For example, permuting the values of $v_{\parallel}$ and $V_{\parallel}$ 10,000 times generates the null hypothesis distribution with $\left\langle{z_{\mathrm{S}}}\right\rangle=0.00\pm 0.01$. The Spearman correlation coefficient $\rho=0.53$ has a $z$-transform of $z_{\mathrm{S}}\left(0.53\right)=0.60$ which is roughly 50 standard deviations from the mean of the null distribution indicating that the parallel velocity correlation is statistically significant and not due to sampling error. However, the correlation $\rho=0.53$ is small and the parallel velocity estimates may not be scientifically significant for accurately predicting the parallel velocity. The plasma velocities may be further constrained by introducing Doppler velocities, but this is beyond the scope of the present discussion. #### 3.1.3 Are $\mbox{\boldmath{$v$}}_{h}$ and $v_{z}$ Redundant? Figure 8: Scatterplots for the $\widehat{\mbox{\boldmath{$x$}}}$ (left) and $\widehat{\mbox{\boldmath{$y$}}}$ (right) components of (2a). The scatterplots indicate the lack of correlation between the terms describing shearing motion ${B}_{z}\,\mbox{\boldmath{$V$}}_{h}$ and emergence $V_{z}\,\mbox{\boldmath{$B$}}_{h}$. Red points indicate the results for DAVE4VM and blue points indicate the results for ANMHD. The Spearman rank order ($\rho$) and Pearson ($C$) correlations between the two terms are very low for both components of the flux transport velocities. DAVE4VM has incorporated an additional component of the velocity over DAVE by introducing three additional variables $\widehat{w}_{0}$, $\widehat{w}_{x}$ and $\widehat{w}_{y}$. Consequently, one may reasonably wonder “are the terms $\mbox{\boldmath{$v$}}_{h}\,B_{z}$ and $v_{z}\,\mbox{\boldmath{$B$}}_{h}$ redundant for DAVE4VM?” The answer is a clear “No” for the ANMHD data. Equation (2a) is composed of two terms ${B}_{z}\,\mbox{\boldmath{$V$}}_{h}$ and $V_{z}\,\mbox{\boldmath{$B$}}_{h}$ describing shearing motion and emergence respectively. Figure 8 shows scatterplots of the two terms for the $\widehat{\mbox{\boldmath{$x$}}}$ (left) and $\widehat{\mbox{\boldmath{$y$}}}$ (right) components of (2a). The scatterplots indicate the lack of correlation between the terms describing shearing motions ${B}_{z}\,\mbox{\boldmath{$V$}}_{h}$ and emergence $V_{z}\,\mbox{\boldmath{$B$}}_{h}$. Red points indicate the results for DAVE4VM and blue points indicate the results for ANMHD. The Spearman rank order ($\rho$) and Pearson ($C$) correlations between the two terms, summarized in Table 3.1.3, are very low for both components of the flux transport velocities from DAVE4VM or ANMHD. These terms describe different physics, that are uncorrelated, and which require independent variables to describe. Table 3: The Spearman rank order ($\rho$) and Pearson ($C$) correlations between the terms in the flux transport velocity describing shearing motion and flux emergence. ### 3.2 Induction Equation and Electric Fields Figure 9: Scatter plots of $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ from the DAVE (left) and $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ from the DAVE4VM (right) versus $\Delta B_{z}/\Delta t$ from ANMHD. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown. Figure 9 shows $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$\vartheta$}}\,B_{z}\right)$ from the DAVE (left) and $\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ from DAVE4VM (right) versus $\Delta B_{z}/\Delta t$ from ANMHD. The derivatives for these plots were estimated from 5-point optimized least squares. These plots indicate how well the two methods satisfy the MHD induction equation globally. The DAVE has higher correlations than the DAVE4VM but the slopes are equivalent. For the DAVE4VM, the most significant deviations from the MHD induction equation occur near $\Delta B_{z}/\Delta t\approx 0$. Neither the DAVE nor DAVE4VM satisfy the induction equation exactly. This is by design, because real magnetogram data are likely to contain significant noise which will contaminate velocity estimates if the induction equation is satisfied exactly. Furthermore, how well a method satisfies the induction equation will generally depend on the differencing template. Consequently, if the velocity estimates are to be used as boundary values for ideal MHD coronal field models, then the velocities of any method will have to be adjusted to satisfy the induction equation on the differencing template implemented by the simulation. Using the Helmholtz decomposition (7), the inductive potential may be computed for the simulation directly from the magnetogram sequence (on the simulation differencing template) $\partial_{t}{B_{z}}=\nabla_{h}^{2}\phi,$ (24a) and the electrostatic potential may be derived from the flux transport vectors determined by the optical flow method Welsch et al. (2004) $\nabla_{h}^{2}\psi=\widehat{\mbox{\boldmath{$z$}}}\cdot\left[\mbox{\boldmath{$\nabla$}}\mbox{\boldmath{$\times$}}\left(\mbox{\boldmath{$u$}}\,B_{z}\right)\right].$ (24b) Incorporating photospheric velocity estimate into boundary conditions for a coronal MHD simulation, in a minimally consistent way with the normal component of the magnetic induction equation, requires solving two Poisson equations on the photospheric boundary using the differencing template of the MHD code. Figure 10: Scatter plots of the estimated perpendicular electric field $\mbox{\boldmath{$e$}}_{\perp}$ from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and from DAVE4VM (right) versus the electric field $\mbox{\boldmath{$E$}}_{\perp}$ from ANMHD. Red, blue, and black correspond to the $x$-, $y$-, and $z$-components respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown for each component of the electric field. Figure 10 shows scatter plots of the estimated perpendicular electric fields $\mbox{\boldmath{$e$}}_{\perp}$ from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and from DAVE4VM (right) versus the electric fields $\mbox{\boldmath{$E$}}_{\perp}$ from ANMHD. Red, blue, and black correspond to the $x$-, $y$-, and $z$-components, respectively. The nonparametric Spearman rank-order correlation coefficients ($\rho$) and Pearson correlation coefficients ($C$) are shown for each component of the electric field. On the present mask the DAVE4VM estimates improve or essentially match the correlation and slopes of the DAVE’s estimates for all three components of the electric field. Particularly dramatic is the improvement in the $\widehat{\mbox{\boldmath{$y$}}}$ component of the electric field which the DAVE does not estimate accurately either on the present mask $\left|\mbox{\boldmath{$B$}}\right|>370$ G or the restricted mask $\left|B_{z}\right|>370$ G Welsch et al. (2007). ### 3.3 Poynting and Helicity Fluxes Figure 11: Scatter plots of the estimated Poynting flux from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus the Poynting flux from ANMHD. The nonparametric Spearman rank- order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown, as is the ratio of the integrated estimated Poynting flux to the integrated ANMHD Poynting flux. Démoulin & Berger (2003) show that the Poynting flux can be expressed concisely in terms of the flux transport vectors $\mbox{\boldmath{$u$}}\,B_{z}$ $s_{z}\left(\mbox{\boldmath{$x$}}\right)=-\frac{1}{4\,\pi}\,\mbox{\boldmath{$B$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=-\frac{\mbox{\boldmath{$B$}}_{h}\cdot\left(\mbox{\boldmath{$u$}}\,\,B_{z}\right)}{4\,\pi}.$ (25) Figure 11 shows scatterplots of the estimated Poynting flux $s_{z}$ from the DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus ANMHD’s Poynting flux $S_{z}$. The correspondence for DAVE4VM, or lack there of for DAVE, indicates the accuracy of the velocity estimates in the direction of the horizontal magnetic field $\mbox{\boldmath{$B$}}_{h}$. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown, as is the ratio of the integrated estimated Poynting flux to the integrated ANMHD Poynting flux $\mathcal{R}_{{s}_{z}}=\sum{s}_{z}/\sum{S}_{z}$. The DAVE4VM’s estimate of Poynting flux is a significant improvement over the DAVE’s. The correlations have improved by roughly a factor of $4\\--6$, the slope has improved by nearly a factor of 18, and the ratio of the totals has improved by nearly a factor of 5. Again, DAVE does not reliably estimate the flux transport velocity in the direction of the horizontal magnetic field $\mbox{\boldmath{$B$}}_{h}$ suggesting that DAVE is insensitive to flux emergence which is proportional to $v_{z}\,\mbox{\boldmath{$B$}}_{h}$. The “ground truth” total power through the mask is $dP/dt=\sum{S_{z}}=7.7\times 10^{28}\,\mathrm{ergs/s}$. Figure 12: Scatter plots of the estimated helicity flux DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus ANMHD’s helicity flux. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown, as is the ratio of the integrated estimated helicity flux to the integrated ANMHD helicity flux. Démoulin & Berger (2003) show that the gauge-invariant helicity flux Berger & Field (1984) can be expressed concisely in terms of the flux transport vectors $\mbox{\boldmath{$u$}}\,B_{z}$ $g_{A}\left(\mbox{\boldmath{$x$}}\right)=-2\,\mbox{\boldmath{$A$}}_{p}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=-2\,\mbox{\boldmath{$A$}}_{p}\cdot\left(\mbox{\boldmath{$u$}}\,B_{z}\right)$ (26) where $\mbox{\boldmath{$A$}}_{p}=\widehat{\mbox{\boldmath{$z$}}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$\nabla$}}\Phi_{p}$ is the potential reference field (with zero helicity) which satisfies $\widehat{\mbox{\boldmath{$z$}}}\cdot\left(\mbox{\boldmath{$\nabla$}}\mbox{\boldmath{$\times$}}\mbox{\boldmath{$A$}}_{p}\right)=\nabla_{h}^{2}\Phi_{p}=B_{z},$ (27) and $\mbox{\boldmath{$\nabla$}}\cdot\mbox{\boldmath{$A$}}_{p}=\widehat{\mbox{\boldmath{$z$}}}\cdot\mbox{\boldmath{$A$}}_{p}=0$. To estimate the helicity flux density, $\Phi_{p}$ was computed on a $257\times 257$ square centered on the region of interest with Dirichlet boundary conditions using MUDPACK Adams (1993). While interpretation of maps of helicity flux $g_{A}\left(\mbox{\boldmath{$x$}}\right)$ through the photosphere is problematic Pariat et al. (2005, 2007), a comparison of $g_{A}\left(\mbox{\boldmath{$x$}}\right)$ estimated from the DAVE or DAVE4VM verses $G_{A}\left(\mbox{\boldmath{$x$}}\right)$ calculated from ANMHD indicates the accuracy of the estimated flux transport vectors in the direction of the vector potential. Figure 12 shows scatter plots of the estimated helicity flux from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (left) and DAVE4VM (right) versus ANMHD’s helicity flux. The nonparametric Spearman rank-order correlation coefficients ($\rho$) and Pearson correlation coefficients ($C$) are shown, as is the ratio of the integrated estimated helicity flux to the integrated ANMHD helicity flux $\mathcal{R}_{g_{A}}=\sum{g_{A}}/\sum{G_{A}}$. The DAVE4VM’s estimates represent a significant improvement over the DAVE’s, improving the correlation coefficients by roughly $0.1$ and the slope by $0.2$. Furthermore, the ratio of totals has improved by roughly a factor of 4 from $0.22$ for the DAVE to $0.94$ for the DAVE4VM. The “ground truth”333This estimate differs by about 10% from the helicity estimate in Welsch et al. (2007). The discrepancy is caused by the different methodologies and boundaries used for computing the vector potential $\mbox{\boldmath{$A$}}_{p}$. MUDPACK was used in this study with (27) whereas Welsch et al. (2007) used a Green’s function scheme to compute $\mbox{\boldmath{$A$}}_{p}$. helicity injected through the surface is $dH_{A}/dt=\sum{G_{A}}=-2.8\times 10^{37}\,\mathrm{Mx}^{2}/\mathrm{s}$. ## 4 Discussion and Conclusions Table 4: Comparison between the DAVE and DAVE4VM over the 3815 pixels that satisfy $\left|B_{z}\right|>370$ G in Figure 2. This corresponds roughly to the mask used in Welsch et al. (2007). Table 5: Comparison of accuracy of the velocity estimates between the DAVE and DAVE4VM over the 3815 pixels that satisfy $\left|B_{z}\right|>370$ G in Figure 2. This corresponds roughly to the mask used in Welsch et al. (2007). For completeness, Tables 4 and 4 provide a summary of metrics and correlation coefficients for the DAVE and DAVE4VM on the original mask $\left|B_{z}\right|>370$ G used by Welsch et al. (2007) in the same format as in Tables 3 and 3. MEF performed the best overall in the original study by Welsch et al. (2007) although there were some metrics where the DAVE outperformed MEF such as in the accuracy of the plasma velocities listed in Table 4 (compare with Figure 8 in Welsch et al. (2007)). The DAVE4VM’s estimates are a substantial improvement over the results of the DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ on this mask. Comparing the rank-order Spearman correlation coefficients for the flux transport vectors, perpendicular plasma velocity, and electric field in Table 4, the DAVE4VM equals or out-performs MEF. Particularly, the DAVE4VM’s estimate of the vertical perpendicular plasma velocity is substantially better than MEF with a rank-order of 0.76 in the former and 0.61 in the latter case. Accurate vertical flows are necessary to diagnose flux emergence and accurately estimate the helicity flux. The one area where MEF exhibits superiority is in the estimate of the Poynting flux where the DAVE4VM captures 76% and MEF captures 100% Welsch et al. (2007). The fractional errors $\left\langle\left|\delta\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$ and $\left\langle\delta\left|\smash{\widetilde{\mbox{\boldmath{$f$}}}}\right|\right\rangle$ are substantially lower than the DAVE for both the flux transport vectors and plasma velocities. The DAVE had the largest vector correlation $C_{\mathrm{vec}}$ and the direction correlation $C_{\mathrm{CS}}$ in the original study and the DAVE4VM improves over this performance exhibiting correlation coefficients of roughly $0.9$. The improvement for the flux transport vectors is particularly dramatic. The plasma velocities are more accurate and exhibit considerably less bias than those reported for MEF in Welsch et al. (2007). The DAVE4VM offers some minor advantages over MEF. The DAVE4VM is somewhat faster than MEF; the DAVE4VM(DAVE) requires 30(10) seconds to process444The routines were all coded in Interactive Data Language IDL (2002) and the computations were performed on a dual processor AMD Opteron 240 running at 1.4 GHz with a one megabyte memory cache and ten gigabytes of Random Access Memory. the full $288\times 288$ pixel frame from ANMHD whereas MEF requires roughly 10 minutes to converge on a reduced mask of the ANMHD data (private communication with Belur Ravindra). The DAVEVM is local and directly estimates velocities across neutral lines and across broader weak field regions whereas MEF is a iterative global method that requires judicious choice of boundaries to ensure convergence. In concert, the DAVE4VM’s velocity estimate might be used with MEF either as an initial guess for the electrostatic potential $\psi$ via (24b) or as ancillary inaccurate velocity measurements in the MEF variational term that constrains the photospheric plasma velocities Longcope (2004). Since the DAVE4VM is fast and does not require supervision beyond choosing a window size (and even this could be automated according to the criteria discussed in § 3), this approach is appropriate for real-time monitoring of helicity and energy fluxes through the photosphere from observatories such as Solar Dynamics Observatory. What is responsible for the DAVE4VM’s improved performance? The only differences between the DAVE and DAVE4VM are the terms $\mathsf{s}_{ij}$ in the structure tensor (14) that describe the local structure of the horizontal magnetic fields necessary for the description of vertical flows. There are two circumstances when line-of-sight methods such as the DAVE, LMSAL’s LCT, and FLCT will produce accurate estimates of the flux transport velocity: 1. 1. $\mbox{\boldmath{$B$}}_{h}=0$: The magnetic field is purely vertical and $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}=\mbox{\boldmath{$v$}}_{\perp{h}}$. If the horizontal magnetic fields and their associated derivatives are zeroed, the DAVE and DAVE4VM produce identical flux transport and perpendicular plasma velocity estimates. The DAVE is consistent with the assumption that the magnetic field is purely vertical $\lim_{\mbox{\boldmath{$B$}}_{h}\rightarrow 0}\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}\right).$ (28a) * 2. $v_{z}=0$: There are no net upflow/downflows and $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}\neq\mbox{\boldmath{$v$}}_{\perp{h}}$. In this situation, there must be projected vertical flows along the magnetic field to cancel any projected vertical flow perpendicular to the magnetic field with $v_{\perp{z}}=-v_{\parallel{z}}=-B_{z}\,\left(\mbox{\boldmath{$v$}}_{h}\cdot\mbox{\boldmath{$B$}}_{h}\right)/B^{2}$. Consequently, $\mbox{\boldmath{$v$}}_{\parallel}\neq 0$. The DAVE is consistent with the assumption that there are no vertical flows $v_{z}=0$: $\lim_{v_{z}\rightarrow 0}\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}-v_{z}\,\mbox{\boldmath{$B$}}_{h}\right)=\partial_{t}B_{z}+\mbox{\boldmath{$\nabla$}}_{h}\cdot\left(B_{z}\,\mbox{\boldmath{$v$}}_{h}\right).$ (28b) Both limits (28a) and (28b) are isomorphic with (6). By induction, (6) is consistent with the assumptions leading to (28a) and (28b). Since DAVE does not consider corrections $\mathsf{s}_{ij}$ due to the horizontal magnetic field, $\vartheta$ should generally be considered a biased estimate of the horizontal plasma velocity $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$ and not the flux transport velocity! Formally the alternative hypothesis $H_{1}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$ may be tested against the null hypothesis $H_{0}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ of Démoulin & Berger (2003). Figure 13: Scatter plots of the estimated velocities $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ from DAVE versus the horizontal plasma velocities $\mbox{\boldmath{$V$}}_{h}\,B_{z}$ from ANMHD. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown. The null hypothesis $H_{0}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ is represented by the left panel in Figure 3. The alternative hypothesis, represented by $H_{1}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$, is characterized by the scatter plot of the estimated velocities $\mbox{\boldmath{$\vartheta$}}\,B_{z}$ for DAVE versus the horizontal plasma velocities $\mbox{\boldmath{$V$}}_{h}\,B_{z}$ from ANMHD in Figure 13. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are all significantly better for the alternative hypothesis than for the null hypothesis. The null hypothesis that the velocities inferred by DAVE represent the flux transport velocities may be rejected in favor555This does not imply that the alternative hypothesis is correct. of the alternative hypothesis $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$. Figure 14: (left) Scatter plots of the estimated Poynting flux (red) and helicity flux (blue) from DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ versus the Poynting and helicity flux from ANMHD. (right) Scatter plots of the estimated Poynting flux (red) and helicity flux (blue) combining DAVE assuming $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$ with the emergence term from ANMHD versus the Poynting and helicity flux from ANMHD. The nonparametric Spearman rank-order correlation coefficients ($\rho$), Pearson correlation coefficients ($C$), and slopes ($S$) estimated by the least absolute deviation method are shown. These results explain why $V_{\perp{z}}$ and $v_{\perp{z}}$ are poorly correlated for the DAVE in Figure 4 and the slope between them is nearly zero — the DAVE is consistent with the assumption $v_{z}=0$ when $\mbox{\boldmath{$B$}}_{h}\neq 0$. Generally, in regions of flux emergence, the accuracy $v_{\perp{z}}$ is critical for estimating the flux transport vectors which in turn is critical for estimating the helicity and Poynting fluxes. When horizontal magnetic fields and vertical flows are present, the flux transport vectors estimated from methods that rely exclusively on the line-of-sight or vertical component (DAVE, LMSAL’s LCT, FLCT) cannot be trusted to provide the total fluxes. This is particularly true along neutral lines where flux is emerging or submerging! Under the best case scenarios, only the shearing or “horizontal fluxes”666Démoulin & Berger (2003) terms these fluxes the “tangential fluxes” but “horizontal” is more appropriate in the context of Welsch’s terminology used in this paper. across the photosphere. $\left.\frac{dp}{dt}\right|_{h}=-\frac{1}{4\,\pi}\,\int_{S}{dx^{2}}\,\mbox{\boldmath{$B$}}_{h}\cdot\left(\mbox{\boldmath{$v$}}_{h}\,B_{z}\right),$ (29a) and $\left.\frac{dh_{A}}{dt}\right|_{h}=-2\,\int_{S}{dx^{2}}\,\mbox{\boldmath{$A$}}_{p}\cdot\left(\mbox{\boldmath{$v$}}_{h}\,B_{z}\right),$ (29b) may be estimated from the line-of-sight tracking methods. In the best case scenarios only the shearing fluxes are captured by line-of-sight tracking methods in partial agreement with the Ansatz of Chae (2001) and in disagreement with the geometrical arguments of Démoulin & Berger (2003) who argue that line-of-sight tracking methods capture both the shearing and emergence. Again, for the energy and helicity, the alternative hypothesis $H_{1}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$ (shearing) may be tested against the null hypothesis $H_{0}:\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$u$}}$ (shearing and emergence). The left panel of Figure 14, representing the null hypothesis is a combination of the left-hand panels of Figure 11 and Figure 12. The right panel of Figure 14, representing the alternative hypothesis, combines the shearing term estimated from DAVE with the emergence term from ANMHD. Generally, in the presence of vertical flows and horizontal magnetic fields, line-of-sight tracking methods do not accurately capture the complete footpoint dynamics and the null hypothesis that the velocities inferred by DAVE represent the flux transport velocities may be rejected in favor of the alternative hypothesis $\mbox{\boldmath{$\vartheta$}}=\mbox{\boldmath{$v$}}_{h}$. The implementation of vector magnetograms in optical flow methods presents practical challenges. First, the transverse magnetic field components are known to be noisier than the line-of-sight component and the noise variance will likely change from pixel to pixel due to variable photon statistics (heteroscedastic errors). Second the line-of-sight component and transverse components are determined from different polarizations and require inter- calibration. Third, the orientation of the transverse component is ambiguous by $180^{\circ}$. The first issue may be addressed within the total least squares framework discussed by Schuck (2006), Branham Jr. (1999) and others (See references in Schuck, 2006). The main obstacle to resolving the first issue is estimating a covariance matrix for the structure tensor $\left\langle\SS\right\rangle$. The second and third issues both may be interpreted as inter-calibration bias where the estimated horizontal magnetic field $\widehat{\mbox{\boldmath{$B$}}}_{h}=\alpha\,\mbox{\boldmath{$B$}}_{h}$ is proportional to the true horizontal magnetic field $\mbox{\boldmath{$B$}}_{h}$; these errors are not random. The flux transport velocities estimated from DAVE4VM are robust to overall inter-calibration errors which include the $180^{\circ}$ ambiguity resolution errors. Changing the overall magnitude or sign of $\mbox{\boldmath{$B$}}_{h}$ has no effect on the flux transport velocities because (5a) is invariant with respect to the transformation $v_{z}\rightarrow{v_{z}}/\alpha$ and $\mbox{\boldmath{$B$}}_{h}\rightarrow\alpha\mbox{\boldmath{$B$}}_{h}$ (private communication with Pascal Demoulin). However, the estimated vertical perpendicular plasma velocity and vertical perpendicular electric field will be anti-correlated with the ground truth when $\alpha<0$. Nonetheless, an overall rescaling of the horizontal magnetic field will have no effect on the helicity flux. However, the Poynting flux will be incorrect by a factor of $\alpha$ including perhaps a sign error because of the rescaling horizontal magnetic field which is inherent in the energy estimate (25). More troublesome are the effects of spatially varying bias errors in inter-calibration or ambiguity resolution. The consequences of these errors, particularly along the boundaries between proper and improper ambiguity resolution, are presently unknown and should be investigated with future end-to-end analysis of synthetic magnetograms. However, local methods such as DAVE4VM are probably more robust than global methods to spatially dependent errors in inter- calibration or ambiguity resolution because local methods inherently localize the effect of bias errors by isolating subregions with the window aperture whereas global methods couple the entire solution region together permitting bias errors in one subregion to influence the solution in other subregions. In light of the DAVE4VM’s dramatic improvement in performance by simply including horizontal magnetic fields, speculation that the ANMHD simulation data are not appropriate for testing tracking methods cannot be correct. Rather, aside from issues of image structure, the ANMHD simulation data represent an ideal case for the line-of-sight methods because the vertical magnetic field is known (not simply the line-of-sight component). The results of this study suggest that horizontal magnetic fields and vertical flows will render velocity estimates from “pure” tracking methods inaccurate if they are treated as the flux transport velocities $u$. This conclusion holds equally true for velocity estimates near disk center as the ANMHD simulations represents disk-center data! The good agreement between the performance of MEF and the DAVE4VM on the ANMHD data implies that incorporating the right physics is more important for producing accurate velocity estimates than is the particular method used the solve the equations. Presently, the only way to explore the “image” physics is by testing the “optical flow” methods on synthetic data from well-designed MHD simulations that attempt to reproduce the physics of the Sun. Naive “moving paint” experiments Schuck (2006) cannot critically test optical flow methods for magnetograms because the test data are consistent with the two circumstances when pure tracking methods will certainly perform well: $\mbox{\boldmath{$B$}}_{h}=0$ and $v_{z}=0$. Consequently, good performance of an optical flow method in naive “moving paint” experiments should not be considered evidence that a method will produce accurate estimates of plasma physics quantities. In the interest of reproducibility Joyner & Stein (2007), all of the software used to perform the calculations, create the figures, and draw the conclusions for this paper are archived with the Astrophysical Journal as a tgz file. Updates to the DAVE/DAVE4VM software are also available.777http://wwwppd.nrl.navy.mil/whatsnew/. I thank the referee for constructive criticism, Graham Barnes for encouraging the publication of this work, Bill Abbett for providing the ANMHD data that formed the core of this research, and Pascal Demoulin and Brian Welsch for encouraging the clarification of several issues discussed in this manuscript. I also gratefully acknowledge useful conversations with George Fisher, Bill Amatucci, and Etienne Pariat. I thank Julie Schuck for editing the manuscript. This work was supported by NASA LWS TR&T grant NNH06AD87I, LWS TR&T Strategic Capability grant NNH07AG26I, and ONR. ## Appendix A Matrix Elements of $\SS$ $\displaystyle\mathcal{G}_{00}$ $\displaystyle=$ $\displaystyle{\left(\partial_{x}{B_{z}}\right)}^{2}$ $\displaystyle\mathcal{G}_{10}$ $\displaystyle=$ $\displaystyle\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right)$ $\displaystyle\mathcal{G}_{11}$ $\displaystyle=$ $\displaystyle{\left(\partial_{y}{B_{z}}\right)}^{2}$ $\displaystyle\mathcal{G}_{20}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right)+{\left(\partial_{x}{B_{z}}\right)}^{2}{\,}x^{\prime}$ $\displaystyle\mathcal{G}_{21}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right)+\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathcal{G}_{22}$ $\displaystyle=$ $\displaystyle B_{z}^{2}+2{\,}{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}+{\left(\partial_{x}{B_{z}}\right)}^{2}{\,}x^{\prime 2}$ $\displaystyle\mathcal{G}_{30}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right)+\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathcal{G}_{31}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right)+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{32}$ $\displaystyle=$ $\displaystyle B_{z}^{2}+{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}+{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}+\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{33}$ $\displaystyle=$ $\displaystyle B_{z}^{2}+2{\,}{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}y^{\prime 2}$ $\displaystyle\mathcal{G}_{40}$ $\displaystyle=$ $\displaystyle{\left(\partial_{x}{B_{z}}\right)}^{2}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{41}$ $\displaystyle=$ $\displaystyle\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathcal{G}_{42}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}+{\left(\partial_{x}{B_{z}}\right)}^{2}{\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{43}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}+\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime 2}$ $\displaystyle\mathcal{G}_{44}$ $\displaystyle=$ $\displaystyle{\left(\partial_{x}{B_{z}}\right)}^{2}{\,}y^{\prime 2}$ $\displaystyle\mathcal{G}_{50}$ $\displaystyle=$ $\displaystyle\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathcal{G}_{51}$ $\displaystyle=$ $\displaystyle{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime}$ $\displaystyle\mathcal{G}_{52}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}+\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime 2}$ $\displaystyle\mathcal{G}_{53}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{54}$ $\displaystyle=$ $\displaystyle\left(\partial_{x}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathcal{G}_{55}$ $\displaystyle=$ $\displaystyle{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime 2}$ $\displaystyle\mathsf{s}_{60}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right)-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right)$ $\displaystyle\mathsf{s}_{61}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right)-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right)$ $\displaystyle\mathsf{s}_{62}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{63}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathsf{s}_{64}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathsf{s}_{65}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{66}$ $\displaystyle=$ $\displaystyle{\left(\partial_{x}{B_{x}}\right)}^{2}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right)+{\left(\partial_{y}{B_{z}}\right)}^{2}$ $\displaystyle\mathsf{s}_{70}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}\left(\partial_{x}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{71}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{72}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}{B_{z}}-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}{\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}{\,}x^{\prime}-{B_{x}}{\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime 2}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime 2}$ $\displaystyle\mathsf{s}_{73}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}{B_{z}}-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}{\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}{\,}x^{\prime}-{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{74}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{75}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime 2}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime 2}$ $\displaystyle\mathsf{s}_{76}$ $\displaystyle=$ $\displaystyle{B_{x}}{\,}\left(\partial_{x}{B_{x}}\right)+{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right)+{\left(\partial_{x}{B_{x}}\right)}^{2}{\,}x^{\prime}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime}$ $\displaystyle\mathsf{s}_{77}$ $\displaystyle=$ $\displaystyle B_{x}^{2}+2{\,}{B_{x}}{\,}\left(\partial_{x}{B_{x}}\right){\,}x^{\prime}+2{\,}{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}+{\left(\partial_{x}{B_{x}}\right)}^{2}{\,}x^{\prime 2}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime 2}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime 2}$ $\displaystyle\mathsf{s}_{80}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}\left(\partial_{x}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathsf{s}_{81}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathsf{s}_{82}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}{B_{z}}-{B_{y}}{\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}{\,}y^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{83}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}{B_{z}}-\left(\partial_{x}{B_{x}}\right){\,}{B_{z}}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}{B_{z}}{\,}y^{\prime}-{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime 2}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime 2}$ $\displaystyle\mathsf{s}_{84}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime 2}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime 2}$ $\displaystyle\mathsf{s}_{85}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{86}$ $\displaystyle=$ $\displaystyle\left(\partial_{x}{B_{x}}\right){\,}{B_{y}}+{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right)+{\left(\partial_{x}{B_{x}}\right)}^{2}{\,}y^{\prime}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{87}$ $\displaystyle=$ $\displaystyle{B_{x}}{\,}{B_{y}}+\left(\partial_{x}{B_{x}}\right){\,}{B_{y}}{\,}x^{\prime}+{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}+{B_{x}}{\,}\left(\partial_{x}{B_{x}}\right){\,}y^{\prime}+{B_{x}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle+{\left(\partial_{x}{B_{x}}\right)}^{2}{\,}x^{\prime}{\,}y^{\prime}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}{\,}y^{\prime}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}x^{\prime}{\,}y^{\prime}$ $\displaystyle\mathsf{s}_{88}$ $\displaystyle=$ $\displaystyle B_{y}^{2}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}{B_{y}}{\,}y^{\prime}+2{\,}{B_{y}}{\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}+{\left(\partial_{x}{B_{x}}\right)}^{2}{\,}y^{\prime 2}+2{\,}\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime 2}+{\left(\partial_{y}{B_{z}}\right)}^{2}{\,}y^{\prime 2}$ $\displaystyle\mathcal{G}_{90}$ $\displaystyle=$ $\displaystyle\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right)$ $\displaystyle\mathcal{G}_{91}$ $\displaystyle=$ $\displaystyle\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right)$ $\displaystyle\mathcal{G}_{92}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{t}{B_{z}}\right)+\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathcal{G}_{93}$ $\displaystyle=$ $\displaystyle{B_{z}}{\,}\left(\partial_{t}{B_{z}}\right)+\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathcal{G}_{94}$ $\displaystyle=$ $\displaystyle\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{x}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathcal{G}_{95}$ $\displaystyle=$ $\displaystyle\left(\partial_{t}{B_{z}}\right){\,}\left(\partial_{y}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{96}$ $\displaystyle=$ $\displaystyle-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{t}{B_{z}}\right)-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{t}{B_{z}}\right)$ $\displaystyle\mathsf{s}_{97}$ $\displaystyle=$ $\displaystyle-{B_{x}}{\,}\left(\partial_{t}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{t}{B_{z}}\right){\,}x^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{t}{B_{z}}\right){\,}x^{\prime}$ $\displaystyle\mathsf{s}_{98}$ $\displaystyle=$ $\displaystyle-{B_{y}}{\,}\left(\partial_{t}{B_{z}}\right)-\left(\partial_{x}{B_{x}}\right){\,}\left(\partial_{t}{B_{z}}\right){\,}y^{\prime}-\left(\partial_{y}{B_{z}}\right){\,}\left(\partial_{t}{B_{z}}\right){\,}y^{\prime}$ $\displaystyle\mathcal{G}_{99}$ $\displaystyle=$ $\displaystyle{\left(\partial_{t}{B_{z}}\right)}^{2}$ The primed coordinates are defined relative to the center of the aperture $\mbox{\boldmath{$x$}}^{\prime}=\mbox{\boldmath{$x$}}-\mbox{\boldmath{$\chi$}}$. ## References * Abbett et al. (2000) Abbett, W. P., Fisher, G. H., & Fan, Y. 2000, ApJ, 540, 548 * Abbett et al. (2004) Abbett, W. P., Fisher, G. H., Fan, Y., & Bercik, D. J. 2004, ApJ, 612, 557 * Adams (1993) Adams, J. C. 1993, Appl. Math. Comput., 53, 235, http://www.cisl.ucar.edu/css/software/mudpack/ * Amari et al. (2003a) Amari, T., Luciani, J. F., Aly, J. J., Mikic, Z., & Linker, J. 2003a, ApJ, 585, 1073 * Amari et al. (2003b) —. 2003b, ApJ, 595, 1231 * Amari et al. (2000) Amari, T., Luciani, J. F., Mikic, Z., & Linker, J. 2000, ApJ, 529, L49 * Antiochos et al. (1999) Antiochos, S. K., DeVore, C. R., & Klimchuk, J. A. 1999, ApJ, 510, 485 * Baker & Matthews (2004) Baker, S., & Matthews, I. 2004, Int. J. Comp. Vision, 56, 221, http://www.ri.cmu.edu/projects/project_515.html * Berger & Field (1984) Berger, M. A., & Field, G. B. 1984, Journal of Fluid Mechanics, 147 * Berger & Ruzmaikin (2000) Berger, M. A., & Ruzmaikin, A. 2000, J. Geophys. Res., 105, 10481 * Branham Jr. (1999) Branham Jr., R. L. 1999, The Astronomical Journal, 117, 1942 * Chae (2001) Chae, J. 2001, ApJ, 560, L95 * Chae (2007) —. 2007, Advances in Space Research, 39, 1700 * Chae et al. (2001) Chae, J., Wang, H., Qiu, J., Goode, P. R., Strous, L., & Yun, H. S. 2001, ApJ, 560, 476 * Chen (1989) Chen, J. 1989, ApJ, 338, 453 * Chen (1996) —. 1996, J. Geophys. Res, 101, 27499 * Démoulin & Berger (2003) Démoulin, P., & Berger, M. A. 2003, Sol. Phys., 215, 203 * DeRosa (2001) DeRosa, M. L. 2001, PhD thesis, Univ. of Colorado * Fan et al. (1999) Fan, Y., Zweibel, E. G., Linton, M. G., & Fisher, G. H. 1999, ApJ, 521, 460 * Fieller et al. (1957) Fieller, E. C., Hartley, H. O., & Pearson, E. S. 1957, Biometrika, 44, 470 * Georgoulis & LaBonte (2006) Georgoulis, M. K., & LaBonte, B. J. 2006, ApJ, 636, 475 * Golub & Van Loan (1980) Golub, G. H., & Van Loan, C. F. 1980, SIAM J. Num. Anal., 17, 883 * Gosling (1993) Gosling, J. T. 1993, J. Geophys. Res., 98, 18937 * Hildreth (1983) Hildreth, E. C. 1983, Proc. Royal. Soc. London, 221, 189 * Hildreth (1984) —. 1984, Measurement of Visual Motion (MIT Press) * IDL (2002) IDL. 2002, IDL Reference Guide (Research Systems, Inc.), http://www.rsinc.com/idl/ * Jähne (2004) Jähne, B. 2004, Practical handbook on image processing for scientific and technical applications, 2nd edn. (Boca Raton, Fla.: CRC Press) * Joyner & Stein (2007) Joyner, D., & Stein, W. 2007, Notices of the AMS, 54, 1279 * Kliem et al. (2004) Kliem, B., Titov, V. S., & Török, T. 2004, A&A, 413, L23 * Kusano et al. (2002) Kusano, K., Maeshiro, T., Yokoyama, T., & Sakurai, T. 2002, ApJ, 577, 501 * Kusano et al. (2004) Kusano, K., Maeshiro, T., Yokoyama, T., & Sakurai, T. 2004, in ASP Conf. Ser. 325: The Solar-B Mission and the Forefront of Solar Physics, 175–+ * LaBonte et al. (2007) LaBonte, B. J., Georgoulis, M. K., & Rust, D. M. 2007, ApJ, 671, 955 * Leese et al. (1971) Leese, J. A., Novak, C. S., & Clark, B. B. 1971, Journal of Applied Meteorology, 10, 118 * Leese et al. (1970) Leese, J. A., Novak, C. S., & Taylor, V. R. 1970, Pattern Recogition, 2, 279 * Leka & Barnes (2003a) Leka, K. D., & Barnes, G. 2003a, ApJ, 595, 1277 * Leka & Barnes (2003b) —. 2003b, ApJ, 595, 1296 * Leka & Barnes (2007) —. 2007, ApJ, 656, 1173 * Linker et al. (2001) Linker, J. A., Lionello, R., Mikić, Z., & Amari, T. 2001, J. Geophys. Res., 106, 25165 * Litvinenko et al. (2007) Litvinenko, Y. E., Chae, J., & Park, S.-Y. 2007, ApJ, 662, 1302 * Longcope (2004) Longcope, D. W. 2004, ApJ, 612, 1181 * Lucas (1984) Lucas, B. 1984, PhD thesis, Carnegie-Mellon University * Lucas & Kanade (1981) Lucas, B., & Kanade, T. 1981, in Proceedings of the 7th International Joint Conference on Artificial Intelligence (IJCAI ’81), Vancouver, BC, Canada, August 1981, ed. P. J. Hayes (Los Altos, CA: William Kaufmann), 674–679 * Marr & Ullman (1981) Marr, D., & Ullman, S. 1981, Proc. R. Soc. Lond. B, 211, 151 * Metcalf et al. (2008) Metcalf, T. R., Derosa, M. L., Schrijver, C. J., Barnes, G., van Ballegooijen, A. A., Wiegelmann, T., Wheatland, M. S., Valori, G., & McTtiernan, J. M. 2008, Sol. Phys., 247, 269 * November & Simon (1988) November, L. J., & Simon, G. W. 1988, ApJ, 333, 427 * Pariat et al. (2005) Pariat, E., Démoulin, P., & Berger, M. A. 2005, A&A, 442, 1105 * Pariat et al. (2007) Pariat, E., Démoulin, P., & Nindos, A. 2007, Advances in Space Research, 39, 1706 * Parker (1957) Parker, E. N. 1957, J. Geophys. Res., 62, 509 * Rust & Kumar (1996) Rust, D. M., & Kumar, A. 1996, ApJ, 464, L199+ * Santos & Büchner (2007) Santos, J. C., & Büchner, J. 2007, Astrophysics and Space Sciences Transactions, 3, 29 * Schrijver et al. (2006) Schrijver, C. J., Derosa, M. L., Metcalf, T. R., Liu, Y., McTiernan, J., Régnier, S., Valori, G., Wheatland, M. S., & Wiegelmann, T. 2006, Sol. Phys., 235, 161 * Schrijver et al. (2005) Schrijver, C. J., DeRosa, M. L., Title, A. M., & Metcalf, T. R. 2005, ApJ, 628, 501 * Schuck (2005) Schuck, P. W. 2005, ApJ, 632, 53 * Schuck (2006) —. 2006, ApJ, 646, 1358 * Stumpf (1911) Stumpf, P. 1911, Zeitschrift fur Psychologie, 321, (Translation in Todorovic, 1996) * Sweet (1958) Sweet, P. A. 1958, Electromagnetic phenomena in cosmical physics, Symposium (International Astronomical Union) No. 6 (Cambridge, England: Cambridge University Press), 123 * Török et al. (2004) Török, T., Kliem, B., & Titov, V. S. 2004, A&A, 413, L27 * Tian & Alexander (2008) Tian, L., & Alexander, D. 2008, The Astrophysical Journal, 673, 532 * Todorovic (1996) Todorovic, D. 1996, Perception, 25, 1235 * Wang et al. (2008) Wang, A., Wu, S., Liu, Y., & Hathaway, D. 2008, The Astrophysical Journal Letters, 674, L57 * Welsch et al. (2007) Welsch, B. T., Abbett, W. P., DeRosa, M. L., Fisher, G. H., Georgoulis, M. K., Kusano, K., Longcope, D. W., Ravindra, B., & Schuck, P. W. 2007, ApJ, 670, 1434 * Welsch et al. (2008) —. 2008, ApJ, 680, 827 * Welsch et al. (2004) Welsch, B. T., Fisher, G. H., & Abbett, W. P. 2004, ApJ, 620, 1148, http://solarmuri.ssl.berkeley.edu/~welsch/public/software * Wildes et al. (2000) Wildes, R. P., Amabile, M. J., Lanzillotto, A.-M., & Leu, T.-S. 2000, Comput. Vis. Image Underst., 80, 246 * Zhang et al. (2008) Zhang, Y., Liu, J., & Zhang, H. 2008, Sol. Phys., 247, 39

arxiv-papers

2024-09-04T02:48:54.521670

0803.3614

# Topological Superconductivity and Superfluidity Xiao-Liang Qi, Taylor L. Hughes, Srinivas Raghu and Shou-Cheng Zhang Department of Physics, McCullough Building, Stanford University, Stanford, CA 94305-4045 ###### Abstract We construct time reversal invariant topological superconductors and superfluids in two and three dimensions which are analogous to the recently discovered quantum spin Hall and three-d $Z_{2}$ topological insulators respectively. These states have a full pairing gap in the bulk, gapless counter-propagating Majorana states at the boundary, and a pair of Majorana zero modes associated with each vortex. We show that the time reversal symmetry naturally emerges as a supersymmetry, which changes the parity of the fermion number associated with each time-reversal invariant vortex. In the presence of external T-breaking fields, non-local topological correlation is established among these fields, which is an experimentally observable manifestation of the emergent supersymmetry. ###### pacs: 74.20.Rp, 73.43.-f, 67.30.he, 74.45.+c The search for topological states of quantum matter has become an active and exciting pursuit in condensed matter physics. The quantum Hall (QH) effect Prange and Girvin (1990) provides the first example of a topologically non- trivial state of matter, where the quantized Hall conductance is a topological invariantThouless et al. (1982). Recently, the quantum spin Hall (QSH) stateC. L. Kane and E. J. Mele (2005a); B.A. Bernevig and S.C. Zhang (2006) has been theoretically predicted B. A. Bernevig et al. (2006) and experimentally observed in HgTe quantum well systemsKönig et al. (2007). The time reversal invariant (TRI) QSH state is characterized by a bulk gap, a $Z_{2}$ topological number C. L. Kane and E. J. Mele (2005b), and gapless helical edge states, where time-reversed partners counter-propagateC. Wu et al. (2006); C. Xu and J. Moore (2006). Chiral superconductors in a time reversal symmetry breaking (TRB) $(p_{x}+ip_{y})$ pairing state in 2d have a sharp topological distinction between the strong and weak pairing regimes Read and Green (2000). In the weak pairing regime, the system has a full bulk gap and gapless chiral Majorana states at the edge, which are topologically protected. Moreover, a Majorana zero mode is trapped in each vortex coreRead and Green (2000), which leads to a ground state degeneracy of $2^{n-1}$ in the presence of $2n$ vortices. When the vortices wind around each other a non-Abelian Berry phase is generated in the $2^{n-1}$ dimensional ground state manifold, which implies non-Abelian statistics for the vorticesIvanov (2001). Chiral superconductors are analogous to the QH state— they both break time reversal (TR) and have chiral edge states with linear dispersion. However, the edge states of a chiral superconductor have only half the degrees of freedom compared to the QH state, since the negative energy quasi-particle operators on the edge of a chiral superconductor describe the same excitations as the positive energy ones. Given the analogy between the chiral superconducting state and the QH state, and with the recent discovery of the TRI QSH state, it is natural to generalize the chiral pairing state to the helical pairing state, where fermions with up spins are paired in the $(p_{x}+ip_{y})$ state, while fermions with down spins are paired in the $(p_{x}-ip_{y})$ state. Such a TRI state have a full gap in the bulk, and counter-propagating helical Majorana states at the edge (in contrast, the edge states of the TRI topological insulator are helical Dirac fermions). Just as in the case of the QSH state, a mass term for an odd number of pairs of helical Majorana states is forbidden by TR symmetry, and therefore, a topologically protected superconducting or superfluid state can exist in the presence of time-reversal symmetry. Recently, a $Z_{2}$ classification of the topological superconductor has been discussed in Refs Roy (a, 2008); Schnyder et al. (2008), by noting the similarity between the Bogoliubov-de Gennes (BdG) superconductor Hamiltonian and the QSH insulator Hamiltonian. The four types of topological states of matter discussed here are summarized in Fig. 1. In this work, we give a $Z_{2}$ classification of both the 2D and 3D cases which has a profound physical implication. In two dimensions, we show that a time-reversal invariant topological defect of a $Z_{2}$ non-trivial superconductor carries a Kramers’ pair of Majorana fermions. Let $N_{F}$ be the operator which measures the number of fermions of a general system, then the fermion-number parity operator is given by $(-1)^{N_{F}}$. This operator is also referred to as the Witten indexWitten (1982), which plays a crucial role in supersymmetric theories. We prove the remarkable fact that in the presence of a topological defect, the TR operator $\cal T$ changes the fermion number parity, ${\cal T}^{-1}(-1)^{N_{F}}{\cal T}=-(-1)^{N_{F}}$ locally around the defect in the $Z_{2}$ non-trivial state, while it preserves the fermion number parity, ${\cal T}^{-1}(-1)^{N_{F}}{\cal T}=(-1)^{N_{F}}$, in the $Z_{2}$ trivial state. This fact gives a precise definition of the $Z_{2}$ topological classification of any TRI superconductor state and is generally valid in the presence of interactions and disorder. A supersymmetric operation can be defined as an operation which changes the fermion number parity; therefore, in this precise sense, we show that the TR symmetry emerges as a supersymmetry in topological superconductors. Though supersymmetry has been studied extensively in high energy physics, it has not yet been observed in Nature. Our proposal offers the opportunity to experimentally observe supersymmetry in condensed matter systems without any fine tuning of microscopic parameters. The physical consequences of such a supersymmetry is also studied. Figure 1: (Top row) Schematic comparison of $2d$ chiral superconductor and the QH state. In both systems, TR symmetry is broken and the edge states carry a definite chirality. (Bottom row) Schematic comparison of $2d$ TRI topological superconductor and the QSH insulator. Both systems preserve TR symmetry and have a helical pair of edge states, where opposite spin states counter- propagate. The dashed lines show that the edge states of the superconductors are Majorana fermions so that the $E<0$ part of the quasi-particle spectra are redundant. In terms of the edge state degrees of freedom, we have ${\rm(QSH)}={\rm(QH)}^{2}={\rm(Helical~{}SC)}^{2}={\rm(Chiral~{}SC)}^{4}$. As the starting point, we consider a TRI $p$-wave superconductor with spin triplet pairing, which has the following $4\times 4$ BdG Hamiltonian: $\displaystyle H=\frac{1}{2}\int d^{2}x\Psi^{\dagger}(x)\left(\begin{array}[]{cc}\epsilon_{\bf p}\mathbb{I}&i\sigma_{2}\sigma_{\alpha}\Delta^{\alpha j}p_{j}\\\ h.c.&-\epsilon_{\bf p}\mathbb{I}\end{array}\right)\Psi(x)$ (3) with $\Psi(x)=\left(c_{\uparrow}(x),c_{\downarrow}(x),c_{\uparrow}^{\dagger}(x),c_{\downarrow}^{\dagger}(x)\right)^{T}$, $\epsilon_{\bf p}={\bf p}^{2}/2m-\mu$ the kinetic energy and chemical potential terms and $h.c.\equiv(i\sigma_{2}\sigma_{\alpha}\Delta^{\alpha j}p_{j})^{\dagger}$. The TR transformation is defined as $c_{\uparrow}\rightarrow c_{\downarrow},~{}c_{\downarrow}\rightarrow- c_{\uparrow}$. It can be shown that the Hamiltonian (3) is time-reversal invariant if $\Delta_{\alpha j}$ is a real matrix. To show the existence of a topological state, consider the TRI mean-field ansatz $\Delta^{\alpha 1}=\Delta(1,0,0),~{}\Delta^{\alpha 2}=\Delta(0,1,0)$. For such an ansatz the Hamiltonian (3) is block diagonal with only equal spin pairing: $\displaystyle H=\frac{1}{2}\int d^{2}x\tilde{\Psi}^{\dagger}\left(\begin{array}[]{cccc}\epsilon_{\bf p}&\Delta p_{+}&&\\\ \Delta p_{-}&-\epsilon_{\bf p}&&\\\ &&\epsilon_{\bf p}&-\Delta p_{-}\\\ &&-\Delta p_{+}&-\epsilon_{\bf p}\end{array}\right)\tilde{\Psi}$ (8) with $\tilde{\Psi}(x)\equiv\left(c_{\uparrow}(x),c_{\uparrow}^{\dagger}(x),c_{\downarrow}(x),c_{\downarrow}^{\dagger}(x)\right)^{T}$, and $p_{\pm}=p_{x}\pm ip_{y}$. From this Hamiltonian we see that the spin up (down) electrons form $p_{x}+ip_{y}$ ($p_{x}-ip_{y}$) Cooper pairs, respectively. In the weak pairing phase with $\mu>0$, the $(p_{x}+ip_{y})$ chiral superconductor is known to have chiral Majorana edge states propagating on each boundary, described by the Hamiltonian $H_{\rm edge}=\sum_{k_{y}\geq 0}v_{F}k_{y}\psi_{-k_{y}}\psi_{k_{y}}$,where $\psi_{-k_{y}}=\psi_{k_{y}}^{\dagger}$ is the quasiparticle creation operator Read and Green (2000) and the boundary is taken to be parallel to the $y$ direction. Thus we know that the edge states of the TRI system described by Hamiltonian (8) consist of spin up and spin down quasi-particles with opposite chirality: $\displaystyle H_{\rm edge}=\sum_{k_{y}\geq 0}v_{F}k_{y}\left(\psi_{-k_{y}\uparrow}\psi_{k_{y}\uparrow}-\psi_{-k_{y}\downarrow}\psi_{k_{y}\downarrow}\right).$ (9) The quasi-particle operators $\psi_{k_{y}\uparrow},~{}\psi_{k_{y}\downarrow}$ can be expressed in terms of the eigenstates of the BdG Hamiltonian as $\displaystyle\psi_{k_{y}\uparrow}$ $\displaystyle=$ $\displaystyle\int d^{2}x\left(u_{k_{y}}(x)c_{\uparrow}(x)+v_{k_{y}}(x)c_{\uparrow}^{\dagger}(x)\right)$ $\displaystyle\psi_{k_{y}\downarrow}$ $\displaystyle=$ $\displaystyle\int d^{2}x\left(u_{-k_{y}}^{*}(x)c_{\downarrow}(x)+v_{-k_{y}}^{*}(x)c_{\downarrow}^{\dagger}(x)\right)$ (10) from which the time-reversal transformation of the quasiparticle operators can be determined to be ${\cal{T}}^{-1}\psi_{k_{y}\uparrow}{\cal{T}}=\psi_{-k_{y}\downarrow},~{}{\cal{T}}^{-1}\psi_{k_{y}\downarrow}{\cal{T}}=-\psi_{-k_{y}\uparrow}$. In other words, $(\psi_{k_{y}\uparrow},~{}\psi_{-k_{y}\downarrow})$ transforms as a Kramers’ doublet, which forbids a gap in the edge states due to mixing of the spin-up and spin-down modes when TR is preserved. To see this explicitly, notice that the only $k_{y}$-independent term that can be added to the edge Hamiltonian (9) is $im\sum_{k_{y}}\psi_{-k_{y}\uparrow}\psi_{k_{y}\downarrow}$ with $m\in\mathbb{R}$. However, such a term is odd under TR, which implies that any back scattering between the quasi-particles is forbidden by TR symmetry. The discussion above is exactly parallel to the $Z_{2}$ topological characterization of the quantum spin Hall system. In fact, the Hamiltonian (8) has exactly the same form as the four band effective Hamiltonian proposed in Ref.B. A. Bernevig et al. (2006) to describe HgTe quantum wells with the QSH effect. The edge states of the QSH insulators consist of an odd number of Kramers’ pairs, which remain gapless under any small TR-invariant perturbationC. Wu et al. (2006); C. Xu and J. Moore (2006). A no-go theorem states that such a “helical liquid” with an odd number of Kramers’ pairs at the Fermi energy can not be realized in any bulk 1d system, but can only appear as an edge theory of a 2d QSH insulatorC. Wu et al. (2006). Similarly, the edge state theory (9) can be called a “helical Majorana liquid”, which can only exist on the boundary of a $Z_{2}$ topological superconductor. Once such a topological phase is established, it is robust under any TRI perturbations. The Hamiltonian (3) can be easily generalized to three dimensions, in which case $\Delta^{\alpha j}$ becomes a $3\times 3$ matrix with $\alpha=1,2,3$ and $j=x,y,z$. An example of such a Hamiltonian is given by the well-known 3He BW phase, for which the order parameter $\Delta^{\alpha j}$ is determined by an orthogonal matrix $\Delta^{\alpha j}=\Delta u^{\alpha j}$, $u\in{\rm SO(3)}$Vollhardt and Wölfle (1990). Here and below we ignore the dipole-dipole interaction term Leggett (1975) since it does not affect any essential topological properties. By applying a spin rotation, $\Delta^{\alpha j}$ can be diagonalized to $\Delta^{\alpha j}=\Delta\delta^{\alpha j}$, in which case the Hamiltonian (3) has the same form as a 3d Dirac Hamiltonian with momentum dependent mass $\epsilon({\bf p})={\bf p}^{2}/2m-\mu$. We know that a band insulator described by the Dirac Hamiltonian is a 3d $Z_{2}$ topological insulator for $\mu>0$Fu et al. (2007); Moore and Balents (2007); Roy (b), and has nontrivial surface states. The corresponding superconductor Hamiltonian describes a topological superconductor with 2d gapless Majorana surface states. The surface theory can be written as $\displaystyle H_{\rm surf}=\frac{1}{2}\sum_{\bf k}v_{F}\psi_{-\bf k}^{T}\left(\sigma_{z}k_{x}+\sigma_{x}k_{y}\right)\psi_{\bf k}$ (11) which remains gapless under any small TRI perturbation since the only available mass term $m\sum_{\bf k}\psi_{-\bf k}^{T}\sigma_{y}\psi_{\bf k}$ is time-reversal odd. We would like to mention that the surface Andreev bound states in 3He-B phase have been observed experimentallyAoki et al. (2005). To understand the physical consequences of the nontrivial topology we study the TRI topological defects of the topological superconductors. We start by considering the equal-spin pairing system with BdG Hamiltonian (8) in which spin up and down electrons form $p_{x}+ip_{y}$ and $p_{x}-ip_{y}$ Cooper pairs, respectively. A TRI topological defect can be defined as a vortex of spin-up superfluid coexisting with an anti-vortex of spin-down superfluid at the same position. In the generic Hamiltonian (3), such a vortex configuration is written as $\Delta^{\alpha j}=[\exp\left(i\sigma_{2}\theta({\bf r-r_{0}})\right)]^{\alpha j},~{}\alpha=1,2$ and $\Delta^{3j}=0$, where $\theta({\bf r-r_{0}})$ is the angle of ${\bf r}$ with respect to the vortex position ${\bf r}_{0}$. Since in the vortex core of a weak pairing $p_{x}+ip_{y}$ superconductor there is a single Majorana zero modeRead and Green (2000); Stone and Chung (2006), one immediately knows that a pair a Majorana zero modes exist in the vortex core we study here. In terms of the electron operators, the two Majorana fermion operators can be written as $\displaystyle\gamma_{\uparrow}$ $\displaystyle=$ $\displaystyle\int d^{2}x\left(u_{0}(x)c_{\uparrow}(x)+u_{0}^{*}(x)c_{\uparrow}^{\dagger}(x)\right)$ $\displaystyle\gamma_{\downarrow}$ $\displaystyle=$ $\displaystyle\int d^{2}x\left(u_{0}^{*}(x)c_{\uparrow}(x)+u_{0}(x)c_{\uparrow}^{\dagger}(x)\right)$ (12) where we have used the fact that the spin-down zero mode wave function can be obtained from the time-reversal transformation of the spin-up one. The Majorana operators satisfy the anti-commutation relation $\left\\{\gamma_{\alpha},\gamma_{\beta}\right\\}=2\delta_{\alpha\beta}$. The TR transformation of the Majorana fermions is $\displaystyle{\cal{T}}^{-1}\gamma_{\uparrow}{\cal{T}}=\gamma_{\downarrow},~{}{\cal{T}}^{-1}\gamma_{\downarrow}{\cal{T}}=-\gamma_{\uparrow}.$ (13) Similar to the case of the edge states studied earlier, the Majorana zero modes are robust under any small TRI perturbation, since the only possible term $im\gamma_{\uparrow}\gamma_{\downarrow}$ which can lift the zero modes to finite energy is TR odd, i.e., ${\cal{T}}^{-1}i\gamma_{\uparrow}\gamma_{\downarrow}{\cal{T}}=-i\gamma_{\uparrow}\gamma_{\downarrow}$. The properties of such a topological defect appear identical to that of a $\pi$-flux tube threading into a TRI topological insulatorQi and Zhang ; Ran et al. , where a Kramers’ pair of complex fermions are trapped by the flux tube. However, there is an essential difference. From the two Majorana zero modes $\gamma_{\uparrow},\gamma_{\downarrow}$ a complex fermion operator can be defined as $a=\left(\gamma_{\uparrow}+i\gamma_{\downarrow}\right)/2$, which satisfies the fermion anticommutation relation $\left\\{a,a^{\dagger}\right\\}=1$. Since $\gamma_{\uparrow},\gamma_{\downarrow}$ are zero modes, we obtain $\left[a,H\right]=0$ which implies that $a$ is the annihilation operator of a zero-energy quasiparticle. Consequently, the ground state of the system is at least two-fold degenerate, with two states $|G_{0}\rangle$ and $|G_{1}\rangle=a^{\dagger}|G_{0}\rangle$ containing $0$ and $1$ $a$-fermions. Since $a^{\dagger}a=\left(1+i\gamma_{\uparrow}\gamma_{\downarrow}\right)/2$, the states $\left|G_{0(1)}\right\rangle$ are eigenstates of $i\gamma_{\uparrow}\gamma_{\downarrow}$ with eigenvalues $-1(+1)$, respectively. Thus from the oddness of $i\gamma_{\uparrow}\gamma_{\downarrow}$ under TR we know that $\left|G_{0}\right\rangle$ and $\left|G_{1}\right\rangle$ are time-reversed partners. Note that superconductivity breaks the charge $U(1)$ symmetry to $Z_{2}$, meaning that the fermion number parity operator $(-1)^{N_{F}}$ is conserved. Thus, all the eigenstates of the Hamiltonian can be classified by the value of $(-1)^{N_{F}}$. If, say, $\left|G_{0}\right\rangle$ is a state with $(-1)^{N_{F}}=1$, then $\left|G_{1}\right\rangle=a^{\dagger}\left|G_{0}\right\rangle$ must satisfy $(-1)^{N_{F}}=-1$. Since $\left|G_{0}\right\rangle$ and $\left|G_{1}\right\rangle$ are time-reversal partners, we know that in the Hilbert space of the zero-energy states the TR transformation changes the fermion number parity: $\displaystyle{\cal{T}}^{-1}(-1)^{N_{F}}{\cal{T}}=-(-1)^{N_{F}}.$ (14) Eq. (14) is the central result of this paper. At a first glance it seems contradict the fundamental fact that the electron number of the whole system is invariant under TR. Such a paradox is resolved by noticing that there are always an even number of topological defects in a closed system without boundary. Under the TR transformation, the fermion number parity around each vortex core is odd, but the total fermion number parity remains even as expected. Once the anomalous transformation property (14) is established for a topological defect in a TRI superconductor, it is robust under any TRI perturbation as long as the bulk quasiparticle gap remains finite and other topological defects are far away. Thus Eq. (14) is a generic definition of TRI topological superconductors: * • Definition I. A two-dimensional TRI superconductor is $Z_{2}$ nontrivial if and only if fermion number parity around a TRI topological defect is odd under TR. A transformation changing fermion number by an odd number is a “supersymmetry”; thus, the TR symmetry emerges as a discrete supersymmetry for each TRI topological defect. The same analysis applies to the edge theory (9), which shows that in the 1d helical Majorana liquid is a theory with TR symmetry as a discrete supersymmetry. All the conclusions above can be generalized to 3d topological superconductors. In the 3He BW phase the Goldstone manifold of the order parameter is $\Delta^{\alpha j}=\Delta u^{\alpha j}\in SO(3)\times U(1)$Vollhardt and Wölfle (1990); Salomaa and Volovik (1987). A time-reversal invariant configuration satisfies $\Delta^{\alpha j}\in\mathbb{R}$, which restricts the order parameter to $SO(3)$. Since $\Pi_{1}(SO(3))=Z_{2}$, the TRI topological defects are 1d “vortex” rings. By solving the BdG equations in the presence of such vortex rings, it can be shown that there are linearly dispersing quasiparticles propagating on each vortex ring, similar to the edge states of the 2d topological superconductor. However, for a ring with finite length the quasi-particle spectrum is discrete. Specifically, there may or may not be a pair of Majorana modes at exactly zero energy. The existence of the Majorana zero modes on the vortex rings turns out to be a topological property determined by the linking number between different vortex rings. Due to the length constraints of the present paper, we will write our conclusion and leave the details for a separate work: There are a pair of Majorana fermion zero modes confined on a vortex ring if and only if the ring is linked to an odd number of other vortex rings. Such a condition is shown in Fig. 2. Consequently, the generalization of Definition I to 3d is: * • Definition II. A 3d TRI superconductor is $Z_{2}$ nontrivial if and only if the fermion number parity around one of the two mutually-linked TRI vortex rings is odd under TR. Figure 2: Illustration of a 3d TRI topological superconductor with two TRI vortex rings which are (a) linked or (b) unlinked. The $E-k_{\parallel}$ dispersion relations show schematically the quasiparticle levels confined on the red vortex ring in both cases. “$\circ$” and “$\times$” stand for the quasiparticle levels that are Kramers’ partners of each other. Only case (a) has a pair of Majorana zero modes located on each vortex ring. Figure 3: (a) Illustration of a 2d TRI topological superconductor with four TRI topological defects coupled to a TR-breaking field. The arrows show the sign of the TR- breaking field ${\rm sgn}(M({\bf r}_{s}))$ at each topological defect. In the two configurations shown, only the field around vortex 1 is flipped, leading to an opposite fermion number parity in the corresponding ground state $|G\rangle$ and $|G^{\prime}\rangle$ (see text). (b) Illustration showing the flow of the energy levels when the upper configuration in figure (a) is deformed to the lower one. The flip of the TR-breaking field $M({\bf r}_{1})$ leads to a level crossing at $M({\bf r}_{1})=0$, where the fermion number parity in the ground state changes sign. Besides providing a generic definition of the $Z_{2}$ topological superconductors, such an emergent supersymmetry also leads to physical predictions. Consider the 2d topological superconductor coupled to a weak TR- breaking field $M({\bf r})$, which is classical but can have thermal fluctuations. This situation can be realized in an isolated superconductor with vortices pinned to quenched weak magnetic impurities. The $n$-point correlation function of $M({\bf r})$ can be obtained by $\displaystyle\left\langle{\prod_{s=1}^{n}M({\bf r}_{s})}\right\rangle\equiv\int\frac{D[M({\bf r})]}{Z}\prod_{s=1}^{n}M({\bf r}_{s}){\rm Tr}\left(e^{-\beta H[M]}\right)_{\rm even}$ (15) in which the trace is restricted to states with an even number of fermions. For a closed system with $N$ vortices, the leading order effect of the TR- breaking field is to lift the degeneracy between the two Majorana fermions in each vortex core. Consequently, the perturbed Hamiltonian $H[M({\bf r})]$ to first order can be written as $\displaystyle H[M({\bf r})]=\sum_{s=1}^{N}iM({\bf r}_{s})a_{s}\gamma_{s\uparrow}\gamma_{s\downarrow}$ (16) in which $\gamma_{s\uparrow(\downarrow)}$ are the Majorana fermion operators, and $a_{s}\in\mathbb{R}$ depend on the details of the perturbation. The important fact is that the mass term induced is linear in $M({\bf r})$ at the defect position ${\bf r}_{s}$, since $i\gamma_{s\uparrow}\gamma_{s\downarrow}$ is TR odd. Since the superconductor has a full gap, naively one would expect all the correlations of $M({\bf r})$ field to be short ranged. However, for a system with $N$ topological defects the $N$-point correlation function has a long range order when ${\bf r}_{s},s=1,2,..N$ are chosen to be the coordinates of the topological defects. In other words, the correlation function $\displaystyle\lim_{|{\bf r}_{i}-{\bf r}_{j}|\rightarrow\infty,~{}\forall i,j}\left\langle{\prod_{s=1}^{N}M({\bf r}_{s})}\right\rangle\neq 0,$ (17) though all the $n$ point correlations in Eq. (15) with $n<N$ remain short ranged. Physically, such a non-local correlation can be understood by comparing two states $|G\rangle$ and $|G^{\prime}\rangle$, which are the ground states of the systems with the field configurations $\mathcal{M}\equiv(M({\bf r}_{1}),M({\bf r}_{2}),...,M({\bf r}_{N}))$ and $\mathcal{M}^{\prime}\equiv(-M({\bf r}_{1}),M({\bf r}_{2}),...,M({\bf r}_{N}))$, respectively. From Hamiltonian (16) it can be seen that $|G\rangle$ and $|G^{\prime}\rangle$ have opposite fermion number parity, since the fermion number parity around the first topological defect $i\gamma_{1\uparrow}\gamma_{1\downarrow}$ is reversed while that of all the other topological defects remains invariant, as shown in Fig. 3. Without loss of generality, we can assume $(-1)^{N_{F}}$ is even for $|G\rangle$ and odd for $|G^{\prime}\rangle$. Since the whole system is required to have an even number of fermions, the lowest energy state in the Hilbert space for the field configuration $\mathcal{M}^{\prime}$ is not $|G^{\prime}\rangle$, but the lowest quasiparticle excitation $a_{\rm min}^{\dagger}|G^{\prime}\rangle$. Thus, the two field configurations $\mathcal{M}$ and $\mathcal{M}^{\prime}$ have different free energies, which leads to the non-vanishing correlation function in Eq. (17). Even when the topological defects are arbitrarily far away, the energy difference between the two configurations remains finite, which shows the non-local topological correlation. Similar non-local correlations can also be obtained for a 3d TRI topological superconductor with linked vortex rings. Acknowledgement.—We acknowledge helpful discussions with S. B. Chung, A. L. Fetter, L. Fu, R. Roy and S. Ryu. This work is supported by the NSF under grant numbers DMR-0342832, the US Department of Energy, Office of Basic Energy Sciences under contract DE-AC03-76SF00515, and the Stanford Institute for Theretical Physics (S.R.). ## References * Prange and Girvin (1990) R. E. Prange and S. M. Girvin, eds., _The Quantum Hall effect_ (Springer-Verlag, USA, 1990). * Thouless et al. (1982) D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). * C. L. Kane and E. J. Mele (2005a) C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005a). * B.A. Bernevig and S.C. Zhang (2006) B.A. Bernevig and S.C. Zhang, Phys. Rev. Lett. 96, 106802 (2006). * B. A. Bernevig et al. (2006) B. A. Bernevig, T. L. Hughes, and S.C. Zhang, Science 314, 1757 (2006). * König et al. (2007) M. König, S. Wiedmann, C. Brüne, A. Roth, H. Buhmann, L. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). * C. L. Kane and E. J. Mele (2005b) C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005b). * C. Xu and J. Moore (2006) C. Xu and J. Moore, Phys. Rev. B 73, 045322 (2006). * C. Wu et al. (2006) C. Wu, B.A. Bernevig, and S.C. Zhang, Phys. Rev. Lett. 96, 106401 (2006). * Read and Green (2000) N. Read and D. Green, Phys. Rev. B 61, 10267 (2000). * Ivanov (2001) D. A. Ivanov, Phys. Rev. Lett. 86, 268 (2001). * Roy (a) R. Roy, arxiv: cond-mat/0608064. * Roy (2008) R. Roy, arxiv: 0803.2868 (2008). * Schnyder et al. (2008) A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig, e-print arXiv: 0803.2786 (2008). * Witten (1982) E. Witten, Nucl. Phys. B 202, 253 (1982). * Vollhardt and Wölfle (1990) D. Vollhardt and P. Wölfle, _The Superfluid Phases of Helium 3_ (Taylor and Francis, USA, 1990). * Leggett (1975) A. J. Leggett, Rev. Mod. Phys. 47, 331 (1975). * Fu et al. (2007) L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). * Moore and Balents (2007) J. E. Moore and L. Balents, Phys. Rev. B 75, 121306 (2007). * Roy (b) R. Roy, arxiv: cond-mat/0607531. * Aoki et al. (2005) Y. Aoki, Y. Wada, M. Saitoh, R. Nomura, Y. Okuda, Y. Nagato, M. Yamamoto, S. Higashitani, and K. Nagai, Phys. Rev. Lett. 95, 075301 (2005). * Stone and Chung (2006) M. Stone and S.-B. Chung, Phys. Rev. B 73, 014505 (2006). * (23) X. L. Qi and S. C. Zhang, arxiv: cond-mat/0801.0252. * (24) Y. Ran, A. Vishwanath, and D.-H. Lee, arxiv: cond-mat/0801.0627. * Salomaa and Volovik (1987) M. M. Salomaa and G. E. Volovik, Rev. Mod. Phys. 59, 533 (1987).

arxiv-papers

2024-09-04T02:48:54.531849

0803.3683

# Asymptotic stability of solitons for the Benjamin-Ono equation C.E. Kenig(1) and Y. Martel(2) ( (1) Department of Mathematics, University of Chicago, 5734 University ave., Chicago, Illinois 60637-1514, cek@math.uchicago.edu (2) Université de Versailles Saint-Quentin-en-Yvelines, Mathématiques, 45, av. des Etats-Unis, 78035 Versailles cedex, France martel@math.uvsq.fr) ###### Abstract In this paper, we prove the asymptotic stability of the family of solitons of the Benjamin-Ono equation in the energy space. The proof is based on a Liouville property for solutions close to the solitons for this equation, in the spirit of [16], [18]. As a corollary of the proofs, we obtain the asymptotic stability of exact multi-solitons. ## 1 Introduction We consider the Benjamin-Ono equation (BO) $u_{t}+\mathcal{H}u_{xx}+uu_{x}=0,\quad(t,x)\in\mathbb{R}\times\mathbb{R},$ (1) where $\mathcal{H}$ denotes the Hilbert transform $\mathcal{H}u(x)=\frac{1}{\pi}\,\mathrm{p.v.}\int_{-\infty}^{+\infty}\frac{u(y)}{y-x}dy=\frac{1}{\pi}\lim_{\varepsilon\to 0}\int_{|y-x|>\varepsilon}\frac{u(y)}{y-x}dy.$ (2) Note that with this notation, $\int u_{x}\mathcal{H}u=\int|D^{\frac{1}{2}}u|^{2}=\|u\|_{\dot{H}^{\frac{1}{2}}}^{2}$. The Cauchy problem for (1) is globally well-posed in $H^{s}$, for any $s\geq 0$ (see Tao [25] for $s\geq 1$ and Ionescu and Kenig [11] for the case $s\geq 0$, see also Burq and Planchon [5] for the case $s>\frac{1}{4}$). Moreover, for solutions in the energy space $H^{\frac{1}{2}}$ the following quantities are invariant $\int u^{2}(t,x)dx=\int u^{2}(0,x)dx,\quad E(t)=\int\Big{(}u_{x}\mathcal{H}u-\tfrac{1}{3}{u^{3}}\Big{)}(t,x)dx=E(0).$ (3) Recall the scaling and translation invariances of equation (1) if $u(t,x)$ is solution then $\forall c>0$, $x_{0}\in\mathbb{R}$, $v(t,x)=c\,u(c^{2}t,c(x-x_{0}))$ is solution. (4) We call soliton any travelling wave solution $u(t,x)=Q_{c}(x-x_{0}-ct)$, where $c>0$, $x_{0}\in\mathbb{R}$, and $Q_{c}(x)=cQ(cx)$ solves: $-\mathcal{H}Q^{\prime}+Q-\tfrac{1}{2}{Q^{2}}=0,\quad Q\in H^{\frac{1}{2}},\quad Q>0.$ (5) It is known that there is a unique (up to translations) solution of (5), which is $Q(x)=\frac{4}{1+x^{2}}.$ (6) (see Benjamin [2] and Amick and Toland [1] for the uniqueness statement). This solution is stable (see Bennet et al. [3] and Weinstein [29]) in the following sense. Stability of soliton in the energy space ([3], [29]). _There exist $C,\alpha_{0}>0$ such that if $u_{0}\in H^{\frac{1}{2}}$ satisfies $\|u_{0}-Q\|_{H^{\frac{1}{2}}}=\alpha\leq\alpha_{0}$ then the solution $u(t)$ of (1) with $u(0)=u_{0}$ satisfies_ $\sup_{t\in\mathbb{R}}\inf_{y\in\mathbb{R}}\|u(t)-Q(.-y)\|_{H^{\frac{1}{2}}}\leq C\alpha.$ See a sketch of proof of this result in Section 5.1. The main result of this paper is the asymptotic stability of the family of solitons of (1). Then, we consider the multisoliton case (see Section 5). ###### Theorem 1 (Asymptotic stability of solitons in the energy space). There exist $C,\alpha_{0}>0$ such that if $u_{0}\in H^{\frac{1}{2}}$ satisfies $\|u_{0}-Q\|_{H^{\frac{1}{2}}}=\alpha\leq\alpha_{0}$, then there exists $c^{+}>0$ with $|c^{+}-1|\leq C\alpha$ and a $C^{1}$ function $\rho(t)$ such that the solution $u(t)$ of (1) with $u(0)=u_{0}$ satisfies $\displaystyle u(t,.+\rho(t))\rightharpoonup Q_{c^{+}}\quad\text{in $H^{\frac{1}{2}}$ weak,}\qquad\|u(t)-Q_{c^{+}}(.-\rho(t))\|_{L^{2}(x>\frac{t}{10})}\to 0,$ (7) $\displaystyle\rho^{\prime}(t)\to c^{+}\quad\text{as $t\to+\infty$.}$ (8) The proof of Theorem 1 is based on the following rigidity result. ###### Theorem 2 (Nonlinear Liouville property). There exist $C,\alpha_{0}>0$ such that if $u_{0}\in H^{\frac{1}{2}}$ satisfies $\|u_{0}-Q\|_{H^{\frac{1}{2}}}=\alpha\leq\alpha_{0}$ and if the solution $u(t)$ of (1) with $u(0)=u_{0}$ satisfies for some function $\rho(t)$ $\forall\varepsilon>0,\exists A_{\varepsilon}>0,\text{ s.t. }\forall t\in\mathbb{R},\quad\int_{|x|>A_{\varepsilon}}u^{2}(t,x+\rho(t))dx<\varepsilon,$ (9) then there exist $c_{1}>0$, $x_{1}\in\mathbb{R}$, such that $u(t,x)=Q_{c_{1}}(x-x_{1}-c_{1}t),\quad|c_{1}-1|+|x_{1}|\leq C\alpha.$ (10) ###### Remark 1. In Theorem 1, the convergence of $u(t)$ to $Q_{c^{+}}$ as $t\to+\infty$ is obtained strongly in $L^{2}$ in the region $x>\frac{t}{10}$. The value $\frac{1}{10}$ is somewhat arbitrary, the result holds for $x>\varepsilon t$, for any $\varepsilon>0$, provided $\alpha_{0}=\alpha_{0}(\varepsilon)>0$ is small enough. Note that this result is optimal in $L^{2}$ since $u(t)$ could contain other small (and then slow) solitons and since in general $u(t)$ does not go to $0$ in $L^{2}$ for $x<0$. For example, if $\|u(t)-Q_{c^{+}}(.-\rho(t))\|_{H^{\frac{1}{2}}(\mathbb{R})}\to 0$ as $t\to+\infty$, then $E(u)=E(Q_{c^{+}})$ and $\int u^{2}=\int Q_{c^{+}}^{2}$ and so by the variational characterization of $Q(x)$ (see [29]), $u(t)=Q_{c^{+}}(x-x_{0}-c^{+}t)$ is exactly a soliton. Under the assumptions of Theorem 1, we expect strong convergence in $H^{\frac{1}{2}}$ to be true as well in the same local sense ($x>\varepsilon t$). This could require some more analysis. By the methods of this paper, we are able to obtain the following weaker result (Section 4.3) $\lim_{t\to+\infty}\int_{t}^{t+1}\|u(s,.+\rho(s))-Q_{c^{+}}\|_{H^{\frac{1}{2}}_{loc}}^{2}ds=0.$ (11) The proof of Theorem 1 follows the approach of [15], [16], concerning the case of the generalized KdV equations, where the asymptotic stability of the family of solitons is deduced from a Liouville type theorem such as Theorem 2. Moreover, similarly as in [16], the proof of Theorem 2 follows from a Liouville property on the linearized equation around $Q$, see Theorem 3 in Section 3. With respect to the gKdV case, there are two main difficulties : (1) $L^{2}$ monotonicity type results, which are similar to the ones for the gKdV equations ([16]), but whose proof are more subtle due to the nonlocal nature of the (BO) operator (see Section 2). For this part, we use a Kato type identity for (1) (see [9] and [23]). (2) The proof of the linear Liouville theorem, which requires the analysis of some linear operators related to $Q$. Note that for this part, we use the fact that $Q(x)$ is explicit, and some known results about the linearized equation around $Q$ ([3], [29]). We point out that except for this part of the analysis, all the arguments are quite flexible and could be applied to generalized versions of the (BO)) equation. In particular, we do not use the integrability property of the equation. As a corollary of the proof of Theorem 1 and of Theorem 2, we obtain stability and asymptotic stability of multisoliton solutions. See Theorem 4 in Section 5 for a precise statement. After the paper was finished and submitted, we learned that S. Gustafson, H. Takaoka, and T-P. Tsai [10] have obtained independently the stability part of Theorem 4. Note that the main result of the present paper, i.e. asymptotic stability of (single or multi-) solitons is not addressed in [10]. The rest of the paper is organized as follows. In Section 2, we prove $L^{2}$ monotonicity type results in the context of Theorem 1. In Section 3, we state and prove the linear Liouville Theorem, which is the main ingredient of the proof of Theorem 2. In Section 4, we prove Theorems 1 and 2 using Sections 2 and 3. Section 5 is devoted to the multisoliton case. In Section 6, we prove some weak convergence and well-posedness results used in the proofs. Finally, Appendix A contains the proof of some technical points. Acknowledgments. The first author is partly supported by the NSF grant DMS-0456583. This work was initiated when the second author was visiting the University of Chicago. He would like to thank the Department of Mathematics for its hospitality. The second author is partly supported by the Agence Nationale de la Recherche (ANR ONDENONLIN). ## 2 Monotonicity arguments for solutions close to $Q$ ### 2.1 Modulation ###### Lemma 1 (Choice of translation parameter). There exist $C,\alpha_{0}>0$ such that for any $0<\alpha<\alpha_{0}$, if $u(t)$ is an $H^{\frac{1}{2}}$ solution of (1) such that $\forall t\in\mathbb{R},\quad\inf_{r\in\mathbb{R}}\|u(t)-Q(.-r)\|_{H^{\frac{1}{2}}}<\alpha,$ (12) then there exists $\rho(t)\in C^{1}(\mathbb{R})$ such that $\eta(t,x)=u(t,x+\rho(t))-Q(x)$ satisfies $\begin{split}\forall t\in\mathbb{R},\quad&\int Q^{\prime}(x)\eta(t,x)dx=0,\quad\|\eta(t)\|_{H^{\frac{1}{2}}}\leq C\alpha,\\\ &|\rho^{\prime}(t)-1|\leq C\left(\int\frac{\eta^{2}(t,x)}{1+x^{2}}dx\right)^{\frac{1}{2}}\leq C\|\eta(t)\|_{L^{2}}.\end{split}$ (13) _Proof of Lemma 1._ This follows from standard arguments (see e.g. [4], Lemma 4.1, [14], Proposition 1 and Lemma 4). _Time independent arguments._ For $u\in H^{\frac{1}{2}}$ and $y\in\mathbb{R}$, set $I_{y}(u)=\int Q^{\prime}(x)(u(x+y)-Q(x))dx\quad\text{so that}\quad{\frac{\partial I_{y}}{\partial y}}~{}_{|y=0,u=Q}=\int(Q^{\prime})^{2}>0.$ Thus, by the implicit function theorem, there exists $\alpha_{1}>0$, $V$ a neighborhood of $0$ in $\mathbb{R}$ and a unique $C^{1}$ map: $y:\\{u\in H^{\frac{1}{2}},~{}\|u-Q\|_{H^{\frac{1}{2}}}\leq\alpha_{1}\\}\to V~{}\text{such that $I_{y(u)}(u)=0$, $|y(u)|\leq C\|u-Q\|_{H^{\frac{1}{2}}}$ .}$ We uniquely extend the $C^{1}$ map $y(u)$ to $U_{\alpha_{1}}=\\{u\in H^{\frac{1}{2}},~{}\inf_{r}\|u(.+r)-Q\|_{H^{\frac{1}{2}}}\leq\alpha_{1}\\}$ so that for all $u$ and $r$, $y(u)=y(u(.+r))+r$. Then, we set $\eta_{u}(x)=u(x+y(u))-Q(x)$, so that $\int\eta_{u}Q^{\prime}=0\quad\text{and}\quad\|\eta_{u}\|_{H^{\frac{1}{2}}}\leq C\|u-Q\|_{H^{\frac{1}{2}}}.$ _Estimates depending on $t$._ For all $t$, we define $\rho(t)=y(u(t))$ and $\eta(t)=\eta_{u(t)}$. To conclude the proof of the lemma, we just have to prove the estimate on $\rho^{\prime}(t)-1$. We perform formal computations which can be justified for $H^{\frac{1}{2}}$ solutions by density and continuous dependence arguments. The function $\eta(t,x)$ satisfies the following equation: $\eta_{t}=(\mathcal{L}\eta-\tfrac{1}{2}\eta^{2})_{x}+(\rho^{\prime}-1)(Q+\eta)_{x}\quad\text{where $\mathcal{L}\eta=-\mathcal{H}\eta_{x}+\eta-Q\eta$.}$ (14) Thus, multiplying the equation of $\eta$ by $Q^{\prime}$ and using $\int\eta Q^{\prime}=0$, we obtain $(\rho^{\prime}-1)\left[\int(Q^{\prime})^{2}-\int\eta Q^{\prime\prime}\right]=\int\eta\mathcal{L}(Q^{\prime\prime})-\tfrac{1}{2}\int\eta^{2}Q^{\prime\prime},$ (15) which finishes the proof for $\alpha_{0}$ small enough. ###### Remark 2. By the proof of Lemma 1, $\rho(t)$ depends continuously on $u(t)$ in $H^{\frac{1}{2}}$. In particular, let $u(t)$ satisfy the assumptions of Lemma 1 with $u(0)=u_{0}$. If $u_{n}(0)\to u_{0}$ in $H^{\frac{1}{2}}$ as $n\to+\infty$, then by continuous dependence (see [11]), we obtain for all $t\in\mathbb{R}$, $\rho_{n}(t)\to\rho(t)$ as $n\to+\infty$, where $\rho_{n}(t)$ is defined from $u_{n}(t)$ ($u_{n}(t)$ is the solution of (1) corresponding to $u_{n}(0)=u_{0n}$). Note also that in the proof of Lemma 1, we can replace the space $H^{\frac{1}{2}}$ by $L^{2}$, so that in the same context if $u_{n}(0)\to u_{0}$ in $L^{2}$ as $n\to+\infty$ then for all $t\in\mathbb{R}$, $\rho_{n}(t)\to\rho(t)$ as $n\to+\infty$ (see continuous dependence in $L^{2}$ also in [11]). Finally, for future reference, we justify that if $u_{n}\rightharpoonup u$ in $H^{\frac{1}{2}}$ weak, then $y(u_{n})\to y(u)$, where $y(u)$ is defined in the proof of Lemma 1. Indeed, in this proof, by the decay of $Q^{\prime}(x)$, we can also replace $H^{\frac{1}{2}}$ by the weighted space $L^{2}(\frac{1}{1+|x|}dx)$, so that if $u_{n}\to u$ in $L^{2}_{loc}$ and $\|u_{n}\|_{L^{2}}+\|u\|_{L^{2}}\leq C$, then $y(u_{n})\to y(u)$ as $n\to+\infty$. In the rest of this section, we present monotonicity arguments on $L^{2}$ quantities for both $u(t)$ and $\eta(t)$, in the context of Lemma 1. These results are reminiscent of similar results for the gKdV equation in [16] and [19], but due to the nonlocal nature of the operator $\mathcal{H}$, the proofs are more involved. ### 2.2 Monotonicity results for $u(t)$ Let $A>1$ to be chosen later and set $\varphi(x)=\varphi_{A}(x)=\frac{\pi}{2}+\arctan\Big{(}\frac{x}{A}\Big{)}\quad\hbox{so that}\quad\varphi^{\prime}(x)=\frac{\frac{1}{A}}{1+(\frac{x}{A})^{2}}>0.$ (16) ###### Proposition 1. Let $0<\lambda<1$. Under the assumptions of Lemma 1, for $\alpha_{0}$ small enough and $A$ large enough, there exists $C>0$ such that for all $x_{0}>1$, $t_{1}\leq t_{2}$, 1. 1. Monotonicity on the right of the soliton: $\int u^{2}(t_{2},x)\varphi(x-\rho(t_{2})-x_{0})dx\leq\int u^{2}(t_{1},x)\varphi(x-\rho(t_{1})-\lambda(t_{2}-t_{1})-x_{0})dx+\frac{C}{x_{0}}.$ (17) 2. 2. Monotonicity on the left of the soliton: $\int u^{2}(t_{2},x)\varphi(x-\rho(t_{2})+\lambda(t_{2}-t_{1})+x_{0})dx\leq\int u^{2}(t_{1},x)\varphi(x-\rho(t_{1})+x_{0})dx+\frac{C}{x_{0}}.$ (18) _Proof of Proposition 1._ First, we note that (18) is a consequence of (17) and the $L^{2}$ norm conservation. Indeed, let $v(t,x)=u(-t,-x)$. Then $v(t)$ is a solution of (1) satisfying the assumptions of Lemma 1 and $\rho_{v}(t)=-\rho(-t)$. Thus, from (17) applied on $v(t,x)$, we deduce $\int u^{2}(-t_{2},x)\varphi(-x+\rho(-t_{2})-x_{0})dx\leq\int u^{2}(-t_{1},x)\varphi(-x+\rho(-t_{1})-\lambda(t_{2}-t_{1})-x_{0})dx+\frac{C}{x_{0}}.$ Since $\varphi(x)=\pi-\varphi(-x)$, from $\int u^{2}(-t_{2})=\int u^{2}(-t_{1})$, we obtain $\int u^{2}(-t_{2},x)\varphi(x-\rho(-t_{2})+x_{0})dx+\frac{C}{x_{0}}\geq\int u^{2}(-t_{1},x)\varphi(x-\rho(-t_{1})+\lambda(t_{2}-t_{1})+x_{0})dx,$ which is exactly formula (18) for $t_{2}^{\prime}=-t_{1}$, $t_{1}^{\prime}=-t_{2}$. We are reduced to prove (17). We perform calculations on regular solutions and then use density arguments and continuous dependence to obtain the result in the framework of Lemma 1. First, we recall a Kato type identity for solutions of the BO equation. By direct computations, we have $\begin{split}\frac{1}{2}\,\frac{d}{dt}\int u^{2}(t,x)\varphi(x)dx&=\int u_{t}u\varphi(x)dx=-\int(\mathcal{H}u_{xx}+uu_{x})u\varphi(x)dx\\\ &=\int(\mathcal{H}u_{x})(u\varphi^{\prime}(x)+u_{x}\varphi(x))dx+\frac{1}{3}\int u^{3}\varphi^{\prime}(x)dx.\end{split}$ (19) For the first term in (19), we prove the following result. ###### Lemma 2. For all $u\in H^{1}(\mathbb{R})$, $\int(\mathcal{H}u_{x})u\varphi^{\prime}(x)dx\leq\frac{C}{A}\int u^{2}\varphi^{\prime}(x)dx.$ (20) _Proof of Lemma 2_. For $f\in L^{2}(\mathbb{R})$, we define the harmonic extension of $f$ on $\mathbb{R}\times\mathbb{R}_{+}=\mathbb{R}^{2}_{+}$, $\forall x\in\mathbb{R},\quad F(x,0)=f(x)\quad\mathrm{and}\quad F(x,y)=\frac{1}{\pi}\int_{-\infty}^{\infty}\frac{y}{(x-x^{\prime})^{2}+y^{2}}\,f(x^{\prime})\,dx^{\prime},\quad\hbox{if $y>0$.}$ (21) In particular, recall that $\mathcal{H}f^{\prime}(x)=\partial_{y}F(x,0)$ (see Stein [24] Chapter III, and the Introduction of Toland [26]). We denote by $\Phi(x,y)$ the harmonic extension of $\varphi^{\prime}(x)$ and $U(x,y)$ the harmonic extension of $u(x)$ on $\mathbb{R}\times\mathbb{R}_{+}$. Note that $\Phi(x,y)$ is explicitly given by $\Phi(x,y)=\frac{1}{A}\frac{1+\frac{y}{A}}{(\frac{x}{A})^{2}+(1+\frac{y}{A})^{2}}.$ (22) Then, by the Green Formula on $\mathbb{R}^{2}_{+}$ (using decay properties of $\Phi(x,y)$ and $\Delta U^{2}=2|\nabla U|^{2}$), we obtain formally $\begin{split}\int(\mathcal{H}u_{x})u\varphi^{\prime}&=\int\partial_{y}U(t,x,0)U(t,x,0)\Phi(x,0)dx=\frac{1}{2}\int_{y=0}\partial_{y}(U^{2})\Phi dx\\\ &=-\frac{1}{2}\iint_{\mathbb{R}^{2}_{+}}(\Delta U^{2})\Phi+\frac{1}{2}\iint_{\mathbb{R}^{2}_{+}}U^{2}\Delta\Phi+\frac{1}{2}\int_{y=0}U^{2}\partial_{y}\Phi\\\ &=-\iint_{\mathbb{R}^{2}_{+}}|\nabla U|^{2}\Phi+\frac{1}{2}\int u^{2}(\mathcal{H}\varphi^{\prime\prime})dx.\end{split}$ (23) See Appendix A.1 for a rigorous proof of (23). Since $\Phi\geq 0$ on $\mathbb{R}^{2}_{+}$, we obtain $\int(\mathcal{H}u_{x})u\varphi^{\prime}\leq\frac{1}{2}\int u^{2}(\mathcal{H}\varphi^{\prime\prime}).$ (24) By explicit computations, since $\mathcal{H}\big{(}\frac{1}{1+x^{2}}\big{)}=-\frac{x}{1+x^{2}}$, we have $\mathcal{H}\varphi^{\prime}=-\frac{1}{A^{2}}\frac{x}{1+(\frac{x}{A})^{2}},\quad\mathcal{H}\varphi^{\prime\prime}=\frac{1}{A}\varphi^{\prime}-2(\varphi^{\prime})^{2}\quad\hbox{and}\quad\mathcal{H}\varphi^{\prime\prime}\leq\frac{1}{A}\varphi^{\prime}.$ (25) Lemma 2 follows. For the second term in (19), we have the following. ###### Lemma 3. For all $u\in H^{1}(\mathbb{R})$, $\left|\int(\mathcal{H}u_{x})u_{x}\varphi dx\right|\leq\frac{C}{A}\int u^{2}\varphi^{\prime}(x)dx.$ (26) _Proof of Lemma 3_. We prove (26) for $u$ smooth and compactly supported in $\mathbb{R}$, the general case will follow by a density argument. Since the limit in (2) holds in $L^{2}$ (see Stein [24], Chapter II), we have $\begin{split}&\int(\mathcal{H}u_{x})u_{x}\varphi dx=\frac{1}{\pi}\int\mathrm{p.v.}\bigg{(}\int\frac{u_{x}(y)}{y-x}dy\bigg{)}u_{x}(x)\varphi(x)dx\\\ &=\frac{1}{\pi}\lim_{\varepsilon\to 0}\iint_{|y-x|>\varepsilon}u_{x}(y)u_{x}(x)\frac{\varphi(x)}{y-x}dydx\\\ &=\frac{1}{2\pi}\iint u_{x}(y)u_{x}(x)\,\frac{\varphi(x)-\varphi(y)}{y-x}dxdy=\frac{1}{2\pi}\iint u(y)u(x)K_{\varphi}(x,y)dxdy,\end{split}$ (27) by symmetry and then integration by parts, where $K_{\varphi}(x,y)=-\frac{\partial^{2}}{\partial x\partial y}\bigg{(}\frac{\varphi(x)-\varphi(y)}{x-y}\bigg{)}=\frac{2(\varphi(x)-\varphi(y))-(\varphi^{\prime}(x)+\varphi^{\prime}(y))(x-y)}{(x-y)^{3}}.$ (28) Note that all the integrals in (27) make sense since $u(x)$ is compactly supported, $(\varphi(x)-\varphi(y))/(x-y)$ is bounded and moreover, by subtracting the following two Taylor formulas: $\begin{split}&\varphi(x)=\varphi(y)+(x-y)\varphi^{\prime}(y)+\frac{1}{2}(x-y)^{2}\varphi^{\prime\prime}(y)+\frac{1}{6}(x-y)^{3}\varphi^{\prime\prime\prime}(x_{1}),\\\ &\varphi(y)=\varphi(x)+(y-x)\varphi^{\prime}(x)+\frac{1}{2}(y-x)^{2}\varphi^{\prime\prime}(x)+\frac{1}{6}(y-x)^{3}\varphi^{\prime\prime\prime}(x_{2}),\end{split}$ where $x_{1},x_{2}\in(y,x)$, we find: $K_{\varphi}(x,y)=\frac{1}{2}\frac{\varphi^{\prime\prime}(y)-\varphi^{\prime\prime}(x)}{x-y}+\frac{1}{6}(\varphi^{\prime\prime\prime}(x_{1})+\varphi^{\prime\prime\prime}(x_{2})),$ (29) which is also bounded on $\mathbb{R}^{2}$. Note also that by explicit computations, we have $\varphi^{\prime\prime\prime}(x)=\frac{\varphi^{\prime}(x)}{A^{2}}\left(\frac{-2}{1+\big{(}\frac{x}{A}\big{)}^{2}}+\frac{8\big{(}\frac{x}{A}\big{)}^{2}}{\big{(}1+\big{(}\frac{x}{A}\big{)}^{2}\big{)}^{2}}\right)=\frac{\varphi^{\prime}(x)}{A}\left(-2\varphi^{\prime}(x)+\frac{8}{A}{x^{2}}(\varphi^{\prime})^{2}\right).$ (30) We are reduced to prove the following estimate $\left|\iint u(y)u(x)K_{\varphi}(x,y)dxdy\right|\leq\frac{C}{A}\int u^{2}\varphi^{\prime}(x)dx.$ (31) We consider only the case $|y|<|x|$ (by symmetry), and we divide $\\{(x,y),\,:\,|y|<|x|\\}$ into the following regions: $\bullet$ $\Sigma_{1}=\\{(x,y)\,:\,x>A,\,0<y<\frac{x}{2}\\}.$ For $(x,y)\in\Sigma_{1}$, by (28) and the fact that $\varphi^{\prime}$ is decreasing on $\mathbb{R}^{+}$, we have $|K_{\varphi}(x,y)|\leq\frac{4}{(x-y)^{2}}\sup_{[y,x]}\varphi^{\prime}\leq\frac{16}{x^{2}}\varphi^{\prime}(y)=\frac{16}{A^{2}}\frac{1}{(\frac{x}{A})^{2}}\varphi^{\prime}(y)\leq\frac{32}{A}\varphi^{\prime}(x)\varphi^{\prime}(y).$ Thus, by Cauchy-Schwarz inequality, since $\int\varphi^{\prime}(x)=\pi$, we obtain $\begin{split}\left|\iint_{\Sigma_{1}}u(y)u(x)K_{\varphi}(x,y)dxdy\right|&\leq\frac{C}{A}\int|u(x)|\varphi^{\prime}(x)dx\int|u(y)|\varphi^{\prime}(y)dy\\\ &\leq\frac{C\pi}{A}\int u^{2}(x)\varphi^{\prime}(x)dx.\end{split}$ The case of the region $\Sigma_{1}^{-}=\\{(x,y)\,:\,x<-A,\,\frac{x}{2}<y<0\\}$ is similar. $\bullet$ $\Sigma_{2}=\\{(x,y)\,:\,x>A,\,-x<y<0\\}.$ For $(x,y)\in\Sigma_{2}$, we have by (28), $|x-y|=x-y>x>\frac{1}{2}(x+A)$, $\varphi^{\prime}(y)>\varphi^{\prime}(x)$ and so by (28) and $\varphi$ bounded, we obtain $|K_{\varphi}(x,y)|\leq\frac{C}{(x+A)^{3}}+\frac{C\varphi^{\prime}(y)}{x^{2}}.$ For the term $\frac{C\varphi^{\prime}(y)}{x^{2}}$, we argue as for $\Sigma_{1}$. For the other term, by Cauchy-Schwarz’ inequality and the expression of $\varphi^{\prime}$, we have $\begin{split}&\iint_{\Sigma_{2}}|u(y)||u(x)|\frac{1}{(x+A)^{3}}dxdy\leq C\left(\iint_{\Sigma_{2}}\frac{u^{2}(x)}{(x+A)^{3}}dxdy\right)^{\frac{1}{2}}\left(\iint_{\Sigma_{2}}\frac{u^{2}(y)}{(x+A)^{3}}dxdy\right)^{\frac{1}{2}}\\\ &\leq C\left(\frac{1}{2}\int_{x>A}\frac{u^{2}(x)}{(x+A)^{2}}dx\right)^{\frac{1}{2}}\left(\frac{1}{2}\int_{y<0}\frac{u^{2}(y)}{(-y+A)^{2}}dy\right)^{\frac{1}{2}}\leq\frac{C^{\prime}}{A}\int u^{2}(x)\varphi^{\prime}(x)dx.\end{split}$ The case of $\Sigma_{2}^{-}=\\{(x,y)\,:\,x<-A,\,0<y<-x\\}$ is similar to $\Sigma_{2}$. $\bullet$ $\Sigma_{3}=\\{(x,y)\,:\,|x|<A,\,|y|<|x|\\}.$ For $(x,y)\in\Sigma_{3}$, and $|s|<|x|$, we have $\frac{1}{2A}\leq\varphi^{\prime}(s)\leq\frac{1}{A}$ and thus, from (29) and (30), we obtain $|K_{\varphi}(x,y)|\leq C\sup_{|s|<|x|}|\varphi^{\prime\prime\prime}(s)|\leq\frac{C}{A^{3}}\leq\frac{C}{A}\varphi^{\prime}(x)\varphi^{\prime}(y)$. We finish as for $\Sigma_{1}$. $\bullet$ $\Sigma_{4}=\\{(x,y)\,:\,x>A,\,\frac{1}{2}x<y<x\\}.$ For $(x,y)\in\Sigma_{4}$, and $y<s<x$, we have from (30): $|\varphi^{\prime\prime\prime}(s)|\leq\frac{10}{A}(\varphi^{\prime}(s))^{2}\leq\frac{10}{A}\varphi^{\prime}(y)\varphi^{\prime}(s)\leq\frac{40}{A}\varphi^{\prime}(y)\varphi^{\prime}(x)$ thus $|K_{\varphi}(x,y)|\leq\frac{C}{A}\varphi^{\prime}(x)\varphi^{\prime}(y),$ and we conclude as for $\Sigma_{1}$. The case of $\Sigma_{4}^{-}=\\{(x,y)\,:\,x<-A,\,x<y<\frac{x}{2}\\}$ is similar In conclusion, we have obtained (31) and Lemma 3 is proved. From (19), Lemmas 2 and 3, there exists $C_{0}>0$ such that $\frac{1}{2}\,\frac{d}{dt}\int u^{2}(t,x)\varphi(x)dx\leq\frac{C_{0}}{A}\int u^{2}(t,x)\varphi^{\prime}(x)dx+\frac{1}{3}\int|u^{3}(t,x)|\varphi^{\prime}(x)dx.$ (32) Now, let $u(t)$ be a solution of (1) satisfying the assumptions of Lemma 1 on $\mathbb{R}$. Let $\eta(t)$, $\rho(t)$ be associated to the decomposition of $u(t)$ on $I$ as in Lemma 1. Let $0<\lambda<1$, $t_{0}\in[t_{1},t_{2}]$ and $x_{0}\geq 1$. For any $t\in[t_{1},t_{0}]$, $x\in\mathbb{R}$, we set $\widetilde{x}=x-x_{0}-\rho(t)-\lambda(t_{0}-t),\quad M_{\varphi}(t)=\frac{1}{2}\int u^{2}(t,x)\varphi(\widetilde{x})dx.$ (33) Then, by (32), we find $M_{\varphi}^{\prime}(t)\leq-\frac{1}{2}\left(\rho^{\prime}(t)-\lambda-\frac{2C_{0}}{A}\right)\int u^{2}(t)\varphi^{\prime}(\widetilde{x})+\frac{1}{3}\int|u(t)|^{3}\varphi^{\prime}(\widetilde{x}).$ (34) Fix now $A>0$ large enough so that $\frac{2C_{0}}{A}\leq\frac{1}{4}(1-\lambda)$. Then, by (13), we choose $\alpha_{0}>0$ small enough so that $\forall t\in I$, $\rho^{\prime}(t)-\lambda>\frac{1}{2}(1-\lambda)$. Therefore, we obtain $M_{\varphi}^{\prime}(t)\leq-\ \frac{1}{8}(1-\lambda)\int u^{2}(t)\varphi^{\prime}(\widetilde{x})+\frac{1}{3}\int|u(t)|^{3}\varphi^{\prime}(\widetilde{x}).$ (35) Finally, we estimate the nonlinear term $\int|u(t)|^{3}\varphi^{\prime}(\widetilde{x})$. We first observe: $\int|u(t)|^{3}\varphi^{\prime}(\widetilde{x})\leq C\int Q^{3}(x{-}\rho(t))\varphi^{\prime}(\widetilde{x})dx+C\int|\eta(t,x)|^{3}\varphi^{\prime}(\widetilde{x})dx.$ (36) For the first term, we distinguish two regions in $x$: $\bullet$ $\Omega_{1}=\\{x\,:\,x<\rho(t)+\frac{1}{2}x_{0}+\frac{1}{2}\lambda(t_{0}-t)\\}$. For $x\in\Omega_{1}$, we have $\widetilde{x}<-\frac{1}{2}x_{0}-\frac{1}{2}\lambda(t_{0}-t)$, and thus $\varphi^{\prime}(\widetilde{x})\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.$ This implies $\int_{\Omega_{1}}Q^{3}(x{-}\rho(t))\varphi^{\prime}(\widetilde{x})\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}\int Q^{3}\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.$ (37) $\bullet$ $\Omega_{2}=\\{x>\rho(t)+\frac{1}{2}x_{0}+\frac{1}{2}\lambda(t_{0}-t)\\}$. For $x\in\Omega_{2}$, we have $x-\rho(t)>\frac{1}{2}x_{0}+\frac{1}{2}\lambda(t_{0}-t)$ and thus $Q^{3}(x{-}\rho(t))\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{6}},\quad\int_{\Omega_{2}}Q^{3}(x{-}\rho(t))\varphi^{\prime}(\widetilde{x})dx\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{6}}.$ Now, we claim $\int|\eta(t,x-\rho(t))|^{3}\varphi^{\prime}(\widetilde{x})dx\leq C\alpha_{0}\int\eta^{2}(t,x-\rho(t))\varphi^{\prime}(\widetilde{x})dx,$ (38) where $C$ is independent of $A$. See proof of (38) in Appendix A.2. Moreover, as before, we find $\begin{split}\int\eta^{2}(t,x-\rho(t))\varphi^{\prime}(\widetilde{x})dx&\leq C\int(u^{2}(t,x)+Q^{2}(x{-}\rho(t)))\varphi^{\prime}(\widetilde{x})dx\\\ &\leq C\int u^{2}(t,x)\varphi^{\prime}(\widetilde{x})dx+\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.\end{split}$ Thus, it follows from (35)–(38) that for $\alpha_{0}>0$ small enough, $\forall t\in[t_{1},t_{0}],$ $\begin{split}M_{\varphi}^{\prime}(t)&\leq-\ \frac{1}{8}(1-\lambda)\int u^{2}(t)\varphi^{\prime}(\widetilde{x})+C\alpha_{0}\int u^{2}(t)\varphi^{\prime}(\widetilde{x})+\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}\\\ &\leq-\ \frac{1}{16}(1-\lambda)\int u^{2}(t)\varphi^{\prime}(\widetilde{x})+\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.\end{split}$ (39) Let $t\in[t_{1},t_{0}]$. By integration of (39) on $[t,t_{0}]$, since $\int_{t}^{t_{0}}\frac{dt^{\prime}}{(x_{0}+\lambda(t_{0}-t^{\prime}))^{2}}=\frac{1}{\lambda x_{0}}\int_{0}^{\frac{\lambda(t_{0}-t)}{x_{0}}}\frac{dt^{\prime\prime}}{(1+t^{\prime\prime})^{2}}\leq\frac{C}{x_{0}},\quad(t^{\prime\prime}=\frac{\lambda}{x_{0}}(t_{0}-t^{\prime}))$ we find: $\begin{split}&\int u^{2}(t_{0},x)\varphi(x-x_{0}-\rho(t_{0}))dx+\frac{1}{C}\int_{t}^{t_{0}}\int u^{2}(t^{\prime},x)\varphi^{\prime}(x-x_{0}-\rho(t^{\prime})-\lambda(t-t^{\prime}))dxdt^{\prime}\\\ &\leq\int u^{2}(t,x)\varphi(x-x_{0}-\rho(t)-\lambda(t_{0}-t))dx+\frac{C}{x_{0}}.\end{split}$ (40) By density and continuous dependence ([11]) estimate (40) also holds for $H^{\frac{1}{2}}$ solutions. ### 2.3 Monotonicity results for $\eta(t)$ Here, we present similar monotonicity arguments for $\eta(t)$. See [19] for similar results in the case of the gKdV equations. ###### Proposition 2. Let $0<\lambda<1$. Under the assumptions of Lemma 1, for $\alpha_{0}$ small enough and $A$ large enough, there exists $C>0$ such that for all $x_{0}>1$, $t_{1}\leq t_{2}$, $\begin{split}&\int\eta^{2}(t_{2},x)(\varphi(x-x_{0})-\varphi(-x_{0}))\,dx\\\ &\leq\int\eta^{2}(t_{1},x)(\varphi(x-\lambda(t_{2}-t_{1})-x_{0})-\varphi(-x_{0}-\lambda(t_{2}-t_{1})))dx+C\int_{t_{1}}^{t_{2}}\frac{\|\eta(t)\|_{L^{2}}^{2}}{(x_{0}+\lambda(t_{2}-t))^{2}}dt.\end{split}$ ###### Remark 3. With respect to Proposition 1, we need to modify slighty the function in the integral ($\varphi(x-x_{0})-\varphi(-x_{0})$ instead of $\varphi(x-x_{0})$) to remove some terms in the second member, see comments in the proof. This estimate is clearly improving Proposition 1 since the remainder term can now be controlled by $\frac{C}{x_{0}}\sup_{t}\|\eta(t)\|_{L^{2}}^{2}$. As for $u(t)$ in the proof of Proposition 1, we have by direct computations using (14), $\displaystyle\frac{1}{2}\frac{d}{dt}\int\eta^{2}(t,x)\varphi(x)dx=\int\eta_{t}\eta\varphi(x)dx$ $\displaystyle=-\int(\mathcal{L}\eta)(\eta\varphi^{\prime}+\eta_{x}\varphi)+\frac{1}{3}\int\eta^{3}\varphi^{\prime}+(\rho^{\prime}-1)\left(\int Q^{\prime}\eta\varphi-\tfrac{1}{2}\int\eta^{2}\varphi^{\prime}\right)$ $\displaystyle=\int(\mathcal{H}\eta_{x})\eta\varphi^{\prime}+\int(\mathcal{H}\eta_{x})\eta_{x}\varphi-\frac{1}{2}\int\eta^{2}\varphi^{\prime}+\frac{1}{2}\int\eta^{2}(Q\varphi^{\prime}-Q^{\prime}\varphi)+\frac{1}{3}\int\eta^{3}\varphi^{\prime}$ $\displaystyle+(\rho^{\prime}-1)\left(\int Q^{\prime}\eta\varphi-\tfrac{1}{2}\int\eta^{2}\varphi^{\prime}\right).$ (41) Let $0<\lambda<1$ and $\overline{x}=x-x_{0}-\lambda(t_{0}-t)$. Then, by Lemmas 2 and 3, we get $\displaystyle\frac{d}{dt}\int\eta^{2}\varphi(\overline{x})$ $\displaystyle\leq-\left(\rho^{\prime}(t)-\lambda-\frac{2C_{0}}{A}\right)\int\eta^{2}\varphi^{\prime}(\overline{x})+\int\eta^{2}(Q\varphi^{\prime}(\overline{x})-Q^{\prime}\varphi(\overline{x}))+\frac{2}{3}\int|\eta|^{3}\varphi^{\prime}(\overline{x})$ $\displaystyle+2(\rho^{\prime}-1)\int Q^{\prime}\eta\varphi(\overline{x}).$ Now, as in the proof of Proposition 1, we fix $A>1$ such that $\frac{2C_{0}}{A}\leq\frac{1}{4}(1-\lambda)$ and $\alpha_{0}$ small enough so that $\rho^{\prime}-\lambda>\frac{1}{2}(1-\lambda)$ by (13). Then, by (38) and (13), we can choose $\alpha_{0}>0$ small enough so that $\frac{2}{3}\int|\eta|^{3}\varphi^{\prime}(\overline{x})\leq\frac{1}{8}(1-\lambda)\int\eta^{2}\varphi^{\prime}(\overline{x}).$ Thus, we obtain $\frac{d}{dt}\int\eta^{2}\varphi(\overline{x})\leq-\frac{1}{8}\left(1-\lambda\right)\int\eta^{2}\varphi^{\prime}(\overline{x})+\int\eta^{2}(Q\varphi^{\prime}(\overline{x})-Q^{\prime}\varphi(\overline{x}))+2(\rho^{\prime}-1)\int Q^{\prime}\eta\varphi(\overline{x}).$ At this point, note that the term $\int\eta^{2}Q^{\prime}\varphi(\overline{x})$ has no sign, and since $\varphi(y)\sim\frac{C}{|y|}$ as $y\to-\infty$, this term can only be controlled by $\frac{C}{(x_{0}+\lambda(t_{0}-t))}\int\eta^{2}$, which is not sufficient for our purposes. We modify slightly the functional to cancel the main order of this term. Indeed, since $\int\eta Q^{\prime}=0$, using (14), we have $\frac{d}{dt}\int\eta^{2}=2\int Q\eta\eta_{x}=-\int Q^{\prime}\eta^{2}.$ Therefore, using also $\int Q^{\prime}\eta=0$, we get $\displaystyle\frac{d}{dt}\int\eta^{2}\left(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t))\right)\leq-\frac{1}{8}\left(1-\lambda\right)\int\eta^{2}\varphi^{\prime}(\overline{x})$ $\displaystyle+\int\eta^{2}\left(Q\varphi^{\prime}(\overline{x})-Q^{\prime}(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t)))\right)$ $\displaystyle+2(\rho^{\prime}-1)\int\eta Q^{\prime}(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t)))-\lambda\varphi^{\prime}(-(x_{0}+\lambda(t_{0}-t)))\int\eta^{2}.$ Now, we claim the following estimate $\forall x\in\mathbb{R},\quad Q(x)\varphi^{\prime}(\overline{x})+\left|Q(x)\left(\varphi(\overline{x})-\varphi(-(x_{0}+\lambda(t_{0}-t)))\right)\right|\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.$ (42) Since $Q(x)\varphi^{\prime}(\overline{x})\leq\frac{C}{(1+x^{2})(1+(x-x_{0}-\lambda(t_{0}-t))^{2})}$ (recall that the value of $A$ has been fixed) estimate (42) is clear for $Q(x)\varphi^{\prime}(\overline{x})$ by considering the two regions $|x|>\frac{1}{2}(x_{0}+\lambda(t_{0}-t))$ and $|x|<\frac{1}{2}(x_{0}+\lambda(t_{0}-t))$. For the other term, we first note that since $|Q(x)|\leq\frac{C}{1+x^{2}}$ and $\varphi$ is bounded, the estimate is clear for $|x|>\frac{1}{2}(x_{0}+\lambda(t_{0}-t))$. For $|x|<\frac{1}{2}(x_{0}+\lambda(t_{0}-t))$, we have $|\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t))|\leq|x|\sup_{[\frac{1}{2}(x_{0}+\lambda(t_{0}-t),\frac{3}{2}(x_{0}+\lambda(t_{0}-t)]}\varphi^{\prime}\leq\frac{C|x|}{(x_{0}+\lambda(t_{0}-t))^{2}};$ thus, for such $x$, we obtain the following estimate which finishes the proof of (42): $\left|Q(x)\left(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t))\right)\right|\leq\frac{C}{(x_{0}+\lambda(t_{0}-t))^{2}}.$ By (13) and (42), and since $|Q^{\prime}(x)|\leq\frac{C}{1+|x|}Q(x)$, we obtain $\left|\int\eta^{2}(Q\varphi^{\prime}(\overline{x})-Q^{\prime}(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t))))\right|\leq\frac{C\|\eta(t)\|_{L^{2}}^{2}}{(x_{0}+\lambda(t_{0}-t))^{2}},$ (43) $\begin{split}\left|(\rho^{\prime}-1)\int Q^{\prime}\eta(\varphi(\overline{x})-\varphi(-x_{0}-\lambda(t_{0}-t)))\right|&\leq\frac{C\|\eta(t)\|_{L^{2}}}{(x_{0}+\lambda(t_{0}-t))^{2}}\int\frac{|\eta|}{1+|x|}\\\ &\leq\frac{C\|\eta(t)\|_{L^{2}}^{2}}{(x_{0}+\lambda(t_{0}-t))^{2}}.\end{split}$ (44) The conclusion is thus: $\displaystyle\frac{d}{dt}\int\eta^{2}\left(\varphi(\overline{x})-\varphi(-(x_{0}+\lambda(t_{0}-t)))\right)\leq-\frac{1}{8}\left(1-\lambda\right)\int\eta^{2}\varphi^{\prime}(\overline{x})+\frac{C\|\eta(t)\|_{L^{2}}^{2}}{(x_{0}+\lambda(t_{0}-t))^{2}}.$ By integration on $[t,t_{0}]$, we get $\begin{split}&\int\eta^{2}(t_{0},x)\left(\varphi(x-x_{0})-\varphi(-x_{0})\right)dx+\frac{1}{C}\int_{t}^{t_{0}}\int\eta^{2}(t^{\prime},x)\varphi^{\prime}(x-x_{0}-\lambda(t_{0}-t^{\prime}))dxdt^{\prime}\\\ &\leq\int\eta^{2}(t,x)\left(\varphi(x-x_{0}-\lambda(t_{0}-t))-\varphi(-x_{0}-\lambda(t_{0}-t))\right)dx+C\int_{t}^{t_{0}}\frac{\|\eta(t^{\prime})\|_{L^{2}}^{2}dt^{\prime}}{(x_{0}+\lambda(t_{0}-t^{\prime}))^{2}}.\end{split}$ ## 3 Linear Liouville property In this section, we prove the following result. ###### Theorem 3. Let $w\in C(\mathbb{R},L^{2}(\mathbb{R}))\cap L^{\infty}(\mathbb{R},L^{2}(\mathbb{R}))$ be a solution of $w_{t}=(\mathcal{L}w)_{x}+\beta(t)Q^{\prime},\quad(t,x)\in\mathbb{R}^{2},\quad\text{where $\beta$ is continuous,}$ (45) satisfying $\forall t\in\mathbb{R},\quad\int w(t,x)Q(x)dx=\int w(t,x)Q^{\prime}(x)dx=0,$ (46) $\forall t\in\mathbb{R},~{}\forall x_{0}>1,\quad\int_{|x|>x_{0}}w^{2}(t,x)dx\leq\frac{C}{x_{0}}.$ (47) Then $w\equiv 0\quad\text{on $\mathbb{R}^{2}$.}$ (48) This result is similar to Theorem 3 in [15]. For the proof, we follow the strategy of [13], [18], introducing a dual problem whose operator has better spectral properties. Since $w(t)$ is only $L^{2}$ and has a weak decay at infinity in space, we will need to regularize and localize the dual solution. For the sake of clarity, we now present the formal argument. The complete justification will be presented in Sections 3.1 and 3.2. Multiplying the equation of $w(t)$ by $xw(t)$, we get $\frac{d}{dt}\int xw^{2}=-2\int(\mathcal{H}w)w_{x}-\int w^{2}+\int w^{2}(Q-xQ^{\prime})+2\beta(t)\int xQ^{\prime}w,$ where $(\int(Q^{\prime})^{2})\beta(t)=\int w\mathcal{L}(Q^{\prime\prime})$ (multiply the equation of $w$ by $Q^{\prime}$ and use $\int wQ^{\prime}=0$). But it is not clear how to study the spectral properties of the operator $2\int(\mathcal{H}w)w_{x}+\int w^{2}-\int w^{2}(Q-xQ^{\prime})+\frac{2}{\int(Q^{\prime})^{2}}\left(\int w\mathcal{L}Q^{\prime\prime}\right)\left(\int xQ^{\prime}w\right).$ Moreover, the decay estimate (47) is not quite enough to control $\int xw^{2}$. Therefore, we instead rely on the dual problem, setting $v=\mathcal{L}w$. Since $\mathcal{L}Q^{\prime}=0$ (direct calculation), we obtain the following equation for $v(t)$: $v_{t}=\mathcal{L}(v_{x})$. Multiplying the equation by $xv$, we obtain $-\frac{d}{dt}\int xv^{2}=2\int(\mathcal{H}v)v_{x}+\int v^{2}-\int v^{2}(Q+xQ^{\prime}).$ Note that the operator in $v$ is much easier to study since now the potential $xQ^{\prime}$ has a positive contribution ($xQ^{\prime}\leq 0$), moreover, there is no scalar product. In fact, we will obtain (see Proposition 4) the positivity of this operator under the orthogonality condition $\int v(xQ)^{\prime}=0$. Observe that $\int v(xQ)^{\prime}=\int(\mathcal{L}w)(xQ)^{\prime}=-\int wQ=0$ since $\mathcal{L}((xQ)^{\prime})=-Q$ (see (78)). Provided that $\int|x|v^{2}(t)\leq C$, we would obtain from the above identity $\int_{-\infty}^{+\infty}\|v(t)\|_{H^{\frac{1}{2}}}^{2}dt\leq C,$ which says that for a subsequence $t_{n}\to+\infty$, $v(t_{n})\to 0$, $w(t_{n})\to 0$. Combined with energy conservation ($(\mathcal{L}w(t),w(t))=C$) and Lemma 15 below, this gives $w\equiv 0$. But (47) is not enough to obtain the estimate $\int|x|v^{2}(t)\leq C$ In fact, since $w(t)$ is only in $L^{2}$, we both need to localize and regularize the dual problem. ### 3.1 Proof of Theorem 3 assuming positivity of a quadratic form ###### Lemma 4 (Regularized dual problem). There exists $\gamma_{0}>0$ such that for any $0<\gamma<\gamma_{0}$, the following is true. Let $v=(1-\gamma\partial_{x}^{2})^{-1}(\mathcal{L}w)$. Then, $v\in C(\mathbb{R},H^{1}(\mathbb{R}))\cap L^{\infty}(\mathbb{R},H^{1}(\mathbb{R}))$ and 1. 1. Equation of $v$. $v_{t}=\mathcal{L}(v_{x})-\gamma(1-\gamma\partial_{x}^{2})^{-1}(2v_{xx}Q^{\prime}+v_{x}Q^{\prime\prime}).$ (49) 2. 2. Decay of $v$. $\forall t\in\mathbb{R},x_{0}>1,\quad\int_{|x|>x_{0}}(v_{x}^{2}(t,x)+v^{2}(t,x))dx\leq\frac{C_{\gamma}}{x_{0}^{\frac{3}{4}}}.$ (50) 3. 3. Virial type estimate. $\int_{-\infty}^{+\infty}\frac{1}{(1+t^{2})^{\frac{2}{5}}}\|v(t)\|_{H^{1}}^{2}dt<C.$ (51) _Proof of Lemma 4._ First, since $\sup_{t}\|w(t)\|_{L^{2}}\leq C$, we obtain $\sup_{t}\|v(t)\|_{H^{1}}\leq C_{\gamma}$ (see Claim 1 below). _1\. Equation of $v$._ Let $\widetilde{v}=\mathcal{L}w$ so that $w_{t}=\widetilde{v}_{x}+\beta Q^{\prime}$. Since $\mathcal{L}Q^{\prime}=0$, the function $\widetilde{v}$ satisfies $\widetilde{v}_{t}=\mathcal{L}w_{t}=\mathcal{L}(\widetilde{v}_{x})$. Now, we introduce a regularization of the function $\widetilde{v}$. For $0<\gamma<\frac{1}{2}$ to be chosen later small enough, we set: $v(t,x)=(1-\gamma\partial_{x}^{2})^{-1}\widetilde{v}(t,x)\quad\text{or equivalently}\quad v-\gamma v_{xx}=\widetilde{v}=\mathcal{L}w.$ (52) Then, $v(t,x)$ satisfies the following equation $v_{t}=(1-\gamma\partial_{x}^{2})^{-1}\widetilde{v}_{t}=(1-\gamma\partial_{x}^{2})^{-1}\mathcal{L}(\widetilde{v}_{x})=\mathcal{L}(v_{x})-(1-\gamma\partial_{x}^{2})^{-1}(\widetilde{v}_{x}Q)+v_{x}Q.$ But $-(1-\gamma\partial_{x}^{2})^{-1}(\widetilde{v}_{x}Q)+v_{x}Q=(1-\gamma\partial_{x}^{2})^{-1}(-2\gamma v_{xx}Q^{\prime}-\gamma v_{x}Q^{\prime\prime})$, and so $v_{t}=\mathcal{L}(v_{x})-\gamma(1-\gamma\partial_{x}^{2})^{-1}(2v_{xx}Q^{\prime}+v_{x}Q^{\prime\prime}).$ (53) _2\. Decay estimate on $v$._ By using the decay on $w(t)$, we claim $\forall x_{0}>1,\forall t,\quad\int_{|x|\geq x_{0}}(v_{x}^{2}(t,x)+v^{2}(t,x))dx\leq\frac{C_{\gamma}}{x_{0}^{\frac{3}{4}}}.$ (54) Indeed, let ($x_{0}>1$) $h(x)=h_{x_{0}}(x)=\varphi_{\sqrt{x_{0}}}^{2}(x-x_{0})=\left(\frac{\pi}{2}+\arctan\Big{(}\frac{x-x_{0}}{\sqrt{x_{0}}}\Big{)}\right)^{2}.$ Note that $0\leq|h^{\prime}|+|h^{\prime\prime}|\leq Ch$. Since $v-\gamma v_{xx}=\mathcal{L}w$, multiplying by $vh$, we have $\int v^{2}h+\gamma\int v_{x}^{2}h-\frac{1}{2}\gamma\int v^{2}h^{\prime\prime}=\int w\mathcal{L}(vh)=\int wD(vh)+\int wvh-\int Qwvh.$ (55) First, from $\left|\int wvh\right|+\left|\int Qwvh\right|\leq C\|w\sqrt{h}\|_{L^{2}}\|v\sqrt{h}\|_{L^{2}}$ and $\int w^{2}h\leq\int_{x<\frac{x_{0}}{2}}w^{2}h+\int_{x>\frac{x_{0}}{2}}w^{2}\leq C\frac{1}{x_{0}}$ (using the definition of $h$ and (47)) it follows that $\left|\int wvh\right|+\left|\int Qwvh\right|\leq\frac{C}{x_{0}^{\frac{1}{2}}}\|v\sqrt{h}\|_{L^{2}}.$ Second, by Lemma 14, we have $\displaystyle\left|\int wD(vh)-\int D(v\sqrt{h})\sqrt{h}w\right|\leq\|w\|_{L^{2}}\|v\sqrt{h}\|_{L^{4}}\|D(\sqrt{h})\|_{L^{4}}\leq\frac{C}{x_{0}^{\frac{3}{8}}}\|v\sqrt{h}\|_{H^{1}}.$ Since $\left|\int D(v\sqrt{h})\sqrt{h}w\right|\leq\|w\sqrt{h}\|_{L^{2}}\|v\sqrt{h}\|_{H^{1}}$ and $\|v\sqrt{h}\|_{H^{1}}^{2}=\int(v_{x}^{2}+v^{2})h+O(\frac{1}{x_{0}})$, we obtain from (55) $\int_{x>x_{0}}(v_{x}^{2}+v^{2})\leq\int(v_{x}^{2}+v^{2})h\leq\frac{C_{\gamma}}{x_{0}^{\frac{3}{4}}}.$ _3\. Virial type estimate on $v(t)$._ Let $\frac{1}{3}<\theta<\frac{1}{2}$, $B>1$ to be chosen later and set $I(t)=\frac{1}{2}\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}(t,x)dx,\quad z=v\sqrt{g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}}\quad\text{where $g(x)=\arctan(x)$,}$ $(\widetilde{\mathcal{L}}z,z)=-2(\mathcal{L}(z_{x}),xz)=2\int|D^{\frac{1}{2}}z|^{2}+\int z^{2}-\int(xQ^{\prime}+Q)z^{2}.$ (56) For any $0<\sigma_{0}<1$, we claim $\begin{split}&\left|2I^{\prime}(t)+\frac{1}{(B+t^{2})^{\theta}}(\widetilde{\mathcal{L}}z,z)\right|\leq\frac{\sigma_{0}}{(B+t^{2})^{\theta}}\|z\|_{L^{2}}^{2}+\frac{C}{\sigma_{0}(B+t^{2})^{1-\theta}}\|v\|_{L^{2}}^{2}\\\ &+\frac{C}{(B+t^{2})^{\frac{7}{4}\theta}}\|z\|_{H^{\frac{1}{2}}}\|v\|_{L^{2}}+\frac{C}{(B+t^{2})^{\theta}}\gamma^{\frac{1}{4}}\|z\|_{H^{\frac{1}{2}}}\|v\|_{L^{2}}+\frac{C}{(B+t^{2})^{2\theta}}\|z\|_{L^{2}}^{2}.\end{split}$ (57) _Proof of ( 57)._ We compute $I^{\prime}(t)$: $\displaystyle I^{\prime}(t)=-\frac{\theta t}{(B+t^{2})^{\theta+1}}\int xg^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}+\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}vv_{t}.$ First, note that by Cauchy-Schwarz’ inequality, for any $\sigma_{0}>0$, $\displaystyle\left|\frac{\theta t}{(B+t^{2})^{\theta+1}}\int xg^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}\right|$ $\displaystyle\leq\frac{\sigma_{0}}{(B+t^{2})^{\theta}}\int g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}$ $\displaystyle+\frac{\theta^{2}t^{2}}{4\sigma_{0}(B+t^{2})^{2-\theta}}\int\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}^{2}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}.$ Since $s^{2}g^{\prime}(s)\leq 1$, we obtain $\displaystyle\left|\frac{\theta t}{(B+t^{2})^{\theta+1}}\int xg^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v^{2}\right|\leq\frac{\sigma_{0}}{(B+t^{2})^{\theta}}\int z^{2}+\frac{C\theta^{2}}{\sigma_{0}(B+t^{2})^{1-\theta}}\int v^{2}.$ Second, we use the equation of $v$ to compute the term $\int g\big{(}\frac{x}{(B+t^{2})^{\theta}}\big{)}vv_{t}$. $\displaystyle\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}vv_{t}$ $\displaystyle=\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v\mathcal{L}v_{x}-\gamma\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v(1-\gamma\partial_{x}^{2})^{-1}(2v_{xx}Q^{\prime}+v_{x}Q^{\prime\prime})=\mathbf{A}+\mathbf{B}.$ Estimate on $\mathbf{A}$. $\displaystyle\mathbf{A}$ $\displaystyle=\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v(-\mathcal{H}v_{xx}+v_{x})-\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}Qvv_{x}$ $\displaystyle=-\int\bigg{(}\frac{1}{(B+t^{2})^{\theta}}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v+g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v_{x}\bigg{)}(-\mathcal{H}v_{x}+v)$ $\displaystyle+\frac{1}{2}\int\bigg{(}\frac{1}{(B+t^{2})^{\theta}}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}Q+g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}Q^{\prime}\bigg{)}v^{2}.$ Next, $\displaystyle\mathbf{A}$ $\displaystyle=-\frac{1}{(B+t^{2})^{\theta}}\int|D^{\frac{1}{2}}z|^{2}+v\Big{(}D\big{(}vg^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}\big{)}-D\big{(}v\sqrt{g^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}\big{)}}\big{)}\sqrt{g^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}\big{)}}\Big{)}$ $\displaystyle+\int(\mathcal{H}v_{x})v_{x}g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}-\frac{1}{2}\frac{1}{(B+t^{2})^{\theta}}\int z^{2}$ $\displaystyle+\frac{1}{2}\frac{1}{(B+t^{2})^{\theta}}\int(xQ^{\prime}+Q)z^{2}+\frac{1}{2}\int\bigg{(}g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}-\frac{x}{(B+t^{2})^{\theta}}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}\bigg{)}Q^{\prime}v^{2}$ $\displaystyle=-\frac{1}{2}\frac{1}{(B+t^{2})^{\theta}}(\widetilde{\mathcal{L}}z,z)+\mathbf{A_{1}}+\mathbf{A_{2}}+\mathbf{A_{3}},$ where $\mathbf{A_{1}}=-\frac{1}{(B+t^{2})^{\theta}}\int v\Big{(}D\big{(}z\sqrt{g^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}}\big{)}\big{)}-(Dz)\sqrt{g^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}\big{)}}\big{)},$ $\mathbf{A_{2}}=\int(\mathcal{H}v_{x})v_{x}g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)},$ $\mathbf{A_{3}}=\frac{1}{2}\int\bigg{(}g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}-\frac{x}{(B+t^{2})^{\theta}}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}\bigg{)}Q^{\prime}v^{2}.$ Estimate on $\mathbf{A_{1}}$. By Lemma 14, we have $|\mathbf{A}_{1}|\leq\frac{C}{(B+t^{2})^{\theta}}\|v\|_{L^{2}}\|z\|_{L^{4}}\big{\|}D\sqrt{g^{\prime}\big{(}\tfrac{x}{(B+t^{2})^{\theta}}\big{)}}\big{\|}_{L^{4}}\leq\frac{C}{(B+t^{2})^{\frac{7\theta}{4}}}\|v\|_{L^{2}}\|z\|_{H^{\frac{1}{2}}}.$ Estimate on $\mathbf{A_{2}}$. Since $\int(\mathcal{H}v_{x})v_{x}=0$, Lemma 3, applied to $A=(B+t^{2})^{\theta}$ gives $|\mathbf{A}_{2}|\leq\frac{C}{(B+t^{2})^{2\theta}}\|z\|_{L^{2}}^{2}.$ Estimate on $\mathbf{A_{3}}$. Since for all $y\in\mathbb{R}$, $|\arctan y-\frac{y}{1+y^{2}}|\leq Cy^{2}$, we have, for all $x\in\mathbb{R}$, $\left|\bigg{(}g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}-\frac{x}{(B+t^{2})^{\theta}}g^{\prime}\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}\bigg{)}Q^{\prime}(x)\right|\leq\frac{x^{2}|Q^{\prime}(x)|}{(B+t^{2})^{2\theta}}\leq\frac{C}{(B+t^{2})^{2\theta}}\frac{1}{1+|x|}.$ Thus, $|\mathbf{A}_{3}|\leq\frac{C}{(B+t^{2})^{2\theta}}\|v\|_{L^{2}}\|z\|_{L^{2}}.$ Estimate on $\mathbf{B}$. First, we claim the following. ###### Claim 1. (i) $x(1-\gamma\partial_{x}^{2})^{-1}f=(1-\gamma\partial_{x}^{2})^{-1}(xf)-2\gamma(1-\gamma\partial_{x}^{2})^{-2}(f^{\prime})$. (ii) $\|(1-\gamma\partial_{x}^{2})^{-1}f\|_{L^{2}}+\gamma^{\frac{1}{2}}\|(1-\gamma\partial_{x}^{2})^{-1}(f^{\prime})\|_{L^{2}}+\gamma\|(1-\gamma\partial_{x}^{2})^{-1}(f^{\prime\prime})\|_{L^{2}}\leq C\|f\|_{L^{2}},$ $\|(1-\gamma\partial_{x}^{2})^{-1}(f^{\prime\prime})\|_{L^{2}}\leq C\gamma^{-\frac{3}{4}}\|f\|_{\dot{H}^{\frac{1}{2}}}.$ _Proof of Claim 1._ (i) Let $h=(1-\gamma\partial_{x}^{2})^{-1}f$. Then, $xh-\gamma(xh)^{\prime\prime}=xf-2\gamma h^{\prime}$ and so $xh=(1-\gamma\partial_{x}^{2})^{-1}(xf-2\gamma(1-\gamma\partial_{x}^{2})^{-1}f^{\prime})$. (ii) $\int|f|^{2}=\int|h-\gamma h^{\prime\prime}|^{2}=\int h^{2}+2\gamma\int(h^{\prime})^{2}+\gamma^{2}\int(h^{\prime\prime})^{2}$, which proves the first estimate. Next, $\|(1-\gamma\partial_{x}^{2})^{-1}f^{\prime\prime}\|_{L^{2}}\leq C\|(\frac{\xi^{2}}{1+\gamma\xi^{2}})\hat{f}\|_{L^{2}}\leq C{\gamma^{-\frac{3}{4}}}\||\xi|^{\frac{1}{2}}\hat{f}\|_{L^{2}}$, since $\forall\xi\in\mathbb{R}$, $\forall\gamma>0$, $\frac{\xi^{2}}{1+\gamma\xi^{2}}\leq{\gamma^{-\frac{3}{4}}}{|\xi|^{\frac{1}{2}}}$. The claim is proved. Using (i) of Claim 1, we obtain $\mathbf{B}=-\gamma\int\frac{1}{x}g\big{(}\frac{x}{(B+t^{2})^{\theta}}\big{)}v\,(1-\gamma\partial_{x}^{2})^{-1}H,$ where $H=2xv_{xx}Q^{\prime}+xv_{x}Q^{\prime\prime}-2\gamma(1-\gamma\partial_{x}^{2})^{-1}(2v_{xx}Q^{\prime}+v_{x}Q^{\prime\prime})_{x}.$ Since $|g(y)|\leq C|y|$, for all $y$, we have $|\mathbf{B}|\leq\frac{C\gamma}{(B+t^{2})^{\theta}}\|v\|_{L^{2}}\|(1-\gamma\partial_{x}^{2})^{-1}H\|_{L^{2}}$. Now, we use Claim 1 (ii) to estimate $\|(1-\gamma\partial_{x}^{2})^{-1}H\|_{L^{2}}$. We can rewrite $H$ under the form: $H=(2vxQ^{\prime})^{\prime\prime}+(vF_{1})^{\prime}+vF_{2}-2\gamma(1-\gamma\partial_{x}^{2})^{-1}((2vQ^{\prime})^{\prime\prime}+(vF_{3})^{\prime}+vF_{4})_{x},$ where for $j=1,\ldots,4$, $|F_{j}(x)|\leq C\frac{1}{1+x^{2}}$. Thus, $\displaystyle\|(1-\gamma\partial^{2})^{-1}H\|_{L^{2}}$ $\displaystyle\leq C\gamma^{-\frac{3}{4}}\|vxQ^{\prime}\|_{\dot{H}^{\frac{1}{2}}}+C\gamma^{-\frac{1}{2}}\|v\tfrac{1}{1+x^{2}}\|_{L^{2}}+\gamma^{\frac{1}{2}}\|(1-\gamma\partial_{x}^{2})^{-1}((2vQ^{\prime})^{\prime\prime}+(vF_{3})^{\prime}+vF_{4})\|_{L^{2}}$ $\displaystyle\leq C\gamma^{-\frac{3}{4}}\|vxQ^{\prime}\|_{\dot{H}^{\frac{1}{2}}}+C\gamma^{-\frac{1}{2}}\|v\tfrac{1}{1+x^{2}}\|_{L^{2}}.$ Now, we claim $\|v\tfrac{1}{1+x^{2}}\|_{L^{2}}\leq C\|z\|_{L^{2}},\quad\|vxQ^{\prime}\|_{\dot{H}^{\frac{1}{2}}}\leq C\|z\|_{H^{\frac{1}{2}}}.$ (58) The first estimate is clear since $\tfrac{1}{1+x^{2}}\leq C\sqrt{g^{\prime}}$. Let $f(x)=\frac{xQ^{\prime}(x)}{\sqrt{g^{\prime}(\frac{x}{(B+t^{2})^{\theta}})}}$. Then, by Lemma 14, $\|D^{\frac{1}{2}}(vxQ^{\prime})\|_{L^{2}}=\|D^{\frac{1}{2}}(zf)\|_{L^{2}}\leq\|(D^{\frac{1}{2}}z)f\|_{L^{2}}+C\|z\|_{L^{4}}\|D^{\frac{1}{2}}f\|_{L^{4}}\leq C\|z\|_{H^{\frac{1}{2}}},$ since $\|f\|_{L^{\infty}}+\|D^{\frac{1}{2}}f\|_{L^{4}}\leq\|f\|_{H^{1}}\leq C.$ Thus, $\|(1-\gamma\partial^{2})^{-1}H\|_{L^{2}}\leq C\gamma^{-\frac{3}{4}}\|z\|_{H^{\frac{1}{2}}}$ and in conclusion for the term $\mathbf{B}$: $|\mathbf{B}|\leq\frac{C\gamma^{\frac{1}{4}}}{(B+t^{2})^{\theta}}\|z\|_{H^{\frac{1}{2}}}\|v\|_{L^{2}}.$ Putting together the above estimates, we obtain (57). We now claim the following (see proof in Section 3.2): ###### Proposition 3. There exist $\lambda>0$, $\gamma_{0}>0$ and $B_{0}>1$ such that, for $0<\gamma<\gamma_{0}$, $B\geq B_{0}$, $\forall t,\quad(\widetilde{\mathcal{L}}z(t),z(t))\geq\lambda\|z(t)\|_{H^{\frac{1}{2}}}^{2},\quad\text{where $z$ is as above}.$ ###### Remark 4. The operator $\widetilde{\mathcal{L}}$ does not depend on $\gamma$ and $B$, but the orthogonality conditions on $w$ imply almost orthogonality conditions on $z$ that depend on $\gamma$, $B$, see proof of Proposition 3. Choose $\theta=\frac{2}{5}$ and fix $\sigma_{0}=\frac{\lambda}{4}$. Then, $-2I^{\prime}(t)\geq\frac{\lambda}{2(B+t^{2})^{\theta}}\|z(t)\|_{H^{\frac{1}{2}}}^{2}-\frac{C}{(B+t^{2})^{\theta}}\bigg{(}\frac{1}{(B+t^{2})^{\frac{1}{5}}}+\gamma^{\frac{1}{2}}\bigg{)}\|v\|^{2}_{L^{2}}.$ By the decay property (50), $\int v^{2}(t)\leq\int_{|x|\leq\frac{1}{2}(B+t^{2})^{\theta}}v^{2}(t)+\frac{C_{\gamma}}{(B+t^{2})^{\frac{3}{4}\theta}}\leq C\int z^{2}(t)+\frac{C_{\gamma}}{(B+t^{2})^{\frac{3}{10}}}.$ For $\gamma>0$ small enough and $B$ large enough, and by $\|v\|_{H^{\frac{1}{2}}}\leq C$, we get $-2I^{\prime}(t)\geq\frac{\lambda}{4(B+t^{2})^{\theta}}\|z(t)\|_{H^{\frac{1}{2}}}^{2}-\frac{C_{\gamma}}{(B+t^{2})^{\frac{3}{5}}}.$ Since $I(t)$ is bounded, we obtain by integration $\int_{-\infty}^{+\infty}\frac{1}{(B+t^{2})^{\theta}}\|z(t)\|_{H^{\frac{1}{2}}}^{2}dt<C_{\gamma}.$ (59) We claim that (59) and (50) imply $\int_{-\infty}^{+\infty}\frac{1}{(B+t^{2})^{\theta}}\|v(t)\|_{H^{\frac{1}{2}}}^{2}dt<C.$ (60) Indeed, by (50) and the expression of $g^{\prime}$, and considering the two regions $x>\frac{1}{(B+t^{2})^{\frac{\theta}{2}}}$, $x<\frac{1}{(B+t^{2})^{\frac{\theta}{2}}}$, we have $\|v-z\|_{H^{1}}^{2}=\|v(1-\sqrt{g^{\prime}})\|_{H^{1}}^{2}\leq\frac{C}{(B+t^{2})^{\frac{3}{8}\theta}}=\frac{C}{(B+t^{2})^{\frac{3}{20}}}.$ (61) Thus, by $\|v\|_{H^{\frac{1}{2}}}\leq\|z\|_{H^{\frac{1}{2}}}+\|v-z\|_{H^{\frac{1}{2}}}$, and (59) $\int_{-\infty}^{+\infty}\frac{1}{(B+t^{2})^{\theta}}\|v(t)\|_{H^{\frac{1}{2}}}^{2}dt\leq 2C_{\gamma}+\int_{-\infty}^{+\infty}\frac{1}{(B+t^{2})^{\frac{11}{20}}}dt\leq C.$ Using another virial argument, we claim $\int_{-\infty}^{+\infty}\frac{1}{(B+t^{2})^{\theta}}\|z(t)\|_{H^{\frac{3}{2}}}^{2}dt<C.$ (62) Proof of (62). We set $J(t)=\frac{1}{2}\int g\bigg{(}\frac{x}{(B+t^{2})^{\theta}}\bigg{)}v_{x}^{2}(t).$ Proceeding as in the proof of (57) (the equation for $v_{x}$ is very similar to the one for $v$), we obtain $\left|J^{\prime}(t)+\frac{1}{(B+t^{2})^{\theta}}\int(D^{\frac{3}{2}}z)^{2}\right|\leq\frac{C}{(B+t^{2})^{\theta}}\|v\|_{H^{1}}(\|v\|_{H^{1}}+\|z\|_{H^{\frac{3}{2}}}).$ Using $\|v\|_{H^{1}}\leq\|z\|_{H^{1}}+\|v-z\|_{H^{1}},$ (61) and the following estimate $\|z\|_{H^{1}}\leq\varepsilon\|D^{\frac{3}{2}}z\|_{L^{2}}+C_{\varepsilon}\|z\|_{L^{2}},$ we obtain, for $\varepsilon>0$ small enough, $-J^{\prime}(t)\geq\frac{1}{2}\frac{1}{(B+t^{2})^{\theta}}\|D^{\frac{3}{2}}z\|_{L^{2}}^{2}+C\frac{1}{(B+t^{2})^{\theta}}\|z\|_{L^{2}}^{2}.$ Since $J(t)$ is bounded and using (59), we obtain (62). Finally, by (59), (61) and (62), we get (51). Lemma 4 is proved. ###### Lemma 5 (Decay estimate on $w(t)$). The following hold $\int_{-\infty}^{+\infty}\frac{1}{(1+t^{2})^{\frac{2}{5}}}\|w(t)\|_{L^{2}}^{2}dt<C,$ (63) $\sup_{t\in\mathbb{R}}\int|x|w^{2}(t,x)dx\leq C.$ (64) _Proof of Lemma 5._ Estimate (63) is a consequence of Lemma 4 by comparing $v$ and $w$. Let $\gamma>0$ small. We have by the definition of $v$: $(1-\gamma\partial_{x}^{2})v=\mathcal{L}w.$ Let $\widetilde{w}=(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}w$. Then, $\int w(1-\gamma\partial_{x}^{2})^{\frac{1}{2}}v=\int\widetilde{w}(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}(\mathcal{L}w).$ On the one hand, we have $\left|\int w(1-\gamma\partial_{x}^{2})^{\frac{1}{2}}v\right|\leq C\|w\|_{L^{2}}\|v\|_{H^{1}}.$ On the other hand, as in the proof of Claim 1 $\|(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}(\mathcal{L}w)-\mathcal{L}\widetilde{w}\|_{L^{2}}\leq\|(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}(Qw)-Qw\|_{L^{2}}+\|Q(w-\widetilde{w})\|_{L^{2}}\leq\gamma^{\frac{1}{4}}\|w\|_{L^{2}}.$ Thus, $\left|\int\widetilde{w}(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}(\mathcal{L}w)-(\mathcal{L}\widetilde{w},\widetilde{w})\right|\leq C\gamma^{\frac{1}{4}}\|w\|_{L^{2}}^{2}$ and since $(\mathcal{L}\widetilde{w},\widetilde{w})\geq\frac{1}{2}\lambda\|\widetilde{w}\|_{H^{\frac{1}{2}}}$ for $\gamma>0$ small enough (this is a consequence of Lemma 15 and the orthogonality conditions on $w$ – see Section 3.2, in particular the proof of Proposition 3), we obtain $\int\widetilde{w}(1-\gamma\partial_{x}^{2})^{-\frac{1}{4}}(\mathcal{L}w)\geq\frac{\lambda}{2}\|\widetilde{w}\|_{H^{\frac{1}{2}}}^{2}-C\gamma^{\frac{1}{4}}\|w\|_{L^{2}}^{2}\geq\lambda_{1}\|w\|_{L^{2}}^{2}.$ In conclusion, we have obtained $\|w\|_{L^{2}}\leq C\|v\|_{H^{1}},$ and Lemma 4 then implies (63). Now, we prove (64). Indeed, the integrability property (63) allows us to obtain the decay on $w(t,x)$ by monotonicity properties. By the proof of Proposition 2, we have, for any $\lambda\in(0,1)$, for any $t_{0}$, $t\in(-\infty,t_{0}]$, $x_{0}>1$, $\begin{split}&\int w^{2}(t_{0},x)\left(\varphi(x{-}x_{0})-\varphi(-x_{0})\right)dx\\\ &\leq\int w^{2}(t,x)\left(\varphi(x{-}x_{0}{-}\lambda(t_{0}{-}t))-\varphi(-x_{0}-\lambda(t_{0}{-}t))\right)dx+C\int_{t}^{t_{0}}\frac{\|w(t^{\prime})\|_{L^{2}}^{2}dt^{\prime}}{(x_{0}+\lambda(t_{0}-t^{\prime}))^{2}}.\end{split}$ (65) The last term in (65) is treated as follows $(x_{0}>1)$ $\int_{t}^{t_{0}}\frac{\|w(t^{\prime})\|_{L^{2}}^{2}dt^{\prime}}{(x_{0}+\lambda(t_{0}-t^{\prime}))^{2}}\leq Cx_{0}^{-\frac{6}{5}}\int_{-\infty}^{+\infty}\frac{\|w(t^{\prime})\|_{L^{2}}^{2}dt^{\prime}}{(1+(t_{0}-t^{\prime}))^{\frac{4}{5}}}$ Thus, by (63) (applied to $w(t+t_{0})$) and (47), letting $t\to-\infty$ in (65), we obtain $\int w^{2}(t_{0})\left(\varphi(x-x_{0})-\varphi(-x_{0})\right)dx\leq C{x_{0}^{-\frac{6}{5}}}.$ By the change of variable $x\to-x$, $t\to-t$, which leaves the equation invariant, we get: $\int w^{2}(t_{0})\left(\varphi(x_{0})-\varphi(x+x_{0})\right)dx\leq C{x_{0}^{-\frac{6}{5}}},$ and thus, summing up the two estimates, $\int w^{2}(t_{0})\left(\varphi(x-x_{0})-\varphi(x+x_{0})+\varphi(x_{0})-\varphi(-x_{0})\right)dx\leq C{x_{0}^{-\frac{6}{5}}}.$ We verify easily that for all $|x|>x_{0}\geq 1$, $\begin{split}\varphi(x-x_{0})-\varphi(x+x_{0})+\varphi(x_{0})-\varphi(-x_{0})&\geq\varphi(0)-\varphi(2x_{0})+\varphi(x_{0})-\varphi(-x_{0})\\\ &\geq\tfrac{\pi}{2}-\arctan(2)>0.\end{split}$ (66) Thus, for all $x_{0}>1$, $\int_{|x|\geq x_{0}}w^{2}(t_{0})\leq C{x_{0}^{-\frac{6}{5}}}.$ (67) By integrating in $x_{0}$, we obtain the following estimate $\forall t\in\mathbb{R},\quad\int|x|w^{2}(t)\leq C.$ (68) Thus Lemma 5 is proved. Now, we claim that estimate (64) implies a gain of regularity on $w(t)$. ###### Lemma 6 (Gain of regularity on $w(t)$). Let $w\in C(\mathbb{R},L^{2}(\mathbb{R}))\cap L^{\infty}(\mathbb{R},L^{2}(\mathbb{R}))$ be a solution of (45) satisfying (64). Then, $w(t)\in C(\mathbb{R},H^{\frac{1}{2}}(\mathbb{R}))$ and the following identity holds $\int xw^{2}(t_{2})-\int xw^{2}(t_{1})=-\int_{t_{1}}^{t_{2}}\int\big{(}2|D^{\frac{1}{2}}w|^{2}+w^{2}+w^{2}(xQ^{\prime}-Q)\big{)}+2\int_{t_{1}}^{t_{2}}\beta(t)\int xQ^{\prime}w.$ (69) _End of the proof of Theorem 3 assuming Lemma 6._ Note first that multiplying the equation of $w(t)$ by $Q^{\prime}$ and using $\int wQ^{\prime}=0$, we find $\left(\int(Q^{\prime})^{2}\right)\beta(t)=\int w\mathcal{L}(Q^{\prime\prime})$, so that $|\beta(t)|\leq C\|w\|_{L^{2}}.$ (70) Multiplying the equation of $w(t)$ by $\mathcal{L}w$ and using $\mathcal{L}Q^{\prime}=0$, we also have $\forall t\in\mathbb{R},\quad(\mathcal{L}w(t),w(t))=(\mathcal{L}w(0),w(0)).$ By (69), the estimates on $\int|x|w^{2}(t)$ and on $\beta(t)$, and Lemma 5, we have $\int_{-\infty}^{+\infty}\frac{1}{(1+t^{2})^{\frac{2}{5}}}\|w(t)\|_{H^{\frac{1}{2}}}^{2}dt<C.$ This implies that for a sequence $t_{n}\to+\infty$, we have $\|w(t_{n})\|_{H^{\frac{1}{2}}}\to 0$ as $n\to+\infty$. Since $(\mathcal{L}w(t),w(t))=\lim_{t_{n}\to\infty}(\mathcal{L}w(t_{n}),w(t_{n}))$, we obtain $(\mathcal{L}w(t),w(t))=0$ and so by the orthogonality conditions on $w(t)$ and Lemma 15, we finally obtain $\forall t$, $w(t)=0$. _Proof of Lemma 6._ Formally, identity (69) follows from multiplying equation (45) by $xw$, integration by parts and properties of the Hilbert transform. To justify (69), we use a regularization of $w(t)$. We set $w_{n}=(1-\frac{1}{n}\partial_{x}^{2})^{-1}w$, so that for all $t$, $w_{n}(t)\to w(t)$ in $L^{2}(\mathbb{R})$ as $n\to+\infty$. Then, $w_{n}$ satisfies the following equation $w_{nt}=(\mathcal{L}w_{n})_{x}-\tfrac{1}{n}(1-\tfrac{1}{n}\partial_{x}^{2})^{-1}(2Q^{\prime}w_{nx}+w_{n}Q^{\prime\prime})_{x}+\beta(1-\tfrac{1}{n}\partial_{x}^{2})^{-1}Q^{\prime}.$ (71) Let $h:\mathbb{R}\to\mathbb{R}$ be a smooth nondecreasing function such that $h(x)=x$ if $x>1$ and $h(x)=0$ if $x<0$. Then, $\begin{split}\int h(x)w^{2}&=\int h(x)(w_{n}-\tfrac{1}{n}w_{nxx})^{2}=\int h(x)w_{n}^{2}-\tfrac{2}{n}\int w_{nxx}w_{n}h(x)+\frac{1}{n^{2}}\int w_{nxx}^{2}h(x)\\\ &=\int h(x)w_{n}^{2}+\tfrac{2}{n}\int w_{nx}^{2}h(x)-\tfrac{1}{n}\int w_{n}^{2}h^{\prime\prime}(x)+\frac{1}{n^{2}}\int w_{nxx}^{2}h(x)\end{split}$ implies that $\int_{x>0}xw_{n}^{2}\leq C\quad\text{and}\quad\int_{x>0}x(w-w_{n})^{2}\to 0\quad\text{as $n\to+\infty$.}$ (72) The same holds true in the region $x<0$. For the functions $w_{n}$, we have the following identity, for any $t_{1}<t_{2}$: $\begin{split}&\int xw_{n}^{2}(t_{2})-\int xw_{n}^{2}(t_{1})=-\int_{t_{1}}^{t_{2}}\int\Big{(}2|D^{\frac{1}{2}}w_{n}|^{2}+w_{n}^{2}+w_{n}^{2}(xQ^{\prime}-Q)\Big{)}dxdt\\\ &+\int_{t_{1}}^{t_{2}}\int\big{(}-\tfrac{2}{n}x(1-\tfrac{1}{n}\partial_{x}^{2})^{-1}(2Q^{\prime}w_{nx}+w_{n}Q^{\prime\prime})_{x}w_{n}\big{)}+2\int_{t_{1}}^{t_{2}}\beta\int x((1-\tfrac{1}{n}\partial_{x}^{2})^{-1}Q^{\prime})w_{n}dxdt.\end{split}$ (73) Indeed, multiplying the equation of $w_{n}$ by $Ag(\tfrac{x}{A})w_{n}$ where $g(x)=\arctan(x)$, we find $\begin{split}&\int Ag(\tfrac{x}{A})w_{n}^{2}(t_{2})-\int Ag(\tfrac{x}{A})w_{n}^{2}(t_{1})\\\ &=-\int_{t_{1}}^{t_{2}}\int\Big{(}2|D^{\frac{1}{2}}w_{n}|^{2}g^{\prime}(\tfrac{x}{A})+2D^{\frac{1}{2}}w_{n}(D^{\frac{1}{2}}(w_{n}g^{\prime}(\tfrac{x}{A}))-D^{\frac{1}{2}}(w_{n})g^{\prime}(\tfrac{x}{A}))+2Dw_{n}w_{nx}Ag(\tfrac{x}{A})\\\ &+w_{n}^{2}g^{\prime}(\tfrac{x}{A})+w_{n}^{2}(Ag(\tfrac{x}{A})Q^{\prime}-g^{\prime}(\tfrac{x}{A})Q)\Big{)}dxdt-2\int_{t_{1}}^{t_{2}}\beta(t)\int x((1-\tfrac{1}{n}\partial_{x}^{2})^{-1}Q)Ag(\tfrac{x}{A})w_{n}\\\ &-\frac{2}{n}\int_{t_{1}}^{t_{2}}\int((1-\tfrac{1}{n}\partial_{x}^{2})^{-1}Q)(2Q^{\prime}w_{nx}+w_{n}Q^{\prime\prime})_{x}Ag(\tfrac{x}{A})w_{n}.\end{split}$ Then, (73) is proved using Lemmas 3 and 14 (see the proof of Lemma 4 for similar arguments) and then passing to the limit as $A\to+\infty$ applying the Lebesgue convergence theorem. From (73), we claim that for any $t_{1},t_{2}$, $\limsup_{n\to+\infty}\int_{t_{1}}^{t_{2}}\|w_{n}(t)\|_{H^{\frac{1}{2}}}^{2}dt<+\infty.$ (74) Proof of (74). By Claim 1 (i), we have $\begin{split}&\tfrac{1}{n}\int x(1-\tfrac{1}{n}\partial_{x}^{2})^{-1}(2Q^{\prime}w_{nx}+w_{n}Q^{\prime\prime})_{x}w_{n}\\\ &=\tfrac{1}{n}\int w_{n}(1-\tfrac{1}{n}\partial_{x}^{2})^{-1}(2xQ^{\prime}w_{nxx}+3xQ^{\prime\prime}w_{nx}+xQ^{(3)}w_{n})\\\ &-\tfrac{2}{n^{2}}\int w_{n}(1-\tfrac{1}{n}\partial_{x}^{2})^{-2}(2Q^{\prime}w_{nx}+w_{n}Q^{\prime\prime})_{xx}=\mathbf{I}+\mathbf{II}.\end{split}$ As in the proof of Lemma 4 (control of $\mathbf{B}$), we have $|\mathbf{I}|\leq\frac{C}{n^{\frac{1}{4}}}\|w_{n}\|_{H^{\frac{1}{2}}}\|w_{n}\|_{L^{2}},\quad|\mathbf{II}|\leq\frac{C}{n^{\frac{1}{2}}}\|w_{n}\|_{L^{2}}^{2}.$ (75) From (73), (70), the $L^{2}$ bounds on $w(t)$ and $w_{n}(t)$ and (75) we obtain $\int_{t_{1}}^{t_{2}}\|w_{n}(t)\|_{H^{\frac{1}{2}}}^{2}dt\leq C|t_{2}-t_{1}|+\sup_{t}\int|x|w_{n}^{2}(t)+\frac{C}{n^{\frac{1}{4}}}\int_{t_{1}}^{t_{2}}\|w_{n}\|_{H^{\frac{1}{2}}}^{2}dt.$ For $n$ large enough, we get $\int_{t_{1}}^{t_{2}}\|w_{n}(t)\|_{H^{\frac{1}{2}}}^{2}dt\leq C.$ Thus (74) is proved. By the well-posedness of the equation of $w(t)$ in $H^{\frac{1}{2}}$, we obtain $\forall t$, $w(t)\in H^{\frac{1}{2}}$ and $w_{n}\to w$ in $H^{\frac{1}{2}}$. Finally, from (72) and (75), we obtain (69) by passing to the limit as $n\to\infty$ in (73). ### 3.2 Positivity of a quadratic form related to the dual problem In this section, we prove Proposition 3. The main ingredient is the following result. ###### Proposition 4. There exists $\lambda_{0}>0$ such that for all $z\in H^{\frac{1}{2}}$, $\int z(xQ)^{\prime}=0\quad\Rightarrow\quad(\widetilde{\mathcal{L}}z,z)=2\int|D^{\frac{1}{2}}z|^{2}+\int z^{2}-\int(xQ^{\prime}+Q)z^{2}\geq\lambda_{0}\|z\|_{H^{\frac{1}{2}}}^{2}.$ (76) _Proof of Proposition 4._ First, we introduce some notation. Recall that $\mathcal{L}f=-\mathcal{H}f^{\prime}+f-Qf.$ (77) We define $S=(xQ)^{\prime}$. Note that $S=\frac{d}{dc}{Q_{c}}_{|c=1}$ and thus by differentiating the equation of $Q_{c}$ with respect to $c$, and taking $c=1$, we find $\mathcal{L}S=-Q.$ Observe also that $\mathcal{L}Q=-\mathcal{H}Q^{\prime}+Q-Q^{2}=-\frac{1}{2}Q^{2}$, by the equation of $Q$. Now, we set $T=S-Q$. Then, $\mathcal{L}T=-Q+\frac{1}{2}Q^{2}=(xQ)^{\prime}=S$, by using the explicit expression $Q(x)=\frac{4}{1+x^{2}}$. We compute $\int TS=\int S^{2}-\int QS$. Since $S=\mathcal{H}Q^{\prime}=\frac{1}{2}Q^{2}-Q$ (explicit computation), we have $\int S^{2}=\int(Q^{\prime})^{2}$ and $(Q^{\prime})^{2}=\frac{64x^{2}}{(1+x^{2})^{4}}=Q^{3}-\frac{1}{4}Q^{4}$, thus $\int(Q^{\prime})^{2}=\int Q^{3}-\frac{1}{4}\int Q^{4}=\int S^{2}=\int(\frac{1}{2}Q^{2}-Q)^{2}=\frac{1}{4}\int Q^{4}-\int Q^{3}+\int Q^{2}$, we find $\int S^{2}=\frac{1}{2}\int Q^{2}$. Moreover, $\int SQ=-\int xQQ^{\prime}=\frac{1}{2}\int Q^{2}$, and so $\int TS=0$. Finally, $\int TQ=-\int T\mathcal{L}S=-\int S^{2}$. In conclusion, we have proved ($(.,.)$ denotes the $L^{2}$ scalar product): $\begin{split}&S=\tfrac{1}{2}Q^{2}-Q=(xQ)^{\prime},~{}T=S-Q,~{}\mathcal{L}Q=-\tfrac{1}{2}Q^{2},~{}\mathcal{L}S=-Q,~{}\mathcal{L}T=S,\\\ &(S,Q)=\tfrac{1}{2}\int Q^{2},~{}(S,T)=0,~{}(T,Q)=-\int S^{2}.\end{split}$ (78) Now, we claim the following. ###### Lemma 7. There exists $\lambda>0$ such that, for all $\varepsilon>0$, if $\int wS_{\varepsilon}=0$, where $S_{\varepsilon}=S+\varepsilon Q$, then $(\mathcal{L}w,w)\geq 0$ and $(\widetilde{\mathcal{L}}w,w)\geq\lambda\|w\|_{H^{\frac{1}{2}}}^{2}$. _Proof of Lemma 7._ Let $T_{\varepsilon}=T-\varepsilon S$ and $S_{\varepsilon}=S+\varepsilon Q$, then by (78) : $\mathcal{L}T_{\varepsilon}=S_{\varepsilon}$ and $(\mathcal{L}T_{\varepsilon},T_{\varepsilon})=(S_{\varepsilon},T_{\varepsilon})=(S,T)+\varepsilon(-(S,S)+(T,Q))-\varepsilon^{2}(S,Q)\leq-2\varepsilon(S,S)<0.$ Moreover, it is clear that if $f_{0}$, $\lambda_{0}$ denote respectively the first eigenfunction and first eigenvalue of $\mathcal{L}$ (see Lemma 15) we have $(S,f_{0})=(\mathcal{L}T,f_{0})=(T,\mathcal{L}f_{0})=\lambda_{0}(xQ^{\prime},f_{0})\neq 0$, since $f_{0}>0$. Thus, by Lemma E.1 in [27], we obtain the first part of Lemma 7. Now, we note that since $xQ^{\prime}>0$, $\begin{split}(\widetilde{\mathcal{L}}w,w)&=2\int|D^{\frac{1}{2}}w|^{2}+\int w^{2}-\int(xQ^{\prime}+Q)w^{2}\\\ &\geq 2\int|D^{\frac{1}{2}}w|^{2}+\int w^{2}-\int Qw^{2}=\int|D^{\frac{1}{2}}w|^{2}+(\mathcal{L}w,w).\end{split}$ (79) Using the inequality $\|w\|_{L^{4}}^{2}\leq C\|w\|_{L^{2}}\|D^{\frac{1}{2}}w\|_{L^{2}}$ (see (133)) and Cauchy-Schwarz’ inequality, we have, for some constant $C_{0}>0$, $\int Qw^{2}\leq C\|w^{2}\|_{L^{2}}\leq C_{0}\int|D^{\frac{1}{2}}w|^{2}+\frac{1}{2}\int w^{2}.$ Thus, for $\delta_{0}>0$ such that $2-C_{0}\delta_{0}>1-\frac{\delta_{0}}{2}$, we have $\begin{split}(\widetilde{\mathcal{L}}w,w)&\geq(2-C_{0}\delta_{0})\int|D^{\frac{1}{2}}w|^{2}+(1-\tfrac{1}{2}\delta_{0})\int w^{2}-(1-\delta_{0})\int Qw^{2}\\\ &\geq(1-\delta_{0})(\mathcal{L}w,w)+\tfrac{\delta_{0}}{2}\|w\|_{H^{\frac{1}{2}}}^{2}\geq\tfrac{\delta_{0}}{2}\|w\|_{H^{\frac{1}{2}}}^{2},\end{split}$ provided $\int wS_{\varepsilon}=0$. Now, we finish the proof of Proposition 4. Let $z\in H^{\frac{1}{2}}$ be such that $\int zS=\int z(xQ)^{\prime}=0$. Let $w=z+aQ$, where $\int wS_{\varepsilon}=0$, $0<\varepsilon<\varepsilon_{0}$, where $\varepsilon_{0}$ is to be chosen small enough. In particular, we have $\int wS_{\varepsilon}=\int zS_{\varepsilon}+a\int QS_{\varepsilon}=\varepsilon\int zQ+a\int SQ+a\varepsilon\int Q^{2}=\varepsilon\int zQ+a\left(\tfrac{1}{2}+\varepsilon\right)\int Q^{2}=0,$ and so $|a|\leq\frac{2}{\|Q\|_{L^{2}}}\varepsilon\|z\|_{L^{2}}$, and $\|w\|_{L^{2}}\leq 2\|z\|_{L^{2}}$ for $\varepsilon_{0}$ small enough. Similarly, we have $\|z\|_{L^{2}}\leq 2\|w\|_{L^{2}}$, by possibly choosing a smaller $\varepsilon_{0}$. By Lemma 7, we obtain $\begin{split}\frac{\lambda}{2}\|z\|_{H^{\frac{1}{2}}}^{2}\leq\lambda\|w\|_{H^{\frac{1}{2}}}^{2}\leq(\widetilde{\mathcal{L}}w,w)=(\widetilde{\mathcal{L}}z,z)+a^{2}(\widetilde{\mathcal{L}}Q,Q)+2a(\widetilde{\mathcal{L}}Q,z).\end{split}$ For $\varepsilon_{0}$ small, we get $(\widetilde{\mathcal{L}}z,z)\geq\frac{\lambda}{4}\|z\|_{H^{\frac{1}{2}}}^{2}$. Now, we are in a position to prove Proposition 3. _Proof of Proposition 3._ In Proposition 3, we want to prove that for $B$ large and $\gamma$ small, and for some $\lambda_{1}>0$, for all $t$, $(\widetilde{\mathcal{L}}z(t),z(t))\geq\lambda_{1}\|z(t)\|_{H^{\frac{1}{2}}}^{2}$, for $z(t)=v(t)\sqrt{g^{\prime}(\frac{x}{(B+t^{2})^{\alpha}})}$ , where $v=(1-\gamma\partial_{x}^{2})^{-1}(\mathcal{L}w).$ Formally, if $B=+\infty$ and $\gamma=0$, we have $z(t)=v(t)=\mathcal{L}w$ and $0=\int wQ=-\int w\mathcal{L}S=-\int zS$, and the result follows from Proposition 4. Now, we justify that the result persists for large values of $B$ and small values of $\gamma$. Let $S_{B,\gamma}(t)=(g^{\prime}(\frac{x}{(B+t^{2})^{\alpha}}))^{-\frac{1}{2}}(S-\gamma S^{\prime\prime})$. Then, $\mathcal{L}((1-\gamma\partial_{x}^{2})^{-1}(\sqrt{g^{\prime}(\frac{x}{(B+t^{2})^{\alpha}})}S_{B,\gamma}(t))=-Q$ and so $\int S_{B,\gamma}(t)z=-\int wQ=0.$ Now, we control $S_{B,\gamma}(t)-S$: $S_{B,\gamma}(t)-S=\sqrt{1{+}\tfrac{x^{2}}{(B+t^{2})^{\alpha}}}(S-\gamma S^{\prime\prime})-S=\Big{(}\sqrt{1{+}\tfrac{x^{2}}{(B+t^{2})^{\alpha}}}-1\Big{)}S-\gamma\sqrt{1{+}\tfrac{x^{2}}{(B+t^{2})^{\alpha}}}S^{\prime\prime}.$ Thus, by elementary estimates and the expression of $S$, we obtain: $|S_{B,\gamma}(t,x)-S(x)|\leq\big{(}B^{-\frac{\alpha}{2}}+\gamma\big{)}\frac{1}{1+|x|}.$ It follows that $\left|\int Sz(t)\right|=\left|\int(S-S_{B,\gamma}(t))z(t)\right|\leq\big{(}B^{-\frac{\alpha}{2}}+\gamma\big{)}\|z\|_{L^{2}}.$ Setting $z=z_{1}+aQ$, where $\int z_{1}S=0$ and $|a|\leq\big{(}B^{-\frac{\alpha}{2}}+\gamma\big{)}\|z\|_{L^{2}}$, we conclude the proof of Proposition 3 as at the end of the proof of Proposition 4, for $B$ large enough and $\gamma$ small enough. ## 4 Proof of asymptotic stability - Theorem 1 In this section, we first prove that Theorem 2 implies Theorem 1. Then, we prove that Theorem 3 (proved in Section 3) implies Theorem 2. ### 4.1 Proof of Theorem 1 assuming Theorem 2 We follow the strategy of [15], [16], the main idea being to use monotonicity type arguments (such as Proposition 1) to prove that a limiting solution of (1) has uniform decay in space. See also [17] for similar use of monotonicity arguments. We consider a solution $u(t)$ of (1) in $H^{\frac{1}{2}}$ which satisfies $\|u_{0}-Q\|_{H^{\frac{1}{2}}}=\alpha<\alpha_{0}$, for $\alpha_{0}>0$ small enough. By the stability property, for all $t\in\mathbb{R}$, $\inf_{y}\|u(t)-Q(.-y)\|_{H^{\frac{1}{2}}}\leq C\alpha$. _1\. Decomposition of $u(t)$ around the asymptotic soliton._ First, we determine the parameter $c^{+}>0$. It is given by the amount of $L^{2}$ norm that remains on the region $x>\frac{t}{10}$ asymptotically as $t\to+\infty$. Let $\varphi$ be as in (16), with $A>1$ so that Proposition 1 holds. Let $c^{+}=\frac{1}{\pi\int Q^{2}}\,\limsup_{t\to+\infty}\int u^{2}(t,x)\varphi(x-\tfrac{t}{10})dx.$ (80) From the stability property, $|c^{+}-1|\leq C\alpha_{0}$ ($\lim_{+\infty}\varphi=\pi$). Using Lemma 1 to decompose $u(t)$ around $Q_{c^{+}}$, we consider the following decomposition of $u(t)$ $\begin{split}&u(t,x)=Q_{c^{+}}(x-\rho(t))+\eta(t,x-\rho(t)),\\\ &\int Q_{c^{+}}^{\prime}\eta(t,x)dx=0,\quad\sup_{t}\|\eta(t)\|_{H^{\frac{1}{2}}}\leq K\alpha_{0}.\end{split}$ (81) In what follows, we consider $\alpha_{0}>0$ small enough, so that the following holds (by (13)): $\forall t,\quad\frac{99}{100}\leq\rho^{\prime}(t)\leq\frac{101}{100},\quad\frac{99}{100}\leq c^{+}\leq\frac{101}{100}.$ (82) _2\. Monotonicity arguments._ We claim the following estimates: ###### Lemma 8 (Asymptotics on $u(t)$). $\forall y_{0}>1,\quad\limsup_{t\to+\infty}\int u^{2}(t,x)\varphi(x-y_{0}-\rho(t))dx\leq\frac{C}{y_{0}},$ (83) $\forall y_{0}>1,\quad\limsup_{t\to+\infty}\int u^{2}(t,x)(\varphi(x-\rho(t)+\tfrac{t}{10})-\varphi(x-\rho(t)+y_{0}))dx\leq\frac{C}{y_{0}},$ (84) $\lim_{t\to+\infty}\int u^{2}(t,x)(\varphi(x-\rho(t)+\tfrac{19}{20}t)-\varphi(x-\rho(t)+\tfrac{t}{10}))dx=0,$ (85) $\lim_{t\to+\infty}\int u^{2}(t,x)\varphi(x-\rho(t)+\tfrac{t}{10})dx=c^{+}\pi\int Q^{2}.$ (86) _Proof of Lemma 8._ Monotonicity property on the right of the soliton. By (17), with $\lambda=\frac{1}{2}$, we have, for all $y_{0}>1$, $\int u^{2}(t,x)\varphi(x-y_{0}-\rho(t))dx\leq\int u^{2}(0,x)\varphi(x-y_{0}-\rho(0)-\tfrac{1}{2}t)dx+\frac{C}{y_{0}}.$ Since $\lim_{t\to+\infty}\int u^{2}(0,x)\varphi(x-y_{0}-\rho(0)-\tfrac{1}{2}t)dx=0$, we obtain (83). Monotonicity property on the left of the soliton. By (18), with $\lambda=\frac{19}{20}$ and $x_{0}=\frac{19}{20}t^{\prime}$, we have for all $0\leq t^{\prime}\leq t$, $\int u^{2}(t,x)\varphi(x-\rho(t)+\tfrac{19}{20}t)dx\leq\int u^{2}(t^{\prime},x)\varphi(x-\rho(t^{\prime})+\tfrac{19}{20}t^{\prime})dx+\frac{C}{t^{\prime}}.$ It follows that $\int u^{2}(t,x)\varphi(x-\rho(t)+\tfrac{19}{20}t)dx$ has a limit as $t\to+\infty$. Set $\ell=\lim_{t\to+\infty}\int u^{2}(t,x)\varphi(x-\rho(t)+\tfrac{19}{20}t)dx,\qquad\ell\geq c^{+}\pi\int Q^{2}.$ Applying (18) with $\lambda=\frac{100}{99}(\frac{19}{20}-\frac{1}{1000})<1$ and $x_{0}=\frac{t}{1000}$, we find $\int u^{2}(t,x)\varphi(x-\rho(t)+\tfrac{19}{20}t)dx\leq\int u^{2}(\tfrac{t}{100},x)\varphi(x-\rho(\tfrac{t}{100})+\tfrac{t}{1000})dx+\frac{C}{t}.$ Since $\limsup_{t\to+\infty}\int u^{2}(\tfrac{t}{100},x)\varphi(x-\rho(\tfrac{t}{100})+\tfrac{1}{10}\tfrac{t}{100})dx\leq c^{+}\pi\int Q^{2},$ we obtain $c^{+}\pi\int Q^{2}=\ell$ and (85). Fix $y_{0}>1$, pick $\lambda=\frac{1}{2}$. Consider $t_{2}>t$ and define $t_{1}=\frac{4}{5}t_{2}+2y_{0}$, so that for $t$ large, $t_{1}<t_{2}$. But then, by (18), $\begin{split}\int u^{2}(t_{2},x)\varphi(x-\rho(t_{2})+\tfrac{t_{2}}{10})dx&=\int u^{2}(t_{2},x)\varphi(x-\rho(t_{2})+\lambda(t_{2}-t_{2})+y_{0})dx\\\ &\leq\int u^{2}(t_{1},x)\varphi(x-\rho(t_{1})+y_{0})dx+\frac{C}{y_{0}}.\end{split}$ In light of (85) and the existence of $\ell$, (84) follows. Thus Lemma 8 is proved. _3\. Construction of a compact limit object._ Let $t_{n}\to+\infty$. By the uniform bound on $u(t)$ in $H^{\frac{1}{2}}$, there exist $\widetilde{u}_{0}\in H^{\frac{1}{2}}$ and a subsequence, still denoted by $(t_{n})$, such that $u(t_{n},.+\rho(t_{n}))\rightharpoonup\widetilde{u}_{0}\quad\text{in $H^{\frac{1}{2}}$ weak as $n\to+\infty$.}$ Consider $\widetilde{u}(t)$ the global $H^{\frac{1}{2}}$ solution of (1) such that $\widetilde{u}(0)=\widetilde{u}_{0}$. By (81), $\|\widetilde{u}_{0}-Q_{c^{+}}\|\leq C\alpha_{0}$ and thus by the stability property, $\sup_{t}\inf_{y}\|\widetilde{u}(t)-Q(.-y)\|_{H^{\frac{1}{2}}}\leq C\alpha_{0}$. Let $\widetilde{\rho}(t)$, $\widetilde{\eta}(t)$ correspond to the decomposition of $\widetilde{u}(t)$ around $Q_{c^{+}}$ given by Lemma 1. By Theorem 5 below and Remark 2, for all $t\in\mathbb{R}$, we have $\displaystyle u(t_{n}+t,.+\rho(t_{n}))\rightharpoonup\widetilde{u}(t)\quad\text{in $H^{\frac{1}{2}}$ weak,}$ $\displaystyle\rho(t_{n}+t)-\rho(t_{n})\to\widetilde{\rho}(t)\quad\text{as $n\to+\infty$}.$ From weak convergence and Lemma 8, we claim the following decay estimate on $\widetilde{u}(t)$: $\forall y_{0}>1,\forall t\in\mathbb{R},\quad\int_{|x|>y_{0}}\widetilde{u}^{2}(t,x+\widetilde{\rho}(t))dx\leq\frac{C}{y_{0}}.$ (87) Indeed, first, from (83), for any fixed $y_{0}>1$, $t\in\mathbb{R}$, we have $\limsup_{n\to+\infty}\int u^{2}(t+t_{n},x+\rho(t_{n}))\varphi(x-\rho(t_{n}+t)+\rho(t_{n})-y_{0})dx\leq\frac{C}{y_{0}},$ and so by weak convergence $\int\widetilde{u}^{2}(t,x)\varphi(x-\widetilde{\rho}(t)-y_{0})dx\leq\frac{C}{y_{0}}.$ Second, from (84), for fixed $t\in\mathbb{R}$, $\limsup_{n\to+\infty}\int u^{2}(t+t_{n},x+\rho(t_{n}))(\varphi(x-\rho(t_{n}+t)+\rho(t_{n})+\tfrac{t+t_{n}}{10})-\varphi(x-\rho(t_{n}+t)+\rho(t_{n})+y_{0}))dx\leq\frac{C}{y_{0}}.$ Note that for fixed $t$, $y_{0}$, we have $\displaystyle\lim_{n\to+\infty}\varphi(x-\rho(t_{n}+t)+\rho(t_{n})+\tfrac{t+t_{n}}{10})-\varphi(x-\rho(t_{n}+t)+\rho(t_{n})+y_{0})$ $\displaystyle=\pi-\varphi(x-\widetilde{\rho}(t)+y_{0})$ $\displaystyle=\varphi(-x+\widetilde{\rho}(t)-y_{0}).$ Thus, we obtain $\int\widetilde{u}^{2}(t,x)\varphi(-x+\widetilde{\rho}(t)-y_{0})dx\leq\frac{C}{y_{0}}.$ Finally, from (83)–(86), for any $y_{0}>1$, we have $\lim_{n\to+\infty}\bigg{|}\int u^{2}(t_{n},x)(\varphi(x-\rho(t_{n})+y_{0})-\varphi(x-\rho(t_{n})-y_{0}))dx-c^{+}\pi\int Q^{2}\bigg{|}\leq\frac{C}{y_{0}}.$ Thus, by $L^{2}_{loc}$ convergence, for any $y_{0}>1$, $\bigg{|}\int\widetilde{u}^{2}_{0}(x)(\varphi(x+y_{0})-\varphi(x-y_{0}))dx-c^{+}\pi\int Q^{2}\bigg{|}\leq\frac{C}{y_{0}}.$ Passing to the limit $y_{0}\to+\infty$, we obtain $\|\widetilde{u}_{0}\|_{L^{2}}=\|\widetilde{u}(t)\|_{L^{2}}=\sqrt{c^{+}}\|Q\|_{L^{2}}=\|Q_{c^{+}}\|_{L^{2}}.$ _4\. Conclusion by Theorem 2._ From Theorem 2, it follows that for some $c_{1}$ close to $c^{+}$ and $x_{1}$ close to $0$, we have $\widetilde{u}(t,x)\equiv Q_{c_{1}}(x-x_{1}-c_{1}t).$ But $\|\widetilde{u}(t)\|_{L^{2}}=\|Q_{c^{+}}\|_{L^{2}}$ implies that $c_{1}=c^{+}$. Moreover, $\widetilde{\rho}(0)=0$ and $\widetilde{u}(0)=Q_{c^{+}}(x-x_{1})=Q_{c^{+}}(x)+\widetilde{\eta}(0,x)$ where $x_{1}$ is small and $\int\widetilde{\eta}(0)Q^{\prime}_{c^{+}}=0$ imply $x_{1}=0$. In conclusion, $\widetilde{u}_{0}=Q_{c^{+}}.$ By a standard argument and (83), (86), we have obtained $u(t,.+\rho(t))\rightharpoonup Q_{c^{+}}\quad\text{in $H^{\frac{1}{2}}$ weak as $t\to+\infty$,}$ $\lim_{t\to+\infty}\int_{x>\frac{t}{10}}|u(t,x)-Q_{c^{+}}(x-\rho(t))|^{2}dx=0.$ (88) Thus Theorem 1 is a consequence of Theorem 2. ### 4.2 Proof of Theorem 2 First, we note that it is sufficient to prove Theorem 2 in the case $\int u_{0}^{2}=\int Q^{2}$. Indeed, for $u_{0}$ satisfying the assumptions of Theorem 2, set $c_{1}=\int u_{0}^{2}/\int Q^{2}$ and $\widetilde{u}(t)=\frac{1}{c_{1}}u(\frac{1}{c_{1}^{2}}t,\frac{1}{c_{1}}x).$ Then, $|c_{1}-1|\leq C\alpha_{0}$ and $\widetilde{u}$ satisfies (1), $\int\widetilde{u}^{2}=\int Q^{2}$ and $\|\widetilde{u}_{0}-Q\|_{H^{\frac{1}{2}}}\leq C\alpha_{0}$. Thus, by the stability property – see Introduction – for all $t$, there exists $y(t)$ such that $\sup_{t}\|\widetilde{u}(t)-Q(.-y(t))\|_{H^{\frac{1}{2}}}\leq C^{\prime}\alpha_{0}$. Moreover, $\widetilde{u}(t)$ also satisfies (9). If we prove $\widetilde{u}(t,x)=Q(x-t-x_{0})$, with $|x_{0}|\leq C\alpha_{0}$, the result follows for $u(t)$. The proof of Theorem 2 is by contradiction. Assume that there exists a sequence $u_{n}(t)$ of $H^{\frac{1}{2}}$ solutions of (1) such that $\displaystyle\sup_{t\in\mathbb{R}}\|u_{n}(t)-Q(.-\rho_{n}(t))\|_{L^{2}}\to 0\quad\text{as $n\to+\infty$,}$ (89) $\displaystyle\int u_{n}^{2}(0)=\int Q^{2},\qquad\eta_{n}\not\equiv 0,$ (90) $\displaystyle\forall n,\forall\varepsilon>0,\exists A_{n,\varepsilon}>0,\text{ s.t. }\forall t\in\mathbb{R},\quad\int_{|x|>A_{n,\varepsilon}}u_{n}^{2}(t,x+\rho_{n}(t))dx<\varepsilon,$ (91) where $\rho_{n}(t)$ and $\eta_{n}(t)$ are defined from $u_{n}(t)$ by Lemma 1. Note that $\int u_{n}^{2}(0)=\int Q^{2}$ implies $\forall n,\forall t,\quad\int\eta_{n}^{2}(t)=-2\int\eta_{n}(t)Q.$ (92) Define $0\not\equiv b_{n}=\sup_{t}\|\eta_{n}(t)\|_{L^{2}}\to 0\quad\text{as $n\to+\infty$.}$ (93) Then, there exists $t_{n}$ such that $\|\eta_{n}(t_{n})\|_{L^{2}}\geq\frac{1}{2}b_{n}$. We set $w_{n}(t,x)=\frac{\eta_{n}(t_{n}+t,x)}{b_{n}}.$ For such a sequence $w_{n}$, we claim the following result. ###### Proposition 5 (Weak convergence of the sequence of renormalized solutions). There exists $(w_{n^{\prime}})$ a subsequence of $(w_{n})$ and $w\in C(\mathbb{R},L^{2}(\mathbb{R}))\cap L^{\infty}(\mathbb{R},L^{2}(\mathbb{R}))$ such that $\forall t\in\mathbb{R},\quad w_{n^{\prime}}(t)\rightharpoonup w(t)\quad\text{in $L^{2}$ weak as $n\to+\infty$.}$ Moreover, $w(t)$ satisfies for some continuous function $\beta(t)$: $\displaystyle w_{t}=(\mathcal{L}w)_{x}+\beta(t)Q^{\prime}\quad\text{on $\mathbb{R}\times\mathbb{R}$},$ $\displaystyle w(0)\neq 0,\quad\int wQ=\int wQ^{\prime}=0,$ $\displaystyle\forall t\in\mathbb{R},\forall x_{0}>1,\quad\int_{|x|>x_{0}}w^{2}(t,x)dx\leq\frac{C}{x_{0}}.$ Proposition 5 is in contradiction with Theorem 3. Thus, for $\alpha_{0}>0$ small, for $u(t)$ satisfying the assumptions of Theorem 2, we have $\eta\equiv 0$ so that $\rho^{\prime}(t)=1$ (by Lemma 1) and $u(t,x)=Q(x-\rho(0)-t)$, with $|\rho(0)|\leq C\alpha_{0}$. Therefore, we are reduced to prove Proposition 5. _Proof of Proposition 5._ One can actually prove a strong $L^{2}$ convergence result. See the end of the proof. Note that the main point in Proposition 5 is the fact $w\neq 0$. For this, we need to obtain a strong convergence in $L^{2}$ for some suitable $t$. _Decay estimate._ From Proposition 2, we have $\begin{split}&\int\eta_{n}^{2}(t_{0},x)(\varphi(x-x_{0})-\varphi(-x_{0}))dx\\\ &\leq\int\eta_{n}^{2}(t,x)(\varphi(x-x_{0}-\lambda(t_{0}-t))-\varphi(-x_{0}-\lambda(t_{0}-t)))dx+\frac{Cb_{n}^{2}}{x_{0}}.\end{split}$ Letting $t\to-\infty$ and using (91), we obtain, for any $x_{0}>1$, $\int\eta_{n}^{2}(t_{0},x)(\varphi(x-x_{0})-\varphi(-x_{0}))dx\leq\frac{Cb_{n}^{2}}{x_{0}}.$ Similarly, arguing on $\eta_{n}(-t,-x)$, for any $x_{0}>1$, $\int\eta_{n}^{2}(t_{0},x)(\varphi(x_{0})-\varphi(x+x_{0}))dx\leq\frac{Cb_{n}^{2}}{x_{0}},$ which gives, by (66), similarly as in the proof of (67): $\forall x_{0}>1,\quad\int_{|x|>x_{0}}\eta_{n}^{2}(t,x)dx\leq\frac{Cb_{n}^{2}}{x_{0}}\quad\text{and}\quad\int_{|x|>x_{0}}w_{n}^{2}(t,x)dx\leq\frac{C}{x_{0}}.$ (94) _Local smoothing estimate on $w_{n}$._ Let $\varphi$ be defined in (16) for a fixed value of $A$ ($A=1$ for example). Then, $\int_{0}^{1}\int|D^{\frac{1}{2}}(w_{n}(t,x)\sqrt{\varphi^{\prime}(x)})|^{2}dxdt\leq C.$ (95) Proof of (95). First, we claim the following estimate: $\frac{1}{2}\frac{d}{dt}\int\eta_{n}^{2}\varphi\leq-\frac{1}{2}\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}+C\int\eta_{n}^{2}\leq-\frac{1}{2}\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}+Cb_{n}^{2}.$ (96) Thus, by integration, $\forall t\in\mathbb{R},\quad\int_{t}^{t+1}\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}dxdt\leq Cb_{n}^{2}\quad\text{and}\quad\int_{t}^{t+1}\int|D^{\frac{1}{2}}(w_{n}\sqrt{\varphi^{\prime}})|^{2}dxdt\leq C.$ (97) Now, we justify (96). Using direct computations, Lemma 3, (13) and then $|\int\eta_{n}^{3}\varphi^{\prime}|\leq C\int|\eta|^{3}\leq C\int\eta^{2}$ (by (133)), we get $\displaystyle\frac{1}{2}\frac{d}{dt}\int\eta_{n}^{2}\varphi$ $\displaystyle=-\int(\mathcal{L}\eta_{n}-\tfrac{1}{2}\eta_{n}^{2})(\eta_{nx}\varphi+\eta_{n}\varphi^{\prime})+(\rho_{n}^{\prime}-1)\int Q^{\prime}\eta_{n}\varphi-\frac{1}{2}(\rho_{n}^{\prime}-1)\int\eta_{n}^{2}\varphi^{\prime}$ $\displaystyle=\int\left((\mathcal{H}\eta_{nx})\eta_{nx}\varphi+(\mathcal{H}\eta_{nx})\eta_{n}\varphi^{\prime}\right)-\frac{1}{2}\int\eta_{n}^{2}\varphi^{\prime}+\frac{1}{2}\int\eta_{n}^{2}(-Q^{\prime}\varphi+Q\varphi^{\prime})$ $\displaystyle\quad+\frac{1}{3}\int\eta_{n}^{3}\varphi^{\prime}+(\rho_{n}^{\prime}-1)\int Q^{\prime}\eta_{n}\varphi-\frac{1}{2}(\rho_{n}^{\prime}-1)\int\eta_{n}^{2}\varphi^{\prime}$ $\displaystyle\leq\int(\mathcal{H}\eta_{nx})\eta_{n}\varphi^{\prime}+C\int\eta_{n}^{2}.$ Using (135) and then (133), we have $\begin{split}&-\int(\mathcal{H}\eta_{nx})\eta_{n}\varphi^{\prime}=\int\eta_{n}D(\eta_{n}\varphi^{\prime})=\int\eta_{n}\sqrt{\varphi^{\prime}}D(\eta_{n}\sqrt{\varphi^{\prime}})+\int\eta_{n}\left(D(\eta_{n}\varphi^{\prime})-\sqrt{\varphi^{\prime}}D(\eta_{n}\sqrt{\varphi^{\prime}})\right)\\\ &\geq\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}-C\|\eta_{n}\|_{L^{2}}\|D(\eta_{n}\varphi^{\prime})-\sqrt{\varphi^{\prime}}D(\eta_{n}\sqrt{\varphi^{\prime}})\|_{L^{2}}\\\ &\geq\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}-C\|\eta_{n}\|_{L^{2}}\|\eta_{n}\sqrt{\varphi^{\prime}}\|_{L^{4}}\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}\geq\frac{1}{2}\int|D^{\frac{1}{2}}(\eta_{n}\sqrt{\varphi^{\prime}})|^{2}-C\|\eta_{n}\|_{L^{2}}^{2}.\end{split}$ (98) (Note that we have used $\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}<+\infty$.) Thus (96) is proved. _Compactness in $L^{2}$ for some time._ From the equation of $\eta_{n}$ and (13), it follows that $\frac{d}{dt}\int\eta_{n}^{2}=-\frac{1}{2}\int Q^{\prime}\eta_{n}^{2}+(\rho_{n}^{\prime}-1)\int Q^{\prime}\eta_{n}\quad\text{and so}\quad\left|\frac{d}{dt}\int\eta_{n}^{2}\right|\leq C_{0}\int\eta_{n}^{2}.$ In particular, by the definition of $t_{n}$, $\forall t\in[0,1],$ $\int\eta_{n}^{2}(t+t_{n})\geq e^{-C_{0}}b_{n}^{2}$ and so $\forall t\in[0,1],\quad\|w_{n}(t)\|_{L^{2}}\geq e^{-\frac{1}{2}C_{0}}=\delta>0.$ (99) It follows from (95) that for all $n$, there exists $\tau_{n}\in[0,1]$ such that $\int|D^{\frac{1}{2}}(w_{n}(\tau_{n})\sqrt{\varphi^{\prime}})|^{2}\leq C$. Thus, there exists a subsequence of $(w_{n})$ (still denoted by $(w_{n})$) and $s_{0}\in[0,1]$, $W\in H^{\frac{1}{2}}$ such that $w_{n}(\tau_{n})\sqrt{\varphi^{\prime}}\rightharpoonup W\quad\text{in $H^{\frac{1}{2}}$ weak,}\qquad\tau_{n}\to s_{0}\quad\text{as $n\to+\infty$.}$ But (by possibly extracting a further subsequence), there exists $w_{s_{0}}\in L^{2}$ such that $\tau_{n}\to s_{0},\quad w_{n}(\tau_{n})\rightharpoonup w_{s_{0}}\quad\text{in $L^{2}$ weak as $n\to+\infty$.}$ It follows that $W=w_{s_{0}}\sqrt{\varphi^{\prime}}$. Since $\sqrt{\varphi^{\prime}}>0$ on $\mathbb{R}$, we get $w_{n}(\tau_{n})\to w_{s_{0}}\quad\text{in $L^{2}_{loc}$ as $n\to+\infty$.}$ By (94) and (99), we finally get $w_{n}(\tau_{n})\to w_{s_{0}}\quad\text{in $L^{2}$ as $n\to+\infty$},\quad\int w_{s_{0}}Q^{\prime}=0,\quad w_{s_{0}}\neq 0.$ (100) Note also that from (92) and $\int\eta_{n}Q^{\prime}=0$, we have $\int w_{s_{0}}Q=\int w_{s_{0}}Q^{\prime}=0.$ (101) _Weak convergence for all time._ Consider $\widetilde{w}(t)\in C(\mathbb{R},L^{2}(\mathbb{R}))$ the unique solution of $\widetilde{w}_{t}=(\mathcal{L}\widetilde{w})_{x}\quad\text{on $\mathbb{R}\times\mathbb{R}$},\quad\widetilde{w}(s_{0})=w_{s_{0}},\quad\text{on $\mathbb{R}$.}$ (It is clear by a standard energy estimate and regularization arguments that the corresponding Cauchy problem is well-posed in $L^{2}$). Now, to obtain weak convergence, we need to remove some terms from the equation of $w_{n}$, following some arguments in [15], Lemma 8 and beginning of proof of Lemma 11. We write $\displaystyle w_{nt}$ $\displaystyle=(\mathcal{L}w_{n}-\tfrac{b_{n}}{2}w_{n}^{2})_{x}+\frac{1}{b}_{n}(\rho_{n}^{\prime}-1)(Q+b_{n}w_{n})_{x}$ $\displaystyle=(\mathcal{L}w_{n})_{x}-\tfrac{b_{n}}{2}(w_{n}^{2})_{x}+\beta_{n}Q^{\prime}+b_{n}F_{n}^{\prime}+b_{n}\widetilde{\beta}_{n}(w_{n})_{x},$ where $\beta_{n}=\frac{1}{\int(Q^{\prime})^{2}}\int w_{n}\mathcal{L}(Q^{\prime\prime}),\quad\widetilde{\beta}_{n}=\frac{1}{b_{n}}(\rho_{n}^{\prime}-1),\quad F_{n}=\frac{1}{b_{n}}(\widetilde{\beta}_{n}-\beta_{n})Q.$ Set $\widetilde{w}_{n}(t)=w_{n}(t)-Q^{\prime}\int_{\tau_{n}}^{t}\beta_{n}(s)ds.$ Then, the equation of $\widetilde{w}_{n}(t)$ writes $\widetilde{w}_{nt}=(\mathcal{L}\widetilde{w}_{n})_{x}-\tfrac{b_{n}}{2}(w_{n}^{2})_{x}+b_{n}F_{n}^{\prime}+b_{n}\widetilde{\beta}_{n}(\widetilde{w}_{n})_{x}+b_{n}\widetilde{\beta}_{n}Q^{\prime\prime}\int_{\tau_{n}}^{t}\beta_{n}(s)ds.$ We claim the following weak convergence result. ###### Lemma 9. For all $t\in\mathbb{R}$, $\widetilde{w}_{n}(t)\rightharpoonup\widetilde{w}(t)\quad\text{in $L^{2}$ weak.}$ Assuming this lemma, from (15), we have, for all $t$, $\widetilde{\beta}_{n}(t)\to\widetilde{\beta}(t)=\frac{1}{\int(Q^{\prime})^{2}}\int\widetilde{w}\mathcal{L}(Q^{\prime\prime}),\quad\int_{\tau_{n}}^{t}\widetilde{\beta}_{n}(s)ds\to\int_{s_{0}}^{t}\widetilde{\beta}(s)ds.$ Set $w(t)=\widetilde{w}(t)+Q^{\prime}\int_{s_{0}}^{t}\widetilde{\beta}(s)ds$. Then, $w(t)$ solves $w_{t}=(\mathcal{L}w)_{x}+\widetilde{\beta}Q^{\prime},$ and $w(s_{0})=w_{s_{0}}\neq 0$. Moreover, for all $t\in\mathbb{R}$, $w_{n}(t)\rightharpoonup w(t)\quad\text{in $L^{2}$ weak.}$ Finally, from (92) and $\int\eta_{n}Q^{\prime}=0$, we have $\int w(t)Q=\int w(t)Q^{\prime}=0,$ and by weak convergence and (94), we have $\forall x_{0}>1,\forall t,\quad\int_{|x|>x_{0}}w^{2}(t,x)dx\leq\frac{C}{x_{0}}.$ Thus, we are reduced to prove Lemma 9. _Proof of Lemma 9._ Set $G_{1,n}=-\frac{1}{2}w_{n}^{2},\quad G_{2,n}=F_{n}+\widetilde{\beta}_{n}\widetilde{w}_{n}+\widetilde{\beta}_{n}Q^{\prime}\int_{\tau_{n}}^{t}\beta_{n}(s)ds,\quad G_{n}=G_{1,n}+G_{2,n}.$ Observe that $\|G_{1,n}\|_{L^{1}}+\|G_{2,n}\|_{L^{2}}\leq C(t),\quad\text{with $C(t)$ bounded on bounded intervals.}$ Let $T\in\mathbb{R}$. By $\sup_{t}\|w_{n}(t)\|_{L^{2}}\leq C$ and the expression of $\widetilde{w}_{n}$, we have $\sup_{[-T,T]}\|\widetilde{w}_{n}(t)\|_{L^{2}}\leq C_{T}$. Let $g\in C_{0}^{\infty}(\mathbb{R})$ and let $v$ solve the problem $\left\\{\begin{aligned} &\partial_{t}v=\mathcal{L}(v_{x})\\\ &v_{|t=T}=g.\end{aligned}\right.$ Then $\displaystyle\int(\widetilde{w}_{n}-\widetilde{w})(T)g(x)dx-\int(w_{n}(\tau_{n})-w(\tau_{n}))v(\tau_{n})dx=\int_{\tau_{n}}^{T}\int\partial_{t}((\widetilde{w}_{n}-\widetilde{w})(t)v(t,x))dxdt$ $\displaystyle\int_{\tau_{n}}^{T}\int((\mathcal{L}\widetilde{w}_{n})_{x}-(\mathcal{L}\widetilde{w})_{x}+b_{n}(G_{n})_{x})v(t,x)+(\widetilde{w}_{n}-\widetilde{w})(\mathcal{L}v)_{x}dxdt$ $\displaystyle=-b_{n}\int_{\tau_{n}}^{T}\int G_{n}v_{x}(t,x)dxdt.$ The energy method gives $\|v\|_{L^{\infty}([\tau_{n},T],L^{2}(\mathbb{R}))}+\|v_{x}\|_{L^{\infty}([\tau_{n},T]\times\mathbb{R})}+\|v_{x}\|_{L^{\infty}([\tau_{n},T],L^{2}(\mathbb{R}))}\leq C.$ Moreover, by continuity of $t\mapsto w(t)$ in $L^{2}$, $\begin{split}&\lim_{n\to+\infty}\int(w_{n}(\tau_{n})-w(\tau_{n}))v(\tau_{n})dx\\\ &=\lim_{n\to+\infty}\int(w_{n}(\tau_{n})-w(s_{0}))v(\tau_{n})dx+\lim_{n\to+\infty}\int(w(s_{0})-w(\tau_{n}))v(\tau_{n})dx=0.\end{split}$ Thus, $\widetilde{w}_{n}(T)\rightharpoonup\widetilde{w}(T)\quad\text{as $n\to+\infty$.}$ and the proof of Lemma 9 is concluded. _Alternate proof by strong $L^{2}$ convergence for all time._ Now, we use Theorem 6 in Section 6 to prove strong $L^{2}$ convergence of the sequence $(w_{n}(t))$ for all $t$. Let $T>0$. Set $\zeta_{n}(t,x)=w_{n}(t,x-\rho_{n}(t)+\rho_{n}(0))+\frac{1}{b_{n}}[Q(x-(\rho_{n}(t)-\rho_{n}(0)))-Q(x-t)],$ (102) so that $u_{n}(t,x+\rho_{n}(0))=Q(x-\rho_{n}(t)+\rho_{n}(0))+b_{n}w_{n}(t,x-\rho_{n}(t)+\rho_{n}(0))=Q(x-t)+b_{n}\zeta_{n}(t,x),$ and $\zeta_{n}$ satisfies $\displaystyle(\zeta_{n})_{t}=(-\mathcal{H}(\zeta_{n})_{x}-Q(x{-}t)\zeta_{n})_{x}-\frac{b_{n}}{2}(\zeta_{n}^{2})_{x},$ $\displaystyle\|\zeta_{n}(t)\|_{L^{2}}\leq C_{T},\quad\forall t\in[-T,T].$ Indeed, since $|\rho^{\prime}_{n}(t)-1|\leq C\|\eta_{n}\|_{L^{2}}\leq Cb_{n}$, we have $|\rho_{n}(t)-\rho_{n}(0)-t|\leq Cb_{n}|t|,$ (103) and the estimate on $\zeta_{n}$ follows. On the one hand, Theorem 6 applied to $\frac{1}{C_{T}}\zeta_{n}$ for $n$ large enough (so that $b_{n}$ is small enough) implies that $t\in[-T,T]\mapsto\zeta_{n}(t)\in L^{2}$ is equicontinuous in $n$. On the other hand, from (95), we have $\int_{[-T,T]}\int\left|D^{\frac{1}{2}}(\zeta_{n}(t,x)\sqrt{\varphi^{\prime}(x)})\right|^{2}dxdt\leq C_{T},$ and the decay property (94) also holds for $\zeta(t)$ on $[-T,T]$ with constant depending on $T$. In particular, there exists $N\subset[-T,T]$ of zero Lebesgue measure such that for all $t\in[-T,T]\setminus N$, $\int|D^{\frac{1}{2}}(\zeta_{n}(t,x)\sqrt{\varphi^{\prime}(x)})|^{2}dxdt<+\infty.$ Now, we choose a dense and countable subset $I$ of $[-T,T]$ such that for all $t\in I$, $\int|D^{\frac{1}{2}}(\zeta_{n}(t,x)\sqrt{\varphi^{\prime}(x)})|^{2}dxdt<+\infty.$ Arguing as in the proof of (100), and using a diagonal argument, there exists a subsequence of $(\zeta_{n})$ which we will still denote by $(\zeta_{n})$ such that for any $t\in I$, $\zeta_{n}(t)\to\zeta(t)$ in $L^{2}$ strong as $n\to+\infty$. Using the equicontinuity, we obtain $\forall t\in[-T,T],\quad\zeta_{n}(t)\to\zeta(t)\quad\text{in $L^{2}$ strong as $n\to+\infty$.}$ (104) By (103) and $|\rho^{\prime}_{n}-1|\leq Cb_{n}$, we may also assume that for the same subsequence $\forall t\in[-T,T],\quad\frac{1}{b_{n}}(\rho_{n}(t)-\rho_{n}(0)-t)\to\kappa(t).$ (105) Now, we deduce from (102), (104) and (105) that $\forall t\in[-T,T],\quad w_{n}(t)\to w(t)=\eta(t,.+t)+\kappa(t)Q^{\prime}\quad\hbox{in $L^{2}$ strong as $n\to+\infty$.}$ ### 4.3 Proof of Remark 1 Let $u(t)$ be a solution satisfying the assumptions of Theorem 1. Let $c^{+}$, $\rho(t)$ and $\eta(t)$ be defined as in the proof of Theorem 1. In particular, by (88), we have $\lim_{t\to+\infty}\int_{x>\frac{t}{10}-\rho(t)}|\eta(t,x)|^{2}dx=0.$ (106) To prove (11), we use the identity (41) on $\eta$, where $\varphi=\frac{\pi}{2}+\arctan(\frac{x}{A})$, $A>1$ large enough be to defined later: $\displaystyle\frac{1}{2}\frac{d}{dt}\int\eta^{2}\varphi$ $\displaystyle=\int(\mathcal{H}\eta_{x})\eta\varphi^{\prime}+\int(\mathcal{H}\eta_{x})\eta_{x}\varphi-\frac{1}{2}\int\eta^{2}\varphi^{\prime}+\frac{1}{2}\int\eta^{2}(-Q^{\prime}\varphi+Q\varphi^{\prime})$ $\displaystyle\quad+\frac{1}{3}\int\eta^{3}\varphi^{\prime}+(\rho^{\prime}-1)\int Q^{\prime}\eta\varphi-\frac{1}{2}(\rho^{\prime}-1)\int\eta^{2}\varphi^{\prime}.$ We claim that for $A$ large enough and $\alpha_{0}$ small enough, for $C>0$ independent of $A$, $\frac{1}{2}\frac{d}{dt}\int\eta^{2}\varphi\leq\int(\mathcal{H}\eta_{x})\eta\varphi^{\prime}+C\int\frac{\eta^{2}}{1+x^{2}}.$ (107) Indeed, by Lemma 3, we have $\int(\mathcal{H}\eta_{x})\eta_{x}\varphi\leq\frac{C}{A}\int\eta^{2}\varphi^{\prime}$. By the definition of $Q$, $\int\eta^{2}(-Q^{\prime}\varphi+Q\varphi^{\prime})\leq C\int\frac{\eta^{2}}{1+x^{2}}$. By (38) (note that the constant in (38) is independent of $A$) $|\int\eta^{3}\varphi^{\prime}|\leq C\alpha_{0}\int\eta^{2}\varphi^{\prime}$. Finally, the last two terms are controlled using (13), so that (107) is proved for $A$ large enough, $\alpha_{0}$ small enough. Now, we use (LABEL:page28bis) on $\eta$. We obtain $\frac{1}{2}\frac{d}{dt}\int\eta^{2}\varphi\leq-\int|D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}+C\|\eta\|_{L^{2}}\|\eta\sqrt{\varphi^{\prime}}\|_{L^{4}}\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}+C\int\frac{\eta^{2}}{1+x^{2}}.$ (108) Note that for $A>1$, we have $\frac{1}{1+x^{2}}\leq C\varphi$ on $\mathbb{R}$. Let $t_{0}>0$. Integrating the above estimate on $[t_{0},t_{0}+1]$, we get $\int_{t_{0}}^{t_{0}+1}\int|D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}dt\leq C\sup_{t\in[t_{0},t_{0}+1]}\left(\int\eta^{2}(t)\varphi+C\|\eta\sqrt{\varphi^{\prime}}\|_{L^{4}}\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}\right).$ On the other hand, by (135), we have $\begin{split}\int|D^{\frac{1}{2}}\eta|^{2}\varphi^{\prime}&\leq 2\int|D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}+2\int|(D^{\frac{1}{2}}\eta)\sqrt{\varphi^{\prime}}-D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}\\\ &\leq 2\int|D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}+C\|\eta\|_{L^{4}}^{2}\|D^{\frac{1}{2}}\sqrt{\varphi^{\prime}}\|_{L^{4}}^{2}\leq 2\int|D^{\frac{1}{2}}(\eta\sqrt{\varphi^{\prime}})|^{2}+C\|D^{\frac{1}{2}}\sqrt{\varphi^{\prime}}\|_{L^{4}}^{2}.\end{split}$ Thus, we obtain $\int_{t_{0}}^{t_{0}+1}\int|D^{\frac{1}{2}}\eta|^{2}\varphi^{\prime}dt\leq C\sup_{t\in[t_{0},t_{0}+1]}\left(\int\eta^{2}(t)\varphi+C\|\eta\sqrt{\varphi^{\prime}}\|_{L^{4}}\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}\right)+C\|D^{\frac{1}{2}}\sqrt{\varphi^{\prime}}\|_{L^{4}}^{2}.$ We have $\|D^{\frac{1}{2}}\sqrt{\varphi^{\prime}}\|_{L^{4}}^{2}\leq CA^{-\frac{3}{2}}$, $\|D\sqrt{\varphi^{\prime}}\|_{L^{4}}\leq CA^{-\frac{5}{4}}$ and $\|\eta\sqrt{\varphi^{\prime}}\|_{L^{4}}\leq\|\eta\|_{L^{8}}\|\sqrt{\varphi^{\prime}}\|_{L^{8}}\leq CA^{-\frac{3}{8}}$. Therefore, $\int_{t_{0}}^{t_{0}+1}\int|D^{\frac{1}{2}}\eta(t,x)|^{2}\frac{dxdt}{1+(\frac{x}{A})^{2}}\leq A\sup_{t\in[t_{0},t_{0}+1]}\left(\int\eta^{2}(t)\varphi\right)+CA^{-\frac{1}{2}}.$ We now choose $A$ depending on $t_{0}$: $A=A_{t_{0}}=\min\left(\frac{\sqrt{t_{0}}}{2},\left(\sup_{t\in[t_{0},t_{0}+1]}\int_{x\geq\frac{t}{10}-\rho(t)}\eta^{2}(t,x)dx\right)^{-\frac{1}{2}}\right).$ For this choice of $A_{t_{0}}$, we have $\lim_{t_{0}\to+\infty}A_{t_{0}}=+\infty$ and, since $\frac{t}{10}-\rho(t)\leq-\frac{t}{2}$, $A\sup_{t\in[t_{0},t_{0}+1]}\left(\int\eta^{2}(t)\varphi\right)\leq CA\sup_{t\in[t_{0},t_{0}+1]}\left(\int_{x\geq\frac{t}{10}-\rho(t)}\eta^{2}(t)\right)+\frac{CA}{t_{0}}\leq CA^{-1}.$ so that $\lim_{t_{0}\to+\infty}A\sup_{t\in[t_{0},t_{0}+1]}\left(\int\eta^{2}(t)\varphi\right)=0$. It follows that $\lim_{t_{0}\to+\infty}\int_{t_{0}}^{t_{0}+1}\int|D^{\frac{1}{2}}\eta(t,x)|^{2}\frac{dxdt}{1+x^{2}}=0.$ ## 5 Multi-soliton case Using the previous arguments and the strategy of [20] for the gKdV equation, we obtain the following result concerning multi-soliton solutions of (1). ###### Theorem 4 (Asymptotic stability of a sum of decoupled solitons). Let $N\geq 1$ and $0<c_{1}^{0}<\ldots<c_{N}^{0}$. There exist $L_{0}>0$, $A_{0}>0$ and $\alpha_{0}>0$ such that if $u_{0}\in H^{\frac{1}{2}}$ satisfies for some $0\leq\alpha<\alpha_{0}$, $L\geq L_{0}$, $\bigg{\|}u_{0}-\sum_{j=1}^{N}Q_{c_{j}^{0}}(.-y_{j}^{0})\bigg{\|}_{H^{\frac{1}{2}}}\leq\alpha\quad\text{where}\quad\forall j\in\\{2,\ldots,N\\},\quad y_{j}^{0}-y_{j-1}^{0}\geq L,$ (109) and if $u(t)$ is the solution of (1) corresponding to $u(0)=u_{0}$, then there exist $\rho_{1}(t),\ldots,\rho_{N}(t)$ such that the following hold (a) Stability of the sum of $N$ decoupled solitons. $\forall t\geq 0,\quad\bigg{\|}u(t)-\sum_{j=1}^{N}Q_{c^{0}_{j}}(x-\rho_{j}(t))\bigg{\|}_{H^{\frac{1}{2}}}\leq A_{0}\left(\alpha+\frac{1}{L}\right).$ (110) (b) Asymptotic stability of the sum of $N$ solitons. There exist $c_{1}^{+},\ldots,c_{N}^{+}$, with $|c_{j}^{+}-c_{j}^{0}|\leq A_{0}\left(\alpha+\frac{1}{L}\right)$, such that $\forall j,\quad u(t,.+\rho_{j}(t))\rightharpoonup Q_{c^{+}_{j}}\quad\text{in $H^{\frac{1}{2}}$ weak as $t\to+\infty$},$ (111) $\bigg{\|}u(t)-\sum_{j=1}^{N}Q_{c^{+}_{j}}(.-\rho_{j}(t))\bigg{\|}_{L^{2}(x\geq\frac{1}{10}{c_{1}^{0}}t)}\to 0,\quad\rho_{j}^{\prime}(t)\to c_{j}^{+}\quad\text{as $t\to+\infty$}.$ (112) Recall that the Benjamin-Ono equation admits explicit multi-soliton solution. We denote by $U_{N}(x;{c_{j}},{y_{j}})$ the explicit family of $N$-soliton profiles, see e.g. [21] formula (1.7) and Appendix A (see also references in [21]). We obtain the following corollary of the above Theorem and the continuous dependence of the solution in $H^{\frac{1}{2}}$. Let $N\geq 1$, $0<c_{1}^{0}<\ldots<c_{N}^{0}$ and set $d_{N}(u)=\inf\big{\\{}\|u-U_{N}(.;c_{j}^{0},y_{j})\|_{H^{\frac{1}{2}}},\ y_{j}\in\mathbb{R}\big{\\}}.$ ###### Corollary 1 (Asymptotic stability in $H^{\frac{1}{2}}$ of multi- solitons). For all $\delta>0$, there exists $\alpha>0$ such that if $d_{N}(u_{0})\leq\alpha$ then for all $t\in\mathbb{R},$ $d_{N}(u(t))\leq\delta$. Recall that a result of stability in $H^{1}$ of double solitons for the BO equation was proved by variational methods in [22]. See also [21] for stability related results. ### 5.1 Sketch of the stability argument [29] For the reader’s convenience, we now sketch the proof of the stability argument for one soliton (see statement in the Introduction). Let $u(t)$ be an $H^{\frac{1}{2}}$ solution of (1) such that $u(0)$ is close to $Q$ in $H^{\frac{1}{2}}$. Let $c^{+}>0$ be close to $1$ such that $\int u^{2}(0)=c^{+}\int Q^{2}$. We use Lemma 1 on $u(t)$ around $Q_{c^{+}}$ so that $\eta(t,x)=u(t,x+\rho(t))-Q_{c^{+}}(x)$ satisfies $\int\eta(t)Q_{c^{+}}^{\prime}=0$ and by $L^{2}$ conservation $\int\eta(t)Q_{c^{+}}=-\frac{1}{2}\int\eta^{2}(t)$. We define the functional $\mathcal{G}(u(t))=E(u(t))+{c^{+}}\int u^{2}(t).$ (113) Observing that $\mathcal{G}(u(t))=\mathcal{G}(u(0))$ and so expanding $u(t)$ in $\mathcal{G}(u(t))$, we obtain $(\mathcal{L}_{c^{+}}\eta(t),\eta(t))+O(\eta^{3}(t))=(\mathcal{L}_{c^{+}}\eta(0),\eta(0))+O(\eta^{3}(0))$ where $\mathcal{L}_{c^{+}}\eta=-\mathcal{H}\eta_{x}+c^{+}\eta-Q_{c^{+}}\eta.$ By the positivity property of $\mathcal{L}_{c^{+}}$, (property (142) of $\mathcal{L}$ and a scaling argument), we then obtain $\|\eta(t)\|_{H^{\frac{1}{2}}}\leq C\|\eta(0)\|_{H^{\frac{1}{2}}}.$ Note that $\left|\int\eta Q_{c^{+}}\right|\leq C\|\eta\|_{L^{2}}^{2}$ replaces the orthogonality condition $\int\eta Q_{c^{+}}=0$. ### 5.2 Sketch of the proof of Theorem 4 The proof is the same as the proof of Theorem 1 in [20]. First, we recall four lemmas (corresponding to Lemmas 1–4 in [20]) which are the main tools in proving Theorem 4. ###### Lemma 10 (Decomposition of the solution). There exist $L_{1},\alpha_{1},K_{1}>0$ such that the following is true. If for $L>L_{1}$, $0<\alpha<\alpha_{1}$, $t_{0}>0$, $\sup_{0\leq t\leq t_{0}}\Big{(}\inf_{y_{j}>y_{j-1}+L}\Big{\\{}\Big{\|}u(t,.)-\sum_{j=1}^{N}Q_{c^{0}_{j}}(.-y_{j})\Big{\|}_{H^{\frac{1}{2}}}\Big{\\}}\Big{)}<\alpha,$ then there exist unique $C^{1}$ functions $c_{j}:[0,t_{0}]\to(0,+\infty),$ $\rho_{j}:[0,t_{0}]\to\mathbb{R}$, such that $\eta(t,x)=u(t,x)-R(t,x)\quad\hbox{where}\quad R(t,x)=\sum_{j=1}^{N}R_{j}(t,x),\quad R_{j}(t,x)=Q_{c_{j}(t)}(x-\rho_{j}(t)),$ satisfies the following orthogonality conditions $\forall j,\forall t\in[0,t_{0}],\quad\int R_{j}(t)\eta(t)=\int(R_{j}(t))_{x}\eta(t)=0.$ Moreover, there exists $C>0$ such that $\forall t\in[0,t_{0}],$ $\|\eta(t)\|_{H^{\frac{1}{2}}}+\sum_{j=1}^{N}|c_{j}(t)-c^{0}_{j}|\leq C\alpha,\quad\forall j,~{}\left|c_{j}^{\prime}(t)\right|+\left|\rho_{j}^{\prime}(t)-c_{j}(t)\right|\leq C\left(\|\eta(t)\|_{L^{2}}+\frac{1}{L}\right).$ ###### Remark 5. In the rest of the argument, the modulation in the scaling parameter for all time (i.e. the introduction of $c_{j}(t)$) is not necessary. Indeed, modulation at $t=0$ would be sufficient since we deal with the subcritical case. However, we have preferred to introduce this modulation to match the strategy of [20]. Expanding $u(t)$ in the energy conservation and using $E(Q_{c})=c^{2}E(Q)$, we have ###### Lemma 11. There exists $C>0$ such that in the context of Lemma 10, $\forall t\in[0,t_{0}]$, $\bigg{|}E(Q)\sum_{j=1}^{N}\left[c_{j}^{2}(t))-c_{j}^{2}(0)\right]+\frac{1}{2}\int\,(\eta_{x}\mathcal{H}\eta-R\eta^{2})(t)\bigg{|}\leq C\left(\|\eta(0)\|_{H^{\frac{1}{2}}}^{2}+\|\eta(t)\|_{H^{\frac{1}{2}}}^{3}+\frac{1}{L}\right).$ We consider $\varphi$ defined as in (16), with $A$ large enough, and we set $\forall j\in\\{2,\ldots,N\\},\quad\mathcal{I}_{j}(t)=\int u^{2}(t,x)\varphi(x-m_{j}(t))dx,\quad m_{j}(t)=\frac{1}{2}(\rho_{j-1}(t)+\rho_{j}(t)).$ Then, proceeding as in the proof of Proposition 1, we obtain the following. ###### Lemma 12. There exists $C>0$ such that in the context of Lemma 10, $\forall j\in\\{2,\ldots,N\\},\ \forall t\in[0,t_{0}],\quad\mathcal{I}_{j}(t)-\mathcal{I}_{j}(0)\leq\frac{C}{L}.$ Finally, setting $c(t,x)=c_{1}(t)+\sum_{j=2}^{N}(c_{j}(t)-c_{j-1}(t))\varphi(x-m_{j}(t)),$ and proceeding as in the proof of Propositions 3 and 4, we have ###### Lemma 13. There exists $\lambda>0$ such that in the context of Lemma 10, $\forall t\in[0,t_{0}],\quad\mathcal{G}_{N}(t):=\int\eta_{x}\mathcal{H}\eta+c(t,x)\eta^{2}-Q\eta^{2}\geq\lambda\|\eta(t)\|_{H^{\frac{1}{2}}}^{2}.$ Recall that the introduction of the functional $\mathcal{G}_{N}(t)$ for the problem of stability of multi-soliton solutions is justified as follows. For the stability of one soliton, the suitable functional is $\mathcal{G}(u(t))$ defined in (113). For the case of $N$ solitons, we introduce the functional $\mathcal{G}_{N}(t)$ which is approximately $E(u(t))+c_{j}(0)\int u^{2}(t)$ around the soliton $Q_{c_{j}}$. Then, we observe (using the energy conservation and Lemma 11) that this quantity is almost decreasing. This is sufficient to conclude the stability argument for several solitons. We now sketch the argument. We refer to [20], Section 3 for more details in the stability proof. _Sketch of the proof of the stability._ Let $\mathcal{V}_{A_{0}}(L,\alpha)=\bigg{\\{}u\in H^{\frac{1}{2}};\inf_{y_{j}-y_{j-1}\geq L}\bigg{\|}u-\sum_{j=1}^{N}Q_{c_{j}^{0}}(.-y_{j})\bigg{\|}_{H^{\frac{1}{2}}}\leq A_{0}\left(\alpha+\frac{1}{L}\right)\bigg{\\}}.$ Part (a) of Theorem 4 is a consequence of the following proposition and continuity arguments. ###### Proposition 6. There exist $A_{0}>0$, $L_{0}>0$ and $\alpha_{0}>0$ such that, for all $u_{0}\in H^{\frac{1}{2}}$, if $\bigg{\|}u_{0}-\sum_{j=1}^{N}Q_{c_{j}^{0}}(.-y_{j}^{0})\bigg{\|}_{H^{\frac{1}{2}}}\leq\alpha,$ where $L\geq L_{0}$, $0<\alpha<\alpha_{0}$, $y_{j}^{0}-y_{j-1}^{0}+L$, and if for $t^{*}>0$, $\forall t\in[0,T^{*}],\quad u(t)\in\mathcal{V}_{A_{0}}(L,\alpha),$ where $u(t)$ is the solution of (1), then $\forall t\in[0,T^{*}],\quad u(t)\in\mathcal{V}_{\frac{1}{2}A_{0}}(L,\alpha).$ The proof of Proposition 6 is exactly the same as the proof of Proposition 1 in [20], using Lemmas 10–13. In particular, we first prove $\forall t\in[0,t^{*}],\quad\sum_{j=1}^{N}|c_{j}(t)-c_{j}(0)|\leq C_{1}\left(\|\eta(t)\|_{H^{\frac{1}{2}}}^{2}+\|\eta(0)\|_{H^{\frac{1}{2}}}^{2}+\frac{1}{L}\right),$ (114) and then $\|\eta(t)\|_{H^{\frac{1}{2}}}^{2}\leq C_{2}\left(\|\eta(0)\|_{H^{\frac{1}{2}}}^{2}+\frac{1}{L}\right),$ (115) where $C_{1},$ $C_{2}>0$ are independent of $A_{0}$, and we then conclude by using the decomposition of $u(t)$ is terms of $\eta(t)$ and $R(t)$. Note that in proving (114), we make use of the following algebraic fact: $E(Q_{c})=c^{2}E(Q),\quad\int Q_{c}^{2}=c\int Q^{2},\quad E(Q)=-\frac{1}{2}\int Q^{2}.$ The last formula is easily obtained from the equation of $Q$ multiplying by $Q$ and then by $xQ^{\prime}$ and using $\int(\mathcal{H}Q^{\prime})(xQ^{\prime})=0$. This allows us to prove the following estimate $\bigg{|}E(Q)\sum_{j=1}^{N}\left(c_{j}(t)-c_{j}(0)\right)+\int Q^{2}\sum_{j=1}^{N}\left\\{c_{j}(0)\left(c_{j}(t)-c_{j}(0)\right)\right\\}\bigg{|}\leq C\sum_{j=1}^{N}\left|c_{j}(t)-c_{j}(0)\right|^{2}.$ which is the analogue of (44) in [20]. The proof of part (b) of Theorem 4 is exactly the same as in [20], Section 4, using Theorem 2, the monotonicity arguments (Proposition 1) and Theorem 5. It follows closely the proof of Theorem 1 in the present paper. The proof of Corollary 1 is omitted since it is the same as the proof of Corollary 1 in [20]. ## 6 Weak convergence and well-posedness results ### 6.1 Weak convergence ###### Theorem 5 (Weak continuity of the BO flow map). Let $(u_{n})$ be a sequence of global $H^{\frac{1}{2}}$ solutions of equation (1). Assume that $u_{n}(0)\rightharpoonup u_{0}$ in $H^{\frac{1}{2}}$ weak and let $u(t)$ be the solution of (1) corresponding to $u(0)=u_{0}$. Then, for all $t\in\mathbb{R}$, $u_{n}(t)\rightharpoonup u(t)$ in $H^{\frac{1}{2}}$ weak. _Proof of Theorem 5._ Let $u_{0,n}=u_{n}(0)$. It is sufficient to prove the result for $T\in[0,1]$. _Step 1. $H^{2}case$._ Here, we assume $u_{0,n}\rightharpoonup u_{0}$ in $H^{2}$. Let $w_{n}=u_{n}-u$. The equation for $w_{n}$ is $\left\\{\begin{aligned} &w_{nt}+\mathcal{H}(w_{n})_{xx}+u_{n}w_{nx}+u_{x}w_{n}=0\\\ &w_{n}(0)=\psi_{n},\quad\psi_{n}=u_{0,n}-u_{0}.\end{aligned}\right.$ (116) Fix $t=T$, $g\in C^{\infty}_{0}(\mathbb{R})$. For a function $\widetilde{u}$ to be determined, we consider the solution $v(t)$ of $\left\\{\begin{aligned} &v_{t}+\mathcal{H}v_{xx}+(\widetilde{u}v)_{x}-u_{x}v=0,\\\ &v(T)=g.\end{aligned}\right.$ Then $\int w_{n}(T,x)g(x)dx-\int\psi_{n}(x)v(0,x)dx=\int_{0}^{T}\int w_{nt}(t)v(t)+\int_{0}^{T}\int w_{n}(t)v_{t}=\mathbf{I}+\mathbf{II}.$ $\mathbf{I}=\int_{0}^{T}\int w_{n}(\mathcal{H}v_{xx}+(u_{n}v)_{x}-u_{x}v),\qquad\mathbf{II}=-\int_{0}^{T}\int w_{n}(\mathcal{H}v_{xx}+(\widetilde{u}v)_{x}-u_{x}v)$ so that $\int w_{n}(T,x)g(x)dx-\int\psi_{n}(x)v(0,x)dx=\int_{0}^{T}\int w_{n}((u_{n}-\widetilde{u})v)_{x}=-\int_{0}^{T}\int w_{nx}(u_{n}-\widetilde{u})v.$ We can assume, after passing to a subsequence, that $u_{n}-\widetilde{u}\to 0$ in $L^{2}_{loc}(\mathbb{R}\times[0,T])$. Next, we will show that given $\varepsilon>0$, there exists $R>0$ such that $\bigg{|}\int_{0}^{T}\int_{|x|>R}w_{nx}(u_{n}-\widetilde{u})v\bigg{|}\leq\varepsilon,\quad\text{uniformly in $n$.}$ In fact, since $\|w_{nx}\|_{L^{\infty}}\leq C$, $\sup_{t}\|v\|_{L^{2}}\leq C$ and $\sup_{t}\|u_{n}-\widetilde{u}\|_{L^{2}}\leq C$, the claim is clear. But then, $\mathbf{I}+\mathbf{II}\to 0$ as $n\to+\infty$. We only needed $\sup_{t}\|v\|_{L^{2}}\leq C$, which needs $\widetilde{u}_{x}\in L^{\infty}$, $u_{x}\in L^{\infty}$, which are both clear. (We use the energy method to bound $v$.) _Step 2. General case._ Fix $N$ large, define $u_{0,n}^{N}$ such that $\widehat{u_{0,n}^{N}}(\xi)=\mathbf{1}_{[-N,N]}(\xi)\widehat{u}_{0,n}(\xi)$, where $\mathbf{1}_{I}$ is the characteristic function of $I$. Note that $\|u_{0,n}^{N}-u_{0,n}\|_{L^{2}}^{2}=\int_{|\xi|\geq N}\big{|}\widehat{u_{0,n}^{N}}(\xi)\big{|}^{2}\leq\frac{1}{N}\|u_{0,n}\|_{H^{\frac{1}{2}}}^{2}\leq\frac{C}{N},$ so that $u_{0,n}^{N}\to u_{0,n}$ in $L^{2}$ as $N\to+\infty$, uniformly in $n$. Fix $g\in C^{\infty}_{0}$, $T\in\mathbb{R}$, $\varepsilon>0$. The proof of the $L^{2}$ continuity of the flow map (see [11]) shows that $\sup_{t\in[0,1]}\|u^{N}(t)-u(t)\|_{L^{2}}\leq C\|u_{0}^{N}-u_{0}\|_{L^{2}},\quad\sup_{t\in[0,1]}\|u_{n}^{N}(t)-u_{n}(t)\|_{L^{2}}\leq C\|u_{0,n}^{N}-u_{0,n}\|_{L^{2}}$ for some universal constant $C>0$. We fix $N$ such that $\bigg{|}\int(u_{n}(T)-u(T))g-\int(u_{n}^{N}(T)-u^{N}(T))g\bigg{|}\leq\frac{\varepsilon}{2},\quad\text{uniformly in $n$.}$ But, for fixed $N$, we let $n\to+\infty$, and use step 1 and the proof is concluded. ### 6.2 Well-posedness result for the nonlinear BO equation with potential In this subsection, for $0<b<b_{0}$, $b_{0}$ small, we consider the IVP $\left\\{\begin{array}[]{l}v_{t}=(-\mathcal{H}v_{x})_{x}-(Q(x{-}t)v)_{x}-\frac{b}{2}(v^{2})_{x}=0\quad\text{on $[-T,T]\times\mathbb{R}$},\\\\[5.0pt] v(t=0,x)=v_{0}(x)\quad\text{on $\mathbb{R}$}.\end{array}\right.$ (117) The well-posedness of the Cauchy problem in $L^{2}$ for this equation is clear from [11] since $u(t,x)=Q(x{-}t)+bv(t,x)$ satisfies the BO equation. Our main concern is a result of equicontinuity of the map $t\mapsto v(t)$ in $L^{2}$ with respect to $b$. To establish such a result we follow the strategy of [11] on equation (117), using the special form of $Q$ and keeping track of the dependency in $b$. ###### Theorem 6. (a) Assume $v_{0}\in H^{\infty}$. Then, there exists $T=T(Q)>0$ and a unique solution $v=S_{b}^{\infty}(v_{0})$ of (117) in $[-T,T]$, $v\in C([-T,T],H^{\infty})$. (b) There exists a constant $C$, independent of $b$ such that $\sup_{t\in[-T,T]}\|v(t)\|_{H^{2}}\leq C\|v_{0}\|_{H^{2}}.$ (118) (c) The mapping $S_{b}^{\infty}$ extends uniquely to a continuous mapping $S_{b}^{0}:L^{2}\to C([-T,T],L^{2})$, and there exists $C$, independent of $b$ such that $\sup_{t\in[-T,T]}\|v(t)\|_{L^{2}}\leq C\|v_{0}\|_{L^{2}}.$ (119) Moreover, given $v_{0}\in L^{2}$, $\|v_{0}\|_{L^{2}}\leq 2$, for any $\varepsilon>0$, there exits $\delta=\delta(v_{0},\varepsilon)>0$ ($\delta$ independent of $b$) such that for any $v_{1}\in L^{2}$, $\|v_{1}\|_{L^{2}}\leq 2,$ $\|v_{0}-v_{1}\|_{L^{2}}\leq\delta\quad\Rightarrow\quad\sup_{t\in[-T,T]}\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{1})(t)\|_{L^{2}}\leq\varepsilon.$ (120) Finally, there exists $\widetilde{\delta}=\widetilde{\delta}(v_{0},\varepsilon)>0$ (independent of $b$) such that for any $t,t^{\prime}\in[-T,T]$, $|t-t^{\prime}|\leq\widetilde{\delta}\quad\Rightarrow\quad\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{0})(t^{\prime})\|_{L^{2}}\leq\varepsilon.$ (121) _Reduction of the proof._ For $0<\lambda\ll 1$, consider $v_{\lambda}(t,x)=\lambda v(\lambda^{2}t,\lambda x)$. Then $v_{\lambda}$ solves $\left\\{\begin{array}[]{l}(v_{\lambda})_{t}=(-\mathcal{H}(v_{\lambda})_{x})_{x}-(\lambda Q(\lambda x{-}\lambda^{2}t)v_{\lambda})_{x}-\frac{b}{2}(v_{\lambda}^{2})_{x}=0\quad\text{on $[-T,T]\times\mathbb{R}$},\\\\[5.0pt] v_{\lambda}(t=0,x)=v_{0,\lambda}(x)\quad\text{on $\mathbb{R}$},\quad v_{0,\lambda}(x)=\lambda v_{0}(\lambda x).\end{array}\right.$ (122) Define $Q_{\lambda}(t,x)=\lambda Q(\lambda x{-}\lambda^{2}t).$ Then the proof of Theorem 6 reduces to prove that for the following (IVP) $\left\\{\begin{array}[]{l}v_{t}=(-\mathcal{H}v_{x})_{x}-(Q_{\lambda}(t,x)v)_{x}-\frac{b}{2}(v^{2})_{x}=0\quad\text{on $[-1,1]\times\mathbb{R}$},\\\\[5.0pt] v(t=0,x)=v_{0}(x)\quad\text{on $\mathbb{R}$},\quad\|v_{0}\|_{L^{2}}\leq\lambda^{\frac{1}{2}},\end{array}\right.$ (123) we have ###### Theorem 7. There exists $b_{0}$, $\lambda>0$ small enough such that if $0<b<b_{0}$, the following hold (a) Assume $v_{0}\in H^{\infty}$. Then, there exists a unique solution $v=S_{b}^{\infty}(v_{0})$ of (123) in $[-1,1]$, $v\in C([-1,1],H^{\infty})$. (b) There exists a constant $C$, independent of $b$ such that $\sup_{t\in[-1,1]}\|v(t)\|_{H^{2}}\leq C\|v_{0}\|_{H^{2}}.$ (124) (c) The mapping $S_{b}^{\infty}$ extends uniquely to a continuous mapping $S_{b}^{0}:L^{2}\to C([-1,1],L^{2})$, and there exists $C$, independent of $b$ such that $\sup_{t\in[-1,1]}\|v(t)\|_{L^{2}}\leq C\|v_{0}\|_{L^{2}}.$ (125) Moreover, given $v_{0}\in L^{2}$, $\|v_{0}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$, for any $\varepsilon>0$, there exits $\delta=\delta(v_{0},\varepsilon)>0$ ($\delta$ independent of $b$) such that for any $v_{1}\in L^{2}$, $\|v_{1}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$, $\|v_{0}-v_{1}\|_{L^{2}}\leq\delta\quad\Rightarrow\quad\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{1})(t)\|_{L^{2}}\leq\varepsilon.$ (126) Finally, there exists $\widetilde{\delta}=\widetilde{\delta}(v_{0},\varepsilon)>0$ (independent of $b$) such that for any $t,t^{\prime}\in[-1,1]$, $|t-t^{\prime}|\leq\widetilde{\delta}\quad\Rightarrow\quad\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{0})(t^{\prime})\|_{L^{2}}\leq\varepsilon.$ (127) The proof of Theorem 7 is based on the following three propositions. ###### Proposition 7. Assume $v_{0}\in H^{\infty}$, then there exists $T=T(\|v_{0}\|_{H^{2}})$ and a unique solution $v$ of (123) in $(-T,T)$. Also, for any $\sigma\geq 2$, $\sup_{t\in(-T,T)}\|u(t)\|_{H^{\sigma}}\leq C(\sigma,\|v_{0}\|_{\sigma},\sup_{t\in(-T,T)}\|v(t)\|_{H^{2}}).$ (128) In particular, the constant $C$ is independent of $b$, ($0<b<b_{0}$) and $\lambda<1$. Proposition 7 is a consequence of the energy method, taking into account that $\|\partial_{x}Q_{\lambda}\|_{L^{1}((-1,1),L^{\infty}_{x})}\leq C.$ ###### Proposition 8. For $\lambda$ small enough, we have that if $T\in(0,1]$, $\|v_{0}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$, $v=S^{\infty}(v_{0})\in C((-T,T),H^{\infty})$ is a solution, then $\sup_{t\in[-T,T]}\|v(t)\|_{H^{2}}\leq C\|v_{0}\|_{H^{2}},$ where $C$ is independent of $b$ $(0<b<b_{0})$. ###### Proposition 9. For $v_{0}\in H^{\infty}$, $N\in[1,\infty)$, $\|v_{0}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$, let $\widehat{v_{0}^{N}}(\xi)=\mathbf{1}_{[-N,N]}(\xi)\hat{v}_{0}(\xi)$, $v_{0}^{N}\in H^{\infty}$. Then, $\sup_{t\in(-1,1)}\|S_{b}^{\infty}(v_{0})(t)-S_{b}^{\infty}(v_{0}^{N})(t)\|_{L^{2}}\leq C\|v_{0}-v_{0}^{N}\|_{L^{2}},\quad\sup_{t\in(-1,1)}\|S_{b}^{\infty}(v_{0})(t)\|_{L^{2}}\leq C\|v_{0}\|_{L^{2}}.$ where $C$ is independent of $b$ $(0<b<b_{0})$. _Proof of Theorem 7 from Propositions 7, 8 and 9_. First, note that Propositions 7 and 8 clearly give (a) and (b) in Theorem 7. Let us turn to the proof of (c): it suffices to show first that if $v_{0,n}\in H^{\infty}$, $\lim_{n\to+\infty}v_{0,n}=v_{0}$ in $L^{2}$, the sequence $S_{b}^{\infty}(v_{0,n})$ is Cauchy in $C([-1,1],L^{2})$. Let $\varepsilon>0$ be given. We want to show that there exists $M_{\varepsilon}$ (independent of $b$) such that $m,\ n\geq M_{\varepsilon}\quad\Rightarrow\quad\sup_{t\in[-1,1]}\|S_{b}^{\infty}(v_{0,n})(t)-S_{b}^{\infty}(v_{0,m})(t)\|_{L^{2}}\leq\varepsilon.$ Observe that $\|v_{0,n}-v_{0,n}^{N}\|_{L^{2}}\leq\|v_{0}-v_{0}^{N}\|_{L^{2}}+\|v_{0}-v_{0,n}\|_{L^{2}}.$ Hence, we can fix $N=N(\varepsilon,v_{0})$ large and $M^{1}_{\varepsilon}$ large such that $\|v_{0,n}-v_{0,n}^{N}\|_{L^{2}}\leq\frac{\varepsilon}{4C}$, for $n\geq M^{1}_{\varepsilon}$, where $C$ is the constant in Proposition 9 ($\|v_{0,n}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$). Then, by Proposition 9, for $n\geq M^{1}_{\varepsilon}$, $\sup_{t\in[-1,1]}\|S_{b}^{\infty}(v_{0,n})(t)-S_{b}^{\infty}(v_{0,n}^{N})(t)\|_{L^{2}}\leq\frac{\varepsilon}{4}.$ It remains to estimate $\sup_{t\in[-1,1]}\|S_{b}^{\infty}(v_{0,n}^{N})(t)-S_{b}^{\infty}(v_{0,m}^{N})(t)\|_{L^{2}}.$ But energy estimates for the difference equation give $\displaystyle\quad\sup_{t\in[-1,1]}\|S_{b}^{\infty}(v_{0,n}^{N})(t)-S_{b}^{\infty}(v_{0,m}^{N})(t)\|_{L^{2}}$ $\displaystyle\leq\|v_{0,n}^{N}-v_{0,m}^{N}\|_{L^{2}}\exp\left(C\int_{-1}^{1}\|\partial_{x}(S_{b}^{\infty}(v_{0,n}^{N})(t)\|_{L^{\infty}_{x}}+C\|\partial_{x}(S_{b}^{\infty}(v_{0,m}^{N})(t)\|_{L^{\infty}_{x}}\right)$ $\displaystyle\leq\|v_{0,n}-v_{0,m}\|_{L^{2}}\exp\left(C\sup_{t\in(-1,1)}\|S_{b}^{\infty}(v_{0,n}^{N}(t)\|_{H^{2}}++C\sup_{t\in(-1,1)}\|S_{b}^{\infty}(v_{0,m}^{N}(t)\|_{H^{2}}\right)$ $\displaystyle\leq\|v_{0,n}-v_{0,m}\|_{L^{2}}\exp\left(CN^{2}\|v_{0,n}\|_{L^{2}}+CN^{2}\|v_{0,n}\|_{L^{2}}\right)\leq\|v_{0,n}-v_{0,m}\|_{L^{2}}\exp(CN^{2})\leq\frac{\varepsilon}{2},$ for $n$, $m$ large (we have used the estimate of Proposition 8). Also, by Proposition 9, we have $\sup_{t\in(-1,1)}\|S_{b}^{\infty}(v_{0,n})(t)\|_{L^{2}}\leq C$. Thus, we obtain the unique extension $S_{b}^{0}$ and (125) holds. To check (126), fix $v_{0}$, $\|v_{0}\|_{L^{2}}\leq\lambda^{\frac{1}{2}}$, let $\varepsilon>0$ be given. With $C$ as in Proposition 9, find $N$ ($N=N(\varepsilon,v_{0})$) so large that $\|v_{0}-v_{0}^{N}\|_{L^{2}}\leq\frac{\varepsilon}{8C}.$ Now find $\delta_{1}=\delta_{1}(\varepsilon,v_{0})$ so small that if $\|v_{0}-v_{1}\|_{L^{2}}\leq\delta_{1}$, then $\|v_{1}-v_{1}^{N}\|_{L^{2}}\leq\frac{\varepsilon}{4C}$. We have $\displaystyle\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{1})(t)\|_{L^{2}}\leq\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{0}^{N})(t)\|_{L^{2}}$ $\displaystyle\quad+\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{1})(t)-S_{b}^{0}(v_{1}^{N})(t)\|_{L^{2}}+\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{1}^{N})(t)-S_{b}^{0}(v_{0}^{N})(t)\|_{L^{2}}.$ By Proposition 9, the first two terms are smaller than $\frac{\varepsilon}{2}$. For the last one, we again use the energy estimate and get, as before $\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{1}^{N})(t)-S_{b}^{0}(v_{0}^{N})(t)\|_{L^{2}}\leq C\|v_{1}-v_{0}\|_{L^{2}}\exp(CN^{2}),$ using Propositions 8 and 9 and (126) follows. For (127), first find $N=N(\varepsilon,v_{0})$ so large that $\|v_{0}-v_{0}^{N}\|_{L^{2}}\leq\frac{\varepsilon}{4C}$, where $C$ is as in Proposition 9. Then, $\sup_{t\in[-1,1]}\|S_{b}^{0}(v_{0})(t)-S_{b}^{0}(v_{0}^{N})(t)\|_{L^{2}}\leq\frac{\varepsilon}{4}$ and we are reduced to showing, for $N$ fixed that if $|t-t^{\prime}|\leq\widetilde{\delta}$, then $\|S_{b}^{0}(v_{0}^{N})(t)-S_{b}^{0}(v_{0}^{N})(t^{\prime})\|_{L^{2}}\leq\frac{\varepsilon}{2}.$ Let $f(t)=\|S_{b}^{0}(v_{0}^{N})(t)\|_{L^{2}}^{2}$. The energy method, combined with Proposition 8 shows that $|f^{\prime}(t)|\leq f(0)\exp(CN^{2})$. But then, for $|t-t^{\prime}|\leq\widetilde{\delta}_{1}$, $|f(t)-f(t^{\prime})|\leq\frac{\varepsilon}{4}$. But $\displaystyle\|S_{b}^{0}(v_{0}^{N})(t)-S_{b}^{0}(v_{0}^{N})(t^{\prime})\|_{L^{2}}^{2}=f(t)+f(t^{\prime})-2\int S_{b}^{0}(v_{0}^{N})(t).S_{b}^{0}(v_{0}^{N})(t^{\prime})dx$ $\displaystyle=f(t^{\prime})-f(t)+2\int S_{b}^{0}(v_{0}^{N})(t)[S_{b}^{0}(v_{0}^{N})(t)-S_{b}^{0}(v_{0}^{N})(t^{\prime})]dx.$ Let $v^{N}(t)=S_{b}^{0}(v_{0}^{N})(t)$. The second term equals $2\int v^{N}(t)\int_{t^{\prime}}^{t}\partial_{s}v^{N}(s)dsdx=2\int^{t}_{t^{\prime}}\int v_{N}(t)[-\mathcal{H}\partial_{x}^{2}v^{N}(s)-(Q_{\lambda}v^{N})_{x}-\frac{b}{2}((v^{N})^{2})_{x}(s)]dxds.$ But by Proposition 8, $\sup_{t\in[-1,1]}\|v^{N}(t)\|_{H^{2}}\leq C\|v_{0}^{N}\|_{H^{2}}\leq CN^{2}.$ Thus, the second term is controlled by $C|t-t^{\prime}|N$, and the proof is complete, provided we prove Propositions 8 and 9. _Proof of Propositions 8 and 9._ _Step 1._ Assume $v_{0}\in H^{\infty}$, $\|v_{0}\|_{H^{2}}\leq M$ and $0<T\leq 2$, $v=S_{b}^{\infty}(t)$ exists in $[-T,T]$. Then, there exist $\lambda_{0}=\lambda_{0}(M)$, $b_{0}=b_{0}(M)$ such that for $0<\lambda<\lambda_{0}$, $0\leq b<b_{0}$, we have $\sup_{t\in[-T,T]}\|v(t)\|_{H^{2}}\leq 2\|v_{0}\|_{H^{2}}.$ (129) Proof of (129). Note that $\|\partial_{x}^{k}Q\lambda\|_{L^{\infty}}\leq C_{k}\lambda^{k+1}$. Let $f(t)=\|v(t)\|_{H^{2}}^{2}$. The standard energy method shows that $|f^{\prime}(t)|\leq C(\lambda_{0}^{2}+b_{0}\|\partial_{x}v(t)\|_{L^{\infty}_{x}})f(t)\leq(\lambda_{0}^{2}+b_{0}(f(t))^{\frac{1}{2}})f(t).$ Integrating the ODE gives the result. As a corollary, we obtain under the circumstances of Step 1 that $v$ exists in $(-1,1)$ and $\sup_{t\in[-2,2]}\|v(t)\|_{H^{2}}\leq 2\|v_{0}\|_{H^{2}}.$ _Step 2._ From now on, we will follow closely [11]. Some of the ideas used before were developed in a forthcoming paper [8]. We have now reduced everything to _a priori_ estimates. We will change notation slightly to match [11]. We then study the problem $\left\\{\begin{aligned} &u_{t}+\mathcal{H}u_{xx}+(Q_{\lambda}u)_{x}+b(\tfrac{1}{2}u^{2})_{x}=0\quad(t,x)\in(-1,1)\times\mathbb{R},\\\ &u_{|t=0}=\phi,\quad\|\phi\|_{L^{2}}\leq\lambda^{\frac{1}{2}},\end{aligned}\right.$ (130) We use the notation $P_{\rm low}$, $P_{\rm\pm high}$ as in [11]: $P_{\rm low}\text{ defined by the Fourier multiplier }\xi\to\mathbf{1}_{[-2^{10},2^{10}]}(\xi);$ $P_{\rm\pm high}\text{ defined by the Fourier multiplier }\xi\to\mathbf{1}_{[2^{10},\infty)}(\pm\xi);$ $P_{\pm}\text{ defined by the Fourier multiplier }\xi\to\mathbf{1}_{[0,\infty)}(\pm\xi).$ Let $\phi_{0}=P_{\rm low}\phi\in H^{\infty}$, real-valued, $\|\phi_{0}\|_{H^{2}}\leq 2^{20}=M$. We choose $\lambda_{0}$, $b_{0}$ as in Step 1 and its corollary, so that Proposition 7 and these results gives, with $u_{0}^{(1)}=S^{\infty}_{b}(\phi_{0})(t)$ that $\sup_{t\in[-2,2]}\|\partial_{t}^{\sigma_{1}}\partial_{x}^{\sigma_{2}}u_{0}^{(1)}\|_{L^{2}_{x}}\leq C_{\sigma_{1},\sigma_{2}}\|\phi\|_{L^{2}},\quad\sigma_{i}\geq 0.$ Let $\widetilde{u}=u-u_{0}^{(1)}$. The equation for $\widetilde{u}$ is $\left\\{\begin{aligned} &\widetilde{u}_{t}+\mathcal{H}\widetilde{u}_{xx}+(Q_{\lambda}\widetilde{u})_{x}+b(u_{0}^{(1)}\widetilde{u})_{x}+b(\tfrac{1}{2}\widetilde{u}^{2})_{x}=0,\\\ &\widetilde{u}_{|t=0}=P_{\rm+high}\phi+P_{\rm-high}\phi.\end{aligned}\right.$ (131) Let now $u_{0}(t,x)=Q_{\lambda}(t,x)+bu_{0}^{(1)}(t,x)$. Then $\sup_{t\in[-2,2]}\|\partial_{t}^{\sigma_{1}}\partial_{x}^{\sigma_{2}}u_{0}\|_{L^{2}_{x}}\leq C_{\sigma_{1},\sigma_{2}}(\lambda_{0}^{\frac{1}{2}}+b_{0}).$ We now want to construct $U_{0}$ similarly to [11], with the following properties $\partial_{x}U_{0}(t,x)=\frac{1}{2}u_{0}(t,x)$, $U_{0}(0,0)=0$ and $\sup_{t\in[-2,2]}\|\partial_{t}^{\sigma_{1}}\partial_{x}^{\sigma_{2}}U_{0}(t,.)\|_{L^{2}_{x}}\leq C_{\sigma_{1},\sigma_{2}}(\lambda_{0}^{\frac{1}{2}}+b_{0})$ where $\sigma_{1},\sigma_{2}\geq 0$, $(\sigma_{1},\sigma_{2})\neq(0,0)$. Since $Q_{\lambda}(t,x)=\frac{4\lambda}{1+(\lambda x-\lambda^{2}t)^{2}}$, we set $U_{0}^{(2)}(t,x)=2\arctan(\lambda x-\lambda^{2}t)$. We next recall the equation $u_{0}^{(1)}(t,x)$ verifies: $\partial_{t}(\tfrac{1}{2}{u_{0}^{(1)}})+\mathcal{H}\partial_{x}^{2}(\tfrac{1}{2}{u_{0}^{(1)}})+\partial_{x}(Q_{\lambda}\tfrac{1}{2}{u_{0}^{(1)}})+b\partial_{x}((\tfrac{1}{2}{u_{0}^{(1)}})^{2})=0.$ We then define first $U_{0}^{(1)}(t,0)$ by the formula $\partial U_{0}^{(1)}(t,0)+\mathcal{H}\partial_{x}(\tfrac{1}{2}{u_{0}^{(1)}(t,0)})+Q_{\lambda}(t,0)\tfrac{1}{2}{u_{0}^{(1)}(t,0)}+b(\tfrac{1}{2}{u_{0}^{(1)}(t,0)})^{2}=0,\quad U_{0}^{(1)}(0,0)=0.$ We then construct $U_{0}^{(1)}(t,x)$ by $\partial_{x}U_{0}^{(1)}(t,x)=\frac{1}{2}u_{0}^{(1)}(t,x)$. Notice that $U_{0}^{(1)}$ is real-valued. Using the equation for $u_{0}^{(1)}$, we have $\partial_{x}\left(\partial_{t}U_{0}^{(1)}+\mathcal{H}\partial_{x}^{2}U_{0}^{(1)}+Q_{\lambda}\partial_{x}U_{0}^{(1)}+b(\partial_{x}U_{0}^{(1)})^{2}\right)=0\quad\text{on $\mathbb{R}\times[-2,2]$}.$ But then, on $\mathbb{R}\times[-2,2]$, we have $\partial_{t}U_{0}^{(1)}(t,x)+\frac{1}{2}\mathcal{H}\partial_{x}u_{0}^{(1)}(t,x)+Q_{\lambda}(t,x)\frac{1}{2}u_{0}^{(1)}(t,x)+\frac{b}{4}(u_{0}^{(1)}(t,x))^{2}.$ We then define $U_{0}(t,x)=bU_{0}^{(1)}(t,x)+U_{0}^{(2)}(t,x)$, and all our properties hold. We recall that $\left\\{\begin{aligned} &\widetilde{u}_{t}+\mathcal{H}\widetilde{u}_{xx}+(u_{0}\widetilde{u})_{x}+b(\tfrac{1}{2}\widetilde{u}^{2})_{x}=0,\\\ &\widetilde{u}_{|t=0}=P_{\rm+high}\phi+P_{\rm-high}\phi.\end{aligned}\right.$ (132) We now proceed as in Section 2 of [11]. We define $P_{+\rm high}\widetilde{u}=e^{-iU_{0}}w_{+}$, $P_{-\rm high}\widetilde{u}=e^{iU_{0}}w_{-}$ and $P_{\rm low}\widetilde{u}=w_{0}$. Applying $P_{+\rm high}$, $P_{-\rm high}$, $P_{\rm low}$ to the above equation and using the definitions above, we have (we write the equation for $w_{+}$, the one for $w_{-}$ is analoguous, the one for $w_{0}$ will be written later). Following the argument in [11], one gets: $\displaystyle(w_{+})_{t}+\mathcal{H}\partial_{x}^{2}w_{+}=-\frac{b}{2}e^{iU_{0}}P_{+\rm high}\partial_{x}((e^{-iU_{0}}w_{+}+e^{iU_{0}}w_{-}+w_{0})^{2})$ $\displaystyle-e^{-iU_{0}}P_{+\rm high}\partial_{x}(u_{0}(e^{iU_{0}}w_{-}+w_{0}))+e^{iU_{0}}(P_{-\rm high}+P_{\rm low})(e^{iU_{0}}u_{0}\partial_{x}w_{+})+2iP_{-}\partial_{x}^{2}w_{+}$ $\displaystyle-e^{iU_{0}}P_{+\rm high}(\partial_{x}(u_{0}e^{-iU_{0}}w_{+}))+iw_{+}\left[(U_{0})_{t}-i(U_{0})_{xx}-((U_{0})_{x})^{2}\right],$ and so after more calculations, we get $\displaystyle(w_{+})_{t}+\mathcal{H}\partial_{x}^{2}w_{+}=-\frac{b}{2}e^{iU_{0}}P_{+\rm high}\partial_{x}((e^{-iU_{0}}w_{+}+e^{iU_{0}}w_{-}+w_{0})^{2})$ $\displaystyle-e^{-iU_{0}}P_{+\rm high}\left[\partial_{x}(u_{0}P_{-\rm high}(e^{iU_{0}}w_{-})+u_{0}P_{\rm low}(w_{0}))\right]$ $\displaystyle+e^{iU_{0}}(P_{-\rm high}+P_{\rm low})\left[\partial_{x}(u_{0}P_{+\rm high}(e^{-iU_{0}}w_{+}))\right]$ $\displaystyle+2iP_{-}\left[\partial_{x}^{2}(e^{iu_{0}}P_{+\rm high}(e^{-iU_{0}}w_{+}))\right]+iw_{+}\left[(U_{0})_{t}+\mathcal{H}\partial_{x}^{2}U_{0}+(\partial_{x}U_{0})^{2}+iP_{+}\partial_{x}U_{0}\right],$ We recall $\partial_{x}U_{0}^{(2)}=\frac{1}{2}Q_{\lambda}$ and that $Q_{\lambda}$ solves $\partial_{t}Q_{\lambda}+\mathcal{H}\partial_{x}^{2}Q_{\lambda}+\partial_{x}(\tfrac{1}{2}Q_{\lambda}^{2})=0$ or $\partial_{t}U_{0}^{(2)}+\mathcal{H}\partial_{x}^{2}U_{0}^{(2)}=-\tfrac{1}{4}Q_{\lambda}^{2}$ and $\partial_{t}U_{0}^{(1)}+\mathcal{H}\partial_{x}^{2}U_{0}^{(1)}=-Q_{\lambda}\partial U_{0}^{(1)}-b(\partial_{x}U_{0}^{(1)})^{2}.$ Hence, $\partial_{t}U_{0}+\mathcal{H}\partial_{x}^{2}U_{0}+(\partial_{x}U_{0})^{2}=0$ and we get $\partial w_{+}+\mathcal{H}w_{+}=E_{+}(w_{+},w_{-},w_{0})$, where $E_{+}$ is defined as in [11], p. 756, except that the first term is multiplied now by $b$. The equation for $w_{-}$ and $E_{-}$ is similar. The equation for $w_{0}$ writes $\partial_{t}(P_{\rm low}\widetilde{u})+\mathcal{H}\partial_{x}^{2}P_{\rm low}\widetilde{u}+P_{\rm low}\partial_{x}(u_{0}\widetilde{u})+\tfrac{b}{2}P_{\rm low}\partial_{x}((\widetilde{u})^{2})=0,$ where $\widetilde{u}=e^{-iU_{0}}w_{+}+e^{iU_{0}}w_{-}+w_{0}$. Next, we note that, with $\delta=(\lambda_{0}^{\frac{1}{2}}+b_{0})$, the estimates (10.19) in [11] hold. Because of this and the form of $E_{+}$, $E_{-}$, $E_{0}$, just as in Proposition 10.5 in [11], we have $\displaystyle\|\psi(t)(\mathbf{E}(\mathbf{w})-\mathbf{E}(\mathbf{w}^{\prime}))\|_{N^{\sigma}}$ $\displaystyle\leq Cb_{0}\|\mathbf{w}-\mathbf{w^{\prime}}\|_{F^{\sigma}}(\|\mathbf{w}\|_{F^{0}}+\|\mathbf{w}^{\prime}\|_{F^{0}})$ $\displaystyle\quad+Cb_{0}\|\mathbf{w}-\mathbf{w^{\prime}}\|_{F^{0}}(\|\mathbf{w}\|_{F^{\sigma}}+\|\mathbf{w}^{\prime}\|_{F^{\sigma}})+C\delta\|\mathbf{w}-\mathbf{w^{\prime}}\|_{F^{\sigma}}.$ Note that $\mathbf{w}=(w_{+},w_{-},w_{0})$ and $\mathbf{E}(\mathbf{w})=(E_{+}(w_{+},w_{-},w_{0}),E_{-}(w_{+},w_{-},w_{0}),E_{0}(w_{+},w_{-},w_{0}))$ as in [11]. The rest of the notation (the norm $\|.\|_{N^{\sigma}}$ and the function $\psi$) is also taken from [11]. We have a slightly different formula for $E_{0}$, but (10.27) in [11] gives the estimate in our case also. We then construct a solution to $\left\\{\begin{aligned} &\mathbf{v}_{t}+\mathcal{H}\mathbf{v}_{xx}=\mathbf{E}(\mathbf{v})\quad\text{on }\mathbb{R}\times[-\tfrac{5}{4},\tfrac{5}{4}],\\\ &\mathbf{v}(0)=\Phi,\end{aligned}\right.$ as in (10.32)-(10.37) in [11]. Note that (10.35) and $\|v(\Phi)-v(\Phi^{\prime})\|_{F^{0}([-\frac{5}{4},\frac{5}{4}])}\leq C\|\Phi-\Phi^{\prime}\|_{\widetilde{H}^{0}}$ hold here too. Next, with $\Phi=(\phi_{+},\phi_{-},\phi_{0})=(e^{iU_{0}(0,.)}P_{+\rm high}\phi,e^{-iU_{0}(0,.)}P_{-\rm high}\phi,0)$, $\Phi\in\widetilde{H}^{20}$, by Lemma 10.1 in [11]. We next show $(w_{+},w_{-},w_{0})=\mathbf{v}(\Phi)$ in $\mathbb{R}\times[-1,1]$. This is as in [11]. Proposition 8, and the second estimate in Proposition 9 now follow from the bounds on $\mathbf{v}(\Phi)$ i.e. (10.35). For Proposition 9, note that for $N$ large, $U_{0}$ corresponding to $\phi$ and to $\phi_{N}$ defined by $\hat{\phi}_{N}=\mathbf{1}_{[-N,N]}(\xi)\hat{\phi}(\xi)$ are the same. We then have $u(t,x)=u_{0}^{(1)}+u-u_{0}^{(1)}=u_{0}^{(1)}+\widetilde{u}=u_{0}^{(1)}+e^{-iU_{0}}w_{+}+e^{iU_{0}}w_{-}+w_{0}$ and similarly, $u^{N}(t,x)=u_{0}^{(1)}+u^{N}-u_{0}^{(1)}=u_{0}^{(1)}+e^{-iU_{0}}w_{+}^{N}+e^{iU_{0}}w_{-}^{N}+w_{0}^{N}$. Hence, $\displaystyle\sup_{t\in[-1,1]}\|u(t,.)-u^{N}(t,.)\|_{L^{2}}$ $\displaystyle\leq\sup_{t\in[-1,1]}\|w(t)-w^{N}(t)\|_{L^{2}}$ $\displaystyle\leq C\|\psi(t)[w-w^{N}]\|_{F^{0}}\leq C\|\phi-\phi^{N}\|_{L^{2}}$ as desired, giving Proposition 9. ## Appendix A Appendix First, we recall the following inequalities: ###### Lemma 14. $\forall 2\leq p<+\infty,\quad\|f\|_{L^{p}}\leq C_{p}\|f\|_{L^{2}}^{\frac{2}{p}}\|D^{\frac{1}{2}}f\|_{L^{2}}^{\frac{p-2}{p}},$ (133) $\|D(fg)-gDf\|_{L^{2}}\leq C\|f\|_{L^{4}}\|Dg\|_{L^{4}},$ (134) $\|D^{\frac{1}{2}}(fg)-gD^{\frac{1}{2}}f\|_{L^{2}}\leq C\|f\|_{L^{4}}\|D^{\frac{1}{2}}g\|_{L^{4}}.$ (135) Recall that (133) is the Gagliardo-Nirenberg inequality, which follows from complex interpolation and Sobolev embedding. Estimate (134) is due to Calderón [6], see also Coifman and Meyer [7], formula (1.1). Estimate (135) is a consequence of Theorem A.8 in [12] for functions depending only on $x$, with the following choice of parameters: $\alpha=\frac{1}{2}$, $\alpha_{1}=0$, $\alpha_{2}=\frac{1}{2}$, $p=2$, $p_{1}=p_{2}=4$. ### A.1 Proof of (23) We claim that for a function $u(x)$ fixed in $H^{2}(\mathbb{R})$ $\int_{y=0}\partial_{y}(U^{2})\Phi=-2\iint_{\mathbb{R}^{2}_{+}}|\nabla U|^{2}\Phi+\int_{y=0}U^{2}\partial_{y}\Phi$ (136) where $U(x,y)$ is the harmonic extension of $u(x)$ in $\mathbb{R}^{2}_{+}$ and $\Phi(x,y)$ is defined in (22). First, we observe that $U,\nabla U\in L^{\infty}(\mathbb{R}^{2}_{+})\quad\text{and}\quad\sup_{y>0}|U(x,y)|\to 0\text{ as $|x|\to+\infty$}.$ (137) Indeed, from [24], Theorem 1, p. 62, we have $\sup_{y>0}|U(x,y)|\leq Mu(x)$, where $Mu(x)$ is the maximal function of $u$ (see [24] Chapter 1), and similarly, $\sup_{y>0}|\partial_{x}U(x,y)|\leq Mu_{x}(x)$, $\sup_{y>0}|\partial_{y}U(x,y)|\leq M(\mathcal{H}u_{x})(x)$. Moreover, from [24] Theorem 1, p. 5, since $u,u_{x},\mathcal{H}u_{x}\in H^{1}\subset L^{\infty}$, we obtain $Mu,Mu_{x},M(Hu_{x})\in L^{\infty}$. Finally, since $u\in H^{1}$, we have $|u(x)|\to 0$ as $|x|\to+\infty$, which implies by the definition of the maximal function (see [24], page 4) that $Mu(x)\to 0$ as $|x|\to+\infty$. Thus (137) is proved. Let $R>0$. We use the Green formula on $D_{R}^{+}=\\{(x,y)\in\mathbb{R}_{+}^{2}~{}|~{}x^{2}+y^{2}<R^{2}\\}$. Let $\Gamma_{R}^{+}=\\{(x,y)\in\mathbb{R}_{+}^{2}~{}|~{}x^{2}+y^{2}=R^{2}\\}$ and $I_{R}={(x,0)~{}|~{}x\in[-R,R]}$. Then: $\begin{split}\int_{\Gamma_{R}^{+}\cup I_{R}}\partial_{n}(U^{2})\Phi&=-\iint_{D_{R}^{+}}(\Delta U^{2})\Phi+\iint_{D_{R}^{+}}U^{2}\Delta\Phi+\int_{\Gamma_{R}^{+}\cup I_{R}}U^{2}\partial_{n}\Phi\\\ &=-2\iint_{D_{R}^{+}}|\nabla U|^{2}\Phi+\int_{\Gamma_{R}^{+}\cup I_{R}}U^{2}\partial_{n}\Phi,\end{split}$ (138) where $\partial_{n}$ denotes the inward normal derivative since $\Delta\Phi=0$ and $\Delta U^{2}=2|\nabla U|^{2}$. Therefore, we only have to prove the following convergence results: $\displaystyle\lim_{R\to+\infty}\int_{\Gamma_{R}^{+}}\partial_{n}(U^{2})\Phi=0,\quad\lim_{R\to+\infty}\int_{I_{R}}\partial_{n}(U^{2})\Phi=\int_{y=0}\partial_{y}(U^{2})\Phi=2\int(\mathcal{H}u_{x})u\varphi^{\prime}$ (139) $\displaystyle\lim_{R\to+\infty}\int_{\Gamma_{R}^{+}}U^{2}\partial_{n}\Phi=0,\quad\lim_{R\to+\infty}\int_{I_{R}}U^{2}\partial_{n}\Phi=\int_{y=0}U^{2}\partial_{y}\Phi=\int u^{2}(\mathcal{H}\varphi^{\prime\prime}).$ (140) The limits $\lim_{R\to+\infty}\int_{-R}^{R}(\mathcal{H}u_{x})u\varphi^{\prime}=\int(\mathcal{H}u_{x})u\varphi^{\prime}$ and $\lim_{R\to+\infty}\int_{-R}^{R}u^{2}(\mathcal{H}\varphi^{\prime\prime})=\int u^{2}(\mathcal{H}\varphi^{\prime\prime})$ are clear since $u\in H^{1}$. Next, from the expression of $\Phi(x,y)$ in (22), we have $\Phi(x,y)\leq C(1+y)R^{-2}$ on $\Gamma_{R}^{+}$. Therefore, from (137), ($d\sigma$ denotes the unit lenght element on $\Gamma_{R}^{+}$) $\begin{split}\int_{\Gamma_{R}^{+}}|\partial_{n}(U^{2})\Phi|&\leq\frac{1}{R^{2}}\int_{\Gamma_{R}^{+}}|\nabla U||U|(1+y)d\sigma\\\ &\leq\frac{C}{R^{2}}\int_{\Gamma_{R}^{+}\cap\\{|x|\leq\sqrt{R}\\}}(1+y)d\sigma+C\sup_{|x|>\sqrt{R},y>0}|U(x,y)|\\\ &\leq\frac{C}{\sqrt{R}}+C\sup_{|x|>\sqrt{R},y>0}|U(x,y)|\end{split}$ and so (139) is proved. Estimate (140) is proved similarly and is easier since $\partial_{y}\Phi$ has more decay than $\Phi$. ### A.2 Proof of (38) In the proof of (38), the time $t$ is fixed, so we set $y_{0}=x_{0}+\lambda(t_{0}-t)$. Let $\chi:\mathbb{R}\to\mathbb{R}$ be a $C^{\infty}$ function such that $\chi=1$ on $[0,1]$, $\chi=0$ on $(-\infty,-1]\cap[2,+\infty)$ and $\chi\leq 1$ on $\mathbb{R}$. Let $\chi_{n}(x)=\chi(x-n)$. Then, by the Gagliardo Nirenberg inequality (133), we obtain $\begin{split}\int|\eta|^{3}\varphi^{\prime}(x-y_{0})&\leq\sum_{n\in\mathbb{Z}}\int_{n}^{n+1}|\eta|^{3}\varphi^{\prime}(x-y_{0})\leq\sum_{n\in\mathbb{Z}}\bigg{(}\int|\eta|^{3}\chi_{n}^{3}\bigg{)}\sup_{[n-y_{0},n+1-y_{0}]}\varphi^{\prime}\\\ &\leq\sum_{n\in\mathbb{Z}}\bigg{(}\int|D^{\frac{1}{2}}(\eta\chi_{n})|^{2}\bigg{)}^{\frac{1}{2}}\bigg{(}\int(\eta\chi_{n})^{2}\bigg{)}\sup_{[n-y_{0},n+1-y_{0}]}\varphi^{\prime}.\end{split}$ By Lemma 14 and (13), we get $\|D^{\frac{1}{2}}(\eta\chi_{n})\|_{L^{2}}\leq C\|(D^{\frac{1}{2}}\eta)\chi_{n}\|_{L^{2}}+C\|\eta\|_{L^{4}}\|D^{\frac{1}{2}}\chi_{n}\|_{L^{4}}\leq C\|\eta\|_{H^{\frac{1}{2}}}\leq C\alpha_{0}.$ Thus, $\begin{split}\int|\eta|^{3}\varphi^{\prime}(x-y_{0})&\leq C\alpha_{0}\sum_{n\in\mathbb{Z}}\bigg{(}\int(\eta\chi_{n})^{2}\bigg{)}\sup_{[n-y_{0},n+1-y_{0}]}\varphi^{\prime}\leq C\alpha_{0}\int\eta^{2}\varphi^{\prime}(x-y_{0})\end{split}$ by the properties of $\chi$ and the following elementary remark: $\forall y\in\mathbb{R},\quad\sup_{[y,y+4]}\varphi^{\prime}\leq C\inf_{[y,y+4]}\varphi^{\prime}.$ (141) Note that the constant $C$ is independent of $A$, for $A>1$. ### A.3 Properties of the operator $\mathcal{L}$ We recall from [27]–[29] and [3] the following properties of $\mathcal{L}$ (recall $\mathcal{L}\eta=-\mathcal{H}\eta_{x}+\eta-Q\eta$). ###### Lemma 15. The operator $\mathcal{L}$ is self-adjoint on $L^{2}$ and satisfies the following properties. (i) The operator $\mathcal{L}$ has exactly one negative eigenvalue $\lambda_{0}$ of multiplicity $1$ with corresponding eigenfunction $f_{0}$, which can be chosen so that $f_{0}>0$. (ii) ${\rm Ker}\,\mathcal{L}={\rm span}\\{Q^{\prime}\\}$. (iii) There exists $\lambda>0$ such that, for all $z\in H^{\frac{1}{2}}$, $(z,Q)=(z,Q^{\prime})=0\quad\Rightarrow\quad(\mathcal{L}z,z)\geq\lambda(z,z).$ (142) ###### Remark 6. Recall from Bennett et al. ([3], Appendix B) that the spectrum of $\mathcal{L}$ is completely understood. Indeed, the operator $\mathcal{L}$ has exactly four eigenvalues, $\lambda_{0}=-\frac{1}{2}(1+\sqrt{5})$, $0$, $\frac{1}{2}(-1+\sqrt{5})$, $1$ and a continuous spectrum $[1,+\infty)$. Now, we sketch a proof of Lemma 15 using general arguments from [27]–[29]. _Sketch of proof._ One easily checks that $\mathcal{L}Q^{\prime}=0$ (differentiate the equation of $Q(x+x_{0})$ with respect to $x_{0}$ and take $x_{0}=0$), and that $\mathcal{L}f_{0}=-\lambda_{0}f_{0}$, where $f_{0}=Q+\frac{1}{4}(1+\sqrt{5})Q^{2}$ (by (78)). Moreover, the proof of (i) follows from the variational characterization of $Q$, see Proposition 4.2 of [29]. Recall that $\frac{d}{dc}\int Q_{c}^{2}=\int Q^{2}>0$ (subcriticality) implies that $\inf\\{(\mathcal{L}f,f);\ (f,Q)=0,\ \|f\|_{L^{2}}=1\\}=0$ (see proof of Proposition 5.1 in [29] and Proposition 3.1 in [28]). Now, we give a new proof for (ii). Let $f\in L^{2}$ be such that $\mathcal{L}f=0$. First, we remark that $f\in H^{s}$, for all $s\geq 0$. Moreover, by similar estimates as in [1], we have $|f(x)|\leq\frac{C}{1+x^{2}}.$ Integrating $\mathcal{L}f=0$ on $\mathbb{R}$, we obtain $\int(f-fQ)=0$. But, we also have $(f,Q)=-(f,\mathcal{L}S)=-(\mathcal{L}f,S)=0$ (see (78)). Thus, $\int f=0$ and we can define $g(x)=\int_{-\infty}^{x}f(s)ds\in L^{2}$, which satisfies $\mathcal{L}(g^{\prime})=0$. Let now $\widetilde{g}=g-aQ$ be such that $(\widetilde{g},Q)=0$ and $\mathcal{L}(\widetilde{g}^{\prime})=0$. From (56) and (79), we obtain $\int|D^{\frac{1}{2}}\widetilde{g}|^{2}+(\mathcal{L}\widetilde{g},\widetilde{g})\leq 0$. But, since $(\widetilde{g},Q)=0$, we have $(\mathcal{L}\widetilde{g},\widetilde{g})\geq 0$. Thus, $\int|D^{\frac{1}{2}}\widetilde{g}|^{2}=0$ and $\widetilde{g}\equiv 0$, so that $g=aQ$ and $f=aQ^{\prime}$. Finally, we sketch the proof of (iii), which follows from the arguments of the proof of Proposition 2.9 in [27] (see also Section 6, example 4 in [29]). By contradiction, assuming that $\inf\\{(\mathcal{L}f,f);\ (f,Q)=(f,Q^{\prime})=0,\ \|f\|_{L^{2}}=1\\}=0,$ and using compactness arguments as in Proposition 2.9 in [27], we obtain the existence of $f\in H^{\frac{1}{2}}$, $\lambda,\beta,\gamma\in\mathbb{R}$ (Lagrange multipliers) such that $(\mathcal{L}f,f)=0,\quad(\mathcal{L}-\lambda)f=\beta Q+\gamma Q^{\prime},\quad(f,Q)=(f,Q^{\prime})=0,\quad\|f\|_{L^{2}}=1.$ But, taking the scalar product by $f$, we find $\lambda=0$. Then, taking the scalar product by $Q^{\prime}$, we find $\gamma=0$. Taking the scalar product with $S$ (see (78)), using $(S,Q)=\frac{1}{2}(Q,Q)$ and $\mathcal{L}(S)=-Q$, we find $\beta=0$, so that $\mathcal{L}f=0$ and $(f,Q^{\prime})=0$. This implies $f=0$ by (ii), a contradiction. ## References * [1] C.J. Amick and J.F. Toland, Uniqueness and related analytic properties for the Benjamin-Ono equation—a nonlinear Neumann problem in the plane. Acta Math. 167 (1991), 107–126. * [2] T.B. Benjamin, Internal waves of permanent form in fluids of great depth, Journal of Fluid Mechanics 29 (1967), 559–592. * [3] D.P. Bennett, R.W. Brown, S.E. Stansfield, J.D. Stroughair, J.L. Bona, The stability of internal solitary waves. Math. Proc. Cambridge Philos. Soc. 94 (1983), 351–379. * [4] J.L. Bona, P.E. Souganidis and W.A. Strauss, Stability and instability of solitary waves of Korteweg-de Vries type, Proc. R. Soc. Lond. A 411 (1987), 395–412. * [5] N. Burq and F. Planchon, On well-posedness for the Benjamin-Ono equation, to appear in Math. Annalen. * [6] A.-P. Calderon, Commutators of singular integral operators. Proc. Nat. Acad. Sci. U.S.A. 53 (1965), 1092 1099. * [7] R.R. Coifman and Y. Meyer, On commutators of singular integrals and bilinear singular integrals. Trans. Amer. Math. Soc. 212 (1975), 315–331. * [8] S. Herr, A.D. Ionescu, C.E. Kenig and H. Koch, Global solutions to dispersive nonlinear equations, preprint. * [9] J. Ginibre and G. Velo, Commutator expansions and smoothing properties of generalized Benjamin-Ono equations. Ann. Inst. H. Poincar Phys. Th or. 51 (1989), 221–229. * [10] S. Gustafson, H. Takaoka and T.-P. Tsai, Stability in $H^{\frac{1}{2}}$ of the sum of $K$ solitons for the Benjamin-Ono equation. Preprint. * [11] A.D. Ionescu and C.E. Kenig, Global well-posedness of the Benjamin–Ono equation in low-regularity spaces, J. Amer. Math. Soc. 20 (2007), 753–798. * [12] C.E. Kenig, G. Ponce and L. Vega, Well-posedness and scattering results for the generalized Korteweg-de Vries equation via the contraction principle, Comm. Pure Appl. Math. 46 (1993), 527–620. * [13] Y. Martel, Linear problems related to asymptotic stability of solitons of the generalized KdV equations, SIAM J. Math. Anal. 38 (2006), 759–781. * [14] Y. Martel and F. Merle, Instability of solitons for the critical generalized Korteweg-de Vries equation. Geom. Funct. Anal. 11 (2001), 74–123. * [15] Y. Martel and F. Merle, A Liouville theorem for the critical generalized Korteweg–de Vries equation. J. Math. Pures Appl. 79 (2000), 339–425. * [16] Y. Martel and F. Merle, Asymptotic stability of solitons for subcritical generalized KdV equations, Arch. Ration. Mech. Anal. 157, (2001) 219–254. * [17] Y. Martel and F. Merle, Asymptotic stability of solitons of the subcritical gKdV equations revisited. Nonlinearity 18 (2005), no. 1, 55–80. * [18] Y. Martel and F. Merle, Asymptotic stability of solitons of the gKdV equations with a general nonlinearity. To appear in Math. Annalen. http://arxiv.org/abs/0706.1174 * [19] Y. Martel and F. Merle, Refined asymptotics around soliton for gKdV equations. DCDS 20 (2008), 177–218. * [20] Y. Martel, F. Merle and Tai-Peng Tsai, Stability and asymptotic stability in the energy space of the sum of $N$ solitons for the subcritical gKdV equations, Commun. Math. Phys. 231, (2002) 347–373. * [21] Y. Matsuno, The Lyapunov stability of the N-soliton solutions in the Lax hierarchy of the Benjamin-Ono equation, Journal of Mathematical Physics 47 (2006), 103505. * [22] A. Neves and O. Lopes, Orbital Stability of Double Solitons for the Benjamin-Ono Equation. Commun. Math. Phys. 262 (2006), 757–791. * [23] G. Ponce, Smoothing properties of solutions to the Benjamin-Ono equation. Analysis and partial differential equations, 667–679, Lecture Notes in Pure and Appl. Math., 122, Dekker, New York, 1990. * [24] E. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton Univ. Press, Princeton, NJ, 1970. * [25] T. Tao, Global well-posedness of the Benjamin-Ono equation in $H^{1}(\mathbb{R})$, Journal of Hyperbolic Differential Equations 1 (2004), 27–49. * [26] J.F. Toland, The Peierls-Nabarro and Benjamin-Ono equations, J. Funct. Anal. 145 (1997), 136–150. * [27] M.I. Weinstein, Modulational stability of ground states of nonlinear Schrödinger equations, SIAM J. Math. Anal. 16, (1985) 472–491. * [28] M.I. Weinstein, Lyapunov stability of ground states of nonlinear dispersive evolution equations. Comm. Pure Appl. Math. 39 (1986), 51–68. * [29] M.I. Weinstein, Existence and dynamic stability of solitary wave solutions of equations arising in long wave propagation. Comm. Partial Differential Equations 12 (1987), 1133–1173.

arxiv-papers

2024-09-04T02:48:54.539054

0803.3744

# The Berry phase in frustrated spin glass Dipti Banerjee∗ Department of Physics,Rishi Bankim Chandra College Naihati,$24$-Parganas(N),Pin-$743165$, West Bengal, INDIA and The Abdus Salam International Center for Theoretical Physics, Trieste, ITALY (10.09.07) ###### Abstract In this letter we have pointed out that frustration in spin glass is realized through the Berry phase due to the conflict between the spin ordering in the course of parallel transport of spinor. We have came to the point that the Berry phase depicting the chiral change of helicity of a quantized spinor is prominent only in the presence of frustration. PACS No:-$71.55$.Jv *email:deepbancu@homail.com;dbanerje@ictp.it In the theory of spin glasses, the concept of frozen spin configuration and frustration have played an important role [1]. The frozen spin gives the emphasis on the rigidity opposed to the spatial ordering of spin. Interaction between the spins are in conflict with each other due to some quenched disorder leading to frustration. It is known that various classes of randomness exist for the behavior of spin glass and this randomness leads to frustration [2]. From the topological point of view, constraints prevent the neighboring spins to have the minimum bond energy in spin glass. The geometry of the spin-ordering in a spin glass has a similarity with the ”parallel transport” of a tangent vector on a curved surface [3]. The misfit between the various lines of transport can be expressed by the frustration and curvature respectively. In this sense frustrated plaquette are curved whereas un- frustrated are flat. The above ideas imply that ’the frustration in spin glass’ can be realized through the curvature of space measured by the Berry phase[4]. This topological phase is developed by the parallel transport of the spinor over a closed path. The analogy at the deeper level lies in gauge symmetries where gauge potentials known as Berry connections are the source of curvature of space time. The spin Berry phase plays an important role in the quantum transport of strongly correlated spin system [5]. This phase is also the very cause of net change of spin chirality visualized through chiral anomaly in the field theoretic aspect [6]. In this paper we will focus our attention to study the frustration of quantum spin glass from the view point of Berry phase. In continuous rotations of spins the Hamiltonian of granular spin glass system is [7], $H=-J_{ij}\Sigma_{ij}cos(\phi_{i}-\phi_{j})$ (1) where $J_{ij}$ is a coupling depending on the nature of host (metal, insulator and superconductor etc.) material. Here the complex energy gap of ith grain is $\psi_{i}=\Delta_{i}exp(i\phi_{i})$ This is similar to the Hamiltonian of a XY spin ferromagnet. Due to short ranged interaction the energy of all domain walls becomes [1] $\Delta E(C)=\Sigma_{ij}J_{ij}cos(\theta_{i}-\theta_{j})$ (2) where $(\theta_{i}-\theta_{j})$ is the angle between the two spins at ith and jth site respectively. In absence of frustration the spinors being strongly correlated and $cos(\theta_{i}-\theta_{j})$ is almost decided by $J_{ij}$. The presence of frustration results weak correlation determining $(\theta_{i}-\theta_{j})$ not only by $J_{ij}$ but also by the rest of the neighbors. The above Hamiltonian changes in presence of magnetic field $H=-J_{ij}\Sigma_{ij}cos(\phi_{i}-\phi_{j}-A_{ij})$ (3) where $A_{ij}=\frac{2\pi}{\Phi_{0}}{\int_{i}}^{j}\vec{A}.\vec{dl}$ (4) is the gauge potential generated by the interaction of two spinors. Here $\Phi_{0}$ is the elementary flux quantum $hc/2e$. This shows that a magnetic field can act as a source of frustration. Replacing the element of the above spin vectors by Pauli matrices, we get a quantized model. The above Hamiltonian in eq.(3) becomes [7] $H=J\Sigma({S_{i}}^{*}U_{ij}S_{j}+h.c)$ (5) where $U_{ij}=exp(iA_{ij})$ represent the link gauge degree of freedom and $S_{i}=\exp(i\phi_{i})$ the spin vector respectively. Here the randomness arises from the difference angle $A_{ij}$ rather than the exchange bonds. The Hamiltonian remains invariant under the local gauge transformation. ${S_{i}}^{\prime}\rightarrow V_{i}S_{i}$ and $~{}~{}~{}~{}~{}~{}{U_{ij}}^{\prime}\rightarrow V_{i}U_{ij}{V_{j}}^{*}$ where $V_{i}=exp(i\theta_{i})$. In fact under local gauge transformation applied at the ith site the corresponding spin gets rotated by an angle $\theta_{i}$ and each of the connecting links get rotated by the difference of angle $(\theta_{i}-\theta_{j})$ such that the Hamiltonian remains invariant under such transformation. For the conventional XY model the matrix $U_{ij}=\pm 1$ restricting the transformation angles $\phi_{i}$ to $(0,\pi)$. In a frustrated spin system the relative orientations of neighboring spins are not only decided by their interaction alone but also by the rest of the spin society. For any closed path in a lattice spin the sign of product of the exchange integral is known as frustrated function. Here the angles $\psi_{ij}$ are treated as continuous variables, which correspond to complex bonds $J_{ij}$. In a frustrated system the exchange integral around any closed contour is equal to $-1$ whereas for the un-frustrated system it is $+1$. The quantity $\exp[2\pi i\phi_{ijkl}]=U_{ij}U_{jk}U_{kl}U_{li}$ (6) is called the frustration function defined for any closed path in the lattice spins [7]. Considering $U_{ij}=\exp{iA_{ij}}$ the above equation changes to $\phi_{ijkl}=A_{ij}+A_{jk}+A_{kl}+A_{li}$ This is the case for the frustrated square plateaus. If we consider the triangular frustrated lattice then the frustrated function will be $\phi_{ijk}=A_{ij}+A_{jk}+A_{ki}=\sum_{ij}A_{ij}$ (7) It seems that this sum over link gauge degree of freedom $\sum_{ij}A_{ij}$ gives rise in the continuum limit the connection over a closed path.This implies that a frustrated function is equivalent to the net change of curvature over the closed path measured through Berry phase. A frustrated system is described by a chiral spin liquid where the signature of chiral spinor $\psi_{L}$ or $\psi_{R}$ may be considered to represent the order parameter. For a system with an odd number of anti ferromagnetic links this change in chirality of the above two spinors will lead to a change in chirality in a frustrated loop [8]. The order parameter of a frustrated spin system can be depicted by chiral fermions represented by two opposites orientation of helicities. It has been pointed out earlier [9] that chiral fermion may be depicted by a scalar particle moving with $l=1/2$ in an anisotropic space. In three space dimension, in an axis-symmetric system where the anisotropy is introduced along a particular direction, the components of the linear momentum satisfy a commutation relation of the form $[p_{i},p_{j}]=i\mu\varepsilon_{ijk}\frac{x^{k}}{r^{3}}$ (8) Here $\mu$ corresponds to the measure of anisotropy and behaves like the strength of a magnetic monopole. Indeed the angular momentum relation in this space is given by $\vec{J}=\vec{r}\times\vec{p}-\mu\hat{r}$ (9) with $\mu=0,\pm 1/2,\pm 1...$. This corresponds to the motion of a charged particle in the field of a magnetic monopole [10]. The spherical harmonics incorporating the term $\mu$ becomes $\displaystyle{Y_{l}}^{m,\mu}=(1+x)^{-(m-\mu)/2}.(1+x)^{-(m+\mu)/2}$ $\displaystyle\times\frac{d^{l-m}}{d^{l-m}x}\left((1+x)^{l-\mu}.(1-x)^{l+\mu}\right)e^{im\phi}e^{i\mu\chi}$ (10) with $x=cos\theta$. Since the chirality is associated with the angle $\chi$ denoting the rotational orientation around the direction vector $\xi_{\mu}$, the variation of the angle $\chi$ i.e. the change of rotational orientation around the direction vector will correspond to the change in chirality.In spherical harmonics given by eqn.(10) the spin angular part associated with the angle $\chi$ is given by $e^{-i\mu\chi}$. Thus when $\chi$ is changed to $\chi+\delta\chi$, we have $i\frac{\partial}{\partial(\chi+\delta\chi)}e^{-i\mu\chi}=i\frac{\partial}{\partial(\chi+\delta\chi)}e^{-i\mu(\chi+\delta\chi)}e^{i\mu\delta\chi}$ (11) which implies that the wave function will acquire the extra phase $e^{i\mu\delta\chi}$ due to an infinitesimal change of the angle $\chi$ to $\chi+\delta\chi$. When the angle $\chi$ is changed over the closed path $0\leq\chi\leq 2\pi$, for one complete rotation, the wave function will acquire the phase [10] $exp[i\mu{{\int_{0}}^{2\pi}}\delta\chi]=e^{2i\pi\mu}$ (12) which represents the spin dependent Berry phase. Indeed in this formalism, a fermion is depicted as a scalar particle moving in the field of a magnetic monopole and when a scalar field(particle) traverses a closed path with one flux quantum ($\mu=1/2$) enclosed, we have the phase $e^{i\pi}$, suggesting the system a fermion. For the specific case of $l=1/2,|m|=|\mu|=1/2$ for half orbital/spin angular momentum, we can construct from the spherical harmonics ${Y_{l}}^{m,\mu}$, the instantaneous eigenstates $|\uparrow,t\rangle$, representing the two component up-spinor as $\displaystyle|\uparrow,t\rangle$ $\displaystyle=$ $\displaystyle{u\choose v}={{Y_{1/2}}^{1/2,1/2}\choose{Y_{1/2}}^{-1/2,1/2}}$ (13) $\displaystyle=$ $\displaystyle{\sin\frac{\theta}{2}\exp i(\phi-\chi)/2\choose\cos\frac{\theta}{2}\exp-i(\phi+\chi)/2}$ and the charge conjugate state, down-spinor by $|\downarrow,t\rangle={{-Y_{1/2}}^{-1/2,1/2}\choose{Y_{1/2}}^{-1/2,-1/2}}={-\cos\frac{\theta}{2}\exp i(\phi+\chi)/2\choose\sin\frac{\theta}{2}\exp-i(\phi-\chi)/2}$ (14) In an arbitrary superposition of elementary qubits $|0\rangle and|1\rangle$ the up-spinors becomes $|\uparrow,t\rangle=\left(\sin\frac{\theta}{2}e^{i\phi}|0\rangle+\cos\frac{\theta}{2}|1\rangle\right)e^{-i/2(\phi+\chi)}$ (15) The time evolution of a two state system is governed by an unitary $SU(2)$ $2\times 2$ transformation matrix $U(g)$ as follows $U(g)=\pmatrix{{\alpha~{}~{}~{}~{}-\beta^{*}}\cr{\beta~{}~{}~{}~{}~{}\alpha^{*}}}$ (16) where $|\alpha|^{2}+|\beta|^{2}=1$ with $|g\rangle=U(g)|0\rangle$= $\alpha\choose\beta$. These states $|\uparrow,t\rangle$ and $|\downarrow,t\rangle$ can be generated by the unitary matrix $U(\theta,\phi,\chi)$ $U(\theta,\phi,\chi)=\pmatrix{{\sin\frac{\theta}{2}e^{i/2(\phi-\chi)}~{}~{}~{}~{}~{}~{}~{}~{}~{}-\cos\frac{\theta}{2}e^{i/2(\phi+\chi)}}\cr\cos\frac{\theta}{2}e^{-i/2(\phi+\chi)}~{}~{}~{}~{}~{}~{}~{}~{}~{}{\sin\frac{\theta}{2}e^{-i/2(\phi-\chi)}}}$ (17) from the basic qubits $|0\rangle$ and $|1\rangle$ as follows $\displaystyle|\uparrow,t\rangle=U(\theta,\phi,\chi)|0\rangle,|\downarrow,t\rangle=U(\theta,\phi,\chi)|1\rangle$ (18) Over a closed path, the single quantized up spinor acquires the geometrical phase [11] $\displaystyle\gamma_{\uparrow}$ $\displaystyle=$ $\displaystyle i\oint\langle\uparrow,t|\nabla|\uparrow,t\rangle.d\lambda$ (19) $\displaystyle=$ $\displaystyle i\oint\langle 0|U{\dagger}dU|0\rangle.d\lambda$ (20) $\displaystyle=$ $\displaystyle i\oint A_{\uparrow}(\lambda)d\lambda$ (21) $\displaystyle=$ $\displaystyle i\oint L^{\uparrow}_{eff}dt$ (22) $\displaystyle=$ $\displaystyle i\frac{1}{2}(\oint d\chi-\cos\theta\oint d\phi)$ (23) $\displaystyle=$ $\displaystyle i\pi(1-\cos\theta)$ (24) This shows that for quantized spinor, the Berry Phase is a solid angle subtended about the quantization axis. For conjugate state, the down spinor becomes $|\downarrow(t)\rangle=(-\cos\frac{\theta}{2}|0\rangle+\sin\frac{\theta}{2}e^{-i\phi}|1\rangle)e^{i/2(\phi+\chi)}$ (25) giving rise in similar manner the Berry phase over the closed path $\gamma_{\downarrow}=-i\pi(1-\cos\theta)$ (26) The fermionic or the antifermionic nature of the two spinors (up/down) can be identified by the maximum value of topological phase $\gamma_{\uparrow/\downarrow}=\pm\pi$ at an angle $\theta=\pi/2$. For $\theta=0$ we get the minimum value of $\gamma_{\uparrow}=0$ and at $\theta=\pi$ no extra effect of phase is realized. This Berry phase visualized by the solid angle is acquired by the parallel transport of the quantized spinor over a closed path resulting the reunion of the final point with the initial in the absence of local frustration. The Berry phase in connection with chirality as in eq.(12) is not visible here. One can verify that this Berry phase in eq.(24) remains same if we neglect the overall phase $e^{-i(\phi+\chi)/2}$ from the quantized spinors in eq.(15). This is possible when there is no local frustration causing any spin conflict in the system. In case of spin glass, the frustrated spinor acquire different Berry phase. Due to nontrivial frustration by the disorder in the glassy system, the quantized spinor does not reach the initial point. In other words the path traced out by the spinor is not closed. Intuitively the initial and final points are connected by the fibre representing the gauge due to randomness in the spin direction. Murakami et.al. [5] pointed out that when an electron hops from site i to j coupled to a spin at each site then the spin wave function is effectively $|\chi_{i}\rangle=t\left(e^{ib_{i}}\cos\frac{\theta_{i}}{2},e^{i(b_{i}+\phi_{i})}\sin\frac{\theta_{i}}{2}\right)$ (27) The overall phase $b_{i}$ corresponding to the gauge degree of freedom does not appear as physical quantities. The effective transfer integral $t_{ij}$ is given by $\displaystyle t_{ij}$ $\displaystyle=$ $\displaystyle t\langle\chi_{i}|\chi_{j}\rangle$ (28) $\displaystyle=$ $\displaystyle te^{(b_{j}-b_{i})}\left(\cos\frac{\theta_{i}}{2}\cos\frac{\theta_{j}}{2}+e^{i(\phi_{j}-\phi_{i})}\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}\right)$ $\displaystyle=$ $\displaystyle te^{ia_{ij}}\cos\frac{\theta_{ij}}{2}$ where $\theta_{ij}$ is the angle between the two spins $\vec{S_{i}}$ and $\vec{S_{j}}$. The phase $a_{ij}$ is the vector potential generated by the spin,and corresponds to the Berry phase felt by the hopping electron. It has been pointed out [5], that the total phase obtained by an electron hopping along a loop $1\longrightarrow 2\longrightarrow 3\longrightarrow 1$ is also the solid angle subtended by the three spins. This $a_{ij}$ measures the spin chirality in the context of quantum spin liquid where the spins fluctuate quantum mechanically [12]. In the light of above works we concentrate in finding the Berry phase of a quantized spinor residing on a spherical frustrated surface. Rotation takes place once through points having the same solid angle in terms of $\theta$. The transfer integral of the quantized spinor as in eq.(15) will be $\displaystyle\langle{\uparrow,t}_{j}|{\uparrow,t}_{i}\rangle=e^{i/2(\phi_{j}-\phi_{i})}.e^{i/2(\chi_{j}-\chi_{i})}$ $\displaystyle\left(\cos\frac{\theta_{i}}{2}\cos\frac{\theta_{j}}{2}+e^{i(\phi_{i}-\phi_{j})}\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}\right)$ (29) Comparing with eq.(28) we have the transfer integral $t_{ij}$ expressing the variation of $\chi$ along with the angle $\theta_{ij}$. $\langle{\uparrow,t}_{j}|{\uparrow,t}_{i}\rangle=e^{i/2(\phi_{j}-\phi_{i})}e^{i/2(\chi_{j}-\chi_{i})}\cos\frac{\theta_{ij}}{2}$ (30) Representing the change of helicity $(\chi_{i}-\chi_{j})=\chi+\delta\chi-\chi=\delta\chi$ the corresponding phase $e^{i/2(\chi_{j}-\chi_{i})}=e^{i/2\delta\chi}=e^{i/2a_{ij}}$ is the Berry phase (eq.12) visualize the chiral change of helicity of quantized fermion. The following phase $\langle{\uparrow,t}_{j}|{\uparrow,t}_{i}\rangle=e^{i/2(\phi_{j}-\phi_{i}+a_{ij})}\cos\frac{\theta_{ij}}{2}$ (31) represents the difference of inclination of helicity over the virtual closed path. Following the local gauge transformation $a_{ij}\longrightarrow a_{ij}+\phi_{i}-\phi_{j}$ (32) the two eqs.(28) and (31) are equivalent. Hence we have a similar form of transfer integral for quantized spinor as in [5]. In a frustrated system the quantized spinors fix up their helicity. Transportation around a closed loop represents only the variation of $\phi$ values from $0\rightarrow 2\pi$ where the slight shift of $\chi$ values is visualized as chiral gauge due to some conflicts between the spins offered by the disorders in the system. The required Berry connection of a quantized up spinor in the frustrated spin system can be obtained after few mathematical steps using eq.(15) $\displaystyle\langle\uparrow_{j}|d|\uparrow_{i}\rangle$ $\displaystyle=$ $\displaystyle e^{i/2(\phi_{j}-\phi_{i})}.e^{i/2(\chi_{j}-\chi_{i})}.$ (33) $\displaystyle\left(\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}e^{i(\phi_{i}-\phi_{j})}d\phi_{i}\right.$ $\displaystyle\left.-\frac{i}{2}\cos\frac{\theta_{ij}}{2}(d\chi_{i}+d\phi_{i})\right)$ and similar connection for the down-spinor from eq.(25), $\displaystyle\langle\downarrow_{j}|d|\downarrow_{i}\rangle$ $\displaystyle=$ $\displaystyle e^{i/2(\phi_{i}-\phi_{j})}.e^{i/2(\chi_{i}-\chi_{j})}.$ (34) $\displaystyle\left(\frac{i}{2}\cos\frac{\theta_{ij}}{2}(d\chi_{i}+d\phi_{i})-\right.$ $\displaystyle\left.-i\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}e^{i(\phi_{j}-\phi_{i})}d\phi_{i}\right)$ Geometrically in a frustrated spin system the parallel transport of a spinor over a closed path on a sphere parameterized by $\theta,\phi$ and $\chi$ implies open curve because the site at the final point do not coincide with initial one. The Berry phase for both frustrated up and down spinors in the spin glass system will be obtained after integration over variation of $\phi$ by $0\leq\phi\leq 2\pi$. ${\Gamma^{\uparrow}}_{F}=-2i\pi e^{i/2(\chi_{j}-\chi_{i})}\left(\cos\frac{\theta_{ij}}{2}-\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}\right)$ (35) and ${\Gamma^{\downarrow}}_{F}=2i\pi e^{i/2(\chi_{i}-\chi_{j})}\left(\cos\frac{\theta_{ij}}{2}-\sin\frac{\theta_{i}}{2}\sin\frac{\theta_{j}}{2}\right)$ (36) These Berry phases ${\Gamma^{\uparrow}}_{F}$ or ${\Gamma^{\downarrow}}_{F}$ for frustrated system, are products of the helicity ($\chi$) dependent phase and the solid angle of spinor between the spinors based not only on the individual angles $\theta_{i}$ and $\theta_{j}$ of the spinors but also on $\theta_{ij}$, the angle between the two. In the absence of local frustration, $\cos\frac{\theta_{ij}}{2}=0$, no spin conflict arises, indicating the transport of spin vectors ideally parallel. As a result the final site coincide with the initial leading to choose $\chi_{i}=\chi_{j}$, $\phi_{i}=\phi_{j}$ and $\theta_{i}=\theta_{j}$ that gives rise the phase for un-frustrated system. $\gamma^{\uparrow}=i\pi(1-\cos\theta_{i})$ (37) and $\gamma^{\downarrow}=-i\pi(1-\cos\theta_{i})$ These are the usual solid angles identical with eqs.(24) and (26) respectively visualizing the Berry phase for the up/down spinor in an isolated system. In a frustrated system, $\cos\frac{\theta_{ij}}{2}=\pm 1$, acts as a signature of two chirality that may act also an order parameter in the system. For an un- frustrated system even in the presence of a magnetic field which is one of the very source of quantization, the helicity/internal helicity depending Berry phase is not visible. This is only realized in a frustrated spin glass system, where disorders offer spin conflict to realize helicity depending phase along with the solid angle. At the end we would like to point out that the frustrated and un-frustrated Berry phase could be a very source in developing the nontrivial matrix Berry phase of two qubit state. In a very recent communication [13] we have pointed out that due to frustration in Quantum Hall system, the lowest Landau level LLL ($\nu=1$) is a two qubit singlet state $\displaystyle\Phi_{1}(z)$ $\displaystyle=$ $\displaystyle\pmatrix{{u_{i}~{}~{}~{}~{}~{}u_{j}}\cr{v_{i}~{}~{}~{}~{}~{}v_{j}}}=(u_{i}v_{j}-u_{j}v_{i})$ (38) $\displaystyle=$ $\displaystyle(u_{i}~{}~{}~{}~{}v_{i})\pmatrix{{0~{}~{}~{}~{}~{}1}\cr{-1~{}~{}~{}~{}~{}0}}{u_{j}\choose v_{j}}$ that has been identified as Hall qubit constructed from the up-spinor $|\uparrow_{i}>={u_{i}\choose v_{i}}$. The non-abelian nature of the connection on the Hall surface will remain if $i\neq j$. $\displaystyle B_{\uparrow}$ $\displaystyle=$ $\displaystyle\pmatrix{{(u_{i}^{*}du_{i}+v_{i}^{*}dv_{i})~{}~{}~{}~{}(u_{i}^{*}du_{j}+v_{i}^{*}dv_{j})}\cr{(u_{j}^{*}du_{i}+v_{j}^{*}dv_{i})~{}~{}~{}~{}(u_{i}^{*}du_{i}+v_{i}^{*}dv_{i})}}$ (39) $\displaystyle=$ $\displaystyle\pmatrix{{\mu_{i}~{}~{}~{}\mu_{ij}}\cr{\mu_{ji}~{}~{}~{}~{}\mu_{j}}}$ This is visualizing the spin conflict during the parallel transport leading to non-abelian Berry phase. We realize in the light of Hwang et.al [14] that non- abelian matrix Berry phase created by frustration is responsible for the pumped charge flow over a cycle by the singlet states in the Quantum Hall system. ${\gamma^{H}}_{\uparrow}=\pmatrix{{\gamma_{i}~{}~{}~{}~{}\Gamma_{ij}}\cr{\Gamma_{ji}~{}~{}~{}~{}\gamma_{j}}}$ (40) Here $\gamma_{i}$ and $\gamma_{j}$ are the respective un-frustrated Berry Phases for the ith and jth spinor as seen in eqs.(15) and (17). $\Gamma_{ij}$ represents the off-diagonal Berry Phase developed by the local frustration in the spin system. In the absence of frustration there will be no development of matrix Berry phase. Hence we would like to conclude that the matrix Berry phase that is responsible for pumped charge to flow can be well realized in the frustrated system. Acknowledgements I like to acknowledge my home institute and specially The Abdus Salam International Center for Theoretical Physics, Trieste,Italy for giving me the full support for doing this work. Also I am highly motivated by participating in the ”School and Workshop on Highly Frustrated Magnets and Strongly Correlated Systems: From Non-Perturbative Approaches to Experiments” held during 30th July to 17th August.07 at ICTP. ## References * [1] G. Baskaran in Physics of Disordered Solids by Prabodh Shukla. P.W.Anderson and H. Hasegawa, Phys.Rev.100, 675 (1955). * [2] G. Toulouse, Commun.Phys. 2, 115 (1977). * [3] Spin Glasses and Other Frustrated Systems by Debashish Chowdhury,World Scientific. * [4] M.V.Berry, Proc.R.Soc.London A392,45(1984). * [5] K. Ohgushi, S. Murakami and N. Nagaosa; cond-mat/9912206. * [6] D.Banerjee; Fort.der Physik 44 (1996) 323. * [7] Spin Glasses by K.H.Fischer and J.A. Hertz, Cambridge University Press. * [8] B.Basu; J.Math.Phys. 34, 737(1993). * [9] P.Bandyopadhyay;Int. J.Mod.Phys. A4,4449(1989). * [10] D.Banerjee and P.Bandyopadhyay ; J.Math.Phys.33, 990 (1992). * [11] D.Banerjee and P.Bandyopadhyay, Physica Scripta, 73,571(2006), ICTP preprint. * [12] G.Baskaran and P.W.Anderson, Phys.Rev.B37, 580 (1988). * [13] D.Banerjee;”The Qubit rotation in QHE” communicated to PRB. * [14] N.Y.Hwang,S.C.Kim, P.S.Park and S.R.Eric Yang;arXiv; cond. mat/0706.0947.

arxiv-papers

2024-09-04T02:48:54.550802

0803.3745

# Qubit rotation in QHE Dipti Banerjee∗ Department of Physics,Rishi Bankim Chandra College Naihati,$24$-Parganas(N),Pin-$743165$, West Bengal INDIA (5.09.07) ###### Abstract In Quantum Hall effect the ground state wave function at $\nu=1$ is the building block of all other states at different filling factors. It is developed by the entanglement of two spinors forming a singlet state. The inherent frustration visualized by the non-abelian matrix Berry phase is responsible for the quantum pumped charge to flow in the Hall surface. The Physics behind the Quantum Hall states is studied here from the view point of topological quantum computation. Key words: spin echo, Berry phase. PACS No:-$73.43$-f $*$This work of the author (Regular Associate) is partially supported by ICTP,Trieste,Italy. email: deepbancu@homail.com,dbanerje@ictp.it. ## 1 Introduction Entanglement is one of the basic aspects of quantum mechanics. It was known long ago that quantum mechanics exhibits very peculiar correlations between two physically distant parts of the total system. Afterwards, the discovery of Bell’s inequality (BI) [1] showed that BI can be violated by quantum mechanics but has to be satisfied by all local realistic theories. The violation of BI demonstrates the presence of entanglement [2]. The theorem of BI may be interpreted as incompatibility of requirement of locality with the statistical predictions of quantum mechanics. So to study the Bell state, the role of local spatial observations, apart from spin correlations, should also be taken into account [3]. This indicates that the spatial variation of a quantum mechanical state would carry its memory through some geometric phase known as Berry phase(BP) [4]. It is expected that the influence of BP on an entangled state could be linked up with the local observations of spins. To have a comprehensive view of the quantum mechanical correlation between two spin $1/2$ particles in an entangled state, we should take into account the role of the Berry phase related to a spinor. This study belong to the field of geometric quantum computation where noticeably the geometrical and topological gates are resistant to local disturbances. In quantum mechanical entanglement of two spin 1/2 particles the Berry phase plays an important role during spin echo method [5]. Kitaev described [6] the topological and quantum computer as a device in which quantum numbers carried by quasiparticles residing in two dimensional electron gas have long range Aharonov-Bohm (AB) interactions between one another. These AB interactions are responsible for nontrivial phase values during interwinding of quasiparticles trajectories in course of time evolution of qubits in Quantum Hall Effects (QHE). Quantization plays an important role to realize the Physics behind different states of QHE from the view point of Berry phase [7]. Recently we have studied the rotation of a quantized spinor identified as qubit in presence of magnetic field under the spin echo method [8]. We will aim at understanding the rotation of QHE qubits specially in the lowest Landau level $\nu=1$ and then parent states $\nu=1/m$ from the view point of geometric quantum computation. ## 2 Quantization of Fermi field and qubits of singlet states The quantization of Fermi field can be achieved assuming anisotropy in the internal space through the introduction of direction vector as an internal variable at each space-time point [9]. The opposite orientations of the direction vector correspond to particle and antiparticle. Incorporation of spinorial variables $\theta(\bar{\theta})$ in the coordinate result the enlargement of manifold from $S^{2}$ to $S^{3}$.This helps us to consider a relativistic quantum particle as an extended one, where the extension involves gauge degrees of freedom. As a result the position and momentum variables of a quantized particle becomes $Q_{\mu}=i\left(\frac{\partial}{{\partial p}_{\mu}}+A_{\mu}\right),~{}~{}~{}~{}~{}~{}~{}~{}~{}P_{\mu}=i\left(\frac{\partial}{{\partial q}_{\mu}}+\tilde{A_{\mu}}\right)$ (1) where $q_{\mu}$ and $p_{\mu}$ are related to the position and momentum coordinates in the sharp point limit and $A_{\mu}(\tilde{A_{\mu}})$ are non- Abelian matrix valued gauge fields belonging to the group $SL(2C)$. In three space dimension, in an axis-symmetric system where the anisotropy is introduced along a particular direction, the components of the linear momentum satisfy a commutation relation of the form [10] $[p_{i},p_{j}]=i\mu\varepsilon_{ijk}\frac{x^{k}}{r^{3}}$ (2) Here $\mu$ corresponds to the measure of anisotropy and behaves like the strength of a magnetic monopole. Indeed in this anisotropic space the conserved angular momentum is given by $\vec{J}=\vec{r}\times\vec{p}-\mu\hat{r}$ (3) with $\mu=0,\pm 1/2,\pm 1...$. This corresponds to the motion of a charged particle in the field of a magnetic monopole. For the specific case of $l=1/2,|m|=|\mu|=1/2$ for half orbital/spin angular momentum, we can construct from the spherical harmonics ${Y_{l}}^{m,\mu}$, the instantaneous eigenstates $\left|\uparrow,t\right\rangle$, representing the two component up-spinor as $\displaystyle\left|\uparrow,t\right\rangle$ $\displaystyle=$ $\displaystyle{u\choose v}={{Y_{1/2}}^{1/2,1/2}\choose{Y_{1/2}}^{-1/2,1/2}}$ (4) $\displaystyle=$ $\displaystyle{\sin\frac{\theta}{2}\exp i(\phi-\chi)/2\choose\cos\frac{\theta}{2}\exp-i(\phi+\chi)/2}$ and the conjugate state is a down-spinor given by $\left|\downarrow,t\right\rangle={{-Y_{1/2}}^{-1/2,1/2}\choose{Y_{1/2}}^{-1/2,-1/2}}={-\cos\frac{\theta}{2}\exp i(\phi+\chi)/2\choose\sin\frac{\theta}{2}\exp-i(\phi-\chi)/2}$ (5) These two spinors (up/down) represent quantized fermi field originated by an arbitrary superposition of elementary qubits $\left|0\right\rangle and\left|1\right\rangle$ as for up spinor $\left|\uparrow,t\right\rangle=\left(\sin\frac{\theta}{2}e^{i\phi}\left|0\right\rangle+\cos\frac{\theta}{2}\left|1\right\rangle\right)e^{-i/2(\phi+\chi)}$ (6) and the down spinor becomes $\left|\downarrow(t)\right\rangle=(-\cos\frac{\theta}{2}\left|0\right\rangle+\sin\frac{\theta}{2}e^{-i\phi}\left|1\right\rangle)e^{i/2(\phi+\chi)}$ (7) The states $\left|\uparrow,t\right\rangle$ and $\left|\downarrow,t\right\rangle$ can be generated by the unitary transformation matrix $U(\theta,\phi,\chi)$ [11] $U(\theta,\phi,\chi)=\pmatrix{{\sin\frac{\theta}{2}e^{i/2(\phi-\chi)}~{}~{}~{}~{}~{}~{}~{}~{}~{}-\cos\frac{\theta}{2}e^{i/2(\phi+\chi)}}\cr\cos\frac{\theta}{2}e^{-i/2(\phi+\chi)}~{}~{}~{}~{}~{}~{}~{}~{}~{}{\sin\frac{\theta}{2}e^{-i/2(\phi-\chi)}}}$ (8) in association with the basic qubits $\left|0\right\rangle$ and $\left|1\right\rangle$ $\displaystyle\left|\uparrow,t\right\rangle=U(\theta,\phi,\chi)\left|0\right\rangle,\left|\downarrow,t\right\rangle=U(\theta,\phi,\chi)\left|1\right\rangle$ (9) Over a closed path, the single quantized up spinor acquires the geometrical phase [8] $\displaystyle\gamma_{\uparrow}$ $\displaystyle=$ $\displaystyle i\oint\left\langle\uparrow,t\right|\nabla\left|\uparrow,t\right\rangle.d\lambda$ (10) $\displaystyle=$ $\displaystyle i\oint\left\langle 0\right|U{\dagger}dU\left|0\right\rangle.d\lambda$ (11) $\displaystyle=$ $\displaystyle i\oint A_{\uparrow}(\lambda)d\lambda$ (12) $\displaystyle=$ $\displaystyle\oint L^{\uparrow}_{eff}dt$ (13) $\displaystyle=$ $\displaystyle\frac{1}{2}(\oint d\chi-\cos\theta\oint d\phi)$ (14) $\displaystyle=$ $\displaystyle\pi(1-\cos\theta)$ (15) representing a solid angle subtended about the quantization axis. For the conjugate state the Berry phase over the closed path becomes $\gamma_{\downarrow}=-\pi(1-\cos\theta)$ (16) The fermionic or the antifermionic nature of the two spinors (up/down) can be identified by the maximum value of topological phase $\gamma_{\uparrow/\downarrow}=\pm\pi$ at an angle $\theta=\pi/2$. For $\theta=0$ we get the minimum value of $\gamma_{\uparrow}=0$ and at $\theta=\pi$ no extra effect of phase is realized. It can be verified that this Berry phase remains the same if we neglect the overall phase $e^{\pm i(\phi-\chi)/2}$ from the quantized spinors as in eqs.(6) and (7) respectively. The identical value of Berry phase $\gamma_{\uparrow/\downarrow}$ in both the approaches is only possible if we consider no local frustration in the spin system otherwise the conflict between the parameters of quantized spinor will cause to have different BP [12]. In the language of quantum computation, the rotation of qubit or quantized spinor can be studied well in the background of geometric phase. Any electronic state at any instant can be written as a linear combination of the instantaneous eigenstates. $\left|\Psi(t)\right\rangle=c_{1}(t)\left|\Phi_{1}(t)\right\rangle+c_{2}(t)\left|\Phi_{2}(t)\right\rangle$ (17) In a cyclic change of the time period T, the instantaneous basis states $\left|\Phi_{1}(T)\right\rangle$ and $\left|\Phi_{2}(T)\right\rangle$ might return to their initial states $\left|\Phi_{1}(0)\right\rangle$ and $\left|\Phi_{2}(0)\right\rangle$ where the coefficients $c_{1}(T)$ and $c_{2}(T)$ may not. This doubly degenerate energy level, a $2\times 2$ matrix Berry phase $\Phi_{c}$ connects the final amplitudes- $c_{1}(T),c_{2}(T)$ with the initial amplitudes $c_{1}(0),c_{2}(0)$ ${c_{1}(T)\choose c_{2}(T)}=\Phi_{c}{c_{1}(0)\choose c_{2}(0)}$ (18) Hwang et.al [13] pointed out that this non abelian matrix Berry phase is responsible for pumped charges where the charge transport in a cycle of the pump in a jth optimal channel becomes $Q_{j}=-i/2\pi\oint\left\langle\psi_{j}\middle|d\psi_{j}\right\rangle$ (19) This charge will be only of topological in nature during transport of qubits, if the influence of dynamical phase can be eliminated. Spin echo method is a popular technic for this removal of dynamical phase where two cyclic evolutions are applied on a spinor with the second application followed by a pair of fast $\pi$ transformations. Vedral et.al.[14] showed the application of spin echo to a spinor (eq.(6)). $\displaystyle\left|\uparrow\right\rangle$ $\displaystyle\longrightarrow^{C_{R}}$ $\displaystyle e^{i(\delta_{\uparrow}-\gamma)}\left|\uparrow\right\rangle{\longrightarrow}^{\pi}e^{i(\delta_{\uparrow}-\gamma)}\left|\downarrow\right\rangle$ (20) $\displaystyle\longrightarrow^{C_{L}}$ $\displaystyle e^{i(\delta_{\uparrow}+\delta_{\downarrow}-2\gamma)}\left|\downarrow\right\rangle{\longrightarrow}^{\pi}e^{i(\delta_{\uparrow}+\delta_{\downarrow}-2\gamma)}\left|\uparrow\right\rangle$ Here ${\longrightarrow}^{C_{R}}$ introduces the dynamical and geometrical phases, $\delta_{\uparrow}$ and $\gamma_{\uparrow}$ through right cyclic evolution of $\left|\uparrow\right\rangle$ spinor respectively. Similar phases of opposite orientations are developed by $\longrightarrow^{C_{L}}$. Referring back to eqs.(15) and (16), we see that $\gamma_{\uparrow}$=$\gamma$ and $\gamma_{\downarrow}$=$-\gamma$ for $\gamma=\pi(1-cos\theta)$. Thus two cyclic evolutions accompanied by two $\pi$ rotations eliminate the net dynamical phases doubling the geometric phase of the original state (up/down spinor) according to. $\left|\uparrow\right\rangle\longrightarrow e^{2i\gamma_{\uparrow}}\left|\uparrow\right\rangle,\left|\downarrow\right\rangle\longrightarrow e^{2i\gamma_{\downarrow}}\left|\downarrow\right\rangle$ (21) For two half periods of spin echo rotation we have $\left|\uparrow\right\rangle\longrightarrow e^{i\gamma_{\uparrow}}\left|\uparrow\right\rangle,\left|\downarrow\right\rangle\longrightarrow e^{i\gamma_{\downarrow}}\left|\downarrow\right\rangle$ (22) where the total effect of dynamical phase disappear. The spin echo method is very fruitful [15] in the construction of two qubit through rotation of one qubit (spin 1/2) in the vicinity of another. Incorporating the spin-echo for half period (as in eqn.22) we find the antisymmetric Bell’s state after one cycle $(t=\tau)$, $\left|\Psi_{-}(t=\tau)\right\rangle=\frac{1}{\sqrt{2}}(e^{i\gamma_{\uparrow}}\left|\uparrow\right\rangle_{1}\otimes\left|\downarrow\right\rangle_{2}-e^{-i\gamma_{\uparrow}}\left|\downarrow\right\rangle_{1}\otimes\left|\uparrow\right\rangle_{2})$ (23) and symmetric state becomes $\left|\Psi_{+}(t=\tau)\right\rangle=\frac{1}{\sqrt{2}}(e^{-i\gamma_{\uparrow}}\left|\uparrow\right\rangle_{1}\otimes\left|\downarrow\right\rangle_{2}+e^{i\gamma_{\uparrow}}\left|\downarrow\right\rangle_{1}\otimes\left|\uparrow\right\rangle_{2})$ (24) where $\gamma_{\downarrow}=-\gamma_{\uparrow}=-\gamma$. Splitting up these above two eqs.(23) and (24) into the symmetric and antisymmetric states and rearranging we have $\displaystyle\left|\Psi_{+}\right\rangle_{\tau}$ $\displaystyle=$ $\displaystyle\cos\gamma\left|\Psi_{+}\right\rangle_{0}-i\sin\gamma\left|\Psi_{-}\right\rangle_{0}$ (25) $\displaystyle\left|\Psi_{-}\right\rangle_{\tau}$ $\displaystyle=$ $\displaystyle i\sin\gamma\left|\Psi_{+}\right\rangle_{0}+\cos\gamma\left|\Psi_{-}\right\rangle_{0}$ (26) the doublet acquiring the matrix Berry phase-$\Sigma$ as rotated from $t=0$ to $t=\tau$. ${\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{\tau}=\Sigma{\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{0}$ (27) $\Sigma=\pmatrix{{\cos\gamma~{}~{}~{}-i\sin\gamma}\cr{i\sin\gamma~{}~{}~{}~{}\cos\gamma}}=\cos 2\gamma$ (28) This non-abelian matrix Berry phase $\Sigma$ is developed from the abelian Berry phase $\gamma$. For $\gamma=0$ there is symmetric rotation of states, but for $\gamma=\pi$ the return is antisymmetric as the values of $\Sigma$=I and -I (where I=identity matrix) respectively. The instantaneous quantum state can be represented by the linear combination of degenerate symmetric and antisymmetric states. Symmetric state will return to antisymmetric state over one half period of spin echo apart from a matrix valued Berry phase [16]. It may be noted that two half period rotations will complete one spin echo resulting the return of the state to itself apart from a geometrical phase factor. ${\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{2\tau}=\pmatrix{{\cos\gamma~{}~{}~{}-i\sin\gamma}\cr{i\sin\gamma~{}~{}~{}~{}\cos\gamma}}\pmatrix{{\cos\gamma~{}~{}~{}i\sin\gamma}\cr{-i\sin\gamma~{}~{}~{}~{}\cos\gamma}}{\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{0}$ Following the notion of one complete spin echo here, the state $\left|\Psi_{+}\right\rangle_{T=2\tau}$ also return to its initial state ${\left|\Psi_{+}\right\rangle}_{0}$ apart from the phase $\cos 2\gamma$. ${\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{2\tau}=\pmatrix{{\cos 2\gamma~{}~{}~{}0}\cr{0~{}~{}~{}~{}\cos 2\gamma}}{\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{0}$ (29) In any even number of half period $\tau$, the symmetric state will return to itself apart from Berry phase factor with increased power of $\cos 2\gamma$. ${\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{2n\tau}=\pmatrix{{\cos 2\gamma~{}~{}~{}0}\cr{0~{}~{}~{}~{}\cos 2\gamma}}^{n}{\left|\Psi_{+}\right\rangle\choose\left|\Psi_{-}\right\rangle}_{0}$ (30) where $n=1,2,3...$ are natural integers. For odd number of the half periods rotations there will be mixture of both the states. On the other hand with the value of $\gamma=\pi$,the symmetric/antisymmetric state remains same after one rotation. In this connection we have shown recently [8] that the singlet state between two spinors at a particular instant is connected with the singlet state of elementary qubits $\left|0\right\rangle$ and $\left|1\right\rangle$ and the Berry phase of the initial antisymmetric Bell’s state is $\gamma_{ent}=\pi(1+\cos 2\theta)$ where if we introduce spin echo in the two qubit then the topological phase $\Sigma=cos2\gamma$ is of matrix valued. By varying the magnetic field angle $\theta:0\longrightarrow\pi/3\longrightarrow\pi/2$, the Berry phase(BP) of a qubit changes to, $\gamma:0\longrightarrow\pi/2\longrightarrow\pi$, that in turn change the two qubit BP, $\Sigma:I\longrightarrow\sigma^{y}\longrightarrow-I$. This explains the physics behind the change from the antisymmetric Bell singlet state $\Psi_{-}$ to the symmetric Bell state $\Psi_{+}$ and back to $\Psi_{-}$. We will now proceed to apply the above idea of entanglement in the field of Quantum Hall effect to study the state formation from one filling factor to another in the light of Geometric Quantum Computation. ## 3 Qubit formation of Quantum Hall state Quantum Hall effect shows a prominent appearance of quantization of Hall particles involving gauge theoretic extension of coordinate by $C_{\mu}\epsilon SL(2C)$ visualized by the field strength $F_{\mu\nu}$ acting as background external magnetic field. It is noted that the gauge field theoretic extension for a Fermi field associated with the direction vector $\xi_{\mu}$ attached to the space-time point $x_{\mu}$ results the field function $\phi(x_{\mu},\xi_{\mu})$ describing a particle moving in an anisotropic space [7]. The external magnetic field introduces frustration in the Hall system. We have considered a two-dimensional frustrated electron gas of N particles on the spherical surface of a three dimensional sphere of large radius R in a strong radial (monopole) magnetic field. In such a 3D anisotropic space we can construct the N-particle wave-function from the spherical harmonics ${Y_{l}}^{m,\mu}$ with $l=1/2$, $|m|=|\mu|=1/2$ (when the angular momentum in the anisotropic space is given by eq.(3)). With the description of a two component up spinor $\left|\uparrow\right\rangle={u\choose v}$ as in eq.(6) we can construct the $N$ particles wave function of Hall states ${\Psi_{N_{\uparrow}}}^{(m)}=\prod(u_{i}v_{j}-u_{j}v_{i})^{m}$ (31) for parent states $m=1/\nu$ where $\nu$ is the Landau filling factor and this $m=J_{ij}=J_{i}+J_{j}$ is the two particle angular momentum equivalent to $m=\mu_{i}+\mu_{j}=2\mu$ (when $i=j$). Similar manner the same Hall state with opposite polarization can be constructed by using the down spinor $\left|\downarrow\right\rangle={\tilde{v}\choose\tilde{u}}$ ${\Psi_{N_{\downarrow}}}^{(m)}=\prod(\tilde{u}_{i}\tilde{v}_{j}-\tilde{u}_{j}\tilde{v}_{i})^{m}$ (32) Here the two states ${\Psi_{N_{\uparrow}}}^{(m)}$ and ${\Psi_{N_{\downarrow}}}^{(m)}$ belong to the same parent filling factor but with opposite polarization of the spinors. The above states are grouped into a family depending on the value of $m$. With $m=3$ the states are the same family of the Laughlin $\nu=1/3$ state etc. In the light of Jain [17] that regarding the filling factor the IQHE of composite fermions are the FQHE of fermions, any FQHE state can be expressed in terms of the IQHE state. It seems that for LLL $\nu=1$, IQHE state $\Phi_{1}(z)$ $\Phi_{1}(z)=(u_{i}v_{j}-u_{j}v_{i})$ (33) is the basic building block for constructing any other IQHE/FQHE state. The lowest level Hall state $\Phi_{1}(z)$ has a similarity with two-qubit singlet state formed by a pair of one qubit states. There is a deep analogy between FQHE and superfluidity [18]. The ground state of anti-ferromagnetic Heisenberg model on a lattice introduce frustration giving rise to the resonating valence bond(RVB) states corresponding spin singlets where two nearest-neighbor bonds are allowed to resonate among themselves. It is suggested that RVB states [6] is a basis of fault tolerant topological quantum computation. Since these spin singlet states forming a RVB gas is equivalent to fractional quantum Hall fluid, its description through quantum computation will be of ample interest. This resonating valence bond(RVB) where two nearest-neighbour bonds are allowed to resonate among themselves has equivalence with entangled state of two one-qubit. The antisymmetric Hall state $\Phi_{1}(z)$ for $\nu=1$ is formed as one spinor at ith site rotating with Berry phase $\gamma=\pm i\pi$ in the vicinity of another at jth site captures the image of spin echo $\displaystyle\left|\Phi_{1}(z)\right\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{2}}(\left|\uparrow\right\rangle_{1}~{}~{}\left|\downarrow\right\rangle_{1})\pmatrix{{0~{}~{}~{}-e^{-i\pi}}\cr{e^{i\pi}~{}~{}~{}0}}{\left|\uparrow\right\rangle_{2}\choose\left|\downarrow\right\rangle_{2}}$ (34) $\displaystyle=$ $\displaystyle(\left|\uparrow\right\rangle_{1}~{}~{}~{}\left|\downarrow\right\rangle_{1})\pmatrix{{0~{}~{}~{}1}\cr{-1~{}~{}~{}0}}{\left|\uparrow\right\rangle_{2}\choose\left|\downarrow\right\rangle_{2}}$ (35) $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{2}}(\left|\uparrow\right\rangle_{1}~{}~{}\left|\downarrow\right\rangle_{2}-\left|\downarrow\right\rangle_{1}~{}~{}\left|\uparrow\right\rangle_{2})$ (36) $\displaystyle=$ $\displaystyle\left|\Psi_{-}\right\rangle$ (37) Due to symmetry, the singlet state can be written on any basis with the same form. We can rotate the spin vector by an arbitrary angle $\theta$ with the following transformation. ${\left|\uparrow\right\rangle\choose\left|\downarrow\right\rangle}=\pmatrix{{\sin\theta e^{i\phi}~{}~{}~{}\cos\theta}\cr{-\cos\theta~{}~{}~{}~{}\sin\theta e^{-i\phi}}}{\left|0\right\rangle\choose\left|1\right\rangle}$ (38) The Quantum Hall systems are so highly frustrated that the ground state $\Phi_{1}(z)$ is an extremely entangled state visualized by the formation of antisymmetric singlet state between a pair of $i,j$th spinors in the Landau filling factor $(\nu=1)$. $\displaystyle\Phi_{1}(z)$ $\displaystyle=$ $\displaystyle\pmatrix{{u_{i}~{}~{}~{}~{}~{}u_{j}}\cr{v_{i}~{}~{}~{}~{}~{}v_{j}}}=(u_{i}v_{j}-u_{j}v_{i})$ (39) $\displaystyle=$ $\displaystyle(u_{i}~{}~{}~{}~{}v_{i})\pmatrix{{0~{}~{}~{}~{}~{}1}\cr{-1~{}~{}~{}~{}~{}0}}{u_{j}\choose v_{j}}$ We identify this two qubit singlet state as Hall qubit constructed from the up-spinor shown in the previous section $\Phi_{1}(z)=\left\langle\uparrow_{i}\right|\pmatrix{{0~{}~{}~{}~{}~{}1}\cr{-1~{}~{}~{}~{}~{}0}}\left|\uparrow_{j}\right\rangle=\left\langle 0\right|U_{i}{\dagger}\pmatrix{{0~{}~{}~{}~{}~{}1}\cr{-1~{}~{}~{}~{}~{}0}}U_{j}\left|0\right\rangle$ (40) The down spinor can construct the opposite polarization of Hall qubit $\Phi_{1}(\tilde{z})=(\tilde{u}_{i}\tilde{v}_{j}-\tilde{u}_{j}\tilde{v}_{i})$ (41) that has a similar representation as eq.(41) $\Phi_{1}(\tilde{z})=\left\langle 1\right|U{\dagger}\pmatrix{{0~{}~{}~{}~{}~{}1}\cr{-1~{}~{}~{}~{}~{}0}}U\left|1\right\rangle$ (42) Now these two Hall qubits of two opposite polarizations representing the state of same lowest Landau level $\nu=m=1$ will automatically generate two respective non-abelian Berry connections. The Hall connection for up spinor becomes $B_{\uparrow}=\Phi_{1}(z)^{*}d\Phi_{1}(z)=\pmatrix{{{u_{i}}^{*}~{}~{}~{}~{}{v_{i}}^{*}}\cr{{u_{j}}^{*}~{}~{}~{}~{}{v_{j}}^{*}}}\pmatrix{{du_{i}~{}~{}~{}~{}du_{j}}\cr{dv_{i}~{}~{}~{}~{}dv_{j}}}$ (43) and similarly for down-spinor $\tilde{B_{\downarrow}}=\Phi_{1}(\tilde{z})^{*}d\Phi_{1}(\tilde{z})=\pmatrix{{\tilde{u_{i}}^{*}~{}~{}~{}~{}\tilde{v_{i}}^{*}}\cr{\tilde{u_{j}}^{*}~{}~{}~{}~{}\tilde{v_{j}}^{*}}}\pmatrix{{d\tilde{u_{i}}~{}~{}~{}~{}d\tilde{u_{j}}}\cr{d\tilde{v_{i}}~{}~{}~{}~{}d\tilde{v_{j}}}}$ (44) The non-abelian nature of the connection or Berry phase on the Hall surface for the lowest Landau level LLL ($\nu=1$) will remain if $i\neq j$. $\displaystyle B_{\uparrow}$ $\displaystyle=$ $\displaystyle\pmatrix{{(u_{i}^{*}du_{i}+v_{i}^{*}dv_{i})~{}~{}~{}~{}(u_{i}^{*}du_{j}+v_{i}^{*}dv_{j})}\cr{(u_{j}^{*}du_{i}+v_{j}^{*}dv_{i})~{}~{}~{}~{}(u_{i}^{*}du_{i}+v_{i}^{*}dv_{i})}}$ (45) $\displaystyle=$ $\displaystyle\pmatrix{{\mu_{i}~{}~{}~{}\mu_{ij}}\cr{\mu_{ji}~{}~{}~{}~{}\mu_{j}}}$ This is visualizing the spin conflict during parallel transport leading to matrix Berry phase. In the light of Hwang et.al [13] our realization includes that in Quantum Hall effect this non-abelian matrix Berry phase is responsible for the charge flow by pumping. In this QHE matrix Berry phase ${\gamma^{H}}_{\uparrow}=\pmatrix{{\gamma_{i}~{}~{}~{}~{}\gamma_{ij}}\cr{\gamma_{ji}~{}~{}~{}~{}\gamma_{j}}}$ (46) $\gamma_{i}$ and $\gamma_{j}$ are the BPs for the ith and jth spinor as seen in eq. (15) and the off-diagonal BP $\gamma_{ij}$ arises due to local frustration in the spin system. Over a closed period $t=\tau$ the QHE state $\Phi_{1}(z)$ at $\nu=1$ filling factor will acquire the matrix Berry phase. $\left\langle\Phi_{1}(z)\right|_{\tau}=e^{i{\gamma^{H}}_{\uparrow}}\left\langle\Phi_{1}(z)\right|_{0}$ (47) Berry connection gets modified as the quantum state differ after one rotation. Usually when any state changes by $\left|\psi^{\prime}\right\rangle=\left|\psi\right\rangle e^{i\Omega(c)}$ the corresponding changed gauge becomes ${A_{\psi}}^{\prime}=A_{\psi}+id\Omega(c)$ provided $\left\langle\psi\middle|\psi\right\rangle=1$. We have pointed out earlier [19] that each Quantum Hall state for a particular filling factor has its distinct Berry phase. Hence BP is constant for a filling factor. The rotation shifts the BP from ground to excited level once. With these ideas we have the topological phase difference between the first excited and the ground state acquired by the rate of change of Berry phase $\Gamma^{1}-\Gamma^{0}=i\oint{\left\langle\Phi_{1}(z)\right|d\gamma^{H}/d\lambda\left|\Phi_{1}(z)\right\rangle}_{0}d\lambda$ (48) The rotation of singlet state by ’n’ number of turns will be ${\left\langle\Phi_{1}(z)\right|^{n}}_{\tau}=e^{i{n\gamma^{H}}_{\uparrow}}{\left\langle\Phi_{1}(z)\right|^{n}}_{0}$ (49) where $n=1,2,3..$ are the natural numbers associated with the number of rotations of the singlet states. We should point out here that the antisymmetric nature of FQHE states would be visualized through the rotation of singlet states. This automatically imposes the following constraint in the topological phase $e^{{in\gamma}^{H}}=e^{im\pi}=-1,$ (50) $for~{}~{}~{}~{}\left\langle\Phi_{1}(z)\right|_{n\tau}=-\left\langle\Phi_{1}(z)\right|_{0}$ where $m=1,3,5..$ being the odd numbers to maintain the antisymmetric nature of wave function. So any number of rotations of the matrix Berry phase lead to odd multiple of $\pi$ angles provided the every state remains antisymmetric. It seems that BP act as a local order parameter of QHE states. ${\left\langle\Phi_{1}(z)\right|}^{m\pi/\gamma}_{\tau}=e^{im\pi}{\left\langle\Phi_{1}(z)\right|}^{m\pi/\gamma}_{0}$ (51) Earlier we showed [7] that the Berry phase for $\nu=1/m$ state is $\gamma==m\pi\theta=2\pi\mu\theta$ where $\theta$ is a coupling constant.This motivated us to write ${\left\langle\Phi_{1}(z)\right|}_{\tau}=e^{im\pi\theta}{\left\langle\Phi_{1}(z)\right|}_{0}$ (52) This makes the experimental observation of parent state in FQHE at $m=odd(3,5,7)$ more transparent. It also shows that the topological phase is responsible for controlling the statistics of the Hall state. In absence of frustration, the role of matrix Berry phase is trivial. In other words $\gamma_{ij}$ becomes zero leading to diagonal matrix Berry phase provided the two particle have identical $\theta$ and $\phi$ values. ${\gamma^{H}}_{\uparrow}=\pi(1-\cos\theta_{i})\pmatrix{{1~{}~{}~{}~{}0}\cr{0~{}~{}~{}~{}1}}$ (53) The non-abelian matrix Berry phase in Quantum Hall effect is originated due to the frustration offered by the magnetic field and the disorder of spins. In the absence of local frustration (latter) this complexity of connection will be removed. We would like to mention that spin echo between two single qubit has the equivalence of RVB state in FQHE and topological quantum computation with BP is responsible for the formation of higher states considering the Hall qubit at $\nu=1$ as a building block of any QHE state. ## 4 Discussion In this paper we have studied the Physics behind the singlet state entangled by the two qubits where one is rotating in the field of the other with the Berry phase only. This image of spin echo has been reflected in the field of Quantum Hall effect. The Hall state for the lowest Landau level at $\nu=1$ is highly frustrated. They are the singlet states identified as the Hall qubit, the building block of other higher IQHE/FQHE states at different filling factors. These states have matrix Berry phase which are responsible for pumped charge flow. In other words the Berry phase acts as a local order parameter of singlet states. Further we pointed out that the antisymmetric nature of $\nu=1/m$ FQHE states depend on their acquired Berry phase. Since these spin singlet states forming a RVB gas is equivalent to fractional quantum Hall fluid, the description of background Physics through quantum computation will be of ample interest. We will proceed to study the hierarchies of FQHE in the light of quantum communication in the future. Acknowledgements: This work is partly supported by The Abdus Salam International Center for Theoretical Physics, Trieste, Italy. The author would like to acknowledge the help from all the works cited in the reference. Moreover the author is thankful to the referees and Editor of my paper in PRB for fruitful comments. ## References * [1] J.Bell, Physics 1, 95(1964); Rev.Mod.Phys.38,447 (1966). * [2] M.A.Nielse and I.L.Chuang, Quantum Computation and Quantum Information, Cambridge University Press, Cambridge (2000). * [3] J.Preskill;Lecture Notes for Physics 219: Quantum Computation (2004). * [4] M.V.Berry, Proc.R.Soc.London A392,45 (1984). * [5] V.Vedral, Cent. Eur. J. Phys.2, 289 (2003);quant-phys/0505029. V.Vedral,quantum-ph/0212133. * [6] A.Y.Kitaev; Ann. Phys. 303,2 (2003). * [7] D.Banerjee; Phys. Rev.-B58,4666 (1998). * [8] D.Banerjee and P.Bandyopadhyay, Physica Scripta, 73,571(2006), ICTP preprint. * [9] P.Bandyopadhay and K.Hajra, J. math. Phys.28, 711 (1987). * [10] P.Bandyopadhyay; Int.J.Mod.Phys.A4, 4449 (1989). * [11] D.Banerjee and P.Bandyopadhyay;Nuovo Cimneto,113(1998)921; D.Banerjee; Fort.der Physik 44 (1996) 323 * [12] D.Banerje;”The Berry phase in frustrated spin glass”,ICTP preprint and communicated to ”Euro Physics Journals-B”. * [13] N.Y.Hwang,S.C.Kim, P.S.Park and S.R.Eric Yang;arXiv; cond-mat/0706.0947. * [14] A.Ekert, M. Ericsson, P.Hayden, H. Inamori, J. A. Jones, D.K.L.Oi and V. Vedral,arXiv: quantum-ph/0004015. * [15] R.A.Bertlmann, K. Durstberger, Y. Hasegawa and B. C. Heismayr; Phys. Rev.-A69, 032112 (2004). * [16] B.Basu ; arXiv: quant-ph/0602089. * [17] J.K. Jain and R.K.Kamilla in ”Composite Fermions” edited by Olle Heironen(World Scientific, New York, 1998.) * [18] V.Kalmeyer and R.B.Laughlin; Phys.Rev.Lett. 59, 2095 (1987). * [19] D.Banerjee and P.Bandyopadhyay; Mod.Phys.Lett. B8, 1643, (1994).

arxiv-papers

2024-09-04T02:48:54.555777

0803.3787

# New Proof of the Equation $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}=0.$ Edmund Landau Translated by Michael J. Coons Preliminary Version: September 2007 [Neuer Beweis der Gleichung $\sum\mu(k)/k=0$, Inaugural-Dissertation, Berlin, 1899.] Translation © M. J. Coons 2007. New Proof of the Equation $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}=0.$ Inaugural–Dissertation for the acquisition of the title of Doctor from the Faculty of Philosophy of Friedrich–Wilhelms University in Berlin defended publicly and approved together with the attached theses on 15 July, 1899 by Edmund Landau of Berlin Opponents: Mr. Rudolf Zeigel, Student of Mathematics Mr. Fritz Hartoge, Student of Mathematics Ernst Steinitz, Ph.D. , Privatdozent at the Royal Technical high school in Charlottenburg. Berlin 1899. For my dear parents. The function $\mu(k)$ is normally defined as the number–theoretic function for which 1. 1. $\mu(1)=1$, 2. 2. $\mu(k)=0$ when $k>1$ is divisible by a square, 3. 3. $\mu(k)=(-1)^{r}$ when $k$ is the product of $r$ distinct primes. This statement was first expressed by Euler, that $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}=0$ holds; that is, $\lim_{x\to\infty}\sum_{k=1}^{x}\frac{\mu(k)}{k}$ exists and equals $0$, the recent proof of which is due to von Mangoldt111) “Beweis der Gleichung $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}=0$”; Proceedings of the Royal Prussian Academy of Science of Berlin, 1897, pp. 835–852.). The same goes for the investigations of Hadamard and de la Vallée Poussin over the Riemann $\zeta$–function, and it seems also, that without the use of these works, the present means of analysis is not enough to give a proof of Euler’s statement. However, if one expects the results of those investigations to be in agreement with those of von Mangoldt, then one, as will be executed in the following, can arrive at the target along a quite short path. The proof, which forms the contents of this dissertation, uses first the theorem222) This theorem is proven without the use of von Mangoldt’s proof.) of Hadamard333) Bulletin de la société mathématique de France, Volume 24, 1896, p. 217.) and de la Vallée Poussin444) Annales de la société scientifique de Bruxelles, Volume 20, Part 2, p. 251.): “If $\vartheta(x):=\sum_{p\leq x}\log p$, then $\lim_{x\to\infty}\frac{\vartheta(x)}{x}=1.$” However, apart from the use of this theorem, it is as elementary as can be for such a transcendent statement. ## 1 Denote by $g(x)$ the sum $\sum_{k=1}^{[x]}\frac{\mu(k)}{k}$, where $[x]$ denotes the greatest integer less than or equal to $x$; more simply we write $g(x)=\sum_{k=1}^{x}\frac{\mu(k)}{k},$ (1) where $k$ ranges over all positive integers less than or equal to $x$. The sum has meaning only for $x\geq 1$; thus, for $x<1$, set $g(x)=0$. With the above notation, we read the two lemmas, which von Mangoldt proves in a simple way at the start of his paper555) 1. c., pp. 837–839.) and which are also applied in the following one, as: For all $x$ $|g(x)|\leq 1\ )$ (2) and for all $x\geq 1$ $\left|\log x\cdot g(x)-\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|\leq 3+\gamma$ (3) where $\gamma$ denotes Euler’s constant. The inequality (3), which von Mangoldt only derived in order to apply it in a certain place in his proof777) 1. c., p. 843.), serves in the following one as the basis of the whole investigation. Concerning the sum $\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}$, Möbuis888) “Über eine besondere Art von Umkehrung Reihen,” Journal für die reine und angewandte Mathematik, Volume 9, p. 122.) believed he had proved that for sufficiently large $x$, its difference from $-1$ is arbitrarily small: however, his proof is not sound. Although new writers consider it probable999) E.g., Mertens proved, which the general validity of the inequality condition assumes: $\left|\sum_{k=1}^{x}\mu(k)\right|\leq\sqrt{x},$ that this theorem is correct, if this relation is generally fulfilled (Proceedings of the Vienna Academy, math.-nat. Kl., Volume 106, Dept. 2a, p. 774.)) that $\lim_{x=\infty}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}$ exists and equals $-1$, it has yet to be proven that for all $x$, $\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}$ is contained between two finite boundaries. Now since (3) yields $\left|g(x)-\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|\leq\frac{3+\gamma}{\log x},$ it follows, with use of the Euler–v. Mangoldt Theorem, that $\lim_{x=\infty}g(x)=0$ so that $\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}$ approaches $0$ as $x\to\infty$. If, in reverse, it was successfully proven that $\lim_{x=\infty}\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}$ exists and equals 0, then one would trivially have that $\sum_{k=1}^{x}\frac{\mu(k)}{k}=0,$ since for any $\delta$ there is a $G$ such that for all $x\geq G$ $\left|\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|\leq\frac{\delta}{2}$ and $0<\frac{3+\gamma}{\log x}\leq\frac{\delta}{2},$ thus it follows for $x\geq G$: $\displaystyle|g(x)|$ $\displaystyle=\left|\left(g(x)-\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right)+\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|$ $\displaystyle\leq\left|\left(g(x)-\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right)\right|+\left|\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|$ $\displaystyle\leq\frac{3+\gamma}{\log x}+\left|\frac{1}{\log x}\sum_{k=1}^{x}\frac{\mu(k)\log k}{k}\right|$ $\displaystyle\leq\frac{\delta}{2}+\frac{\delta}{2}=\delta,$ also $\lim_{x=\infty}g(x)=0.$ The proof, that for $f(x)=\sum_{k=1}^{x}\frac{\mu(k)\log k}{k},$ (4) one has $\lim_{x=\infty}\frac{f(x)}{\log x}=0$ (5) will be supplied in what follows. In order not to have to interrupt the course of the investigation, we note the following simple lemma, which was already known to Gram101010) 1. c., p. 197, where separately for all valid $r$ in equation (43) set $r=1$.): it is $\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)=g(x)+\frac{1}{2}g\left(\frac{x}{2}\right)+\frac{1}{3}g\left(\frac{x}{3}\right)+\cdots+\frac{1}{[x]}g\left(\frac{x}{[x]}\right)=1.$ (6) The $\nu$–th summand $\frac{1}{\nu}g\left(\frac{x}{\nu}\right)$ contains the sum of the terms $\frac{1}{\nu}\frac{\mu(1)}{1}=\frac{\mu(1)}{\nu},\ \frac{1}{\nu}\frac{\mu(2)}{2}=\frac{\mu(2)}{2\nu},\cdots,\ \frac{1}{\nu}\frac{\mu(n)}{n}=\frac{\mu(n)}{n\nu},\cdots,\ \frac{1}{\nu}\frac{\mu\left[\frac{x}{\nu}\right]}{\left[\frac{x}{\nu}\right]}=\frac{\mu\left[\frac{x}{\nu}\right]}{\left[\frac{x}{\nu}\right]\nu};$ the sum $\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)$ consists of terms of the form $\frac{\mu(n)}{t}$, where $n$ is a divisor of $t$, and $t$ runs through the integers from 1 to $[x]$; that is, $\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)=\sum_{t=1}^{x}\frac{1}{t}\sum_{n|t}\mu(n);$ now since $\sum_{n|t}\mu(n)$ is 1 for $t=1$ and $0$ otherwise, we have $\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)=1.$ ## 2 If one lets $x=p_{1}p_{2}\cdots p_{r}$ in the defining equation (4) of $f(x)$, replaces111111) The $\log k$ is multiplied by the factor $\frac{\mu(k)}{k}$, which occurs only for $k$ that are the product of distinct primes.) $\log(x)$ by $\log p_{1}+\cdots+\log p_{r}$ and gathers like terms in which the logarithm is applied to the same prime number, then (4) becomes an equation of the form $f(x)=\sum F(p,x)\log p$ where the sum extends over all prime numbers $p\leq x$. As was easily given by von Mangoldt121212) 1. c., p. 840.) for another purpose, $F(p,x)=-\left(\frac{1}{p}\sum_{k=1}^{\frac{x}{p}}\frac{\mu(k)}{k}+\frac{1}{p^{2}}\sum_{k=1}^{\frac{x}{p^{2}}}\frac{\mu(k)}{k}+\frac{1}{p^{3}}\sum_{k=1}^{\frac{x}{p^{3}}}\frac{\mu(k)}{k}+\cdots\right),$ a series which has only a finite number of non-zero summands, since the summation index of the $i$–th sum runs from 1 to $\left[\frac{x}{p^{i}}\right]$, so that $p^{i}\leq x$ gives $i\leq\frac{\log x}{\log p}$. Therefore $f(x)=-\sum_{p\leq x}\log p\left(\frac{1}{p}g\left(\frac{x}{p}\right)+\frac{1}{p^{2}}g\left(\frac{x}{p^{2}}\right)+\frac{1}{p^{3}}g\left(\frac{x}{p^{3}}\right)+\cdots\right),$ $f(x)=-\sum_{p\leq x}\frac{\log p}{p}g\left(\frac{x}{p}\right)-\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}g\left(\frac{x}{p^{2}}\right)+\frac{1}{p^{3}}g\left(\frac{x}{p^{3}}\right)+\cdots\right).$ (7) According to (2), for all $y$ $|g(y)|\leq 1,$ so that the absolute value of the second sum in (7) is $\displaystyle\left|\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}g\left(\frac{x}{p^{2}}\right)+\frac{1}{p^{3}}g\left(\frac{x}{p^{3}}\right)+\cdots\right)\right|$ $\displaystyle\leq\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}\left|g\left(\frac{x}{p^{2}}\right)\right|+\frac{1}{p^{3}}\left|g\left(\frac{x}{p^{3}}\right)\right|+\cdots\right)$ $\displaystyle\leq\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}+\frac{1}{p^{3}}+\frac{1}{p^{4}}+\cdots+\frac{1}{p^{n}}+\cdots\right)$ $\displaystyle\leq\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}+\frac{1}{2p^{2}}+\frac{1}{2^{2}p^{2}}+\cdots+\frac{1}{2^{n-2}p^{2}}+\cdots\right)$ $\displaystyle\leq\sum_{p\leq x}\frac{\log p}{p^{2}}\left(1+\frac{1}{2}+\frac{1}{4}+\frac{1}{8}+\cdots\right)$ $\displaystyle\leq 2\sum_{p\leq x}\frac{\log p}{p^{2}}$ $\displaystyle<2\sum_{\nu=1}^{\infty}\frac{\log\nu}{\nu^{2}}.$ It is well known that $\sum_{\nu=1}^{\infty}\frac{\log\nu}{\nu^{2}}$ is convergent; thus as $x\to\infty$ the sum on the right-hand side of (7) approaches either a certain bound, or its value oscillates between two finite uncertain bounds. In either case, the quotient with $\log x$ approaches $0$ as $x\to\infty$. Denote by $h(x)$, the function defined by $h(x)=\sum_{p\leq x}\frac{\log p}{p}g\left(\frac{x}{p}\right).$ (8) If $\lim_{x=\infty}\frac{h(x)}{\log x}$ exists and equals $0$, then according to (7) $\frac{f(x)+h(x)}{\log x}=\frac{-1}{\log x}\sum_{p\leq x}\log p\left(\frac{1}{p^{2}}g\left(\frac{x}{p^{2}}\right)+\frac{1}{p^{3}}g\left(\frac{x}{p^{3}}\right)+\cdots\right),$ and as we just saw, as $x$ gets large the right-hand side approaches $0$, from which the correctness of statement (5) follows. The proof of Euler’s statement, that $\sum_{k=1}^{x}\frac{\mu(k)}{k}=0,$ thus depends on the proof of the statement $\lim_{x=\infty}\frac{h(x)}{\log x}=0,$ which will be furnished in the following section. ## 3 Recall that the function $\vartheta(x)$ 131313) See the theorem of Hadamard and de la Vallée Poussin in the introduction.) is defined for all positive $\nu$ by $\vartheta(\nu)-\vartheta(\nu-1)=\begin{cases}\log\nu&\mbox{ if $\nu$ is prime,}\\\ 0&\mbox{ if $\nu$ is composite or 1},\\\ \log\nu=0&\mbox{ if $\nu=1$.}\end{cases}$ And $h(x)=\sum_{\nu=1}^{x}\frac{\vartheta(\nu)-\vartheta(\nu-1)}{\nu}g\left(\frac{x}{\nu}\right),$ where the sum ranges over the integers between $1$ and $x$. In the place of the function $\vartheta(x)$, use $\vartheta(x)=x\\{1+\varepsilon(x)\\},$ (9) where the function $\varepsilon(x)$ takes only non-negative values of $x$, and $\varepsilon(0)=0$. We note the following properties of $\varepsilon(x)$: 1. 1. Since by definition, $\vartheta(x)$ is never negative, then always $\varepsilon(x)\geq-1.$ 2. 2. As shown by Mertens141414) “Ein Beitrag zur analytischen Zahlentheorie,” Journal für die reine und angewandte Mathematik, Volume 78, p. 48.), for all $x$ $\vartheta(x)<2x,$ so that always $\varepsilon(x)<1;$ therefore, we gain the inequality $|\varepsilon(x)|\leq 1.$ (10) 3. 3. The theorem cited in the introduction151515) See the theorem of Hadamard and de la Vallée Poussin in the introduction.), that $\lim_{x=\infty}\frac{\vartheta(x)}{x}=1,$ gives $\lim_{x=\infty}\varepsilon(x)=0.$ (11) The introduction of the function $\varepsilon(x)$ yields, for $h(x)$, $\displaystyle h(x)$ $\displaystyle=\sum_{\nu=1}^{x}\frac{\nu+\nu\varepsilon(\nu)-(\nu-1)-(\nu-1)\varepsilon(\nu-1)}{\nu}g\left(\frac{x}{\nu}\right)$ $\displaystyle=\sum_{\nu=1}^{x}\left\\{\frac{1}{\nu}g\left(\frac{x}{\nu}\right)+\left(\varepsilon(\nu)-\frac{\nu-1}{v}\varepsilon(\nu-1)\right)g\left(\frac{x}{\nu}\right)\right\\}$ $\displaystyle=\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)+\sum_{\nu=1}^{x}\left(\varepsilon(\nu)-\varepsilon(\nu-1)+\frac{1}{\nu}\varepsilon(\nu-1)\right)g\left(\frac{x}{\nu}\right).$ Using Eqs. (8) and (9), $\sum_{\nu=1}^{x}\frac{1}{\nu}g\left(\frac{x}{\nu}\right)=1;$ yielding $h(x)-1=\sum_{\nu=1}^{x}\left(\varepsilon(\nu)-\varepsilon(\nu-1)\right)g\left(\frac{x}{\nu}\right)+\sum_{\nu=1}^{x}\frac{1}{\nu}\varepsilon(\nu-1)g\left(\frac{x}{\nu}\right).$ (12) For the first of the two sums in (12) we get $\displaystyle\sum_{\nu=1}^{x}(\varepsilon(\nu)$ $\displaystyle-\varepsilon(\nu-1))g\left(\frac{x}{\nu}\right)$ $\displaystyle=\sum_{\nu=1}^{x}\varepsilon(x)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)+\varepsilon([x])g\left(\frac{x}{[x]+1}\right)$ $\displaystyle=\sum_{\nu=1}^{x}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right),$ where $x<[x]+1$ so that $g\left(\frac{x}{[x]+1}\right)=0$. If in the second sum in (12) we write $\nu+1$ in place of $\nu$, we have $\sum_{\nu=0}^{x-1}\frac{1}{\nu+1}\varepsilon(\nu)g\left(\frac{x}{\nu+1}\right)=\sum_{\nu=1}^{x-1}\frac{1}{\nu+1}\varepsilon(\nu)g\left(\frac{x}{\nu+1}\right)$ and so $h(x)-1=\sum_{\nu=1}^{x}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)+\sum_{\nu=1}^{x-1}\frac{1}{\nu+1}\varepsilon(\nu)g\left(\frac{x}{\nu+1}\right).$ (13) Let $\delta$ be an arbitrary small positive quantity. Then by (11), there is a $G$ such that for all $\nu\geq G$ $|\varepsilon(\nu)|\leq\frac{\delta}{3}.$ (14) This yields $\left|\sum_{\nu=1}^{x}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq\left|\sum_{\nu=1}^{G-1}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ $+\left|\sum_{\nu=G}^{x}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ $\leq\sum_{\nu=1}^{G-1}|\varepsilon(\nu)|\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|+\sum_{\nu=G}^{x}|\varepsilon(\nu)|\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|.$ As the right-hand side is increased, $|\varepsilon(\nu)|$ goes to 1 in the first sum (by (10)), and goes to $\delta/3$ in the second sum (by (14)), yielding $\left|\sum_{\nu=1}^{x}\varepsilon(\nu)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq\sum_{\nu=1}^{G-1}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ (15) $+\frac{\delta}{3}\sum_{\nu=G}^{x}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|.$ Now $\displaystyle\sum_{\nu=1}^{G-1}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ $\displaystyle\leq\sum_{\nu=1}^{G-1}\left\\{\left|\left(g\left(\frac{x}{\nu}\right)\right|-\left|g\left(\frac{x}{\nu+1}\right)\right)\right|\right\\}$ $\displaystyle=\sum_{\nu=1}^{G-1}(1+1)\ \mbox{ (by (2))},$ so that $\sum_{\nu=1}^{G-1}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq 2(G-1)$ (16) and $\displaystyle\sum_{\nu=G}^{x}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ $\displaystyle\leq\sum_{\nu=1}^{x}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|$ $\displaystyle=\sum_{\nu=1}^{x}\left|\sum_{k=1}^{\frac{x}{\nu}}\frac{\mu(k)}{k}-\sum_{k=1}^{\frac{x}{\nu+1}}\frac{\mu(k)}{k}\right|$ $\displaystyle=\sum_{\nu=1}^{x}\left|\sum\frac{\mu(k)}{k}\right|,$ where $k$ ranges over all integers in the interval $\left(\frac{x}{\nu+1},\frac{x}{\nu}\right]$. Therefore $\sum_{\nu=G}^{x}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq\sum_{\nu=1}^{x}\sum_{\frac{x}{\nu}\geq k>\frac{x}{\nu+1}}\frac{|\mu(k)|}{k}\leq\sum_{\nu=1}^{x}\sum_{\frac{x}{\nu}\geq k>\frac{x}{\nu+1}}\frac{1}{k}$ $=\sum_{x\geq k>\frac{x}{2}}\frac{1}{k}+\sum_{\frac{x}{2}\geq k>\frac{x}{3}}\frac{1}{k}+\sum_{\frac{x}{3}\geq k>\frac{x}{4}}\frac{1}{k}+\cdots+\sum_{\frac{x}{[x]-1}\geq k>\frac{x}{[x]}}\frac{1}{k}$ $+\sum_{\frac{x}{[x]}\geq k>1}\frac{1}{k}=\sum_{k=1}^{x}\frac{1}{k},$ and since always $\sum_{k=1}^{x}\frac{1}{k}\leq\log x+1,$ we have $\sum_{\nu=G}^{x}\left|\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq\log x+1.$ (17) Replacing both sums of the right-hand side of inequality (15) by the results gained in (16) and (17) yields $\left|\sum_{\nu=1}^{x}\varepsilon(x)\left(g\left(\frac{x}{\nu}\right)-g\left(\frac{x}{\nu+1}\right)\right)\right|\leq 2(G-1)+\frac{\delta}{3}(\log x+1).$ (18) The handling of the second sum in (13) is somewhat simpler. We have $\displaystyle\left|\sum_{\nu=1}^{x-1}\frac{1}{\nu+1}\varepsilon(\nu)g\left(\frac{x}{\nu+1}\right)\right|$ $\displaystyle\leq\sum_{\nu=1}^{x-1}\frac{1}{\nu+1}|\varepsilon(\nu)|\left|g\left(\frac{x}{\nu+1}\right)\right|$ $\displaystyle\leq\sum_{\nu=1}^{x-1}\frac{|\varepsilon(\nu)|}{\nu+1}$ $\displaystyle=\sum_{\nu=1}^{G-1}\frac{|\varepsilon(\nu)|}{\nu+1}+\sum_{\nu=G}^{x-1}\frac{|\varepsilon(\nu)|}{\nu+1}$ $\displaystyle\leq\sum_{\nu=1}^{G-1}\frac{1}{\nu+1}+\sum_{\nu=G}^{x-1}\frac{1}{\nu+1}\frac{\delta}{3}$ $\displaystyle\leq\sum_{\nu=1}^{G-1}1+\frac{\delta}{3}\sum_{\nu=1}^{x-1}\frac{1}{\nu},$ so that $\left|\sum_{\nu=1}^{x-1}\frac{1}{\nu+1}\varepsilon(\nu)g\left(\frac{x}{\nu+1}\right)\right|\leq G-1+\frac{\delta}{3}(\log x+1).$ (19) With help from the inequalities (18) and (19), (13) becomes $|h(x)|\leq 1+2G-2+\frac{\delta}{3}\log x+\frac{\delta}{3}+G-1+\frac{\delta}{3}\log x+\frac{\delta}{3}$ $=3G-2+\frac{2}{3}\delta+\frac{2}{3}\delta\log x,$ thus for all $x\geq e^{\frac{3G-2+\frac{2}{3}\delta}{\frac{1}{2}\delta}},$ we have $3G-2+\frac{2}{3}\delta\leq\frac{1}{3}\delta\log x,$ so that $|h(x)|\leq\frac{1}{3}\delta\log x+\frac{2}{3}\delta\log x=\delta\log x,$ which yields $\left|\frac{h(x)}{\log x}\right|\leq\delta.$ (20) For such a $\delta$ there is a $\xi$ assignable, such that for all $x\geq\xi,$ (20) is fulfilled; therefore the $\lim_{x=\infty}\frac{h(x)}{\log x}$ exists and equals 0. Thus all the results shown in the first two paragraphs of this work are valid; that is, the $\lim_{x=\infty}\sum_{k=1}^{x}\frac{\mu(k)}{k}$ exists and equals 0, and thus the correctness of the equation named in the title, briefly $\sum_{k=1}^{x}\frac{\mu(k)}{k}=0.$ ## 4 If we define161616) von Mangoldt, 1. c., p. 850.) $M(x)=\sum_{k=1}^{x}\mu(k),$ then with help of the proven result, $\lim_{x=\infty}g(x)=0,$ we have $\lim_{x=\infty}\frac{M(x)}{x}=0.$ Von Mangoldt171717) 1. c., pp. 849–851.) proved this indirectly by use of the identity $g(x)=\sum_{k=1}^{x}\frac{\mu(k)}{k}=\sum_{k=1}^{x}(M(k)-M(k-1))\frac{1}{k}.$ It can be furnished as follows directly. From the equation $M(x)=\sum_{k=1}^{x}\mu(k)=\sum_{k=1}^{x}\frac{\mu(k)}{k}\cdot k=\sum_{k=1}^{x}(g(k)-g(k-1))\cdot k,$ it follows that $\displaystyle M(x)$ $\displaystyle=\sum_{k=1}^{x-1}g(k)(k-(k+1))+g(x)[x]$ $\displaystyle=-\sum_{k=1}^{x-1}g(k)+g(x)[x],$ so that since for $\delta>0$ there is a $G$ such that for all $k\geq G$ $|g(k)|\leq\frac{\delta}{3}$ for all $x\geq G$ $\displaystyle|M(x)|$ $\displaystyle\leq\sum_{k=1}^{G-1}|g(k)|+\sum_{k=G}^{x-1}|g(k)|+|g(x)|\cdot x$ $\displaystyle\leq G-1+\frac{\delta}{3}([x]-G)+\frac{\delta}{3}x,$ so that $\left|\frac{M(x)}{x}\right|\leq\frac{G-1-\frac{\delta}{3}G}{x}+\frac{2}{3}\delta,$ then for $x\geq\frac{G-1-\frac{\delta}{3}G}{\frac{\delta}{3}}$ and at the same time greater than or equal to $G$, $\left|\frac{M(x)}{x}\right|\leq\frac{1}{3}\delta+\frac{2}{3}\delta=\delta,$ with which the statement $\lim_{x=\infty}\frac{M(x)}{x}=0$ is proved. ## Theses 1. 1. It is desirable during every existence proof of a mathematical quantity to be led, at the same time on the way to the result, to the actual existing quantity. 2. 2. A boundary between arithmetic and analytic areas of mathematics cannot be drawn. 3. 3. The concept of the semiconvergent series is a relative concept. 4. 4. Out of the impossibility of perpetual motion of second kind comes the proof of the second law of thermodynamics. 5. 5. It did not succeed, the justifying of psychology on an exactly mathematical basis.

arxiv-papers

2024-09-04T02:48:54.561234

0803.3797

# Condensate fraction in a 2D Bose gas measured across the Mott-insulator transition I. B. Spielman ian.spielman@nist.gov W. D. Phillips J. V. Porto Joint Quantum Institute, National Institute of Standards and Technology, and University of Maryland, Gaithersburg, Maryland, 20899, USA ###### Abstract We realize a single-band 2D Bose-Hubbard system with Rb atoms in an optical lattice and measure the condensate fraction as a function of lattice depth, crossing from the superfluid to the Mott-insulating phase. We quantitatively identify the location of the superfluid to normal transition by observing when the condensed fraction vanishes. Our measurement agrees with recent quantum Monte Carlo calculations for a finite-sized 2D system to within experimental uncertainty. Measurements of condensed matter systems realized by cold atoms in optical lattices are now performed with sufficient accuracy to compare with ab-initio calculations Gerbier et al. (2005a); Spielman et al. (2007); Mun et al. (2007). Bosonic atoms in a lattice nearly perfectly realize the iconic Bose- Hubbard (BH) Hamiltonian. Here, we study the system’s momentum distribution, measure the condensate fraction, and accurately identify the transition point from the the low temperature superfluid (SF) phase by identifying when the condensate fraction vanishes. The SF to Mott insulator (MI) transition can be accessed by changing the depth of the optical potential Jaksch et al. (1998), and has been observed in 1D Köhl et al. (2005), 2D Spielman et al. (2007) and 3D Greiner et al. (2002). A range of studies have verified a detailed understanding of the MI phase in 2D and 3D Gerbier et al. (2005a, b); Fölling et al. (2006); Spielman et al. (2007). In contrast, the SF phase and the details of the transition to MI have gone largely unstudied. Indeed, the only quantitative measurement locating the transition is in 3D and is not in agreement with calculations Mun et al. (2007). Here we focus specifically on the superfluid phase of a 2D system and its transition to a normal state: we observe the expected increasing momentum spread and vanishing condensate fraction as the system leaves the SF phase. Our measured transition point agrees with the best available calculations Wessel et al. (2004), thereby locating a point on the non-zero temperature 2D BH phase diagram. Interestingly, the condensate fraction in our non-zero temperature system vanishes more sharply than expected for a zero temperature inhomogenous system, confirming that the superfluid regions are rapidly driven normal as soon as an insulator appears Ho and Zhou (2007); Gerbier (2007). The physics of interacting systems frequently depends spectacularly on dimensionality: in 3D the SF is a conventional Bose-Einstein condensate (BEC); in 2D, a Berezinskii-Kosterlitz-Thouless (BKT) SF; finally, in 1D there is no true SF. In contrast, only the detailed properties of the MI phase depend on dimensionality. In the $T>0$, 2D case studied here, the very existence of Bose condensation is a consequence of the finite size of our trapped system. We associate the presence of a bimodal momentum distribution with the SF phase, and use fits to the distribution to identify the Bose-condensed fraction and thereby measure the transition point between SF and normal. At low temperature the transition from SF is to a normal state which crosses over to a MI phase as the lattice depth increases Gerbier (2007); Ho and Zhou (2007). As a result any $T>0$ measurement based on condensate fraction will identify the SF to normal transition but be largely insensitive to the subsequent crossover into the MI phase. We study samples of ultra-cold rubidium atoms in a combined sinusoidal plus harmonic potential. For atom occupancy per lattice site larger than unity Greiner et al. (2002); Campbell et al. (2006), the low temperature SF phase (shallow lattice) is expected to evolve into a structure composed of alternating shells of SF and integer-occupied MI (deep lattice). As the lattice deepens, each successive MI region appears and grows, as probed in Ref. Fölling et al. (2006). At $T=0$ the amount of SF varies smoothly with lattice depth giving no abrupt changes in the momentum distribution to indicate a phase transition. In this work, we simplify the situation by working near unit filling, where the only insulating phase is unit occupied MI; thus, any observed signature can only be the transition from SF to normal. Absent the lattice, recent experiments have shown that weak contact interactions lead to a decrease in the 2D BEC transition temperature Kruger et al. (2007). Lattice potentials increase the relative importance of interactions; indeed, the onset of the MI phase corresponds to driving the critical temperature to zero. The BH model describes lattice-bosons that have a hopping matrix element $t$, and an on-site interaction energy $U$. The physics of the BH model depends only on $U/t$ Fisher et al. (1989). In an infinite, homogenous $T=0$ 2D system, the transition from SF to MI occurs at the critical value $(U/t)_{\rm c}\approx 16.5$ Krauth and Trivedi (1991); Elstner and Monien (1999); Wessel et al. (2004); Kato et al. (2007). Remarkably, we observe a sharp transition at $U/t=15.8(20)$ 111All uncertainties herein reflect the uncorrelated combination of single-sigma statistical and systematic uncertainties. in our $T>0$, finite-sized, harmonically trapped system. Our data consists of images of atom density after sudden release and time-of- flight (TOF), approximating the in-situ momentum distribution. Figure 1 shows 2D momentum distribution (right), and cross-sections through each distribution (left). As evidenced by Fig. 1-a and -b, each diffraction order in the momentum distributions consists of a narrow peak on a broad pedestal. Fitting to a bimodal distribution (see below), we determine $f$, the fractional contribution of the narrow component, and identify $f$ as the “condensate” fraction. We associate images with non-negligible $f$ as being in the SF phase Yi et al. (2007). We emphasize, however, that superfluidity is a transport phenomena and cannot unambiguously be associated with features in the momentum distribution Diener et al. (2007); Gerbier (2007); Ho and Zhou (2007). This association is also imperfect at $T>0$ because in our 2D trapped system we expect to observe a discernible “condensate” fraction even after the vortex- pairs of a BKT SF unbind Hadzibabic et al. (2006), destroying the 2D SF. $f$ vanishes only when the resulting phase-fluctuating quasi-condensate vanishes Kruger et al. (2007); Cladé and Helmerson (2008). To characterize the transition from SF to normal, we extract two independent quantities from an analysis of TOF images: $f$, and an “energy scale” $\sigma$. We also measure a related quantity, the full width at half maximum (FWHM) $\Gamma$ of the quasi-momentum distribution, which we compare to theory. As the lattice depth is increased we find that $f$ vanishes concurrently with a sudden increase in $\Gamma$, abrupt signatures that we associate with the transition. Figure 1: Momentum distributions and cross-sections at $U/t$ = 4(1), 8(1), and 20(2). Each row shows a single momentum distribution normalized by the total atom number; the lines in the top-right panel indicate trajectories along which four cross-sections were taken. The left panel shows the average of these four sections (black solid line); the red-dashed lines denote the fit to the bimodal distribution. We produce nearly pure 3D ${}^{87}\rm{Rb}$ BECs with $N_{T}=1.2(4)\times 10^{5}$ atoms in the $\left|F=1,m_{F}=-1\right>$ state Spielman et al. (2007). A pair of linearly polarized, $\lambda=820{\ {\rm nm}}$ laser beams forms a $30(2){{E_{R}}}$ deep vertical optical lattice along $\hat{z}$ that divides the 3D BEC into about $70$ 2D systems (turn-on time = $200{\ {\rm ms}}$). The single photon recoil wave-vector and energy are ${{k_{R}}}=2\pi/\lambda$ and ${{E_{R}}}=\hbar^{2}{{k_{R}}}^{2}/2m=h\times 3.4{\ {\rm kHz}}$; $m$ is the atomic mass and $h$ is Planck’s constant. The largest 2D system, containing $\approx 3000$ atoms, has a chemical potential $\mu_{\rm 2D}=h\times 600(100){\ {\rm Hz}}$ and we measure a temperature $k_{B}T=k_{B}\times 33(4){\ {\rm nK}}=h\times 700(70){\ {\rm Hz}}$. Since the first vibrational spacing $h\times 33(1){\ {\rm kHz}}\gg\mu_{\rm 2D},\ k_{B}T$, this system is well into the 2D regime. In addition, a weaker, square 2D lattice in the $\hat{x}$-$\hat{y}$ plane is produced by a second beam arranged in a folded- retroreflected configuration Sebby-Strabley et al. (2006), linearly polarized in the $\hat{x}$-$\hat{y}$ plane (turn-on-time = $100{\ {\rm ms}}$ 222The beams for the two lattices originate from the same Titanium-Sapphire laser but differ in frequency by about $160{\ {\rm MHz}}$.). The intensities of both lattices follow exponentially increasing ramps, with $50{\ {\rm ms}}$ and $25{\ {\rm ms}}$ time constants respectively, and reach their peak values concurrently. These time-scales are chosen to be adiabatic with respect to mean-field interactions, vibrational excitations, and tunneling within each 2D system. The final depth of the $\hat{x}$-$\hat{y}$ lattice determines $U/t$ and ranges from $V=0$ to $25(2){{E_{R}}}$ 333The depth of the lattice along $\hat{x}$ and $\hat{y}$ differ by 6%, and $V$ is the average.. The lattice depths are calibrated by pulsing the lattice for $3{\ \mu{\rm s}}$ and observing the resulting atom diffraction Ovchinnikov et al. (1998). We calculate $U/t$ using a 2D band-stucture model and the $s$-wave scattering length van Kempen et al. (2002). The uncertainty in $U/t$ stems from the uncertainty in lattice depth 444The uncertainty in the $\hat{x}$-$\hat{y}$ lattice depth affects both $t$ and $U$, while vertical lattice uncertainties affect only $U$. The $\pm 0.2\%$ uncertainty van Kempen et al. (2002) in the ${}^{87}\rm{Rb}$ $s-$wave scattering length is a negligible contribution to the overall uncertainty.. The resulting uncertainty in $U/t$ is $\pm 10\%$. Once both lattices are at their final intensity, the atomic system consists of an array of 2D gasses each in a square lattice of depth $V$ and with a typical density of 1 atom per lattice site. The atoms are held for $30{\ {\rm ms}}$, and all confining potentials are abruptly removed (the lattice and magnetic potentials turn off in $\lesssim 1{\ \mu{\rm s}}$ and $\simeq 300{\ \mu{\rm s}}$, respectively). As a result, the initially confined states are projected onto free particle states which expand for a $20.1{\ {\rm ms}}$ TOF 555Some of the data were not taken under exactly these conditions: some had a $29.1{\ {\rm ms}}$ TOF and in others the lattice depth was rapidly (in $50{\ \mu{\rm s}}$) increased to $30{{E_{R}}}$ before imaging (changing the single-site wave functions, not the correlations which structure to the momentum distribution); these differences do not affect the measurement, and are included in the figures. , when they are detected by resonant absorption imaging. Apart from effects of atomic interactions during expansion and the initial size of the sample, initial momentum maps into final position, so each image approximates the $\hat{x}$-$\hat{y}$ projection of the momentum distribution. We fit the each momentum distribution to a simple function which describes the distributions over the full range of $U/t$ studied here, with just three free parameters. First, we model the broad background as a thermal distribution of non- interacting classical particles in a single 2D sinusoidal band where states are labeled by their quasi-momentum $q_{x}$ and $q_{y}$, $n(q_{x},q_{y})\propto\exp\left[2(\cos\pi q_{x}/{{k_{R}}}+\cos\pi q_{x}/{{k_{R}}})/\sigma\right]$; this contributes two fitting parameters: $\sigma$ and the non-condensed atom-number. In the shallow lattice limit, $\sigma$ gives the temperature, $\sigma=k_{B}T/t$. This fit does not distinguish atoms thermally occupying higher momentum states from atoms occupying these states in the ground state wavefunction, i.e., from the quantum depletion of the SF. In fact, $n(q_{x},q_{y})$ multiplied by a suitable Wannier function, correctly describes the momentum distribution of atoms in the MI phase to first order in $t/U$ where $\sigma$ is unconnected to temperature, and is given by $\sigma=U/4t$. Our function fits the random phase approximation (RPA) momentum distribution fairly well even as higher order terms become important Sengupta and Dupuis (2005); Spielman et al. (2007). The second portion of the momentum distribution consists of a narrow peak, which we interpret as Bose-condensed atoms. We therefore take the narrow peak to be the inverted parabola of a Thomas-Fermi profile (of fixed width for all comparable data 666We do not allow the width of the condensate-peak to vary with each fit; instead we first fit all of the SF data with the condensate- width as a free parameter, and then repeat the fits with it held constant at the average value: for $20.1{\ {\rm ms}}$ TOF we found $R_{\rm TF}=19(2){\ \mu{\rm m}}$, and for $29.1{\ {\rm ms}}$ TOF we found $R_{\rm TF}=26(2){\ \mu{\rm m}}$.), characterized by a single fitting parameter, condensed number. The observed condensate peak width after TOF stems largely from initial system size, not interaction effects during TOF or the initial momentum spread. Here interactions during TOF are reduced due to rapid expansion along $\hat{z}$ after release from the tightly confining vertical lattice. Our analysis further reduces these interaction effects by excluding data inside the $1^{\rm st}$ Brillouin zone, with the highest density. This decreases the measured FWHM of the peak from $30(1){\ \mu{\rm m}}$ to $22(1){\ \mu{\rm m}}$ and the inferred momentum width from $0.26{{k_{R}}}$ to $0.21{{k_{R}}}$. Changing the TOF from $20.1{\ {\rm ms}}$ to $29.1{\ {\rm ms}}$ only increased the FWHM from $22(1){\ \mu{\rm m}}$ to $28(1){\ \mu{\rm m}}$ (decreasing the observed momentum width from $0.21{{k_{R}}}$ to $0.17{{k_{R}}}$). Figure 2: Condensed fraction $f$ and $\sigma$ vs. $V$ (bottom axis) or $U/t$ (top axis). The dots denote values determined from 2D fits to the full momentum distribution: small dots result from one image and the large dots indicate data averaged over about 20 separate images. The uncertainties are their RMS variation, and are indicative of the single-image uncertainties. (a) Condensate fraction. The red-dashed line is computed from our MFT model. (b) fit parameter $\sigma$. There are two distinct regimes: at low $U/t$ it is nearly constant (blue dashed line), from which we infer an initial temperature $k_{B}T\approx 2t$; and at large $U/t$ it monotonically increases, consistent with predictions of perturbation theory in the MI phase (red-dashed line). Figure 2-a shows that as $V$ increases, $f$ vanishes at a critical value $V_{\rm crit}$, while the total atom number remains constant. We verified that this disappearance does not result from excessive irreversible heating of the system by exceeding $V_{\rm crit}$, then lowering the lattice and observing a condensed fraction Greiner et al. (2002). To gain a qualitative understanding of the vanishing condensate fraction, we performed a non-zero temperature mean-field theory (MFT) simulation of an array of 2D BH systems in a 3D harmonic trap Sheshadri et al. (1993). To model the non-zero temperature experimental system, we determine the entropy at small $U/t$ that gives the observed $\approx 45\%$ condensate fraction, and assume this entropy is unchanged as $V$ increases. The red-dashed line in Fig. 2-a shows the MFT condensate fraction vs. $V$ at constant entropy. Given that $T=0$ MFT overestimates the transition ($(U/t)_{\rm MFT}=23.3$, compared to $(U/t)_{\rm c}=16.5$ from more accurate calculations), the curve unexpectedly lies on the data. MFT also gives $f$ as function of $U/t$ in units of $(U/t)_{\rm c}$. We identify the transition point by fitting this function to the data allowing $(U/t)_{\rm c}$ to vary, yielding $(U/t)_{\rm c}=15.8(20)$ (a lattice depth $V_{\rm crit}=9.0(5){{E_{R}}}$). Figure 2-b displays $\sigma$ from the uncondensed background portion of the distribution. At large $V$ we recover the behavior expected in the MI phase; this measurement is equivalent to observations of the modulated momentum distribution in the MI phase Gerbier et al. (2005a); Spielman et al. (2007). At higher total atom number, our system would develop doubly and triply occupied MI shells, expected to manifest as kinks in this curve. $\sigma$ is monotonic with $V$, varying smoothly across $V_{\rm crit}$. This is in agreement with RPA theory where the onset of superfluidity affects only states near zero quasi-momentum. Figure 2-b shows that when $V\lesssim 4{{E_{R}}}$ ($U/t\lesssim 3$), $k_{B}T/t\approx 2.0(3)$. Extrapolating to $V=0$ gives $k_{B}T={{E_{R}}}\sigma/\pi^{2}\approx k_{B}\times 33{\ {\rm nK}}$ (valid when $T\ll{{E_{R}}}$). This temperature is well below the $k_{B}\times 45{\ {\rm nK}}$ expected for non-interacting particles in our 2D harmonic trap with $f=0.45$, this reduction is similar to that observed in Ref. Kruger et al. (2007), which focused on the critical temperature in interacting 2D atomic systems with no 2D lattice. Figure 3: Quasi-momentum width vs. $V$ (bottom axis) or $U/t$ (top axis). The symbols denote the average FWHM of the quasi-momentum distribution along the axes of highest symmetry (Top: averaged along $\hat{x}$+$\hat{y}$ and $\hat{x}$-$\hat{y}$; Bottom: averaged along $\hat{x}$ and $\hat{y}$). The small and large dots and uncertainties are as explained at Fig. 2. The red- dashed line is the horizontally displaced RPA momentum width, as discussed in the text, and the vertical grey line denotes the location of the SF-normal transition identified from the sudden increase in $\Gamma$. A related characterization of the system is the FWHM $\Gamma$ of the full quasi-momentum distribution Greiner et al. (2002); Wessel et al. (2004); Kato et al. (2007); Pollet and Troyer ; Yi et al. (2007). Figure 3 shows the width of the 2D distributions (see Ref. Spielman et al. (2007)) as a function of $V$. In the SF phase $\Gamma$ hardly depends on $V$ since the dominant feature of the distribution is the condensate peak. $\Gamma$ only begins to change very close to the SF to normal phase transition when the heights of the two components of the bimodal distribution become comparable when the condensate disappears, consistent with calculations in homogenous and trapped systems Wessel et al. (2004). We calculate $\Gamma$ in the MI phase from the RPA Sengupta and Dupuis (2005) quasi-momentum distribution which accurately describes the large $U/t$ limit ($\sqrt{2}{{k_{R}}}$ along $\hat{x}+\hat{y}$ and ${{k_{R}}}$ along $\hat{x}$). In the RPA, $\Gamma$ is a function of $U/t$ in units of $(U/t)_{\rm c}$. The red dashed lines are fits to the measured widths using two free parameters (joint between both panels): in the MI region we use the RPA functional form with $(U/t)_{\rm c}$ as the first fit parameter, and the constant width in the SF phase is the second. We obtain $(U/t)_{\rm c}=16.7(20)$, in accord with the $(U/t)_{\rm c}=15.8(20)$ from our fit to the condensate fraction. We identify the point when the condensate fraction vanishes (and $\Gamma$ abruptly increases) with the onset of the SF to normal transition, i.e., when a normal region begins to rapidly expand in our inhomogeneous system. (Our measured visibility, computed as in Refs. Gerbier et al. (2005a, b), abruptly drops from near unity at $U/t\approx 16$.) Increasingly accurate numerical calculations give values of $(U/t)_{\rm c}$: $16.25(10)$ Wessel et al. (2004) and $16.77$ Kato et al. (2007). Perhaps most relevant are QMC calculations which include the effects of harmonic confinement; in this case, Wessel et al. Wessel et al. (2004) find that a MI region first forms at $(U/t)_{\rm c}=17.2$ (the exact value of $(U/t)_{\rm c}$ depends on the details of the harmonic potential). Both values lie within our experimental uncertainty. The calculations Wessel et al. (2004); Kato et al. (2007) are at zero temperature, and while they agree with our observed $(U/t)_{\rm c}$, they do not predict a sudden increase in peak width or a vanishing condensate fraction at $(U/t)_{\rm c}$. At $T=0$ and as $U/t$ increases past $(U/t)_{\rm c}$, where an inhomogeneous system first develops a unit-occupied Mott core, the shell of SF persists to large $U/t$. Thus, at $T=0$, $f$ drops rapidly at $(U/t)_{\rm c}$, but does not vanish. Our system, however, is at small but non-zero temperature, with a reduced condensate fraction of $\approx 45\%$ for small $V$. Our MFT model shows that this temperatures quickly drives the SF shells to the normal phase as $U/t$ increases past $(U/t)_{\rm c}$. As a result the SF shells rapidly disappear as the normal, and then Mott regions form (see Refs. Rey et al. (2006); Ho and Zhou (2007); Gerbier (2007)). That this feature is seen in preliminary non-zero temperature QMC calculations Pollet and Troyer underscores the need for further non-zero temperature calculations to compare with experiment. This experiment constitutes the measurement of a single point of the non-zero temperature 2D BH phase diagram. We expect future experiments will expand on this result at different temperatures, densities, in different dimensions, and in traps with more homogenous density distributions; new theory should aide in the interpretation of these experiments. We appreciate enlightening conversations with C. A. R. Sa de Melo, N. Trivedi, L. Pollet, and C. J. Williams. We acknowledge the financial support of ODNI/IARPA, and ONR; and I.B.S. thanks the NIST/NRC program. ## References * Gerbier et al. (2005a) F. Gerbier, A. Widera, S. Fölling, O. Mandel, T. Gericke, and I. Bloch, Physical Review Letters 95, 050404 (2005a). * Spielman et al. (2007) I. B. Spielman, W. D. Phillips, and J. V. Porto, Physical Review Letters 98, 080404 (2007). * Mun et al. (2007) J. Mun, P. Medley, G. K. Campbell, L. G. Marcassa, D. E. Pritchard, and W. Ketterle, Physical Review Letters 99, 150604 (2007). * Jaksch et al. (1998) D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P. Zoller, Physical Review Letters 81, 3108 (1998). * Köhl et al. (2005) M. Köhl, H. Moritz, T. Stöferle, C. Schori, and T. Esslinger, Journal of Low Temperature Physics 138, 635 (2005). * Greiner et al. (2002) M. Greiner, O. Mandel, T. Esslinger, T. Hänsch, and I. Bloch, Nature 415, 39 (2002). * Gerbier et al. (2005b) F. Gerbier, A. Widera, S. Fölling, O. Mandel, T. Gericke, and I. Bloch, Physical Review A 72, 053606 (2005b). * Fölling et al. (2006) S. Fölling, A. Widera, T. Müller, F. Gerbier, and I. Bloch, Physical Review Letters 97, 060403 (2006). * Wessel et al. (2004) S. Wessel, F. Alet, M. Troyer, and G. G. Batrouni, Physical Review A 70, 053615 (2004). * Campbell et al. (2006) G. K. Campbell, J. Mun, M. Boyd, P. Medley, A. E. Leanhardt, L. Marcassa, D. E. Pritchard, and W. Ketterle, Science 313, 649 (2006). * Hadzibabic et al. (2006) Z. Hadzibabic, P. Krüger, M. Cheneau, B. Battelier, and J. Dalibard, Nature 441 (2006). * Cladé and Helmerson (2008) P. Cladé and K. Helmerson (2008), private communication. * Kruger et al. (2007) P. Kruger, Z. Hadzibabic, and J. Dalibard, Physical Review Letters 99, 040402 (2007). * van Kempen et al. (2002) E. G. M. van Kempen, S. J. J. M. F. Kokkelmans, D. J. Heinzen, and B. J. Verhaar, Phys. Rev. Lett. 88, 093201 (2002). * Fisher et al. (1989) M. P. A. Fisher, P. B. Weichman, G. Grinstein, and D. S. Fisher, Physical Review B 40, 546 (1989). * Krauth and Trivedi (1991) W. Krauth and N. Trivedi, Europhysics Letters 14, 627 (1991). * Elstner and Monien (1999) N. Elstner and H. Monien, arXiv:cond-mat p. 9905367 (1999). * Kato et al. (2007) Y. Kato, N. Kawashima, and N. Trivedi (2007). * Yi et al. (2007) W. Yi, G.-D. Lin, and L.-M. Duan, arXiv p. 0705.4352v1 (2007). * Diener et al. (2007) R. B. Diener, Q. Zhou, H. Zhai, and T.-L. Ho, Physical Review Letters 98, 180404 (2007). * Gerbier (2007) F. Gerbier, Physical Review Letters 99, 120405 (2007). * Ho and Zhou (2007) T.-L. Ho and Q. Zhou, Physical Review Letters 99, 120404 (2007). * Sebby-Strabley et al. (2006) J. Sebby-Strabley, M. Anderlini, P. S. Jessen, and J. V. Porto, Phys. Rev. A 73, 033605 (2006). * Ovchinnikov et al. (1998) Y. B. Ovchinnikov, J. H. Müller, M. R. Doery, E. J. D. Vredenbregt, K. Helmerson, S. L. Rolston, and W. D. Phillips, Phys. Rev. Lett. 83, 284 (1998). * (25) L. Pollet and M. Troyer, private communication. * Sengupta and Dupuis (2005) K. Sengupta and N. Dupuis, Physical Review A 71, 033629 (2005). * Sheshadri et al. (1993) K. Sheshadri, H. R. Krishnamurthy, R. Pandit, and T. V. Ramakrishnan, Europhysics Letters 22, 257 (1993). * Rey et al. (2006) A. M. Rey, G. Pupillo, and J. V. Porto, Physical Review A 73, 023608 (2006).

arxiv-papers

2024-09-04T02:48:54.565122

0803.3838

# Recorded Step Directional Mutation for Faster Convergence Ted Dunning Chief Scientist Aptex Software (25 September, 1995) ###### Abstract Two meta-evolutionary optimization strategies described in this paper accelerate the convergence of evolutionary programming algorithms while still retaining much of their ability to deal with multi-modal problems. The strategies, called directional mutation and recorded step in this paper, can operate independently but together they greatly enhance the ability of evolutionary programming algorithms to deal with fitness landscapes characterized by long narrow valleys. The directional mutation aspect of this combined method uses correlated meta-mutation but does not introduce a full covariance matrix. These new methods are thus much more economical in terms of storage for problems with high dimensionality. Additionally, directional mutation is rotationally invariant which is a substantial advantage over self- adaptive methods which use a single variance per coordinate for problems where the natural orientation of the problem is not oriented along the axes. Step-recording is a subtle variation on conventional meta-mutational algorithms which allows desirable meta-mutations to be introduced quickly. Directional mutation, on the other hand, has analogies with conjugate gradient techniques in deterministic optimization algorithms. Together, these methods substantially improve performance on certain classes of problems, without incurring much in the way of cost on problems where they do not provide much benefit. Somewhat surprisingly their effect when applied separately is not consistent. This paper examines the performance of these new methods on several standard problems taken from the literature. These new methods are directly compared to more conventional evolutionary algorithms. A new test problem is also introduced to highlight the difficulties inherent with long narrow valleys. ## 1 Overview ### 1.1 Arguments for Meta-evolution A number of stochastic optimization procedures have been developed since the electronic computer has made automated optimization possible. Methods which have had substantial recent development include simulated annealing, genetic algorithms, evolutionary strategies and evolutionary programming. One shared property of each of these classes of algorithms is that they trade some degree of convergence speed for a decreased likelihood of avoiding locally optimal but globally suboptimal solutions. In each of these well-known algorithms, there is a parameter or set of parameters which can be manipulated to affect this trade-off between convergence power and speed. In simulated annealing, this parameter is the simulated temperature, while in evolutionary programming, this parameter is the mutation rate. Typically, the temperature or mutation rate is decreased as the optimization progresses. This tactic substantially improves the rate of convergence, often without significantly increasing the likelihood of finding a suboptimal solution. In special cases such as a quadratic bowl, cooling schedules can be derived which satisfy various theoretical constraints regarding the effort needed to have a given probability of finding a solution in a given amount of time, but this cannot be done in general since the derivation of the optimal annealing schedule requires detailed knowledge of properties of the function being optimized. Instead, an arbitrary and hopefully sufficiently conservative cooling schedule is typically invented and used. An alternative to a fixed cooling schedule is to derive the cooling schedule adaptively as the optimization algorithm learns about the fitness landscape that it is exploring. The idea that the mutation rate itself should be a parameter specific to each member of the population to be evolved is not new and has been recently explored in [Fog92] and [Atm91]. This form of meta- evolution is attractive in that no explicit cooling schedule need be given. ### 1.2 Common problems Problems whose solutions are found in long narrow valleys cause severe problems with evolutionary programming algorithm because the narrowness of the valley greatly decreases the probability of finding a solution which improves on a point which is already on the floor of the valley. These problems have been attacked in the past by using a covariance matrix to cause mutations to be correlated as described in [Seb92], [Sch81] and [Fog92]. This method is similar in essence to the conjugate gradient techniques used in conventional numerical optimization codes in that they concentrate the exploration of the fitness landscape in particular directions based on past experience. Various forms of directional mutation has been the subject since the mid 60’s as indicated by the work of Bremermann and others [BR64, BR65]. Combining this method with meta-evolution as is done in the methods presented in this paper raises some interesting problems, however. In particular, in meta-evolution, all parameters which control mutation are included in the genome and must themselves be subject to mutation. But, if there are $n$ real valued parameters in the genome initially, then it takes $n^{2}$ entries in a covariance matrix to describe how those $n$ parameters should be changed. Including the covariance matrix in the genome increases its size to $n^{2}+n$ real valued parameters and raises the serious question of how the mutation of the new parameters should be described. Self-similar mutation schemes can avoid the risk of a recursive explosion in the number of parameters to be optimized. Even so, the original $n^{2}$ covariance parameters in the description of each element of the entire population can be very expensive, even if meta-meta- mutation parameters are not included. Thus, it is desirable to find a more economical method for describing correlated mutation, and to find a way so that this correlated mutation description contains its own description of how it should be changed. One additional desiderata is that the meta-mutation be self similar so that the algorithm is insensitive to changes in scale. The two new strategies described in this report address this need. Other work along these lines can be found in [Fog97] where a directional mutation scheme is described which has overhead similar to the methods described here. It should be noted that the stochastic optimization methods which have used Cauchy distributed mutations as in [FK92, SH87, Yao91, Yao95, YL97] are inherently not rotationally invariant. In evolutionary optimizations dealing with a large number of dependent parameters, this means that the proportion of progeny which change only a few parameters will be much higher than would be expected under the assumption of complete rotational symmetry of the optimization algorithms. For many problems, notably including the synthetic problems often used to evaluate optimization algorithms, this coordinate orientation can be highly advantageous. For example, in the multi-modal test problems explored in [YL97] the local optima are arranged in orthogonal grids parallel to the parameter axes. In other problems of significant practical significance, however, correlated changes in parameters are necessary. Examples of this need for correlated changes occur in artificial neural networks or in electronic filter design. ## 2 Two New Strategies ### 2.1 Step Recording In a conventional meta-mutation algorithm, the mutation rate for each member of the population is mutated independently of the state vector. This method can lead to slower convergence, especially in conjunction with directional mutation if a low probability step leads to improvement in fitness. If this happens, repeating a step like the one that caused the improvement can be advantageous. If the mutation rate (and possibly the mutation direction) was changed independently of the fitness parameters, then similar steps will probably stay unlikely and convergence will be slow. This situation will happen when a very fit solution in a small basin has been found due to taking a very large step. That particular solution will tend to remain in the population, but further improvement is unlikely unless subsequent small steps are taken. Even if an improved solution is found by taking a small step (which is unlikely, but it will happen eventually), it is likely that the mutation rate will still be large (which is why we found this solution in the first place), and further improvements will only come slowly as solutions with both better fitness and lower mutation rates are found. The typical sequence is for the mutation rate to decrease first which then allows smaller steps to be taken resulting in further optimization. With step recording, on the other hand, the mutation rate of an offspring is set to the magnitude of the distance between the parent and child solution. This coupling of mutation rate and change in position means that any solutions which improve fitness by taking small steps will automatically lead to a line of progeny which tend to explore by taking small steps. Similarly, when directional mutation is combined with step recording, once a step is taken along the fitness gradient, further steps similar to that one are likely. This provides algorithms using directional mutation a sense of history in much the same manner as conjugate gradient methods use past history to improve further optimization efforts. ### 2.2 Directional Mutation In order to fully characterize all of the possible correlations between $n$ random variables, it is necessary to use roughly $n^{2}$ quantities. It does not, however, follow that a description this complete is necessary to gain the benefits of directional mutation. In particular, a covariance matrix specifies much more than just mutation biased in a particular direction. Indeed, a covariance matrix contains much more information than can be reliably extracted from the recent pedigree of a single member of a population. Instead, we propose the use of a much more limited model of directional mutation in which the mutation rate has a directional component and an omni- directional component. The total mutation is the sum of mutations derived from each of these components. The directional component of mutation is restricted to a line, while the omni-directional component is sampled from a symmetric gaussian distribution. Together these components give a total mutation distribution which is an ellipsoidal gaussian distribution. In a heuristic attempt to enhance convergence, the directional component is also biased slightly. If we use the notation $N(\mu,\sigma)$ to indicate a normally distributed random variable with mean $\mu$ and standard deviation $\sigma$ and use $U(a,b)$ to indicate a uniformly distributed random variable taken from the half open interval $[a,b)$, then the following mutation algorithm suffices to provide a directional mutation of the vector $\bf x$ $\lambda:=N(1,1)$ For each $x_{i}$, $x_{i}:=x_{i}+N(0,\sigma)+\lambda k_{i}$ Here $\lambda$ is a biased random variable which indicates how far to go along the direction indicated by $\bf k$, while $\sigma$ provides the magnitude of the omni-directional mutation component. The mutation parameters $\bf k$ and $\sigma$ can themselves be mutated by setting $\sigma:=-(\sigma+|{\bf k}|/10)\log(1-U(0,1))$ $\lambda:=N(1,1)$ and then for each $k_{i}$, $k_{i}:=N(0,\sigma)+\lambda k_{i}$ In this algorithm the mutation of $\sigma$ is done by taking a new value from the exponential distribution with mean equal to $\sigma$ augmented by a fraction of the magnitude of $\bf k$. This cross coupling between $\sigma$ and $\bf k$ prevents the mutations from becoming too directional. The mutation of $\bf k$ uses $\sigma$ to provide diversity in direction and $\lambda$ to provide diversity in magnitude in a manner identical with the way that the mutation of $x_{i}$ was done. The use of an exponential distribution is somewhat in contrast with the trend in the literature toward the use of a log- normal distribution, but the motivation is essentially the same. Both the exponential and log-normal distributions allow self-similar mutation which allows the entire algorithm to be scale invariant. It should be noted that the meta-mutation operation described here is self- similar and orientation independent. This means that the distribution of mutation parameters after several generations in the absence of selection is invariant up to the scale and orientation of the original value. This fact also implies that the properties of the resulting meta-evolutionary algorithm are subject to analysis by renormalization methods. To convert this algorithm to a step recording algorithm, the mutation of $\bf x$ is simplified and is done after the mutation of $\sigma$ and $\bf k$ as shown below. $\sigma:=-(\sigma+|{\bf k}|/10)\log(1-U(0,1))$ $\lambda:=N(1,1)$ $k_{i}:=N(0,\sigma)+\lambda k_{i}$ $x_{i}:=x_{i}+k_{i}$ The result is that $\bf k$ records the mutation step which was taken so that if this step results in an improvement, similar steps are likely to be used again. Another notable feature of the algorithms described here are the coupling between the directional parameters ${\bf k}$ and the omni-directional parameter $\sigma$. The coupling from ${\bf k}$ to $\sigma$ allows a population to stop mutating directionally when necessary as well as providing a bias which tends to increase the overall mutation rate in the absence of selection for lower rates. The coupling from $\sigma$ back to ${\bf k}$ allows changes in the preferred direction of mutation to take place. ## 3 Experimental Methods To provide a preliminary test of the efficacy of the proposed algorithms, all four combinations of conventional meta-evolution, meta-evolution with step recording, meta-evolution with directional mutation and conventional meta- evolution with both step recording and directional mutation were tested on three simple problems. These problems included a three dimensional symmetric quadratic bowl (function F1 from [Fog95]), a Bohachevsky multi-modal bowl problem (function F6 from the same work) and a very narrow two dimensional quadratic bowl whose axis was not aligned with either axis (labelled F9 here to avoid conflict with F1 through F8 from [Fog95]). These problems were not intended to provide a comprehensive inventory of the interesting problems, but rather were simply taken as exemplars which would highlight the contrast between previous methods and the directional recorded-step method. The dimensionality of the test functions used here is quite low, but the essential difficulty posed to previous evolutionary algorithms by long narrow valleys which are not aligned along the coordinate axes is independent of dimension. Additional tests with dimensionality as high as 30 show the same results as demonstrated here. The test functions are described by the following functions: $f_{1}(x,y,z)=x^{2}+y^{2}+z^{2}$ $f_{6}(x,y)=x^{2}+2y^{2}-0.3\cos(3\pi x)-0.4\cos(4\pi y)+0.7$ $f_{9}(x,y)=(x+y)^{2}+(100y-100x)^{2}$ For this test, the evolutionary algorithm used 20 survivors each generation, each of which generated 9 progeny to create a population of 200. After evaluating the fitness function for each member of the population, the entire population was sorted to find the best 20 members who would survive into the next generation. Each algorithm was run 10 times and a median fitness at each generation was used to compare algorithms. All programs were limited to 50 generations or less. Generally, convergence to a solution with $10^{-8}$ of the correct value was found within far fewer generations. ## 4 Results The graph in 1 illustrates the convergence for the four algorithms for the symmetric bowl (Function F1). Figure 1: Convergence for Symmetric Bowl As can be seen, the convergence of the conventional meta-evolutionary strategy is slightly faster than for the modified algorithms, but the difference in terms of number generations required to converge is not substantial and the ultimate accuracy of the final solution is essentially identical. It is interesting to note that the omni-directional mutation rate was a close approximation of the square root of the remaining error. This behavior is close to the theoretical optimum cooling for this problem; that it was derived automatically by the meta-mutation was noteworthy. Detailed examination of the population showed that omni-directional mutation was the dominant mechanism of exploration in the case of the symmetric bowl. The graph in 2 illustrates convergence for the Bohachevsky function. Figure 2: Convergence for Bohachevsky Function Again, the difference between the algorithms is not striking, except for the algorithm which used directional mutation without step recording. Even so, the degradation in convergence time was less than a factor of two for directional mutation, and the loss in performance for the other methods was minimal. Finally, the graph in 3 illustrates the convergence rates for the narrow quadratic bowl. Figure 3: Convergence for Narrow Bowl Here, all algorithms except for directional mutation with step recording have severe problems with convergence. The differences here are highly significant. The difference in the case of the non-directional algorithms is due to the fact that with symmetric mutation, if the mutation rate is much larger than the distance to the major axis of the valley, then any mutation is likely to fall outside of the narrow valley and thus not result in any improvement. The effect is that as the population approaches the major axis of the valley, the mutation rate is decreased and progress toward the optimum slows down. Ultimately, solutions very close to the major axis are found, and the mutation rate is reduced to a small value. This low mutation rate makes progress down the major axis of the valley toward the global optimum quite slow. It is not clear why directional mutation without step recording performs so poorly, but early experiments with different meta-mutation operators appeared to perform better, so the problem may have had more to do with the meta- mutation itself than with an inherent defect in the pure directional mutational algorithm. The algorithm which used directional mutation and step recording performed very well in the narrow valley problem. Detailed examination of evolving populations showed that populations far from the major axis of the valley quickly evolved directional mutations which took them to the major axis. Once there, the populations converted their directional mutations into omni- directional mutations which were in turn converted into directional mutations strongly oriented along the major axis of the valley. This quickly lead to bracketing of the solution at which point, the population variance shrank rapidly. Once the population had adapted to the nature of the problem, convergence proceeded essentially identically to the convergence behavior noted for the symmetric bowl problem ($f1$). It is instructive to compare these results with those from the table on page 173 of [Fog95]. The relevant parts of that table are reproduced in 1 and extended with the current results. Note that the meta-evolutionary algorithms (the new columns which are labelled MEP, MEP+RS and MEP+RS+DM) are clearly able to produce results which are comparable with previous evolutionary algorithms (the original columns which were labelled GA, DPE and EP in the original work). It should be remembered that when examining this table that comparing the various forms of evolutionary programming after such an extreme degree of convergence is not terribly meaningful. Function | GA | DPE | EP ---|---|---|--- F1 | $2.8\times 10^{-4}$ | $1.1\times 10^{-11}$ | $3.1\times 10^{-66}$ F6 | $2.629\times 10^{-3}$ | $1.479\times 10^{-9}$ | $5.193\times 10^{-96}$ Function | MEP | MEP+RS | MEP+RS+DM ---|---|---|--- F1 | $3.3\times 10^{-71}$ | $3.2\times 10^{-125}$ | $1.5\times 10^{-51}$ F6 | $0$ | $0$ | $0$ Table 1: Convergence of Various Evolutionary Algorithms (GA = Genetic Algorithm, DPE = GA with Dynamic Parameter Estimation, EP = Evolutionary Programming, MEP = Meta-Evolutionary algorithms, RS = Recorded Step, DM = Directional Mutation) ## 5 Summary and Discussion This work clearly shows that meta-evolutionary strategies can be effective in accelerating the convergence of evolutionary programming algorithms under certain conditions. Furthermore, the directional mutation and recorded step do not significantly degrade this performance on simple problems. They can provide highly signficant improvement in convergence speed on problems which involve long narrow valleys. The directional search algorithm presented here has a number of clear advantages over carrying a full covariance matrix with each member of the population. These include lower storage requirements, lower computational overhead, and an intuitively appealing method for doing meta-mutation. Although the algorithms describe here perform well on moderately multi-modal problems such as the Bohachevsky functions, they probably trade off some of the ability to avoid locally optimal solutions in return for their ability to explore narrow valleys. This ability may need to be recovered for some problems. One way that this might be done is to allow only parent/progeny competition. This would help avoid the situation where a solution is found which is good enough to swamp the survivor pool with progeny before more obscure solutions are found. Another method for attacking this problem would be introduce speciation. In partially or wholly decomposable problems with high dimensionality, narrow valleys can occur which are aligned with the axes rather than aligned arbitrarily. In these cases, it the cost of the directional mutation algorithm given here might be better spent by keeping a separate mutation rate for each dimension. Each of these mutation rates could be subjected to the self-similar exponential mutation described in this paper. Another option would be to keep the directional mutation, and expand the omni-directional mutation rate to one mutation rate per parameter. Whether either of these changes would actually enhance performance is an open question. The use of a Cauchy distribution for the directional mutation is also an intriguing possibility. As was noted earlier, the use of Cauchy distributions in problems similar to the long narrow valley examined here could severely degrade convergence. This is because of the fact that multi-variate Cauchy distributions are not rotationally invariant. One intriguing option for a hybrid approach is to use a Cauchy distribution for the directional mutation while retaining a normal distribution for the omni-directional mutation. Such a hybrid would retain the rotationally invariant properties of the algorithm described here while taking advantage of the desirable aspects of the use of the Cauchy distribution. Another interesting avenue for further research is to combine step recording and directional mutation with more conventional self-adaptation of mutation rates for each parameter. This combination might provide the advantages of recorded step methods when solving largely separable problems without losing the ability of directional mutation to deal with narrow non-separable valleys. The cost of this hybrid would be the requirement to keep $2n$ meta-parameters with each member of the population instead of the $n+1$ meta-parameters required by the methods described here. Overall, step recording and the directional mutation operator described in this report seem to provide strong advantages for optimizing certain classes of problems. The experiments described here provide an initial indication of how large these advantages can be. Further work is needed to characterize the interactions between these innovative techniques and other methods. ## References * [Atm91] David B. Fogel; Larry J. Fogel; J. Wirt Atmar. Meta-evolutionary programming. In R.R. Chen, editor, Proceedings of the 25th Asilomar Conference on Signals, Systems and Computers, pages 542–545. Pacific Grove, CA, 1991. * [BR64] H.J. Bremermann and M. Rogson. An evolution-type search method on convex sets. Technical Report ONR Technical Report, Contracts 222(85) AND 3656(58), ONR, 1964. * [BR65] H.J. Bremermann and M. Rogson. Search by evolution. In M. Maxfield, A. Callahan, and L. Fogel, editors, Biophysics and Cybernetic Systems, pages 157–167. Spartan Books, Washington D.C., 1965\. * [FK92] G. S. Fishman and V. G. Kulkarni. Improving monte carlo efficiency by increasing variance. Management Science, 38(10):1432–1444, 1992. * [Fog92] David B. Fogel; Larry J. Fogel; J. Wirt Atmar; Gary B. Fogel. Hierarchic methods of evolutionary programming. In Proc. of the First Annual Conference on Evolutionary Programming, Evolutionary Programming Society, San Diego, CA, 1992. * [Fog95] David B. Fogel. Evolutionary Computation: Toward a New Philosophy of Machine Intelligence. IEEE Press, New York, NY, 1995. * [Fog97] David B. Fogel. A preliminary investigation into directed mutations in evolutionary algorithms. In H.-M. Voigt, W Ebeling, I. RechenBerg, and H.-P. Schwefel, editors, PPSN4, pages 329–335. Springer Verlag, Berlin, 97. * [Sch81] H.-P. Schwefel. Numerical Optimization of Computer Models. John Wiley, Chichester, UK, 1981. * [Seb92] A.V. Sebald. On exploiting the global information generated by evolutionary programs. In Proc. of the First Annual Conference on Evolutionary Programming, Evolutionary Programming Society, San Diego, CA, 1992. * [SH87] H. H. Szu and R. L. Hartley. Nonconvex optimization by fast simulated annealing. Proceedings of the IEEE, 75:1538–1540, 1987. * [Yao91] Xin Yao. Simulated annealing with extended neighborhood. International Journal of Computer Mathematics, 40:169–189, 1991\. * [Yao95] Xin Yao. A new simulated annealing algorithm. International Journal of Computer Mathematics, 56:161–168, 1995\. * [YL97] Xin Yao and Yong Liu. Fast evolutionary programming. In Evolutionary Programming, V. MIT Press, 1997.

arxiv-papers

2024-09-04T02:48:54.569565

0803.3840

# Set families with a forbidden subposet Boris Bukh ###### Abstract We asymptotically determine the size of the largest family $\mathcal{F}$ of subsets of $\\{1,\dotsc,n\\}$ not containing a given poset $P$ if the Hasse diagram of $P$ is a tree. This is a qualitative generalization of several known results including Sperner’s theorem. ## Introduction We say that a poset $P$ is a _subposet_ of a poset $P^{\prime}$ if there is an injective map $f\colon P\to P^{\prime}$ such that $a\leq_{P}b$ implies $f(a)\leq_{P^{\prime}}f(b)$. A poset $P$ is an _induced subposet_ of $P^{\prime}$ if there is an injective map $f\colon P\to P^{\prime}$ for which $a\leq_{P}b$ if and only if $f(a)\leq_{P^{\prime}}f(b)$. For instance, is a subposet of , but not an induced subposet. For a poset $P$ a _Hasse diagram_ , denoted by $H(P)$, is a graph whose vertices are elements of $P$, and $xy$ is an edge if $x<y$ and for no other element $z$ of $P$ we have $x<z<y$. Let $[n]=\\{1,\dotsc,n\\}$, and denote by $2^{[n]}$ the collection of all subsets of $[n]$. One can think of a family $\mathcal{F}$ of subsets of $[n]$ as a poset under inclusion. In this way $\mathcal{F}$ becomes an induced subposet of the Boolean lattice. In this paper we examine the size of the largest family $\mathcal{F}\subset 2^{[n]}$ subject to the condition that $\mathcal{F}$ does not contain a fixed finite subposet $P$. We do not require $P$ to be an induced subposet. A set family $\mathcal{F}$ not containing a subposet $P$ will be called a $P$-free family. We denote by $\operatorname{ex}(P,n)$ the size of the largest $P$-free family $\mathcal{F}\subset 2^{[n]}$. For example, the classical Sperner’s theorem [Spe28] asserts that $\operatorname{ex}(\includegraphics[scale={0.32}]{chain2}\,,n)=\binom{n}{\lfloor n/2\rfloor}$. Erdős [Erd45] extended Sperner’s result, and proved that if $C_{l}$ denotes the chain of length $l$, then $\operatorname{ex}(C_{l},n)$ is equal to the sum of $l-1$ largest binomial coefficients of order $n$. Katona and Tarján[KT83] proved that $\operatorname{ex}(\includegraphics[scale={0.32}]{v2}\,,n)=\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$. A common generalization of results of Erdős and Katona-Tarján was established by Thanh[Tha98] who showed that if $P_{k,l}$ is any fixed poset with vertex set $\\{1,\dotsc,k\\}\cup\\{1^{\prime},\dotsc,l^{\prime}\\}$ in which the relations are $1<2<\dotsb<k$ and $1^{\prime}<1$, $2^{\prime}<1$, …$l^{\prime}<1$, then $\operatorname{ex}(P_{k,l})=k\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$ (the error term was subsequently improved by de Bonis and Katona[DBK07]). For example, $\operatorname{ex}(\raisebox{-3.0pt}{\includegraphics[scale={0.32}]{broom}}\,,n)=2\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$. In [DBKS05] it is shown that $\operatorname{ex}(\includegraphics[scale={0.32}]{butterfly}\,,n)=\binom{n}{\lfloor n/2\rfloor}+\binom{n}{\lfloor n/2\rfloor+1}$. Griggs and Katona [GK08] proved that $\operatorname{ex}(\includegraphics[scale={0.32}]{n}\,,n)=\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$. Recently, Griggs and Lu[GL] proved that $\operatorname{ex}(L_{4k},n)=\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$, where $L_{4k}$ is the loop of length $4k$ on two adjacent levels of the Boolean lattice. Let $\operatorname{Mon}(\mathbb{Z})$ be the set of all functions $f\colon\mathbb{Z}\to\\{0,1\\}$ such that $f(n)=1$ and $f(-n)=0$ for all sufficiently large $n$. The elements of $\operatorname{Mon}(\mathbb{Z})$ will be called eventually monotone functions. For $f,g\in\operatorname{Mon}(\mathbb{Z})$ write $f\leq g$ if $f(n)\leq g(n)$ for all $n$. Then $(\operatorname{Mon}(\mathbb{Z}),<)$ is a distributive lattice. Note that $\sum_{n}f(n)-g(n)$ is well-defined for every $f,g\in\operatorname{Mon}(\mathbb{Z})$. We define a _level_ of $\operatorname{Mon}(\mathbb{Z})$ to be a maximal family $L\subset\operatorname{Mon}(\mathbb{Z})$ satisfying $\sum_{n}f(n)-g(n)=0$ for all $f,g\in L$. Note that a level of $\operatorname{Mon}(\mathbb{Z})$ is an antichain. The poset $\operatorname{Mon}(\mathbb{Z})$ can be thought of as the induced subposet of $2^{\mathbb{Z}}$ spanned by the set $\\{X\subset\mathbb{Z}:\lvert X\triangle Y\rvert<\infty\\}$, for some fixed $Y\subset\mathbb{Z}$ which is neither finite nor cofinite. The simplest explanation for all the results above is the following conjecture. ###### Conjecture. For a finite poset $P$ let $l(P)$ be the maximum number of levels in $\operatorname{Mon}(\mathbb{Z})$ so that their union does not contain $P$ as a subposet, then $\operatorname{ex}(P,n)=l(P)\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}.$ Intuitively the conjecture asserts that the largest $P$-free family is essentially the union of the maximum number of middle levels of the Boolean lattice $2^{[n]}$ not containing $P$. If true, the conjecture would be an analogue of Erdős-Stone-Simonovits theorem from the extremal graph theory which asserts that the largest graph not containing a given graph $G$ is essentially the largest complete partite graph not containing $G$ In this paper, we establish the conjecture whenever $H(P)$ is a tree, generalizing several of the results mentioned above. Unlike the papers above we are not concerned with establishing the best possible bounds inside the $O(1/n)$ term, which allows us to give a rather short proof. The rest of the paper is occupied by the proof of the following theorem. ###### Theorem 1. If $P$ is a finite poset and $H(P)$ is a tree, then $\operatorname{ex}(P,n)=(h(P)-1)\binom{n}{\lfloor n/2\rfloor}\bigl{(}1+O(1/n)\bigr{)}$ where $h(P)$ is the height of $P$, i.e., the length of the longest chain in $P$. Moreover, $l(P)=h(P)-1$ for such a $P$. For the case $h(P)=2$, the theorem was also independently proved by Griggs and Lu[GL] by a very different argument. ## Proof idea Before embarking on the proof of Theorem 1, we first non-rigorously sketch a simple proof for the special case $h(P)=2$. The proof unfortunately does not generalize to $h(P)\geq 3$, but it will motivate the otherwise hard-to-follow technical details of the more general proof. We shall need a strengthening of Sperner’s lemma, to the effect that if $\lvert\mathcal{F}\rvert>\binom{n}{\lfloor n/2\rfloor}$, then not only there are pairs of comparable sets, but a plentitude of such pairs. The following statement is easy and can be proved similarly to Lemma 4 below (see [Kle68] for a sharper result for small $\epsilon$). ###### Lemma 2. If $\lvert\mathcal{F}\rvert\geq(1+\epsilon)\binom{n}{\lfloor n/2\rfloor}$, then there are at least $(1/10)n\epsilon\lvert\mathcal{F}\rvert$ pairs of sets $F_{1}\subset F_{2}$ contained in $\mathcal{F}$. ###### Proof of the case $h(P)=2$ of Theorem 1 assuming Lemma 2. Suppose $\mathcal{F}\subset 2^{[n]}$ is a family of $\lvert\mathcal{F}\rvert\geq(1+20\lvert P\rvert/n)\binom{n}{\lfloor n/2\rfloor}$ sets, where $P$ is a poset. We will show that $\mathcal{F}$ contains a copy of $P$. Let $G$ be the graph with vertex set $\mathcal{F}$ where a pair of distinct sets $F_{1},F_{2}\in\mathcal{F}$ connected by an edge if $F_{1}\subset F_{2}$ or $F_{2}\subset F_{1}$. By Lemma 2 the average degree of $G$ is at least $4\lvert P\rvert$. Since every graph of of average degree $d$ contains a non-empty subgraph of minimum degree at least $d/2$, the graph $G$ contains a subgraph $G^{\prime}$ of minimum degree at least $2\lvert P\rvert$. Then we shall embed elements of $P$ one-by-one into $V(G^{\prime})$, in such a way that at each step embedding is injective, and preserves order. More precisely, we assume that $P$ is not a subposet of $V(G^{\prime})$, and use this assumption to construct a sequence of embeddings $\pi_{i}\colon P_{i}\to V(G^{\prime})$, where 1. a) $P_{i}$ is an induced subposet of $P$, and $H(P_{i})$ is a tree, 2. b) $P_{i}\setminus P_{i-1}$ consists of a single element, 3. c) $v\leq_{P_{i}}u$ implies $\pi_{i}(v)\leq\pi_{i}(u)$ for all $u,v\in P_{i}$. First, we embed some element of $P$ arbitrarily into $V(G^{\prime})$, and let $P_{1}$ to consist of that single element. Then, for each $i\geq 2$ let $v\in P\setminus P_{i-1}$ be a not-yet-embedded element of $P$, which is comparable to some $u\in P_{i-1}$, and let $P_{i}=P_{i-1}\cup\\{v\\}$. Since $h(P)=2$, and $H(P)$ is a tree, $u$ is the only element of $P_{i-1}$ comparable to $v$. We shall define embedding $\pi_{i}$ that agrees with $\pi_{i-1}$ on $P_{i-1}\setminus\\{u\\}$. Without loss of generality we can assume that $u<v$. Set $\tau_{1}=\pi_{i-1}$, and write $u_{1}=\tau_{1}(u)$. Since $u_{1}\in V(G^{\prime})$, and degree of every vertex in $V(G^{\prime})$ is at least $2\lvert P\rvert\geq\lvert P\rvert+1$ there is at least one neighbor $u_{2}$ of $u_{1}$ in $G^{\prime}$ which is not in the image of $\tau_{1}$. If $u_{1}<u_{2}$, then we let $\pi_{i}$ to be an extension of $\tau_{1}$ sending $v$ to $u_{2}$. Suppose $u_{1}>u_{2}$. Let $\tau_{2}$ be the embedding obtained from $\tau_{1}$ by mapping $u$ to $u_{2}$ instead of $u_{1}$. The embedding $\tau_{2}$ satisfies the same properties (a)–(c) that $\pi_{i-1}$ does. As $2\lvert P\rvert\geq\lvert P\rvert+2$ we can again find a neighbor $u_{3}$ of $u_{2}$ which is neither $u_{1}$ nor in the image of $\pi_{2}$, and we can assume that $u_{2}>u_{3}$. Repeating this process $\lvert P\rvert$ times yields a chain $u_{1}>u_{2}>\dotsc>u_{\lvert P\rvert}$ of elements of $V(G^{\prime})$. However $P$ is a subposet (but not necessarily an induced subposet) of any linear extension of itself, and thus embeds into $\\{u_{1},\dotsc,u_{\lvert P\rvert}\\}\subset\mathcal{F}$. ∎ The proof above is clearly similar to the proof that every tree on $d$ vertices embeds into a graph of average degree $2d$. To extend the proof above to the case $h(P)=3$, for example, it is tempting to replace the graph $G$ by a $3$-uniform hypergraph of triples of sets in a chain. The problem with this approach is lack of any good replacement for the concept of minimum degree. The solution is therefore to eliminate minimum degree from the proof entirely. To see how it is done, we present an alternative way of embedding trees in graphs of large average degree. It is far more wasteful, but avoids minimum degrees. ###### Proposition 3. For every tree $T$ there is a $d_{0}(T)$ such that $T$ embeds into every finite graph of average degree at least $d_{0}(T)$. ###### Proof. The proof is by induction on $T$, with $\lvert T\rvert=1$ being the trivial base case. Assume $\lvert T\rvert\geq 2$. Let $v$ be any leaf of $T$, and $u$ be its unique neighbor. Set $T^{\prime}=T\setminus v$. We will show that we can take $d_{0}(T)=d_{0}(T^{\prime})+2\lvert T\rvert$. Suppose $G$ is of average degree at least $d_{0}(T^{\prime})+2\lvert T\rvert$. Let $B=\\{v\in G:\deg_{G}(v)\leq\lvert T\rvert\\}$. Then the graph $G^{\prime}=G\setminus B$ is only $\lvert B\rvert\lvert T\rvert$ edges smaller than $G$, and thus has has average degree at least $(d_{0}(T^{\prime})+2\lvert T\rvert)-2\lvert B\rvert\lvert T\rvert/\lvert V\rvert\geq d_{0}(T^{\prime})$. By the induction hypothesis there is an embedding $\pi\colon T^{\prime}\to G^{\prime}$. As $\pi(u)\in G^{\prime}$, we infer $\deg_{G}(\pi(u))>\lvert T\rvert$, implying that there is at least one neighbor of $\pi(u)$ that is not in $\pi(T^{\prime})$. Use this neighbor to extend the embedding $\pi$ of $T^{\prime}$ to an embedding of $T^{\prime}\cup\\{v\\}=T$. ∎ For technical reasons, it turns out to be easier to work with marked chains rather than hypergraphs; intuitively this change corresponds to hypergraphs with weighted edges. In the other respects, the proof of Theorem 1 is a straightforward generalization of the two arguments above. ## Proof of Theorem 1 By an _interval_ in a poset $P$ we mean a set of the form $[x,y]=\\{z\in P:x\leq z\leq y\\}$. A _maximal chain_ in a poset $P$ is a chain, which is not contained in any other chain. In particular, a maximal chain in $2^{[n]}$ is a chain of sets $\emptyset=S_{0}\subset S_{1}\subset\dotsb\subset S_{n}=[n]$ with $\lvert S_{i}\rvert=i$. A _$k$ -marked chain_ with markers $F_{1},\dotsc,F_{k}$ is a $k+1$-tuple $(M,F_{1},\dotsc,F_{k})$ where $M$ is a maximal chain in $2^{[n]}$, $F_{1}\supset\dotsb\supset F_{k}$ and $F_{1},\dotsc,F_{k}$ belong to $M$. ###### Lemma 4. If $\mathcal{F}\subset 2^{[n]}$ is of size $\lvert\mathcal{F}\rvert\geq(k-1+\epsilon)\binom{n}{\lfloor n/2\rfloor},$ then there are at least $(\epsilon/k)n!$ $k$-marked chains whose markers belong to $\mathcal{F}$. ###### Proof. Let $C_{i}$ be the number of maximal chains that contain exactly $i$ sets from $\mathcal{F}$. Counting the number of pairs $(F,M)$ where $F\in\mathcal{F}$ is an element of a maximal chain $M$ in two different ways, we obtain $\sum_{i}iC_{i}=\sum_{F\in\mathcal{F}}\frac{n!}{\binom{n}{\lvert F\rvert}}\geq\lvert\mathcal{F}\rvert\frac{n!}{\binom{n}{\lfloor n/2\rfloor}}\geq(k-1+\epsilon)n!.$ From this and $\sum C_{i}=n!$, we infer that $\sum_{i\geq k}iC_{i}\geq\sum_{i}iC_{i}-(k-1)\sum_{i\leq k-1}C_{i}\geq\epsilon n!.$ The number of $k$-marked chains with markers in $\mathcal{F}$ is $\displaystyle\sum_{i\geq k}\binom{i}{k}C_{i}$ $\displaystyle=\sum_{i\geq k}\binom{i-1}{k-1}\frac{i}{k}C_{i}\geq\frac{1}{k}\sum_{i\geq k}iC_{i}\geq\frac{\epsilon}{k}n!.\qed$ Call a poset $P$ of height $k$ _saturated_ if every maximal chain is of length $k$. For us the maximal chains in $P$ play the role analogous to the edges of a tree $T$ in Proposition 3. However, in general the edges might have different sizes, which is analogous to dealing with non-uniform hypergraphs. The saturated posets are the analogues of uniform hypergraphs. The next two lemmas establish a couple of intuitively obvious, but annoyingly hard to rigorously prove facts about saturated posets whose Hasse diagram is a tree. The first lemma will be used to reduce the problem of embedding an arbitrary $P$ to the problem of embedding saturated $P$. The second lemma will allow us to do induction on $\lvert P\rvert$. ###### Lemma 5. If $P$ is a finite poset with $H(P)$ being a tree, then $P$ is an induced subposet of some saturated finite poset $\tilde{P}$ with $H(\tilde{P})$ being a tree, and $h(P)=h(\tilde{P})$. ###### Proof. For the purpose of this proof let $s(P)$ be the number of maximal chains in $P$. Since every element is contained in some maximal chain, $\lvert P\rvert\leq s(P)h(P)$, implying that for fixed $s$ and $k$ there only finitely many posets $P$ with $s(P)=s$ and $h(P)=k$. Assume there is a counterexample to the lemma. Let $P$ be a counterexample with the largest number of elements for given $s(P)$ and $h(P)$. Since $P$ is not saturated there is a maximal chain $v_{1}<\dotsb<v_{t}$ in $P$ of length $t<k$. For each $i=0,\dotsc,t-1$ we can define a new poset $P_{i}$ which is obtained from $P$ by adding a new element $v$ and two new relations $v_{i}<v$ and $v<v_{i+1}$ (in the case $i=0$ we add only one new relation). Clearly, $s(P_{i})=s(P)$ and $P$ is an induced subposet of $P_{i}$. We will show by induction on $i$ that for each $i=0,\dotsc,t-1$ either $h(P_{j})=h(P)$ for some $j\leq i$, or there is a chain in $P$ of length $k-i$ whose smallest element is $v_{i+1}$. If $h(P_{0})>k$, then there is a chain $C$ in $P_{0}$ of length $k+1$. Since $C$ is not a chain of $P$, it contains $v$. Thus $v_{1}\in C$, and $C\setminus\\{v\\}$ contains $v_{1}$ and has length $k$. Now suppose $i\geq 1$ and we have established the inductive claim for all smaller values of $i$. If $h(P_{j})>h(P)$ for all $j=0,\dotsc,i-1$, then by the induction hypothesis there is a chain $C\subset P$ of length $k-i+1$ whose smallest element is $v_{i}$. Therefore, all chains in $P$ whose largest element is $v_{i}$ have lengths not exceeding $i$. Since the length of the chain $v_{1}<\dotsb<v_{i}$ is $i$, the assumption $h(P_{i})>k$ implies that there is a chain in $P$ of length at least $k-i$ whose smallest element is $v_{i+1}$. This establishes the inductive claim. Hence, if $h(P_{i})>h(P)$ for all $i$, there is a chain of length $k-t+1\geq 2$ whose smallest element is $v_{t}$. This contradicts the maximality of $v_{1}<\dotsb<v_{t}$, implying that $h(P_{i})=h(P)$ for some $i$. However, $P$ was assumed to be the largest counterexample with given $s(P)$ and $h(P)$. Therefore, there are no counterexamples to the lemma. ∎ ###### Lemma 6. Suppose $P$ is a saturated finite poset of height $h(P)=k\geq 2$, and $H(P)$ is a tree. Furthermore, assume that $P$ is not a chain. Then there is a $v\in P$, which is a leaf in $H(P)$, and an interval $I$ of length $\lvert I\rvert\leq k-1$ containing $v$ such that $H(P\setminus I)$ is a tree, and the poset $P\setminus I$ is a saturated poset of height $h(P)$. ###### Proof. A sequence $v_{1},\dotsc,v_{l}\in P$ is said to be a _poset path_ of length $l-1$ if $v_{i}$ and $v_{i+1}$ are comparable for all $i$. A _poset distance_ between $v$ and $u$, denoted $\operatorname{pdist}(v,u)$, is the shortest length of a poset path connecting them. [r] Let $v$ and $u$ be a pair of leaves maximizing $\operatorname{pdist}(v,u)$, and let $v=v_{1},v_{2},v_{3},\dotsc,v_{l}=u$ be the shortest poset path between them. Observe that $\operatorname{pdist}(v,u)\geq 2$, for $P$ is not a chain. Without loss of generality we can assume that $v<v_{2}$, for we can consider the opposite poset otherwise. Let $I_{0}$ be the longest interval containing $v$, all of whose elements have degree at most two in $H(P)$. If $\lvert I_{0}\rvert<h(P)$, set $I=I_{0}$. If $\lvert I_{0}\rvert=h(P)$, set $I=I_{0}\setminus\\{\max I_{0}\\}$, where $\max I_{0}$ is the largest element of $I_{0}$. Let $C$ be an arbitrary maximal chain in $P\setminus I$. We will show that $C$ has length $h(P)$. Suppose $C$ contains an element $w$ which is incomparable with $I$. Let then $M$ be a maximal chain in $P$ containing $C$. Since $w$ is comparable to every element of $M$, the chain $M$ is disjoint from $I$, implying $C=M$ and $h(C)=h(M)=h(P)$. So we can suppose that all elements of $C$ are comparable to $I$. If $\lvert I_{0}\rvert=h(P)$, then the only element in $P\setminus I$ which is comparable to $I$ is $\max I_{0}$. Since $P$ is not a chain, there is a chain of length two containing $\max I_{0}$, which is disjoint from $I$, contradicting the maximality of $C$. Hence, we can assume that $\lvert I_{0}\rvert<h(P)$. Let $w$ be any element of $P\setminus I$ comparable to an element $z$ of $I$. If $w<z$, then $\min([w,z]\cap I)$ has degree at least $3$ in $H(P)$. If $z<w$ and $\max I\not<w$, then $\max([z,w]\cap I)$ has degree at least $3$ in $H(P)$. In either case, we reach a contradiction with the choice of $I$. Hence $w$ exceeds all elements of $I$. Taking $w=\min C$, it follows that $\max I<\min C$. By the maximality of $C$, there is no $z\in P\setminus I$ satisfying $\max I<z<\min C$. Upon taking $w=v_{2}$, it also follows that $\max I<v_{2}$, implying $\min C\leq v_{2}$. If $v_{2}=\min C$, then $C\cup\\{v_{3}\\}$ is also a chain, contradicting the maximality of $C$. Let $y=\max C$. The only path from $u$ to $y$ in $H(P)$ goes through $v_{2}$ and $\min C$. Therefore every poset path from $u$ to $y$ has to go through $v_{2}$ and $\min C$, implying $\operatorname{pdist}(y,u)>\operatorname{pdist}(v,u)$. Though the element $y$ needs not to be a leaf itself, if $z$ is any leaf of $H(P)$ such that the path from $u$ to $z$ goes through $y$, then $\operatorname{pdist}(z,u)\geq\operatorname{pdist}(y,u)>\operatorname{pdist}(v,u)$, contradicting the choice of $v$ and $u$. ∎ The core of the proof of Theorem 1 is contained in the following lemma. In the lemma the family of $k$-marked chains $\mathcal{L}$ plays analogous role to the graph $G$ in the Proposition 3, with $\mathcal{F}$ being analogous to the vertex set of $G$. The condition that $\mathcal{F}$ does not contain a chain of length $K$ comes from the fact that every poset embeds into every sufficiently long chain. ###### Lemma 7. Let $P$ be a saturated finite poset of height $h(P)=k\geq 2$, whose Hasse diagram is a tree. Suppose $\mathcal{F}\subset 2^{[n]}$ is a set family, such that no chain contains more than $K$ sets from $\mathcal{F}$, and all sets in $\mathcal{F}$ are of size between $n/4$ and $3n/4$. Moreover, suppose $\mathcal{L}$ is a family of $k$-marked chains with markers in $\mathcal{F}$ of size $\lvert\mathcal{L}\rvert\geq\frac{\binom{\lvert P\rvert+1}{2}4^{K+1}}{n}n!.$ Then there is an embedding of $P$ into $\mathcal{F}$ in which every maximal chain of $P$ is mapped to the set of markers of some $k$-marked chain in $\mathcal{L}$. ###### Proof. The proof is by induction on $\lvert P\rvert$. If $P$ is the chain of length $k$, then finding the required embedding is easy: marked elements on any $L\in\mathcal{L}$ form the desired chain. Now suppose we want to embed $P$, and have already established the lemma for all smaller saturated posets. Use the preceding lemma to obtain a leaf $v$ and an interval $I\ni v$ such that $P\setminus I$ is a still a saturated poset of height $k$. By passing to the opposite poset to $P$ and replacing $\mathcal{F}$ by $\bar{\mathcal{F}}=\\{[n]\setminus F:F\in\mathcal{F}\\}$ if necessary, we can assume that $v$ is smaller than any element that is comparable with $v$. Let $C$ be a maximal chain containing $I$. Let $s=\lvert C\setminus I\rvert=k-\lvert I\rvert$. Note that $s\geq 1$. Call a chain $F_{1}\supset\dotsb\supset F_{s}$ of length $s$ a _bottleneck_ if there is a set $\mathcal{S}\subset\mathcal{F}$ with than $\lvert P\rvert$ elements such that for every $k$-marked chain in $\mathcal{L}$ of the form $(M,F_{1},\dotsc,F_{s},F_{s+1},\dotsc,F_{k})$ we have $\mathcal{S}\cap\\{F_{s+1},\dotsc,F_{k}\\}\neq\emptyset$. Such an $\mathcal{S}$ is said to be a _witness_ to the fact that $F_{1}\supset\dotsb\supset F_{s}$ is a bottleneck. Note that without loss of generality, a witness contains only proper subsets of $F_{s}$. For each bottleneck $F_{1}\supset\dotsb\supset F_{s}$, let $\mathcal{S}(F_{1},\dotsc,F_{s})$ be a fixed witness containing only proper subsets of $F_{s}$. Call a $k$-marked chain $(M,F_{1},\dotsc,F_{k})\in\mathcal{L}$ _bad_ if for some $s$ the chain $F_{1}\supset\dotsb\supset F_{s}$ is a bottleneck. Consider any $s$-element set $R=\\{r_{1},\dotsc,r_{s}\\}$ of integers with $1\leq r_{1}<\dotsb<r_{s}\leq K$. If $M$ is a maximal chain in $2^{[n]}$ containing at least $r_{s}$ elements from $\mathcal{F}$, let $F_{1}\supset F_{2}\supset\dotsb\supset F_{r_{s}}$ be the $r_{s}$ largest of these elements. The subchain of $F_{1}\supset F_{2}\supset\dotsb\supset F_{r_{s}}$ indexed by $R$ is $F_{r_{1}}\supset F_{r_{2}}\supset\dotsb\supset F_{r_{s}}$, and we denote it by $C_{R}(M)$. If $C_{R}(M)$ is a bottleneck, and $\mathcal{L}$ contains a $k$-marked chain of the form $(M,\dotsc)$, whose $s$ largest markers are $C_{R}(M)$, then we say that $M$ is _$R$ -bad_. Intuitively, the $R$-bad chains correspond to the edges adjacent to the vertices of low degree in Proposition 3. Pick a maximal chain $M$ in $2^{[n]}$ uniformly at random. Let $B_{R}$ be the event that $M$ is $R$-bad. We will estimate $\Pr[B_{R}]$ for each fixed $R$ individually. One way to pick a random maximal chain $M$ of $2^{[n]}$ is to start with $[n]$ and remove elements one by one, each step choosing an element uniformly at random among the remaining elements. Thus one can generate chain $M$ in two stages. In the first stage, we remove elements from $[n]$ at random until either we encounter $r_{s}$ sets from $\mathcal{F}$, or until we run out of elements to remove. Denote by $T$ the chain obtained at the end of the first stage (it is not a maximal chain, unless we ran out of elements). In the second stage, we resume removing elements at random from $\min T$, until no elements are left. If $T$ is not maximal, then $C_{R}(M)$ is independent of what happens in the second stage, and $C_{R}(T)$ is defined in the obvious way. If $T$ is a maximal chain, or $C_{R}(T)$ is not a bottleneck, then $B_{R}$ does not hold. Otherwise, let $\mathcal{S}=\mathcal{S}(C_{R}(T))$ be the witness that $C_{R}(T)$ is a bottleneck. Recall that $\mathcal{S}\subset\mathcal{F}$, $\lvert\mathcal{S}\rvert<\lvert P\rvert$ and $\mathcal{S}$ meets every $k$-chain in $\mathcal{L}$ whose top $s$ markers are $C_{R}(T)$. Let $\mathcal{T}_{R}=\\{\text{chain }T_{0}\text{ in }2^{[n]}:\lvert T_{0}\cap\mathcal{F}\rvert=r_{s},C_{R}(T_{0})\text{ is a bottleneck}\\}.$ The probability that $M$ meets $S$ is thus $\displaystyle\Pr[M\cap\mathcal{S}\neq\emptyset\,|\,C_{R}(T)\text{ is a bottleneck}]$ $\displaystyle\leq\max_{T_{0}\in\mathcal{T}_{R}}\Pr[M\cap\mathcal{S}\neq\emptyset|T=T_{0}]$ $\displaystyle\leq\max_{T_{0}\in\mathcal{T}_{R}}\lvert\mathcal{S}(C_{R}(T_{0}))\rvert\max_{\begin{subarray}{c}F\in\mathcal{F}\\\ F<\min T_{0}\end{subarray}}\Pr[F\in M|T=T_{0}]$ $\displaystyle\leq\lvert P\rvert\max_{T_{0}\in\mathcal{T}_{R}}\max_{\begin{subarray}{c}F\in\mathcal{F}\\\ F<\min T_{0}\end{subarray}}\Pr[F\in M|T=T_{0}]$ $\displaystyle\leq\lvert P\rvert\max_{T\in\mathcal{T}_{R}}\max_{\begin{subarray}{c}F\in\mathcal{F}\\\ F<\min T_{0}\end{subarray}}\frac{1}{\lvert F\rvert+1}$ $\displaystyle\leq\lvert P\rvert\max_{F\in\mathcal{F}}\frac{1}{\lvert F\rvert+1},$ $\displaystyle\leq 4\lvert P\rvert/n$ where the fourth inequality follows because at the step before obtaining $F$, we have $\lvert F\rvert+1$ choices as to which element to remove, with at most one choice yielding $F$. If $M\cap\mathcal{S}=\emptyset$, then there is no $k$-marked chain of the form $(M,\dotsc)$ in $\mathcal{L}$, and $B_{R}$ does not hold. Therefore $\displaystyle\Pr[B_{R}]$ $\displaystyle=\Pr[C_{R}(T)\text{ is a bottleneck}]\Pr[B_{R}|C_{R}(T)\text{ is a bottleneck}]$ $\displaystyle\leq\Pr[C_{R}(T)\text{ is a bottleneck}]\Pr[M\cap\mathcal{S}\neq\emptyset\,|\,C_{R}(T)\text{ is a bottleneck}]\leq 4\lvert P\rvert/n.$ Since $R$ is a subset of $[K]$, the number of pairs $(M,R)$ where $M$ is an $R$-bad maximal chain is at most $(4\lvert P\rvert\binom{K}{s}/n)n!$. Since no chain contains more than $K$ elements of $\mathcal{F}$, every bad $k$-marked chain gives rise to one such pair $(M,R)$. Since $R\subset[K]$, every pair $(M,R)$ arises in at most $\binom{K}{s}$ ways, implying that there are no more than $(4\lvert P\rvert\binom{K}{s}^{2}/n)n!\leq(\lvert P\rvert 4^{K+1}/n)n!$ bad $k$-marked chains in $\mathcal{L}$. Let $\mathcal{L}^{\prime}$ be the set of all good $k$-marked chains in $\mathcal{L}$. There are $\lvert\mathcal{L}^{\prime}\rvert\geq\lvert\mathcal{L}\rvert-\frac{\lvert P\rvert 4^{K+1}}{n}n!\geq\frac{4^{K+1}\bigl{[}\binom{\lvert P\rvert+1}{2}-\lvert P\rvert\bigr{]}}{n}n!=\frac{\binom{\lvert P\rvert}{2}4^{K+1}}{n}n!$ of them. By the induction hypothesis there is an embedding $\pi\colon P\setminus I\to\mathcal{F}$. Recall that $C$ was a maximal chain containing $I$, and look at $C\setminus I$. Since $P\setminus I$ is saturated, $C\setminus I$ is contained in some chain $C^{\prime}$ of length $k$ in $P\setminus I$. Therefore $\pi(C^{\prime})$ is contained in some $L\in\mathcal{L}^{\prime}$. Since all $k$-marked chains in $\mathcal{L}^{\prime}$ are good, $\pi(C\setminus I)$ is not a bottleneck. In particular, since $\lvert P\setminus I\rvert<\lvert P\rvert$, we infer that $\pi(P\setminus I)$ is not a witness that $\pi(C\setminus I)$ is a bottleneck. Thus there is a $k$-marked chain $\tilde{L}\in\mathcal{L}$ containing $\pi(C\setminus I)$ as markers, but not containing any other element of $\pi(P\setminus I)$ as a marker. Therefore, we can map $I$ to the bottom $k-s$ markers of $\tilde{L}$, completing the desired embedding. ∎ With most of the work already done, we are ready to prove our main result. ###### Proof of Theorem 1. If $h(P)=1$, then $P$ is a single-element poset, and the theorem is trivially true. So, assume that $h(P)\geq 2$. Consider the case when $P$ is a saturated poset, and suppose $\lvert\mathcal{F}\rvert\geq(h(P)-1)\binom{n}{\lfloor n/2\rfloor}\Bigl{(}1+\frac{h(P)\lvert P\rvert^{2}4^{\lvert P\rvert+2}}{n}\Bigr{)}$ and $n$ is sufficiently large. We will show that $\mathcal{F}$ contains $P$. The number of sets $F\in 2^{[n]}$ with fewer than $n/4$ or more than $3n/4$ elements is equal $2^{n}$ times the probability that for a randomly chosen $F\in 2^{[n]}$ we have $\lvert\lvert F\rvert-n/2\rvert>n/4$. Thus by Chernoff’s inequality the number of such sets $F\in 2^{[n]}$ is at most $2^{n}\cdot 2\exp\bigl{(}-2(n/4)^{2}/n\bigr{)}=o\bigl{(}\binom{n}{\lfloor n/2\rfloor}/n\bigr{)}$. Let $\mathcal{F}^{\prime}=\\{F\in F:\lvert F-n/2\rvert\leq n/4\\}$. As our $n$ is sufficiently large, $\lvert\mathcal{F}^{\prime}\rvert\geq(h(P)-1)\binom{n}{\lfloor n/2\rfloor}\Bigl{(}1+\frac{h(P)\lvert P\rvert^{2}4^{\lvert P\rvert+1}}{n}\Bigr{)}.$ Therefore, from Lemma 4 and Lemma 7 applied with $k=h(P)$ and $K=\lvert P\rvert$ it follows that either there is an embedding of $P$ into $\mathcal{F}^{\prime}$ or $\mathcal{F}^{\prime}$ contains a chain $C$ of length $\lvert P\rvert$. In the latter case, we can find an embedding $\pi$ of $P$ into $\mathcal{F}^{\prime}$ anyway by simply letting $\pi\colon P\to C$ be any linear extension of $P$. If $H(P)$ is a tree and $P$ is not a saturated poset, then by Lemma 5 it is contained in some saturated poset $P^{\prime}$ of height $h(P^{\prime})=h(P)$, such that $H(P^{\prime})$ is a tree. Therefore $\operatorname{ex}(P,n)=(h(P)-1)\binom{n}{\lfloor n/2\rfloor}(1+O(1/n))$ for every poset $P$, for which $H(P)$ is a tree. It remains to prove that $l(P)=h(P)-1$. The inequality $l(P)\geq h(P)-1$ is clear, as a union of $h(P)-1$ levels does not contain a chain of length exceeding $h(P)-1$, and hence does not contain $P$. Let $L_{1},\dotsc,L_{h}$ be $h$ distinct levels of $\operatorname{Mon}(\mathbb{Z})$, and let $L$ be their union. Suppose furthermore that the levels $L_{i}$ are so ordered that for any functions $f_{i}\in L_{i}$, the inequality $\sum_{n}f_{i}(n)-f_{j}(n)>0$ holds whenever $i>j$ (by the definition of a level, if the inequality holds between a pair functions in levels $L_{i}$ and $L_{j}$, then it holds for all pairs). To complete the proof, we need to exhibit an embedding of $P$ into $L$. By Lemma 5 it suffices to treat the case when $P$ is saturated. We will prove the existence of the embedding by induction on $\lvert P\rvert$. If $P$ is a chain of length $h$, then the embedding is obvious. Suppose $P$ is not a chain, we can find embedding for all smaller saturated $P$ of height $h$. By Lemma 6 there is a leaf $v$ and an interval $I$ of the form $I=[v,u)$ such that $P\setminus I$ is a saturated poset of height $h$. By induction $P\setminus I$ is embeddable into $L$. Fix any such embedding. Since $\pi(u)$ is contained in a chain of length $k$ in $L$, and $P$ is saturated, it follows that $\pi(u)\in L_{\lvert I\rvert+1}$. Let $n_{0}$ be a large enough that $(\pi(w))(n)=1$ for all $w\in P\setminus I$ and $n\geq n_{0}$. Complete the embedding by mapping the interval $I$ to the interval of functions $f_{1},\dotsc,f_{\lvert I\rvert}\in\operatorname{Mon}(\mathbb{Z})$ defined by $f_{i}(n)=\begin{cases}0,&\text{if }n_{0}\leq n\leq n_{0}+i-1,\\\ (\pi(w))(n),&\text{otherwise}.\end{cases}\qed$ ## Concluding remarks Though it would be interesting to determine exactly or find very good asymptotic estimates for $\operatorname{ex}(P,n)$ in general, a first step is to find the leading term in the asymptotic. In this paper we found the leading term of $\operatorname{ex}(P,n)$ whenever $H(P)$ is a tree. For some posets $P$ whose Hasse diagram is not a tree, one can find a poset $P^{\prime}$ that contains $P$ and whose Hasse diagram is a tree with $l(P)=l(P^{\prime})$, to obtain that $\operatorname{ex}(P,n)\sim\operatorname{ex}(P^{\prime},n)$. For example, $\operatorname{ex}(\includegraphics[scale={0.32}]{butterfly}\,,n)\sim\operatorname{ex}(\raisebox{-3.0pt}{\includegraphics[scale={0.32}]{supbutterfly}}\,,n)$, and similarly for other complete two-level posets, thus recovering the results from [DBK07, Section 5]. The simplest two posets that are not subposets of trees with the same value of $l(P)$ are and , and the asymptotics of the function $\operatorname{ex}$ for these posets is not known. It is conceivable that the conjecture in this paper is even true if its premise that $\mathcal{F}$ does not contain a subposet $P$ is replaced by the weaker premise that $\mathcal{F}$ does not contain $P$ as an induced subposet. Acknowledgement. I thank Máté Matolcsi for reading a preliminary version of this paper, and two referees for useful suggestions. ## References * [DBK07] Annalisa De Bonis and Gyula O. H. Katona. Largest families without an $r$-fork. Order, 24(3):181–191, 2007. * [DBKS05] Annalisa De Bonis, Gyula O. H. Katona, and Konrad J. Swanepoel. Largest family without $A\cup B\subseteq C\cap D$. J. Combin. Theory Ser. A, 111(2):331–336, 2005. arXiv:math/0407373v1. * [Erd45] P. Erdös. On a lemma of Littlewood and Offord. Bull. Amer. Math. Soc., 51:898–902, 1945. * [GK08] Jerrold R. Griggs and Gyula O. H. Katona. No four subsets forming an N. J. Combin. Theory Ser. A, 115(4):677–685, 2008. http://www.math.sc.edu/~IMI/technical/07papers/0704.pdf. * [GL] Jerrold R. Griggs and Linyuan Lu. On families of subsets with a forbidden subposet. arXiv:0807.3702v1. * [Kle68] D. Kleitman. A conjecture of Erdős-Katona on commensurable pairs among subsets of an $n$-set. In Theory of Graphs (Proc. Colloq., Tihany, 1966), pages 215–218. Academic Press, New York, 1968. * [KT83] G. O. H. Katona and T. G. Tarján. Extremal problems with excluded subgraphs in the $n$-cube. In Graph theory (Łagów, 1981), volume 1018 of Lecture Notes in Math., pages 84–93. Springer, Berlin, 1983. * [Spe28] Emanuel Sperner. Ein Satz über Untermengen einer endlichen Menge. Math. Z., 27(1):544–548, 1928. * [Tha98] Hai Tran Thanh. An extremal problem with excluded subposet in the Boolean lattice. Order, 15(1):51–57, 1998.

arxiv-papers

2024-09-04T02:48:54.574056

0803.3901

# Crystallization of medium length 1-alcohols in mesoporous silicon: An X-ray diffraction study Anke Henschel Patrick Huber p.huber@physik.uni-saarland.de Klaus Knorr knorr@mx.uni-saarland.de Faculty of Physics and Mechatronics Engineering, Saarland University, D-66041 Saarbrücken, Germany ###### Abstract The linear 1-alcohols n-C16H33OH, n-C17H35OH, n-C19H37OH have been imbibed and solidified in lined up, tubular mesopores of silicon with 10 nm and 15 nm mean diameters, respectively. X-ray diffraction measurements reveal a set of six discrete orientation states (”domains”) characterized by a perpendicular alignment of the molecules with respect to the long axis of the pores and by a four-fold symmetry about this direction, which coincides with the crystalline symmetry of the Si host. A Bragg peak series characteristic of the formation of bilayers indicates a lamellar structure of the spatially confined alcohol crystals in 15 nm pores. By contrast, no layering reflections could be detected for 10 nm pores. The growth mechanism responsible for the peculiar orientation states is attributed to a nano-scale version of the Bridgman technique of single-crystal growth, where the dominant growth direction is aligned parallelly to the long pore axes. Our observations are analogous to the growth phenomenology encountered for medium length n-alkanes confined in mesoporous silicon (Phys. Rev. E 75, 021607 (2007)) and may further elucidate why porous silicon matrices act as an effective nucleation-inducing material for protein solution crystallization. ###### pacs: 81.07.-b, 61.46.Hk, , 61.10.-i, 68.18.Jk Molecular ensembles condensed into mesopores usually show melting temperatures reduced with respect to the bulk state Christenson ; AlbaSim . The close packed high-symmetry crystal structures of small globular molecules such as the fcc structure of the heavy rare gases or the hcp structure of N2 and CO are conserved in pore confinement Huber99 . For the medium length n-alkanes the situation is different. These molecules form layered crystals Mueller1932 . Within the layers the molecules are tightly packed side-by-side, thereby forming a quasi-hexagonal 2D array. In the so-called ”crystalline, C” low- temperature phase the molecules are ordered with respect to rotations about the molecule axis. This leads to a long-range azimuthal order of the herringbone-type, accompanied by an uniaxial distortion of the lattice with respect to the hexagonal reference. Furthermore the all-trans zig-zag chains of the C-atoms of lateral neighbours lock-in resulting in well defined collective tilts of the molecule axis with respect to the normal of the layers, the tilt angle is zero for alkanes with an odd number of C-atoms and of the order of 30 deg for even-numbered alkanes. Close to the melting temperatures mesophases (rotator R phases) appear with partial or complete rotational disorder, reduced uniaxial distortion and - in case of even alkanes - reduced tilt angles Sirota1 ; Sirota2 ; Doucet1 ; Doucet2 ; Dirand ; Ewen1980 . When confined in porous varieties of silica (e.g. Vycor) with pore diameters of a few nm, melting and the C-R transition temperature are reduced, the layering reflections are absent, the temperature range over which the R-phases are stable is increased, and the odd-even difference is lifted Huber1 ; Huber2 . Altogether the effects of pore confinement appear plausible. The pore networks of the silica substrates are highly random, the pore solid is subject to random strains which in turn couple to the rotational and translational degrees-of-freedom Hoechli1990 . Thereby the disordered phases are favored and the odd-even effect is blurred. Figure 1: (Color online) X-ray diffraction pole figure of the (20) Bragg reflection of C19OH confined in mesoporous silicon at 298 K. As illustrated in the upper panel, the orientation of the sample sheet with respect to the scattering vector q is specified by the polar angle $\Phi$ and the azimuth $\chi$. The two orientational domains of the molecules about the $\chi$ axis are schematically sketched in the figure: The orientation of the long axes of the molecules about the $\chi$-axis differ by 90 deg resulting in a view on the lateral herringbone-type ordered in-plane structure of the alcohols for the first domain (left pore) and a side view on the lamellar order of the chains for the second domain (right pore). Note, the long axes of the molecules in both $\chi$-domains are perpendicular to the long pore axis, which coincides with the [100] direction of the Si host. Recently we have reported x-ray diffraction results on the n-alkanes Cn, $n=16,17,19,25$ ($n$ stands for the number of C-atoms) in a porous Si (100) sheet with a pore diameter of 10 nm Henschel . In this mesoporous substrate the pores are lined up, perpendicular to the sheet. In most respects the results were found to be analogous to those obtained in mesoporous silica, but the layering reflections do now show up and the crystal lattice of the alkanes has a well defined set of orientation states (”domains”) with respect to the Si lattice. In particular, the long axes of the molecules are arranged perpendicular to the long axes of the pores, similarly as has been reported for the crystallization of folded polymer chains forming lamellae in aligned tubular alumina pores Steinhart . The present article deals with the corresponding 1-alcohols C$n$OH, $n=16,17,19$ in porous Si Canham1995 . In the bulk state, the structural and thermodynamic behavior of C$n$ and C$n$OH are closely related as far as the phase sequence and the odd-even effect is concerned, the main difference being the fact that the monolayers of the alkanes are replaced by tail-to-tail bilayers in which the sublayers are coupled by H-bonds forming an O$-$H$\cdot\cdot$O chain Sirota ; Ventola ; Abrahamsson . The preparation of the substrate and the x-ray diffraction experiment has been described in Ref. Henschel ; Huber2007 . The in-plane diffraction patterns of all three alcohols investigated are basically identical. They are dominated by the fundamental reflections of the quasi-hexagonal in-plane lattice which are indexed (11), (1-1), (20) in terms of the rectangular 2D lattice with the unit mesh containing two molecules. The lattice parameters, $a$ and $b$, can be extracted from the peak positions and converted into the uniaxial distortion $D$ with respect to hexagonal reference lattice, $D=1-a/b\sqrt{3}$, and the area per molecule $A=ab/2$. The scattering vector q forms a polar angle $\Phi$ with the pore axis (=sheet normal) and an azimuth $\chi$ about the pore axis - see also Fig. 1. The origin of the $\chi$ is chosen such that for $\chi$=0, q is along the [011] direction of the Si lattice. The fundamental triple can only be observed for $\Phi$ being close to a multiple of 60 deg and $\chi$ being close to a multiple of 90 deg. As discussed in detail in Ref. Henschel , this observation calls for six domain states, and we have argued that the discrete $\Phi$-values result from a Bridgman type selection process of the propagation of the solidification front along the pores, whereas the texture with respect to $\chi$ signals epitaxy on pore walls formed by (011) facets of the host lattice. Figure 2: (Color online) A series of $\Phi$-”rocking” scans on C17OH crystallized in mesoporous silicon at 245 K. The scattering vector q lies in the $\chi$=90 deg-plane. $\left|\textbf{q}\right|$ is held constant along an individual scan at a value specified in the figure. The three peaks observed are the fundamental triple mentioned in the text. Figure 3: (Color online) A series of radial scans on C17OH solidified in mesoporous silicon showing the (11)/(1-1) and the (20) reflections at selected temperatures $T$ indicated in the figure, both on cooling and heating through the melting/freezing and the C-R transition. The scattering vector is parallel to the pore axis ($\Phi$=0). The formation of the domain states is apparent from Figs. 1 and 2. Fig. 1 is a pole figure of the (20) reflection. The fourfold symmetry with respect to $\chi$-rotations about the pore axis is evident. In Fig. 2 we show the fundamental triple by means of $\Phi$ rocking scans in the $\chi$=90 deg-plane for a series of different Bragg angles 2 $\Theta$ and hence moduli of the scattering vector q. As can be seen the triple is centered at $\Phi$=120 deg. The three peaks represent the three $\Phi$-domains ($\Phi$=0 deg,$\pm$60 deg) which obviously have about equal statistical weights. The changes of the diffraction pattern with temperature, both on cooling and heating, are illustrated by the scans of Fig. 3 where q is parallel to the pore axis ($\Phi$=0). Such a radial scan hits the (20) peak directly, the (11) and (1-1) peaks are slightly off the scan path, but their intensity is picked up because of the finite mosaic width. The freezing and melting temperatures, $T_{\rm f}$ and $T_{\rm m}$, can be determined from the onset of Bragg intensities on cooling and their disappearance on heating, respectively. The peak splitting signals the transition from the R to the C phase which also shows some thermal hysteresis between cooling and heating. The transition temperatures are collected in Table I. The low-$T$ phase shows (21)-reflections characteristic of a herringbone type orientational order and has in-plane lattice parameters $a$ and $b$ that are within experimental error identical to those of prototypic orthorhombic C phase of the alkanes which in turn is closely related to the fully ordered, low tilt monoclinic $\beta$-phase of the odd numbered 1-alcohols Ventola ; Yamamoto . This justifies calling the low-T phase ”C”. Close to $T_{\rm f}$/$T_{\rm m}$ the fundamental triple merges into a single peak which points to the hexagonal in-plane lattice, known from the rotator phase RII. On the other hand the peaks of the mesophases are always asymmetric which could mean that there is some non-resolved peak splitting due to residual distortions with a magnitude similar to what has been observed in the rotator RV and RIV phases of the bulk alkanes and the analogous phases of the alcohols. Altogether the in-plane metric observed does not tolerate tilt angles of the order of 30 deg that have been observed in C phases of the bulk even-numbered alkanes and alcohols. In pores the tilt angle practically vanishes, not only for the odd-numbered molecules, but also in C16OH. The peculiar texture observed not only means that the layer normal, but also that the molecules are oriented perpendicular to the pore axis. alcohol | $T_{\rm f/m}^{\rm cooling}$ | $T_{\rm f/m}^{\rm heating}$ | $T_{\rm R/C}^{\rm cooling}$ | $T_{\rm R/C}^{\rm heating}$ | $T_{\rm f/m}^{\rm Bulk}$ | $T_{\rm R/C}^{\rm Bulk}$ ---|---|---|---|---|---|--- C16OH | 304 K | 312 K | 290 K | 293 K | 322 K | 316 K C17OH | 310 K | 318 K | 294 K | 298 K | 326 K | 314 K C19OH | 320 K | 327 K | 306 K | 313 K | 334 K | 324 K Table 1: Table of melting and freezing temperatures $T_{\rm f/m}$ and phase transition temperatures $T_{\rm R/C}$ on heating and cooling for n=16, 17, 19. The bulk data are given for comparison Sirota . Layering reflections have been searched for, in particular in scans with q perpendicular to the pores, but except for a very weak (001) reflection, no higher layering reflections have been observed, quite in contrast to the corresponding alkanes. Based on space filling arguments we do in fact conclude from the mass uptake of the pores that the alcohols are arranged in tightly packed parallel layers, but that there is a sizeable mean square displacement of the molecules in the $z$-direction perpendicular to layers (larger than the 2 Å thick interlayer gap), such that the electron density contrast between the layers and the interlayer gaps, on which the intensity of such reflections relies, is washed out. The absence of the layering reflections in the alcohols may be related to the fact that the thickness of the bilayers (e.g. 4.8 nm for C17OH) is already getting close to the pore radius, on the other hand one might have thought that the strong H-bonds within the bilayer could help to reduce the $z$-excursions of molecules relative to the situation encountered for the alkanes. Obviously this is not the case. Figure 4: (Color online) Radial scan on C16OH solidified in mesoporous silicon. The scattering vector is perpendicular to the pore axis ($\Phi$=90 deg). The reflection comb characteristic of the equidistant Bragg peaks due to bilayer formation is included as a guide for the eye. Solid lines of the comb mark clearly observed bilayer Bragg peaks, whereas dotted lines indicate allowed bilayer Bragg peak positions, which, within our $q$-resolution, coincide with the in-plane reflections (11)/(1-1), (20), (31), and (3-1), respectively. The diffraction experiments have been repeated after the pore walls had been oxidized by a treatment with H2O2. See Ref. Henschel . The melted alcohols are still sucked into the pores, but are pushed out of the pores upon solidification. The solid then forms epitaxial layers on the surface of the substrate with bulk properties, analogous to what has been observed for the alkanes. In an additional preparation step, the oxide on the pore walls has been removed by etching with HF. This procedure increases the pore diameter to about 15 nm and also reduces the roughness of the pore walls Kumar2008 . Now the pore solid is again stable. The in-plane diffraction patterns are practically identical to those on the original samples, apart from the fact that the Bragg peaks are somewhat sharper and that the transition temperatures are slightly higher. However, in radial scans with q perpendicular to the pore axis the layering peaks show up - see Fig. 4. Note, due to the parallel alignment of the layers to the long pore axes and the 90 deg texture about the $\chi$ axis, three out of six domains and hence half of the entire crystallized sample contributes to the intensity of the layering peaks in this scan geometry rendering them observable up to the 18th order. The bilayer spacings conform to the relation $d=(4+2.56\,n)$ Å which has been established in case of extended, all-trans molecules and vanishing tilt. The coherence length derived from the width of the layering peaks is 14 nm, which is close to the pore diameter. Finally, we would like to speculate that the recently reported advantages of mesoporous silicon matrices in order to induce and speed up bulk crystallization in protein solutions Chayen may not only be attributable to an increased heterogeneous crystal nucleation in the pores and at the substrate surfaces Page2006 , but also, to some extent, to a Bridgman type alignment mechanism of the fast growing crystal nuclei. The anisotropic pore confinement may align the dominant growth direction of protein seeds within the pores parallel to the long axes of the pores, similarly as demonstrated here for n-alcohols. Once these fast growing nuclei reach the pore mouths, they can induce fast, directed bulk protein crystallization in the bulk solution surrounding mesoporous Si. ###### Acknowledgements. We thank P. Leibenguth for taking the pole figure. The work has been supported by the German Research Foundation (DFG) via the Collaborative Research Centre (SFB) 277, Saarbrücken. ## References * (1) H.K. Christenson, J. Phys. Condens. Mat. 13, R95 (2001); K. Knorr, P. Huber, and D. Wallacher, Z. Phys. Chem. 222, 257 (2008). * (2) C. Alba-Simionesco, B. Coasne, G. Dosseh, G. Dudziak, K.E. Gubbins, R. Radhakrishnan, and M.G. Sliwinska-Bartkowiak, J. Phys. Condens. Mat. 18, R15 (2006). * (3) P. Huber and K. Knorr, Phys. Rev. B 60, 12657 (1999); P. Huber and K. Knorr, Mater. Res. Soc. Symp. Proc. 876E, R3.1 (2005) and cond-mat/0508683. * (4) A. Müller, Proc. Roy. Soc. A 138, 514 (1932). * (5) E. B. Sirota, H. E. King, Jr. , D. M. Singer, and Henry H. Shao, J. Chem. Phys. 98 (1993). * (6) E. B. Sirota and A. B. Herhold, Science 283, 529 (1999). * (7) J. Doucet, I. Denicolo, and A. Craievich, J. Chem. Phys. 75, 1523 (1981). * (8) J. Doucet, I. Denicolo, A. Craievich, and A. Collet, J. Chem. Phys. 75, 5125 (1981). * (9) M. Dirand, M. Bouroukba, V. Chevallier, D. Petitjean, E. Behar, and V. Ruffier-Meray, J. Chem. Data 47 115 (2002). * (10) B. Ewen, G. R. Strobl, and D. Richter, Faraday Disc. 69, 19 (1980). * (11) P. Huber, D. Wallacher, J. Albers, and K. Knorr, Europhys. Lett. 65, 351 (2004). * (12) P. Huber, V.P. Soprunyuk, and K. Knorr, Phys. Rev. E 74, 031610 (2006). * (13) U. Höchli, K. Knorr and A. Loidl, Adv. Phys. 39 (1990) 405. * (14) A. Henschel, T. Hofmann, P. Huber, and K. Knorr, Phys. Rev. E 75,021607 (2007). * (15) M. Steinhart, P. Göring, H. Dernaika, M. Prabhukaran, U. Gösele, E. Hempel, and T. Thurn-Albrecht, Phys. Rev. Lett. 97, 027801 (2006); E. Woo, J. Huh, Y.G. Jeong, and K. Shin, Phys. Rev. Lett. 98, 136103 (2007). * (16) L.T. Canham, Adv. Mat. 7, 1033 (1995). * (17) E.B. Sirota and X.Z. Wu, J. Chem. Phys. 105, 7763 (1996). * (18) L. Ventola, M. Ramirez, T. Calvet, X. Solans, M.A. Cuevas-Diarte, P. Negrier, D. Mondieig, J.C. van Miltenburg, and H.A.J. Oonk, Chem. Mater. 14, 508 (2002). * (19) S. Abrahamsson, G. Larsson, and E. von Sydow, Acta Cryst. 13, 770 (1960). * (20) P. Huber, S. Gruener, C. Schaefer, K. Knorr, and A.V. Kityk, Europ. Phys. J. Spec. Top. 141, 101 (2007). * (21) T. Yamamoto, K. Nozaki, and T. Hara, J. Chem. Phys. 92, 631 (1990). * (22) P. Kumar, T. Hofmann, K. Knorr, P. Huber, P. Scheib, and P. Lemmens, J. Appl. Phys. 103, 024303 (2008). * (23) N.E. Chayen, E. Saridakis, R. El-Bahar, Y. Nemirovsky, J. Mol. Biol. 312, 591 (2001); N.E. Chayen, E. Saridakis, and R.P. Sear, Proc. Nat. Acad. Sc. 103, 597 (2006); D. Frenkel, Nature 443, 641 (2006). * (24) A.J. Page and R.P. Sear, Phys. Rev. Lett. 97, 065701 (2006); S. Stolyarova, E. Saridakis, N.E. Chayen, and Y. Nemirovsky, Biophys. J. 91, 3857 (2006).

arxiv-papers

2024-09-04T02:48:54.580548

0803.4093

# Electrodynamic spherical harmonic A V Novitsky Department of Theoretical Physics, Belarusian State University, Nezavisimosti Avenue 4, 220050 Minsk, Belarus andrey.novitsky@tut.by ###### Abstract Electrodynamic spherical harmonic is a second rank tensor in three-dimensional space. It allows to separate the radial and angle variables in vector solutions of Maxwell’s equations. Using the orthonormalization for electrodynamic spherical harmonic, a boundary problem on a sphere can be easily solved. ## 1 Introduction In this paper we introduce new function — electrodynamic spherical harmonic. It is represented as a second rank tensor in three-dimensional space. But the function differs from tensor spherical harmonic [1, 2, 3, 4, 5]. By its definition the electrodynamic spherical function possesses a number of properties of usual scalar and vector harmonics and includes them as component parts. Why electrodynamic spherical harmonic? The word “electrodynamic” implies that the function is applied for solution of vector field problems. We use it in electrodynamics, however one can apply the function for description of spin fields in quantum field theory. Electrodynamic spherical harmonic is not a simple designation of the well- known functions. It satisfies the Maxwell equations and describes the angular dependence of vector fields. The introduced function separates the variables (radial coordinate and angles) in the fields. Moreover, the notation of the fields in terms of electrodynamic spherical harmonic noticeably simplifies the solution of a boundary problem on spherical interface. Just application of the ortonormalization condition allows to find the coefficients of spherical function expansion (e.g. scattering field amplitudes). ## 2 Scalar and vector spherical harmonics Hamilton’s operator $\nabla$ in three-dimensional space contains the derivatives on three coordinates. In spherical coordinates ($r$, $\theta$, $\varphi$) the derivatives on radial coordinate and angles can be separated. This is achieved by representing the unit tensor ${\bf 1}$ in 3D space as the superposition of two projection operators: $\nabla={\bf 1}\nabla=\left(\frac{{\bf r}\otimes{\bf r}}{r^{2}}-\frac{{\bf r}^{\times 2}}{r^{2}}\right)\nabla=\frac{1}{r^{2}}{\bf r}({\bf r}\nabla)-\frac{1}{r^{2}}{\bf r}^{\times}({\bf r}^{\times}\nabla),$ where ${\bf r}\otimes{\bf r}/r^{2}$ is the projector onto the direction ${\bf e}_{r}={\bf r}/r$, $-{\bf r}^{\times 2}/r^{2}$ is the projector onto the plane orthogonal to the unit vector ${\bf e}_{r}$. Tensor ${\bf r}^{\times}$ dual to the vector ${\bf r}$ gives the well-known vector product when acting on a vector ${\bf a}$: ${\bf r}^{\times}{\bf a}={\bf r}\times{\bf a}$ and ${\bf a}{\bf r}^{\times}={\bf a}\times{\bf r}$ [6]. Introducing the vector differential operator ${\bf L}=\frac{1}{{\rm i}}{\bf r}\times\nabla$ (1) the equation (LABEL:nabla1) is rewritten as follows $\nabla=\frac{{\bf r}}{r}\frac{\partial}{\partial r}-\frac{{\rm i}}{r^{2}}{\bf r}\times{\bf L}.$ (2) Vector ${\bf L}$ is called orbital angular momentum operator in quantum mechanics, because it is presented as vector product of radius vector ${\bf r}$ and momentum ${\bf p}=-{\rm i}\nabla$ operators. ${\bf L}$ includes only derivatives on the angles $\theta$ and $\varphi$. Using the definition of ${\bf L}$ one can derive the Laplace operator $\Delta=\nabla^{2}=\frac{1}{r^{2}}\frac{\partial}{\partial r}\left(r^{2}\frac{\partial}{\partial r}\right)-\frac{{\bf L}^{2}}{r^{2}}$ (3) and the following properties: ${\bf r}{\bf L}=0,\qquad({\bf L}{\bf r})=0,\qquad{\bf L}^{2}{\bf L}={\bf L}{\bf L}^{2},\qquad{\bf L}\times{\bf L}={\rm i}{\bf L}.$ (4) Scalar spherical harmonic $Y_{lm}(\theta,\varphi)$ is defined as the eigenfunction of the operator ${\bf L}^{2}$: ${\bf L}^{2}Y_{lm}=l(l+1)Y_{lm},$ (5) where $l$ and $m$ are integer numbers. Number $m$ is the eigenvalue of the operator of projection of angular momentum onto the axis $z$, $L_{z}$: $L_{z}Y_{lm}=mY_{lm}.$ (6) Spherical harmonics $Y_{lm}$ are orthogonal and normalized by the unit: $\int_{0}^{\pi}\int_{0}^{2\pi}Y^{\ast}_{l^{\prime}m^{\prime}}(\theta,\varphi)Y_{lm}(\theta,\varphi)\sin\theta{\rm d}\theta{\rm d}\varphi=\delta_{l^{\prime}l}\delta_{m^{\prime}m}.$ (7) where the sign ∗ denotes the complex conjugate. If we multiply equation (5) by the vector operator ${\bf L}$ and take into account the commutation of ${\bf L}$ and ${\bf L}^{2}$, we will obtain that vector ${\bf L}Y_{lm}$ satisfies the same equation (5), too. The quantity defined as ${\bf X}_{lm}=\frac{1}{\sqrt{l(l+1)}}{\bf L}Y_{lm}$ (8) is called vector spherical harmonic. The coefficient before ${\bf L}Y_{lm}$ is chosen so that the orthonormalization condition is of the form $\int_{0}^{\pi}\int_{0}^{2\pi}{\bf X}_{l^{\prime}m^{\prime}}^{\ast}(\theta,\varphi){\bf X}_{lm}(\theta,\varphi)\sin\theta{\rm d}\theta{\rm d}\varphi=\delta_{l^{\prime}l}\delta_{m^{\prime}m}.$ (9) From the self-conjugacy of the angular momentum operator ${\bf L}$ and properties (4) the orthogonality $\displaystyle\int_{0}^{\pi}\int_{0}^{2\pi}{\bf e}_{r}({\bf X}_{l^{\prime}m^{\prime}}^{\ast}\times{\bf X}_{lm})\sin\theta{\rm d}\theta{\rm d}\varphi$ $\displaystyle=$ $\displaystyle\int_{0}^{\pi}\int_{0}^{2\pi}\frac{Y_{l^{\prime}m^{\prime}}^{\ast}{\bf e}_{r}({\bf L}\times{\bf L})Y_{lm}}{\sqrt{l(l+1)l^{\prime}(l^{\prime}+1)}}\sin\theta{\rm d}\theta{\rm d}\varphi$ (10) $\displaystyle=$ $\displaystyle\int_{0}^{\pi}\int_{0}^{2\pi}\frac{{\rm i}Y_{l^{\prime}m^{\prime}}^{\ast}{\bf e}_{r}{\bf L}Y_{lm}}{\sqrt{l(l+1)l^{\prime}(l^{\prime}+1)}}\sin\theta{\rm d}\theta{\rm d}\varphi=0.$ follows. Below we give some properties of vector spherical harmonics: ${\bf L}{\bf X}_{lm}=\sqrt{l(l+1)}Y_{lm},\qquad{\bf L}({\bf e}_{r}\times{\bf X}_{lm})=0.$ (11) Scalar (vector) spherical harmonics satisfy the scalar (vector) equation for eigenfunctions of the squared orbital angular momentum operator ${\bf L}^{2}$. The orthogonality condition for the scalar (7) and vector (9) spherical harmonics have the same form. Therefore, one can hope to combine them into one mathematical object. ## 3 Electrodynamic spherical harmonic: definition and properties We define an electrodynamic spherical harmonic as a second rank tensor in three-dimensional space $F_{lm}=Y_{lm}{\bf e}_{r}\otimes{\bf e}_{r}+{\bf X}_{lm}\otimes{\bf e}_{\theta}+({\bf e}_{r}\times{\bf X}_{lm})\otimes{\bf e}_{\varphi}.$ (12) The first term of (12) determines the longitudinal part of the tensor. It is calculated by means of the scalar spherical function. The last two summands of (12) fix the transverse solution, in the plane ($\theta$, $\varphi$) perpendicular to the direction ${\bf e}_{r}$. It includes vector spherical harmonics. The left and right vectors in dyads form two sets of orthogonal vectors: $(Y_{lm}{\bf e}_{r},{\bf X}_{lm},{\bf e}_{r}\times{\bf X}_{lm})$ and $({\bf e}_{r},{\bf e}_{\theta},{\bf e}_{\varphi})$. ### 3.1 Orthonormalization Multiplying the electrodynamic spherical harmonic by the Hermitian conjugate $F_{lm}^{+}$ we get to $F_{l^{\prime}m^{\prime}}^{+}F_{lm}=Y^{\ast}_{l^{\prime}m^{\prime}}Y_{lm}{\bf e}_{r}\otimes{\bf e}_{r}+{\bf X}^{\ast}_{l^{\prime}m^{\prime}}{\bf X}_{lm}({\bf e}_{\theta}\otimes{\bf e}_{\theta}+{\bf e}_{\varphi}\otimes{\bf e}_{\varphi})+({\bf e}_{r}({\bf X}_{l^{\prime}m^{\prime}}^{\ast}\times{\bf X}_{lm})){\bf e}_{r}^{\times}.$ (13) The quantity before each dyad is orthogonal or normalized as (7), (9), or (10). Therefore, the orthonormalization condition for electrodynamic spherical harmonics is $\int_{0}^{\pi}\int_{0}^{2\pi}F_{lm}^{+}(\theta,\varphi)F_{lm}(\theta,\varphi)\sin\theta{\rm d}\theta{\rm d}\varphi={\bf 1}\delta_{l^{\prime}l}\delta_{m^{\prime}m}.$ (14) Here we consider the vectors ${\bf e}_{r}$, ${\bf e}_{\theta}$, and ${\bf e}_{\varphi}$ to be constant in dyads and dual tensor ${\bf e}_{r}^{\times}$. If tensors $F_{lm}$ are the solutions of equations, we can always write these solutions in components. For each component of the tensor $F_{lm}^{+}F_{lm}$ the orthonormalization is carried out. We will obtain the same, if the dependence of the orts ${\bf e}_{r}$, ${\bf e}_{\theta}$, ${\bf e}_{\varphi}$ on angles in (13) is omitted, i.e. they are regarded as constants. However, when substituting $F_{lm}$ in equations, we should take into account the angular dependence of the orts. ### 3.2 Explicit form Let us substitute the explicit expression for the vector operator ${\bf L}$ ${\bf L}=-{\rm i}{\bf e}_{\varphi}\frac{\partial}{\partial\theta}+{\rm i}\frac{{\bf e}_{\theta}}{\sin\theta}\frac{\partial}{\partial\varphi}$ (15) into equation (12). Then the electrodynamic spherical harmonic is equal to $F_{lm}=\left[{\bf e}_{r}\otimes{\bf e}_{r}+\frac{{\rm i}}{\sin\theta\sqrt{l(l+1)}}\left(I\frac{\partial}{\partial\varphi}-{\bf e}_{r}^{\times}\sin\theta\frac{\partial}{\partial\theta}\right)\right]Y_{lm},$ (16) where $I=-{\bf e}_{r}^{\times}{\bf e}_{r}^{\times}={\bf 1}-{\bf e}_{r}\otimes{\bf e}_{r}$ is the projection operator onto the plane ($\theta$, $\varphi$). To calculate the derivatives one can replace them by means of operators $L_{z}$ and $L_{\pm}=L_{x}\pm L_{y}$ as $\frac{\partial}{\partial\varphi}={\rm i}L_{z},\qquad\frac{\partial}{\partial\theta}=\frac{1}{2}({\rm e}^{-{\rm i}\varphi}L_{+}-{\rm e}^{{\rm i}\varphi}L_{-}),$ because their action on the scalar spherical harmonic is well-known: $L_{+}Y_{lm}=\sqrt{(l-m)(l+m+1)}Y_{l,m+1},\qquad L_{-}Y_{lm}=\sqrt{(l+m)(l-m+1)}Y_{l,m-1}.$ (17) Hence, the electrodynamic spherical harmonic can be presented as follows $F_{lm}=\left[{\bf e}_{r}\otimes{\bf e}_{r}-\frac{1}{\sin\theta\sqrt{l(l+1)}}\left(IL_{z}+{\bf e}_{r}^{\times}\frac{{\rm i}\sin\theta}{2}({\rm e}^{-{\rm i}\varphi}L_{+}-{\rm e}^{{\rm i}\varphi}L_{-})\right)\right]Y_{lm}.$ (18) It is easy to exclude the operators from (18). The final formula for tensor $F_{lm}$ contains scalar spherical harmonics as angle dependence. Unit vectors ${\bf e}_{r}$, ${\bf e}_{\theta}$, ${\bf e}_{\varphi}$ determine the structure of the tensor in three-dimensional space. $F_{lm}$ is formed by three basic tensors: ${\bf e}_{r}\otimes{\bf e}_{r}$, ${\bf e}_{r}^{\times}$, and $I$. Hence, it commutes with each of these tensors. ### 3.3 Invariants The first invariant of the electrodynamic spherical harmonic as tensor quantity is its trace. The trace of the tensor (12) equals ${\rm tr}(F_{lm})=Y_{lm}+2({\bf e}_{\theta}{\bf X}_{lm}).$ (19) The second invariant is determinant $\det(F_{lm})=Y_{lm}{\bf X}_{lm}^{2}.$ (20) The third invariant of three-dimensional tensor $F_{lm}$ is the trace of the adjoint tensor $\overline{F}_{lm}$. Adjoint tensor is defined by $\overline{F}_{lm}F_{lm}=F_{lm}\overline{F}_{lm}=\det(F_{lm}){\bf 1}$ and equals $\overline{F}_{lm}={\bf X}_{lm}^{2}{\bf e}_{r}\otimes{\bf e}_{r}+Y_{lm}{\bf e}_{\theta}\otimes{\bf X_{lm}}+Y_{lm}{\bf e}_{\varphi}\otimes{{\bf e}_{r}\times\bf X_{lm}}.$ (21) Further the trace is easily calculated: ${\rm tr}(\overline{F}_{lm})={\bf X}_{lm}^{2}+2Y_{lm}({\bf e}_{\theta}{\bf X_{lm}}).$ (22) Using these three invariants one can find other ones. For example, the trace of squared tensor is determined from equation ${\rm tr}(F_{lm}^{2})=({\rm tr}(F_{lm}))^{2}-2{\rm tr}(\overline{F}_{lm})$. ### 3.4 Generalization of electrodynamic spherical harmonic The main condition on electrodynamic spherical harmonic is the orthonormalization (14). There is more general form of the second rank tensor spherical harmonic satisfying this equation. It is $G_{lm}=Y_{lm}{\bf e}_{r}\otimes{\bf a}+{\bf X}_{lm}\otimes{\bf b}+({\bf e}_{r}\times{\bf X}_{lm})\otimes{\bf c}.$ (23) Unit vectors ${\bf a}$, ${\bf b}$, and ${\bf c}$ form the orthogonal basis in three-dimensional space. In equation (12) we have assumed ${\bf a}={\bf e}_{r}$, ${\bf b}={\bf e}_{\theta}$, ${\bf c}={\bf e}_{\varphi}$ because of the spherical symmetry of the function. If we will take ${\bf a}={\bf e}_{r}$, ${\bf b}={\bf e}_{\theta}$, ${\bf c}=-{\bf e}_{\varphi}$, then the electrodynamic spherical harmonic becomes more complex function in explicit form, however its invariants are simplified (e.g., the trace equals ${\rm tr}(F_{lm})=Y_{lm}$). ## 4 Solution of Maxwell’s equations In this section we will find the spherically symmetric solutions (i.e. electric ${\bf E}$ and magnetic ${\bf H}$ fields) of the Maxwell equations $\nabla\times{\bf E}({\bf r})={\rm i}k\mu{\bf H}({\bf r}),\qquad\nabla\times{\bf H}({\bf r})=-{\rm i}k\varepsilon{\bf E}({\bf r})$ (24) for the monochromatic electromagnetic waves in isotropic medium with dielectric permittivity $\varepsilon$ and magnetic permeability $\mu$. The quantity $k=\omega/c$ is called wavenumber in vacuum, and $\omega$ is the wave frequency. Arbitrary time dependence can be obtained by using the linear superposition of monochromatic waves ${\bf E}({\bf r})\exp(-{\rm i}\omega t)$. We will search the solution in the form ${\bf E}({\bf r})=F_{lm}(\theta,\varphi){\bf E}^{l}(r).$ (25) The components of the vector ${\bf E}^{l}$ in spherical coordinates depend only on the radial coordinate $r$. The dependence on the angle coordinates presents only in the basis vectors. So, the vector ${\bf E}^{l}$ looks like ${\bf E}^{l}(r)=E^{l}_{r}(r){\bf e}_{r}+E^{l}_{\theta}(r){\bf e}_{\theta}+E^{l}_{\varphi}(r){\bf e}_{\varphi}.$ (26) Further we should calculate ${\rm rot}{\bf E}$. By substituting Hamilton’s operator (2) one obtains $\nabla\times{\bf E}={\bf e}_{r}^{\times}\frac{\partial{\bf E}}{\partial r}-\frac{{\rm i}}{r}{\bf L}({\bf e}_{r}{\buildrel\downarrow\over{{\bf E}}})+\frac{{\rm i}}{r}{\bf e}_{r}({\bf L}{\bf E}),$ (27) where the arrow $\downarrow$ implies that operator ${\bf L}$ acts only on vector ${\bf E}$, but not ${\bf e}_{r}$. Let us calculate each summand of equation (27) applying the solution (25). The first term is of the form ${\bf e}_{r}^{\times}\frac{\partial{\bf E}}{\partial r}=F_{lm}{\bf e}_{r}^{\times}\frac{\partial{\bf E}^{l}}{\partial r},$ (28) where the commutation relation $[F_{lm},{\bf e}_{r}^{\times}]=0$ is taken into account. The second summand yields ${\bf L}({\bf e}_{r}{\buildrel\downarrow\over{{\bf E}}})={\bf L}({\bf e}_{r}{\bf E})-{\bf L}({\buildrel\downarrow\over{{\bf e}_{r}}}{\bf E})={\bf L}(Y_{lm}E^{l}_{r})-{\bf L}({\buildrel\downarrow\over{{\bf e}_{r}}}{\bf E}).$ (29) The quantity ${\bf L}({\buildrel\downarrow\over{{\bf e}_{r}}}{\bf E})$ is easily calculated using the explicit expression (15) of the operator ${\bf L}$ and the relationships $\partial{\bf e}_{r}/\partial\theta={\bf e}_{\theta}$ and $\partial{\bf e}_{r}/\partial\varphi={\bf e}_{\varphi}\sin\theta$: ${\bf L}({\buildrel\downarrow\over{{\bf e}_{r}}}{\bf E})=-{\rm i}{\bf e}_{r}^{\times}{\bf E}.$ (30) So, we get the formula ${\bf L}({\bf e}_{r}{\buildrel\downarrow\over{{\bf E}}})=F_{lm}\left(\sqrt{l(l+1)}{\bf e}_{\theta}\otimes{\bf e}_{r}+{\rm i}{\bf e}_{r}^{\times}\right){\bf E}^{l}.$ (31) The third term in (27) can be rewritten using the equation (11): ${\bf e}_{r}({\bf L}{\bf E})={\bf e}_{r}\left(E^{l}_{r}{\bf L}(Y_{lm}{\bf e}_{r})+E^{l}_{\theta}{\bf L}{\bf X}_{lm}+E^{l}_{\varphi}{\bf L}({\bf e}_{r}\times{\bf X}_{lm})\right)=\sqrt{l(l+1)}F_{lm}({\bf e}_{r}\otimes{\bf e}_{\theta}){\bf E}^{l}.$ (32) Finally, the curl of the electric field vector ${\bf E}$ equals $\nabla\times{\bf E}=F_{lm}(\theta,\varphi)\left({\bf e}_{r}^{\times}\frac{\partial}{\partial r}+\frac{1}{r}{\bf e}_{r}^{\times}-\frac{{\rm i}\sqrt{l(l+1)}}{r}{\bf e}_{\varphi}^{\times}\right){\bf E}^{l}(r).$ (33) In expression (33) we have took into account the derivatives on the angle variables. Therefore, further the orts of spherical coordinates ${\bf e}_{r}$, ${\bf e}_{\theta}$, and ${\bf e}_{\varphi}$ can be considered as constants. The Maxwell equations are reduced to the set of ordinary differential equations $\displaystyle{\bf e}_{r}^{\times}\frac{{\rm d}{\bf E}^{l}}{{\rm d}r}+\frac{1}{r}{\bf e}_{r}^{\times}{\bf E}^{l}-\frac{{\rm i}\sqrt{l(l+1)}}{r}{\bf e}_{\varphi}^{\times}{\bf E}^{l}={\rm i}k\mu{\bf H}^{l},$ $\displaystyle{\bf e}_{r}^{\times}\frac{{\rm d}{\bf H}^{l}}{{\rm d}r}+\frac{1}{r}{\bf e}_{r}^{\times}{\bf H}^{l}-\frac{{\rm i}\sqrt{l(l+1)}}{r}{\bf e}_{\varphi}^{\times}{\bf H}^{l}=-{\rm i}k\varepsilon{\bf E}^{l}.$ (34) Equations (34) allow to determine the radial dependence of the fields. Multiplying the set of equations (34) by the unit vector ${\bf e}_{r}$ we can express the longitudinal components of the fields as follows $E^{l}_{r}=-\frac{\sqrt{l(l+1)}}{\varepsilon kr}H^{l}_{\theta},\qquad H^{l}_{r}=\frac{\sqrt{l(l+1)}}{\mu kr}E^{l}_{\theta}.$ (35) The tangential field components ${\bf E}^{l}_{\rm t}=I{\bf E}^{l}=E^{l}_{\theta}{\bf e}_{\theta}+E^{l}_{\varphi}{\bf e}_{\varphi}$ and ${\bf H}^{l}_{\rm t}=I{\bf H}^{l}$ are determined from the equations which can be written in matrix form: $\frac{{\rm d}(r{\bf W})}{{\rm d}r}=ikM(r{\bf W}),$ (36) where $\displaystyle{\bf W}=\left(\begin{array}[]{c}{\bf H}^{l}_{\rm t}\\\ {\bf E}^{l}_{\rm t}\end{array}\right),\qquad M=\left(\begin{array}[]{cc}0&\varepsilon A\\\ -\mu A&0\end{array}\right),\qquad A={\bf e}_{r}^{\times}-\frac{l(l+1)}{\varepsilon\mu k^{2}r^{2}}{\bf e}_{\varphi}\otimes{\bf e}_{\theta}.$ (41) Equation (36) is satisfied for inhomogeneous media $\varepsilon(r)$, $\mu(r)$, too. Such matrix equation can be solved numerically for arbitrary medium, or analytically for homogeneous one. Let us find tangential field components ${\bf W}$ when $\varepsilon=const$ and $\mu=const$. The simplest way is to write the equation for projection $W_{\theta}=(H^{l}_{\theta},E^{l}_{\theta})$: $\frac{{\rm d}^{2}(rW_{\theta})}{{\rm d}r^{2}}+\left(k^{2}\varepsilon\mu-\frac{l(l+1)}{r^{2}}\right)(rW_{\theta})=0,$ (42) The solutions of the equation (42) are well-known and can be presented as $W_{\theta}=f^{(1)}\left(\begin{array}[]{c}c_{1}\\\ c^{\prime}_{1}\end{array}\right)+f^{(2)}\left(\begin{array}[]{c}c_{2}\\\ c^{\prime}_{2}\end{array}\right)=(f^{(1)}{\bf c}_{1}+f^{(2)}{\bf c}_{2})\left(\begin{array}[]{c}{\bf e}_{\theta}\\\ {\bf e}_{\varphi}\end{array}\right),$ (43) where ${\bf c}_{1}$ and ${\bf c}_{2}$ are constant vectors. The couples of independent solutions are spherical Bessel functions $f^{(1)}=j_{l}(nkr)$, $f^{(2)}=j_{-l-1}(nkr)$ or spherical Hankel functions of the first and second kind $f^{(1)}=h_{l}^{(1)}(nkr)$, $f^{(2)}=h_{l}^{(2)}(nkr)$. $n=\sqrt{\varepsilon\mu}$ is the refractive index. After determining the $\varphi$-projections of the fields as $W_{\varphi}=\frac{{\rm i}}{kr\varepsilon\mu}\left(\begin{array}[]{cc}0&-\varepsilon\\\ \mu&0\end{array}\right)\frac{{\rm d}(rW_{\theta})}{{\rm d}r}$ (44) one can write the transverse vector field ${\bf W}(r)=\left(\begin{array}[]{cc}\eta_{1}(r)&\eta_{2}(r)\\\ \zeta_{1}(r)&\zeta_{2}(r)\end{array}\right)\left(\begin{array}[]{c}{\bf c}_{1}\\\ {\bf c}_{2}\end{array}\right),$ (45) where $\displaystyle\eta_{1,2}=f^{(1,2)}{\bf e}_{\theta}\otimes{\bf e}_{\theta}-\frac{{\rm i}}{\mu kr}\frac{{\rm d}(rf^{(1,2)})}{{\rm d}r}{\bf e}_{\varphi}\otimes{\bf e}_{\varphi},$ $\displaystyle\zeta_{1,2}=f^{(1,2)}{\bf e}_{\theta}\otimes{\bf e}_{\varphi}+\frac{{\rm i}}{\varepsilon kr}\frac{{\rm d}(rf^{(1,2)})}{{\rm d}r}{\bf e}_{\varphi}\otimes{\bf e}_{\theta}.$ (46) Tangential field vector ${\bf W}$ plays an important part, because it is continuous on the spherical surface. That is why it can be applied for the study of electromagnetic wave diffraction by a sphere. ## 5 Conclusion Thus, the general solution of the Maxwell equations is of the form $\left(\begin{array}[]{cc}{\bf H}({\bf r})\\\ {\bf E}({\bf r})\end{array}\right)=\sum_{l=0}^{\infty}\sum_{m=-l}^{l}F_{lm}(\theta,\varphi)V^{l}(r)\left(\begin{array}[]{cc}\eta^{l}_{1}(r)&\eta^{l}_{2}(r)\\\ \zeta^{l}_{1}(r)&\zeta^{l}_{2}(r)\end{array}\right)\left(\begin{array}[]{c}{\bf c}^{l}_{1}\\\ {\bf c}^{l}_{2}\end{array}\right),$ (47) where $V^{l}$ is the matrix that takes into account the longitudinal components of electric and magnetic fields. This matrix can be easily calculated from the equation (35). In each partial solution included into the general one (47) the radial and angle variables are separated. Using the orthonormalization (14) for the electrodynamic spherical harmonic $F_{lm}$, each partial wave can be easily singled out: $V^{l}(r)\left(\begin{array}[]{cc}\eta^{l}_{1}(r)&\eta^{l}_{2}(r)\\\ \zeta^{l}_{1}(r)&\zeta^{l}_{2}(r)\end{array}\right)\left(\begin{array}[]{c}{\bf c}^{l}_{1}\\\ {\bf c}^{l}_{2}\end{array}\right)=\int_{0}^{\pi}\int_{0}^{2\pi}F^{+}_{lm}(\theta^{\prime},\varphi^{\prime})\left(\begin{array}[]{cc}{\bf H}(r,\theta^{\prime},\varphi^{\prime})\\\ {\bf E}(r,\theta^{\prime},\varphi^{\prime})\end{array}\right)\sin\theta^{\prime}{\rm d}\theta^{\prime}{\rm d}\varphi^{\prime}.$ (48) This property of the electrodynamic spherical harmonic is very useful for the investigation of electromagnetic wave scattering. In scattering, the boundary condition is the single equation for tangential fields ${\bf W}$. It is easily solved, if the orthonormalization is applied. Some attempts of investigation of scattering in the similar manner as described above have been made in [7]. In the general solution (47) the constants ${\bf c}_{1,2}$ determined by initial conditions are explicitly shown. Vectors ${\bf c}_{1}$ and ${\bf c}_{2}$ set independent solutions. For instance, if ${\bf c}_{1}=0$, then the radial dependence is determined by the function $f^{(2)}(r)$, and vice versa. If ${\bf c}_{2}=0$ and $f^{(1)}(r)=h^{(1)}(nkr)$, then the field (47) is the multipole expansion [8]. The amplitude of electric multipole field is equal to $a_{E}(l,m)={\b{c}}_{1}{\bf e}_{\theta}$, the amplitude of magnetic multipole field is equal to $a_{M}(l,m)={\bf c}_{1}{\bf e}_{\varphi}$. So, vector ${\bf c}_{1}$ can be called vector amplitude of multipole fields. In further investigations we will study the scattering and multipole expansion of electromagnetic fields in details. ## References ## References * [1] Newman E and Penrose R 1966 Note on the Bondi-Metzner-Sachs group J. Math. Phys. 7 836–870 * [2] James R W 1976 New Tensor Spherical Harmonics, for Application to the Partial Differential Equations of Mathematical Physics Phil. Trans. R. Soc. London. Series A 281 195–221 * [3] Sandberg V D 1978 Tensor spherical harmonics on $S^{2}$ and $S^{3}$ as eigenvalue problems J. Math. Phys. 19 2441–2446 * [4] Winter J 1982 Tensor spherical harmonics Letters in Mathematical Physics 6 91–96 * [5] Kostelec P, Maslen D K, Rockmore D N, and Healy Jr D 2000 Computational harmonic analysis for tensor fields on the two-sphere J. Comput. Phys. 162 514–535 * [6] Fedorov F I 1976 Theory of Gyrotropy (Minsk: Nauka i Tehnika) * [7] Novitsky A V and Barkovsky L M 2006 Evolution and impedance operators of spherically symmetric bianisotropic media J. Phys. A: Math. Gen. 39 7543-60 * [8] Jackson J D 1998 Classical electrodynamics (New York: John Wiley & Sons)

arxiv-papers

2024-09-04T02:48:54.588747

0803.4129

# Regularity of conjugacies of algebraic actions of Zariski dense groups Alexander Gorodnik, Theron Hitchman, Ralf Spatzier School of Mathematics University of Bristol Bristol BS8 1TW, U.K. a.gorodnik@bristol.ac.uk Department of Mathematics University of Northern Iowa Cedar Falls, IA 50614-0506 theron.hitchman@uni.edu Department of Mathematics University of Michigan Ann Arbor, MI 48109-1043 spatzier@umich.edu ###### Abstract. Let $\alpha_{0}$ be an affine action of a discrete group $\Gamma$ on a compact homogeneous space $X$ and $\alpha_{1}$ a smooth action of $\Gamma$ on $X$ which is $C^{1}$-close to $\alpha_{0}$. We show that under some conditions, every topological conjugacy between $\alpha_{0}$ and $\alpha_{1}$ is smooth. In particular, our results apply to Zariski dense subgroups of $\hbox{SL}_{d}(\mathbb{Z})$ acting on the torus $\mathbb{T}^{d}$ and Zariski dense subgroups of a simple noncompact Lie group $G$ acting on a compact homogeneous space $X$ of $G$ with an invariant measure. A.G. was supported by NSF grants DMS 0400631, 0654413 and RCUK Fellowship, R.S. was supported by NSF Grant DMS 0604857 ###### Contents 1. 1 Introduction 1. 1.1 Acknowledgement 2. 2 Main result 3. 3 Proof of the Main Theorem 1. 3.1 $C^{0}$ implies Hölder 2. 3.2 Invariance of fast stable manifolds 3. 3.3 Convergence of the sequences $f^{-n}gf^{n}$ 4. 3.4 Hölder implies $C^{1}$ along fast stable manifolds 5. 3.5 Completion of the proof of the main theorem 4. 4 Existence of good pairs 1. 4.1 Tori 2. 4.2 Semisimple groups ## 1\. Introduction The investigation of rigidity properties has been at the forefront of research in dynamics in the past two decades. Of particular interest has been the study of higher rank abelian groups and local rigidity of their actions by Hurder, Katok, Lewis, and the last author amongst others. Remarkably, many such actions cannot be perturbed at all, in the sense that any $C^{1}$-close perturbation is $C^{\infty}$-conjugate to the original action. Critically, these groups contain higher rank abelian groups. Similar results were found for higher rank semisimple Lie groups and their lattices by Hurder, Lewis, Fisher, Margulis, Qian and others. We refer to [5] for a more extensive survey of these developments. Smoothness of the conjugacy for these actions came as quite a surprise. Classically, in fact, the stability results of Anosov and later Hirsch, Pugh and Shub guaranteed a continuous conjugacy or orbit equivalence between a single Anosov diffeomorphism or flow and their perturbations [13]. Simple examples however show that such a conjugacy cannot be even $C^{1}$ in general. In the present paper, we investigate similar regularity phenomena for affine actions of a large class of groups. Notably, our results do not require the presence of higher rank subgroups or any assumptions on the structure of the group. In particular they hold for discrete subgroups of rank one semisimple groups. We recall that a group acts affinely on a homogeneous space $H/\Lambda$ for $H$ a Lie group and $\Lambda$ a discrete subgroup if every element acts by an affine diffeomorphism i.e. one which lifts to a composite of a translation and an automorphism on $H$. We denote by $\hbox{\rm Diff}(X)$ the group of $C^{\infty}$-diffeomorphisms of a space $X$. For simplicity let us mention two corollaries of our main theorem in Section 2. ###### Theorem 1.1. Let $\Gamma\subset\hbox{\rm SL}_{d}(\mathbb{Z})$ for $d\geq 2$ be a finitely generated Zariski dense subgroup in $\hbox{\rm SL}_{d}({\mathbb{R}})$, and $\alpha_{0}$ the associated action on the $d$-torus $\mathbb{T}^{d}$. If a perturbation $\alpha_{1}:\Gamma\to\hbox{\rm Diff}(\mathbb{T}^{d})$ is sufficiently $C^{1}$-close to $\alpha_{0}$, then any $C^{0}$-conjugacy $\Phi:\mathbb{T}^{d}\mapsto\mathbb{T}^{d}$ between $\alpha_{0}$ and $\alpha_{1}$ is a $C^{\infty}$-diffeomorphism. For $d=2$, E. Cawley found a $C^{1+\alpha}$-regularity result for Zariski- dense subgroups of $\hbox{\rm SL}_{2}(\mathbb{Z})$ acting on the 2-torus in [4] in the early 1990’s. Her techniques however are restricted to the 2-torus due to the use of $C^{1+\alpha}$-regularity of stable foliations. Subsequently, the second author obtained a general $C^{\infty}$-regularity theorem for groups acting on general tori in his thesis [14]. A second application of our main theorem to actions on homogeneous spaces of semisimple groups is novel. ###### Theorem 1.2. Let $G$ be a connected simple noncompact Lie group, $\Lambda$ a cocompact lattice in $G$, and $\Gamma$ a finitely generated Zariski dense subgroup of $G$. Let $\alpha_{0}$ be the affine action of $\Gamma$ on $G/\Lambda$. If a $C^{\infty}$-action $\alpha_{1}$ is sufficiently $C^{1}$-close to $\alpha_{0}$, then any $C^{0}$-conjugacy $\Phi:G/\Lambda\mapsto G/\Lambda$ is a $C^{\infty}$-diffeomorphism. Let us note that our techniques are based on certain mixing properties of the actions and do not allow the treatment of actions on general nilmanifolds. Fisher and Hitchman recently proved a local rigidity theorem for actions of lattices with the Kazhdan property [8]. We recall that an action $\alpha$ is called $C^{k,l}$-rigid if any $C^{k}$-close perturbation of the action is $C^{l}$-conjugate to $\alpha$. ###### Theorem 1.3 (Fisher-Hitchman). Let $\Gamma$ be a lattice in a semisimple Lie group without compact factors which satisfies Kazhdan’s property. Then any affine action $\alpha$ of $\Gamma$ is $C^{3,0}$-locally rigid. Fisher and Hitchman actually prove this for _quasi-affine_ actions, which are extensions of affine actions by isometries. Their technique is based on a type of heat flow. If $\alpha$ does not admit a common neutral direction, then Fisher and Hitchman’s proof yields $C^{1,0}$-local rigidity. Using our regularity result, we immediately obtain ###### Corollary 1.4. Let $G$ be a simple noncompact Lie group which satisfies Kazhdan’s property, $\Gamma$ a lattice in $G$, and $X$ a compact homogeneous space of $G$ supporting an invariant measure. Then the affine action of $\Gamma$ on $X$ is $C^{1,\infty}$-locally rigid. ###### Remark 1.5. We can also deduce $C^{1,\infty}$-local rigidity for the action of a Kazhdan lattice $\Gamma$, embedded in $\hbox{SL}_{d}(\mathbb{Z})$, on the torus $\mathbb{T}^{d}$ under the assumption that $\Gamma\times\Gamma$ is not contained in the subvarieties $\det([X^{\ell},Y]-id)=0$, $\ell\geq 1$, $\phi(\ell)\leq d^{2}$, where $\phi$ is the Euler totient function (see Lemma 4.2). This assumption is needed to construct good pairs in $\Gamma$ (see Definition 2.1). Fisher and Hitchman proved $C^{\infty,\infty}$-local rigidity for a more general class of actions of cocompact lattices in the same groups [8]. In particular their approach works on nilmanifolds. At the heart of our argument lies the investigation of sequences of the form $\gamma^{-n}\delta\gamma^{n}$ for two hyperbolic elements $\gamma$ and $\delta$ in “general position”. Such elements always exist in Zariski-dense groups. The behavior of these sequences is badly divergent in directions transverse to the fast stable direction of $\gamma$, and cannot be controlled. However, these sequences do converge along the fast stable manifolds of $\gamma$. This is elementary for an affine action. We prove $C^{1}$-convergence for the perturbed action. These limiting maps along the fast stable foliation of $\gamma$ form a rich system which acts transitively along the fast stable leaves under suitable conditions. Moreover, the conjugacy $\Phi$ between the actions will also intertwine these limiting maps along fast stables. It follows that $\Phi$ has to be $C^{1}$ along each of these fast stable manifolds. We prove smoothness in a separate argument. The proof of $C^{1}$-convergence is technically the most difficult piece of the argument. It requires careful estimates which are an adaptation of the proof of Livsic’ theorem for cocycles with non-abelian targets. The use of sequences of the form $\gamma^{-n}\delta\gamma^{n}$ was introduced by Hitchman in his thesis [14]. His argument relied on the idea that the resulting limit maps along fast stable leaves often exhibit higher rank abelian behavior which could then be used to prove regularity similar to the case of actions by higher rank abelian groups. Let us comment that our arguments seem to be of rather general nature. In the weakly hyperbolic setting, the hard part in proving local rigidity results lies in getting a $C^{0}$-conjugacy. Indeed, the common strategy for most of the known local rigidity results has been to show existence of a $C^{0}$-conjugacy and then improve the regularity. Margulis–Qian in higher rank and Fisher-Hitchman for all Kazhdan Lie groups have the most extensive results [19, 9]. The current paper shows regularity under rather general conditions, reducing smooth local rigidity to continuous local rigidity. To pinpoint precisely when local rigidity holds appears difficult. On the one hand, we have the results above for actions of lattices in the Kazhdan rank one groups. On the other hand, Fisher found non-trivial affine deformations of actions of lattices in $SO(n,1)$ resulting from “bending lattices” [6, 7]. Finally, if the action has isometric directions, even regularity becomes difficult as evidenced even in higher rank by the works of Fisher and Margulis [10] and Fisher and Hitchman [8]. ### 1.1. Acknowledgement We would like to thank R. Feres, A. Gogolev, J. Heinonen, B. Kalinin, and B. Schmidt for useful discussions. We also would like to thank a referee for careful reading and for pointing out some deficiencies in our original proofs. A.G. would like to express his thanks for hospitality to Princeton University, where part of this work was completed. ## 2\. Main result Let $G$ be a connected Lie group, $\Lambda$ a cocompact lattice in $G$, and $X=G/\Lambda$. The space $X$ is equipped with a finite invariant Radon measure. The group $\hbox{\rm Aff}(X)$ of affine transformations of $X$ consists of maps of the form $f:x\mapsto L_{g}\circ a(x),\quad x\in X,$ where $L_{g}$ denotes the left mutiplication action of $g\in G$ and $a$ is an automorphism of $G$ preserving $\Lambda$. Every such map $f$ preserves the measure and defines an automorphism $Df$ of $\hbox{Lie}(G)\simeq T_{e\Lambda}(X)$ given by $Df:=\hbox{Ad}(g)\circ D(a)_{e}.$ We denote by $W^{min}_{f}$ the sum of the generalized eigenspaces of $Df$ with eigenvalues of minimal modulus and by $P^{min}_{f}:\hbox{Lie}(G)\to W^{min}_{f}$ the projection map along the other generalized eigenspaces. ###### Definition 2.1. We call a pair $f,g\in\hbox{\rm Aff}(X)$ good if the following conditions are satisfied: 1. (i) The map $Df$ is partially hyperbolic. 2. (ii) The map $Df:W^{min}_{f}\to W^{min}_{f}$ is semisimple. 3. (iii) The map $P^{min}_{f}\circ Dg:W^{min}_{f}\to W^{min}_{f}$ is nondegenerate. 4. (iv) For every subsequence $\\{n_{i}\\}$ , the sequence $\\{f^{-n_{i}}gf^{n_{i}}(x)\\}$ is dense in $X$ for $x$ in a set of full measure. If for $f\in\hbox{\rm Aff}(X)$, there exists $g\in\hbox{\rm Aff}(X)$ so that the pair $f,g$ is good, we say $f$ is a good mapping. ###### Remark 2.2. In the case when the map $Df:W^{min}_{f}\to W^{min}_{f}$ does not have a rotation component of infinite order (e.g., when $\dim W^{min}_{f}=1$), it suffices to assume that the sequence $\\{f^{-n}gf^{n}(x)\\}$ is dense in $X$ for $x$ in a set of full measure. In general, we have to pass to a subsequence to guarantee that the maps $f^{-n}gf^{n}$ converge along the fast stable leaves as $n\to\infty$ (see Proposition 3.13). The theorems stated in the introduction will be deduced from the following general result: Main Theorem. Let $\Gamma$ be a finitely generated discrete group and $\alpha_{0}:\Gamma\to\hbox{\rm Aff}(X)$ an affine action of $\Gamma$ such that * • $(D\alpha_{0})(\Gamma)$ acts irreducibly on $\hbox{\rm Lie}(G)$, * • $\alpha_{0}(\Gamma)$ contains a good pair. Let $\alpha_{1}:\Gamma\to\hbox{\rm Diff}(X)$ be a $C^{\infty}$-action of $\Gamma$ which is sufficiently $C^{1}$-close to $\alpha_{0}$. Then every homeomorphism $\Phi:X\to X$ satisfying $\Phi\circ\alpha_{0}(\gamma)=\alpha_{1}(\gamma)\circ\Phi\quad\hbox{for all $\gamma\in\Gamma$}$ is a $C^{\infty}$-diffeomorphism. ###### Remark 2.3. Irreducibility of the action of $\Gamma$ on $\hbox{Lie}(G)$ is used in the following places: * • In Section 3.1, to deduce weak hyperbolicity (see (1)), * • In Section 3.2, to construct essential sets (see Lemma 3.9), * • In Section 3.5, to deduce that $\Phi$ is $C^{\infty}$ from smoothness on subspaces of the fast stable leaves (see (51)). Existence of good pairs for some classes of affine actions will be proved in Section 4. In particular, Theorem 1.1 follows from the Main Theorem and Proposition 4.1, and Theorem 1.2 follows from the Main Theorem and Proposition 4.4. ###### Outline of the proof of the Main Theorem. Irreducibility of $\Gamma$-action and property (i) of a good pair are used to prove that $\Phi$ is bi-Hölder (Section 3.1). Next, irreducibility of the $\Gamma$-action and property (ii) of a good pair are used to show that $\Phi$ maps fast stable manifolds to fast stable manifolds (Section 3.2). Property (ii) is also used to show that a subsequence of maps $f^{-n}gf^{n}$ restricted to fast stable manifolds is precompact in the $C^{0}$-topology and, in fact, in the $C^{1}$-topology (Section 3.3). Then one utilizes property (iii) of a good pair to deduce that the limits of these maps are homeomorphisms and property (iv) of a good pair to deduce that these limits generate transitive $C^{1}$-action on fast stable manifolds. Using that $\Phi$ is a conjugacy between the constructed $C^{1}$-actions, we show that $\Phi$ is $C^{1}$ along the fast stable leaves (Section 3.4). A more elaborate argument, which is based on the nonstationary Sternberg linearization [16, 12, 11], shows that $\Phi$ is $C^{\infty}$ along some subspaces of fast stable leaves. Finally, we deduce that $\Phi$ is $C^{\infty}$ on $X$ from elliptic regularity using irreducibility of the $\Gamma$-action (Section 3.5). ∎ ## 3\. Proof of the Main Theorem We continue with the notation that $X=G/\Lambda$ is a compact quotient of a connected Lie group $G$ by a discrete subgroup $\Lambda\subset G$. ### 3.1. $C^{0}$ implies Hölder In this section, we will prove that the conjugacy map $\Phi:X\to X$ in the Main Theorem is bi-Hölder. The proof is similar to Proposition 5.7 of [19]. As they do not show that their map is Hölder, and also use somewhat different hypotheses, we will give a proof here for simplicity. Following [19], we say that a $C^{1}$-action $\alpha$ of a discrete group $\Gamma$ on a compact manifold $M$ is _weakly hyperbolic_ when there is a choice of finitely many elements $\gamma_{1},\ldots,\gamma_{k}$ in $\Gamma$ such that each diffeomorphism $\alpha(\gamma_{i})$ is partially hyperbolic and, for each point $x\in M$, (1) $\sum_{i=1}^{k}T_{x}W^{s}_{\alpha(\gamma_{i})}(x)=T_{x}M,$ where $W^{s}_{\alpha(\gamma_{i})}(x)$ denotes the stable manifold of $\alpha(\gamma_{i})$ through $x$. ###### Theorem 3.1. Let $\Gamma$ be a finitely generated discrete group, $\alpha_{0}:\Gamma\to\hbox{\rm Aff}(X)$ be an affine weakly hyperbolic action, and $\alpha_{1}:\Gamma\to\hbox{\rm Diff}^{1}(X)$ a smooth action which is sufficiently $C^{1}$-close to $\alpha_{0}$. Then every homeomorphism $\Phi:X\to X$ such that $\Phi\circ\alpha_{0}(\gamma)=\alpha_{1}(\gamma)\circ\Phi\quad\hbox{for all $\gamma\in\Gamma$}$ is bi-Hölder. The proof is divided into several lemmas. ###### Lemma 3.2. Let $f_{1},\ldots,f_{k}$ be partially hyperbolic diffeomorphisms of $X$ such that $\sum_{i=1}^{k}T_{x}W^{s}_{f_{i}}(x)=T_{x}M\quad\hbox{for all $x\in X$,}$ and $g_{1},\ldots,g_{k}$ are $C^{1}$-close $C^{1}$-diffeomorphisms. Then $g_{i}$’s are partially hyperbolic and $\sum_{i=1}^{k}T_{x}W^{s}_{g_{i}}(x)=T_{x}M\quad\hbox{for all $x\in X$}.$ Lemma 3.2 follows from stability of partial hyperbolicity under perturbations (see, for example, [20, Lemma 3.5]). ###### Lemma 3.3. Let $\Phi$ be a continuous conjugacy between two partially hyperbolic diffeomorphisms of a compact manifold. Then $\Phi$ is bi-Hölder continuous along the stable manifolds of these mappings. Lemma 3.3 follows from the standard argument as in [15, Theorem 19.1.2]. ###### Lemma 3.4. Let $\alpha:\Gamma\to\hbox{\rm Diff}^{1}(X)$ be a smooth weakly hyperbolic action and $\gamma_{1},\ldots,\gamma_{k}\in\Gamma$ satisfy (1). Then there exist $c,\epsilon>0$ such that for every $x,y\in X$ satisfying $d(x,y)<\epsilon$, there exists a path $\ell$ from $x$ to $y$ which consists of $2k$ pieces contained in stable manifolds of $\alpha(\gamma_{1}),\ldots,\alpha(\gamma_{k})$, and $L(\ell)\leq c\,d(x,y)$. ###### Proof. We will use an argument similar to [22, Lemma 3.1]. Let $d=\dim X$. There exists a family of (global) continuous unit vector fields $v_{1},\ldots,v_{d}$ that span the tangent space at every point and for some $1=d_{0}\leq d_{1}\leq\cdots\leq d_{k}=d+1$ and every $i=1,\ldots,k$, the vectors $v_{d_{i-1}},\ldots,v_{d_{i}-1}$ are contained in the stable distribution of $\alpha(\gamma_{i})$. Let $\delta>0$. There exists $\delta^{\prime}>0$ such that $d(u,w)<\delta^{\prime}$ implies that $d(v_{i}(u),v_{i}(w))<\delta$ for all $i$. By [18, Corollary 4.5], for every $x\in X$, there exists $\epsilon(x)>0$ such that every $y\in B_{\epsilon(x)}(x)$ can be connected to $x$ by a path $\ell$ of length at most $\delta^{\prime}/2$, and for some $0=t_{0}\leq t_{1}\leq\cdots\leq t_{d}=L(\ell)$, we have $\ell^{\prime}(t)=v_{i}(\ell(t))$ when $t\in[t_{i-1},t_{i})$. Let $\epsilon>0$ be the Lebesgue number of the cover $\\{B_{\epsilon(x)}(x)\\}$. Then every $y_{1},y_{2}\in X$ such that $d(y_{1},y_{2})<\epsilon$ are connected by a path $\ell$ which consists of $2k$ pieces tangent to $v_{j}$’s and $L(\ell)<\delta^{\prime}$. To estimate the distance $d(y_{1},y_{2})$, we may assume, without loss of generality, that we work in an open neighborhood of $\mathbb{R}^{d}$ equipped with the standard metric. By the triangle inequality, $\displaystyle\|y_{1}-y_{2}\|$ $\displaystyle=\left\|\sum_{i}\int_{t_{i-1}}^{t_{i}}v_{i}(\ell(t))dt\right\|$ $\displaystyle\geq\left\|\sum_{i}(t_{i}-t_{i-1})v_{i}(y_{1})\right\|-\sum_{i}\int_{t_{i-1}}^{t_{i}}\|v_{i}(\ell(t))-v_{i}(y_{1})\|dt$ $\displaystyle\geq(c-\delta)L(\ell)$ where $c=\min\left\\{\left\|\sum_{i}s_{i}v_{i}(y)\right\|:\,\sum_{i}|s_{i}|=1,\,y\in X\right\\}>0.$ Taking $\delta$ sufficiently small, this implies the estimate for $L(\ell)$. Since the stable distributions are uniquely integrable, $\ell([t_{i-1},t_{i}))$ is contained in the stable manifold of $\alpha(\gamma_{i})$. ∎ ###### Proof of Theorem 3.1. Let $\gamma_{1},\ldots,\gamma_{k}\in\Gamma$ be elements satisfying (1). By Lemma 3.3, the map $\Phi$ is bi-Hölder restricted to the stable manifolds of $\alpha_{0}(\gamma_{i})$’s. By Lemma 3.4, for sufficiently close $x,y\in X$, there exist points $x_{0}=x,x_{1},\ldots,x_{2k}=y$ such that $x_{j-1}$ and $x_{j}$ are on the same stable manifold of some $\alpha_{0}(\gamma_{i_{j}})$, and $d(x_{j-1},x_{j})\leq c\,d(x,y)$. Then $\displaystyle d(\Phi(x),\Phi(y))$ $\displaystyle\leq\sum_{j=1}^{2k}d(\Phi(x_{j-1}),\Phi(x_{j}))\leq\sum_{j=1}^{2k}c_{j}d(x_{j-1},x_{j})^{\theta_{j}}$ $\displaystyle\leq\left(\sum_{j=1}^{2k}c_{j}c^{\theta_{j}}\right)d(x,y)^{\theta}$ where $\theta=\min\theta_{j}$. By Lemma 3.2, the action $\alpha_{1}$ is also weakly hyperbolic. Then the proof that $\Phi^{-1}$ is Hölder follows the same argument. ∎ ### 3.2. Invariance of fast stable manifolds Let $f\in\hbox{\rm Diff}(X)$, and the tangent bundle $TX$ has continuous $f$-invariant splitting (2) $TX=E^{-}\oplus E^{+}$ such that for some $\lambda\in(0,1)$ and $\mu>\lambda$,111The notation $A\ll B$ means that there exists $c>0$, independent of other parameters, such that $A\leq c\,B$. (3) $\displaystyle\|D(f^{n})_{x}v\|$ $\displaystyle\ll\lambda^{n}\|v\|\quad\hbox{for all $n\geq 0$, $x\in X$, and $v\in E_{x}^{-}$},$ $\displaystyle\|D(f^{n})_{x}v\|$ $\displaystyle\gg\mu^{n}\|v\|\quad\hbox{for all $n\geq 0$, $x\in X$, and $v\in E_{x}^{+}$}.$ We recall (see, for example, [20, Theorem 4.1]) that the distribution $E^{-}$ is integrable to the fast stable foliation $\\{W^{fs}_{f}(x)\\}_{x\in X}$, and this foliation is Hölder continuous with $C^{\infty}$-leaves. We denote by $d^{fs}$ the induced metrics on the leaves of this foliation. For $\rho>\lambda$ and $x,y\in X$ such that $y\in W^{fs}_{f}(x)$, $d^{fs}(f^{n}(x),f^{n}(y))\ll\rho^{n}d^{fs}(x,y).$ There exists $\epsilon_{0}>0$ such that for every $z,w\in X$ satisfying $w\in W^{fs}_{f}(z)$ and $d^{fs}(z,w)<\epsilon_{0}$, we have (4) $d^{fs}(z,w)\ll d(z,w)\leq d^{fs}(z,w).$ Let $f_{0}\in\hbox{\rm Aff}(X)$ be such that $Df_{0}$ is partially hyperbolic, and $\lambda_{0}<\mu_{0}$ denote the least two absolute values of the eigenvalues of $Df_{0}$. If $f\in\hbox{\rm Diff}(X)$ is a $C^{1}$-small perturbation of $f_{0}$, then we have a splitting as above with $\lambda=\lambda_{0}+\epsilon$ and $\mu=\mu_{0}-\epsilon$ for some small $\epsilon>0$, depending on $d_{C^{1}}(f,f_{0})$ (see [20, Lemma 3.5]). The fast stable manifolds $W^{fs}_{f}(x)$ are defined with respect to this splitting. Note that $W^{fs}_{f_{0}}(x)=\exp(W^{min}_{f_{0}})x$ where $\exp$ is the Lie exponential map, and $W^{min}_{f_{0}}$ is defined as on page 2. The aim of this section is to prove the following theorem. ###### Theorem 3.5. Let $\alpha_{0}:\Gamma\to\hbox{\rm Aff}(X)$ and $\alpha_{1}:\Gamma\to\hbox{\rm Diff}(X)$ be $C^{1}$-close actions of a finitely generated discrete group $\Gamma$, and let $\Phi:X\to X$ be a homeomorphism such that $\Phi\circ\alpha_{0}(\gamma)=\alpha_{1}(\gamma)\circ\Phi\quad\hbox{for all $\gamma\in\Gamma$}.$ Assume that $(D\alpha_{0})(\Gamma)$ acts irreducibly on $\hbox{\rm Lie}(G)$. Then for every partially hyperbolic $f_{0}:=\alpha_{0}(\gamma)$ and $f:=\alpha_{1}(\gamma)$, $\gamma\in\Gamma$, such that $Df_{0}$ is semisimple on $W^{min}_{f_{0}}$, $\Phi(W^{fs}_{f_{0}}(z))=W^{fs}_{f}(\Phi(z))\quad\hbox{for all $z\in X$.}$ Moreover, the map $\Phi$ is bi-Hölder with respect to the induced metrics on fast stable leaves of $f_{0}$ and $f$. Let us start with some preliminary reductions. We will prove that (5) $\Phi^{-1}(W^{fs}_{f}(z))\subset W^{fs}_{f_{0}}(\Phi^{-1}(z))\quad\hbox{for all $z\in X$.}$ This also implies that the equality. Indeed, it follows from (5) that every leaf $W^{fs}_{f_{0}}(\Phi^{-1}(z))$ is a disjoint union of sets of the form $\Phi^{-1}(W^{fs}_{f}(y))$ for some $y\in X$. By [20, Lemma 3.5], the fast stable leaves of $f_{0}$ and $f$ have the same dimension. Hence, by the invariance of domain, every set $\Phi^{-1}(W^{fs}_{f}(y))$ is open in $W^{fs}_{f_{0}}(\Phi^{-1}(z))$. Since $W^{fs}_{f_{0}}(\Phi^{-1}(z))$ is connected, we deduce that $\Phi^{-1}(W^{fs}_{f}(z))=W^{fs}_{f_{0}}(\Phi^{-1}(z)).$ Let (6) $\mathcal{S}_{\epsilon^{\prime}}(x)=\\{\Phi^{-1}(z):\;z\in W^{fs}_{f}(\Phi(x)),\,d^{fs}(z,\Phi(x))<\epsilon^{\prime}\\}.$ We will show that there exists $\epsilon^{\prime}\in(0,\epsilon_{0})$ such that for every $x\in X$, $\mathcal{S}_{\epsilon^{\prime}}(x)\subset W^{fs}_{f_{0}}(x).$ This will imply the theorem. First, we observe the following property of points lying on the same fast stable leaf for affine actions: ###### Proposition 3.6. Let $f_{0},g_{0}\in\hbox{\rm Aff}(X)$ be such that $(Df_{0})|_{W^{min}_{f_{0}}}$ is semisimple. Then there exists $c>0$ such that for every $z,w\in X$ satisfying $w\in W^{fs}_{f_{0}}(z)$ and $n\geq k\geq 0$, $d(f_{0}^{-k}g_{0}f_{0}^{n}(z),f_{0}^{-k}g_{0}f_{0}^{n}(w))\leq c\,\lambda_{0}^{n-k}d^{fs}(z,w)$ where $\lambda_{0}$ is the least absolute value of the eigenvalues of $Df_{0}$. ###### Proof. It suffices to prove the proposition when $d^{fs}(z,w)$ small. Write $w=\exp(v)z$ for $v\in W^{min}_{f_{0}}$. Then $w=\exp(D(f_{0}^{-k}g_{0}f_{0}^{n})v)f_{0}^{-k}g_{0}f_{0}^{n}(z),$ and it suffices to show that for a norm on $\hbox{Lie}(G)$, $\|D(f_{0}^{-k}g_{0}f_{0}^{n})v\|\ll\lambda_{0}^{n-k}\|v\|,$ which is easy to check. ∎ A similar but weaker property also holds for small perturbations of affine actions: ###### Proposition 3.7. Let $f_{0}\in\hbox{\rm Aff}(X)$, $g\in\hbox{\rm Diff}(X)$, and $\nu>1$. Then there exists $c>0$ such that for any sufficiently $C^{1}$-small perturbations $f\in\hbox{\rm Diff}(X)$ of $f_{0}$, $z,w\in X$ satisfying $w\in W^{fs}_{f}(z)$, and $n\geq 0$, (7) $d(f^{-n}gf^{n}(z),f^{-n}gf^{n}(w))\leq c\,\nu^{n}d^{fs}(z,w).$ ###### Proof. Let $\lambda_{0}$ denote the least absolute value of the eigenvalues of $Df_{0}$. Take $\lambda_{-}<\lambda_{0}<\lambda_{+}$ such that $\frac{\lambda_{+}}{\lambda_{-}}<\nu$. For $f$ sufficiently $C^{1}$-close to $f_{0}$, we have $\|D(f^{-n})_{u}\|\ll\lambda_{-}^{-n}\quad\hbox{for all $u\in X$ and $n\geq 0$,}$ and $\left\|D(f^{n})|_{T_{u}(W^{fs}_{f}(u))}\right\|\ll\lambda_{+}^{n}\quad\hbox{for all $u\in X$ and $n\geq 0$.}$ This implies that $\left\|D(f^{-n}gf^{n})|_{T_{u}(W^{fs}_{f}(u))}\right\|\ll\left(\frac{\lambda_{+}}{\lambda_{-}}\right)^{n}\quad\hbox{for all $u\in X$ and $n\geq 0$.}$ Let $\ell$ be a smooth curve in $W^{fs}_{f}(z)$ from $z$ to $w$ such that $L(\ell)=d^{fs}(z,w)$. Then $L(f^{-n}gf^{n}(\ell))\ll\left(\frac{\lambda_{+}}{\lambda_{-}}\right)^{n}L(\ell)<\nu^{n}d^{fs}(z,w)$ for all $n\geq 0$. This proves the proposition. ∎ It turns out that property (7) characterizes points lying on the same fast stable leaves. This observation is crucial for the proof of Theorem 3.5 and is the main point of Theorem 3.10 below. Since the proof of Theorem 3.10 is quite involved, we first present its linear analogue – Proposition 3.8. Although the argument in the proof of Theorem 3.10 follows the same idea, it requires more delicate quantitative estimates because we have to work in injectivity neighborhoods of the exponential map. Let $A\in\hbox{GL}_{l}(\mathbb{R})$. We denote by $\lambda_{1}<\cdots<\lambda_{d}$ the absolute values of the eigenvalues of $A$, and $P_{i}$ denote the projection to the sum of the generalized eigenspaces of $A$ corresponding to $\lambda_{i}$ along the other eigenspaces. ###### Proposition 3.8. Let $B_{1},\ldots,B_{k}\in\hbox{\rm GL}_{l}(\mathbb{R})$ be such that for some $\eta>0$ (8) $\max_{k}\|P_{1}B_{k}v\|>\eta\|v\|\quad\hbox{for all $v\in\mathbb{R}^{l}$.}$ Then there exists $\nu>1$ such that $W^{min}_{A}=\\{v:\,\max_{k}\|A^{-n}B_{k}A^{n}\,v\|=O(\nu^{n})\quad\hbox{as $n\to\infty$}\\}.$ ###### Proof. For every small $\rho>0$ there exists a norm on $\mathbb{R}^{l}$ (see [15, Proposition 1.2.2]) such that $\|v_{1}+v_{2}\|=\|v_{1}\|+\|v_{2}\|$ for $v_{1}$ and $v_{2}$ in different generalized eigenspaces and $\displaystyle(\lambda_{i}-\rho)\|v\|$ $\displaystyle\leq\|Av\|\leq(\lambda_{i}+\rho)\|v\|,\quad v\in\hbox{im}(P_{i}).$ The parameter $\rho$ is fixed, but has to be chosen sufficiently small so that $(\lambda_{1}-\rho)^{-1}(\lambda_{1}+\rho)<\min_{i>1}(\lambda_{1}+\rho)^{-1}(\lambda_{i}-\rho).$ It follows from (8) that $\displaystyle\max_{k}\|A^{-n}B_{k}A^{n}v\|$ $\displaystyle\geq\max_{k}(\lambda_{1}+\rho)^{-n}\|P_{1}B_{k}A^{n}v\|$ $\displaystyle\geq(\lambda_{1}+\rho)^{-n}\eta\|A^{n}v\|$ $\displaystyle\geq(\lambda_{1}+\rho)^{-n}\eta\sum_{i}(\lambda_{i}-\rho)^{n}\|P_{i}v\|.$ We take $\nu>1$ such that $\nu<(\lambda_{1}+\rho)^{-1}(\lambda_{i}-\rho)$ for $i>1$ and $\nu>(\lambda_{1}-\rho)^{-1}(\lambda_{1}+\rho)$. Then $\max_{k}\|A^{-n}B_{k}A^{n}v\|=O(\nu^{n})$ implies that $P_{i}v=0$ for $i>1$. Also, for $v\in W^{min}_{A}$, $\displaystyle\max_{k}\|A^{-n}B_{k}A^{n}v\|$ $\displaystyle\leq(\lambda_{1}-\rho)^{-n}\left(\max_{k}\|B_{k}\|\right)(\lambda_{1}+\rho)^{n}=O(\nu^{n}).$ This proves the proposition. ∎ Proposition 3.7 and (4) imply that uniformly on $z,w\in X$, satisfying $w\in W^{fs}_{f}(z)$ and $d^{fs}(z,w)<\epsilon_{0}$, and $n\geq 0$, we have $d(f^{-n}gf^{n}(z),f^{-n}gf^{n}(w))\ll\nu^{n}d(z,w).$ Now we take $g=\alpha_{1}(\delta)$ and $g_{0}=\alpha_{0}(\delta)$ for some $\delta\in\Gamma$. Since the action of $\Gamma$ on $\hbox{Lie}(G)$ is irreducible, $\alpha_{0}$ is weakly hyperbolic. Hence, by Theorem 3.1, the conjugacy map $\Phi$ and its inverse are Hölder with some exponent $\theta>0$. It follows that uniformly on $x,y\in X$, satisfying $y\in\mathcal{S}_{\epsilon^{\prime}}(x)$, and $n\geq 0$, (9) $\displaystyle d(f_{0}^{-n}g_{0}f_{0}^{n}(x),f_{0}^{-n}g_{0}f_{0}^{n}(y))$ $\displaystyle\ll d(f^{-n}gf^{n}(\Phi(x)),f^{-n}gf^{n}(\Phi(y)))^{\theta}$ $\displaystyle\ll\nu^{\theta n}d(\Phi(x),\Phi(y))^{\theta}$ $\displaystyle\ll\nu^{\theta n}d(x,y)^{\theta^{2}}.$ Let $\lambda_{1}<\cdots<\lambda_{d}$ be the absolute values of the eigenvalues of $Df_{0}$ and $P_{i}$ denote the projection from $\hbox{Lie}(G)$ to the sum of the generalized eigenspaces of $Df_{0}$ corresponding to $\lambda_{i}$ along the other generalized eigenspaces. We say that a set $\\{g_{1},\ldots,g_{l}\\}\subset\hbox{\rm Aff}(X)$ is essential for $f_{0}$ if for some $\eta>0$ and every $v\in\hbox{Lie}(G)$, $\displaystyle\max_{k}\|P_{1}(Dg_{k})v\|>\eta\|v\|.$ Note this definition does not depend on a choice of the norm. Existence of essential sets follows from the following lemma: ###### Lemma 3.9. A set $g_{1},\ldots,g_{l}\in\hbox{\rm Aff}(X)$ is essential if and only if (10) $\bigcap_{k=1}^{l}(Dg_{k})^{-1}\hbox{\rm ker}(P_{1})=0.$ In particular, every subgroup $\Gamma\subset\hbox{\rm Aff}(X)$ such that $D\Gamma$ acts irreducibly on $\hbox{\rm Lie}(G)$ contains an essential set. Although the group $\Gamma$ in the Main Theorem needs to be finitely generated, this assumption is not needed in Lemma 3.9. ###### Proof. Since the map $v\mapsto(P_{1}(Dg_{k})v:\,k=1,\ldots l):\hbox{Lie}(G)\to\hbox{Lie}(G)^{l}$ is injective when (10) holds, one can take $\eta=\min\\{\max_{k}\|P_{1}(Dg_{k})v\|:\|v\|=1\\}>0.$ The converse is also clear. To prove the second claim, we observe that there exists a subset $\\{g_{1},\ldots,g_{l}\\}\subset\Gamma$ such that $\bigcap_{k=1}^{l}(Dg_{k})^{-1}\hbox{ker}(P_{1})=\bigcap_{g\in\Gamma}(Dg)^{-1}\hbox{ker}(P_{1}),$ and this space is zero by irreducibility. ∎ The following theorem is the main ingredient of the proof of Theorem 3.5: ###### Theorem 3.10. There exists $\nu=\nu(\vartheta,f_{0})>1$ such that given constants $a,\vartheta>0$, a map $f_{0}\in\hbox{\rm Aff}(X)$ such that $Df_{0}$ is semisimple on $W^{min}_{f_{0}}$, an essential set $g_{1},\ldots,g_{l}\in\alpha_{0}(\Gamma)$, and a family of subsets $\mathcal{L}_{\epsilon}(x)$, $x\in X$, of $X$ that satisfy 1. (i) $x\in\mathcal{L}_{\epsilon}(x)\subset B_{\epsilon}(x)$, 2. (ii) $f_{0}^{-1}(\mathcal{L}_{\epsilon}(x))\supset\mathcal{L}_{\epsilon}(f_{0}^{-1}(x))$, 3. (iii) for every $y\in\mathcal{L}_{\epsilon}(x)$ and $n\geq 0$, (11) $\max_{k}d(f_{0}^{-n}g_{k}f_{0}^{n}(x),f_{0}^{-n}g_{k}f_{0}^{n}(y))\leq a\nu^{n}d(x,y)^{\vartheta},$ one can choose $\epsilon>0$ such that $\mathcal{L}_{\epsilon}(x)\subset W^{fs}_{f_{0}}(x)\quad\hbox{for every $x\in X$}.$ ###### Outline of the proof of Theorem 3.10. We first observe that the sets $\mathcal{L}_{\epsilon}(x)$ lie in “cones” around $W^{fs}_{f_{0}}(x)$ where the size of the cones is controlled by $\nu$ and can be made sufficiently small (Lemma 3.11). Note that this argument is analogous to the proof of Proposition 3.8, but we can only derive a weaker conclusion because one has to work in injectivity neighborhoods of the exponential map. In the next step, we show that applying the map $f_{0}^{-1}$, the size of the cones can be made arbitrary small (Lemma 3.12). This implies the theorem. ∎ We fix a norm on $\hbox{Lie}(G)$, depending on parameter $\rho>0$, as in the proof of Proposition 3.8 with $A=Df_{0}$. The parameter $\rho$ has to be chosen sufficiently small. It controls the size of the cone in Lemma 3.11. We always take $\rho>0$ so that $\displaystyle\lambda_{i}<\lambda_{j}-\rho\quad\hbox{when $\lambda_{i}<\lambda_{j}$,}$ $\displaystyle\lambda_{i}-\rho>1\quad\hbox{when $\lambda_{i}>1$,}$ $\displaystyle\lambda_{i}+\rho<1\quad\hbox{when $\lambda_{i}<1$.}$ Note that since $(Df_{0})|_{W^{min}_{f_{0}}}$ is semisimple, we also have $\displaystyle\|(Df_{0})v\|=\lambda_{1}\|v\|,\quad v\in\hbox{im}(P_{1}),$ and $\|(Df_{0})^{-n}\|\leq\lambda_{1}^{-n}.$ By the assumption on $g_{k}$’s, there exists $\eta>0$ such that (12) $\displaystyle\max_{k}\|P_{1}(Dg_{k})v\|>\eta\|v\|,\quad v\in\hbox{Lie}(G).$ Let $\mu_{i}=\lambda_{1}^{-1}(\lambda_{i}+\rho)$ and $\sigma_{i}=\frac{\log\mu_{i}}{\log\mu_{d}}$. For $v\in\hbox{Lie}(G)$, we define $N(v)=\max_{i>1}\left\\{\|P_{i}v\|^{\sigma_{i}^{-1}}\right\\}.$ For $\beta,s>0$, we define $C(\beta,s)=\\{v\in\hbox{Lie}(G):\,N(v)\leq\beta\|v\|^{s}\\}.$ ###### Lemma 3.11. There exist $\epsilon,\beta>0$ such that for every $x,y\in X$ satisfying $d(x,y)<\epsilon$ and (11), $y\in\exp(C(\beta,s))x.$ where $s=s(\nu,\rho,\vartheta,f_{0})>0$ is such that $s\to\infty$ as $\nu\to 1^{+}$ and $\rho\to 0^{+}$. ###### Proof. Let $c_{1}=\max_{k}\|Dg_{k}\|$. There exist $\delta_{0}>0$ and $c_{0}>1$ such that for every $x\in X$ and $v\in\hbox{Lie}(G)$ satisfying $\|v\|<\delta_{0}$, we have (13) $c_{0}^{-1}\|v\|\leq d(x,\exp(v)x)\leq c_{0}\|v\|.$ Let $b>0$ such that $\sum_{j>1}b^{\sigma_{j}}=\delta_{0}/(2c_{1})$. We choose $\epsilon>0$ so that $d(x,y)<\epsilon$ implies that $y=\exp(v)x$ where $N(v)<\min\\{1,b\\}\quad\hbox{and}\quad\|v\|<\min\left\\{\delta_{0},\delta_{0}/2c_{1}\right\\}.$ Assuming that the claim fails, we will show that there exists $n\geq 0$ such that (14) $\displaystyle ac_{0}^{\vartheta+1}\nu^{n}\|v\|^{\vartheta}<\max_{k}\|D(f_{0}^{-n}g_{k}f_{0}^{n})v\|<\delta_{0}.$ Since $d(f_{0}^{-n}g_{k}f_{0}^{n}(x),f_{0}^{-n}g_{k}f_{0}^{n}(y))=d(f_{0}^{-n}g_{k}f_{0}^{n}(x),\exp(D(f_{0}^{-n}g_{k}f_{0}^{n})v)f_{0}^{-n}g_{k}f_{0}^{n}(x)),$ we deduce from (13) and (14) that $ac_{0}^{\vartheta+1}\nu^{n}(c_{0}^{-1}\,d(x,y))^{\vartheta}<\max_{k}c_{0}\,d(f_{0}^{-n}g_{k}f_{0}^{n}(x),f_{0}^{-n}g_{k}f_{0}^{n}(y)),$ which contradicts (11). To obtain the upper estimate in (14), we observe that $\displaystyle\max_{k}\|D(f_{0}^{-n}g_{k}f_{0}^{n})v\|$ $\displaystyle\leq\lambda_{1}^{-n}\max_{k}\|D(g_{k}f_{0}^{n})v\|\leq\lambda_{1}^{-n}c_{1}\|D(f_{0}^{n})v\|$ $\displaystyle\leq c_{1}\|P_{1}v\|+\lambda_{1}^{-n}c_{1}\sum_{j>1}(\lambda_{j}+\rho)^{n}\|P_{j}v\|$ $\displaystyle\leq c_{1}\|v\|+c_{1}\sum_{j>1}\mu_{j}^{n}\|P_{j}v\|.$ We choose $n\geq 0$ so that (15) $\mu_{d}^{-1}\frac{b}{N(v)}<\mu_{d}^{n}\leq\frac{b}{N(v)}.$ Then $\mu_{d}^{-\sigma_{j}}\frac{b^{\sigma_{j}}}{N(v)^{\sigma_{j}}}<\mu_{j}^{n}\leq\frac{b^{\sigma_{j}}}{N(v)^{\sigma_{j}}}$ and $\displaystyle\max_{k}\|D(f_{0}^{-n}g_{k}f_{0}^{n})v\|\leq c_{1}\|v\|+c_{1}\sum_{j>1}b^{\sigma_{j}}\frac{\|P_{j}v\|}{N(v)^{\sigma_{j}}}<\delta_{0}.$ The lower estimate in (14) is proved similarly using that $g_{1},\ldots,g_{l}$ is essential (see (12)). Let $\gamma_{j}>0$ be such that $\lambda_{1}^{-1}(\lambda_{j}-\rho)=\mu_{j}^{1-\gamma_{j}}$. We have $\displaystyle\max_{k}\|D(f_{0}^{-n}g_{k}f_{0}^{n})v\|$ $\displaystyle\geq\max_{k}\lambda_{1}^{-n}\|P_{1}D(g_{k}f_{0}^{n})v\|\geq\lambda_{1}^{-n}\eta\|D(f_{0}^{n})v\|$ $\displaystyle\geq\lambda_{1}^{-n}\eta\left(\lambda_{1}^{n}\|P_{1}v\|+\sum_{j>1}(\lambda_{j}-\rho)^{n}\|P_{j}v\|\right)$ $\displaystyle\geq\eta\sum_{j>1}\mu_{j}^{n(1-\gamma_{j})}\left\|P_{j}v\right\|$ $\displaystyle\geq\eta\sum_{j>1}(\mu_{d}^{-1}b)^{\sigma_{j}(1-\gamma_{j})}N(v)^{\sigma_{j}\gamma_{j}}\frac{\|P_{j}v\|}{N(v)^{\sigma_{j}}}$ $\displaystyle\geq\eta(\mu_{d}^{-1}b)^{\sigma_{j_{0}}(1-\gamma_{j_{0}})}N(v)^{\sigma_{j_{0}}\gamma_{j_{0}}}.$ where $j_{0}>1$ is such that $\|P_{j_{0}}v\|^{1/\sigma_{j_{0}}}=N(v)$. This implies that $\max_{k}\|D(f_{0}^{-n}g_{k}f_{0}^{n})v\|\geq\min_{j>1}\eta(\mu_{d}^{-1}b)^{\sigma_{j}(1-\gamma_{j})}N(v)^{\sigma_{j}\gamma_{j}}.$ Let $\omega=\frac{\log\nu}{\log\mu_{d}}$. It follows from (15) that the first inequality in (14) is satisfied provided that $ac_{0}^{\vartheta+1}N(v)^{-\omega}b^{\omega}\|v\|^{\vartheta}<\min_{j>1}\eta(\mu_{d}^{-1}b)^{\sigma_{j}(1-\gamma_{j})}N(v)^{\sigma_{j}\gamma_{j}}.$ Since this gives a contradiction, we deduce that $ac_{0}^{\vartheta+1}b^{\omega}\|v\|^{\vartheta}\geq\min_{j>1}\eta(\mu_{d}^{-1}b)^{\sigma_{j}(1-\gamma_{j})}N(v)^{\omega+\sigma_{j}\gamma_{j}}.$ Hence, $N(v)\leq\beta\|v\|^{s}$ with explicit $\beta>0$ and $s=\vartheta/(\omega+\max_{j>1}(\sigma_{j}\gamma_{j}))$. Clearly, $s\to\infty$ as $\nu\to 1^{+}$ and $\rho\to 0^{+}$. This completes the proof. ∎ For $i=1,\ldots,d$ and $\delta,\beta,s>0$, we define $C^{i}_{\delta}(\beta,s)=\\{v\in\hbox{Lie}(G):\,\|v\|<\delta;\,\|P_{i}v\|^{\sigma_{i}^{-1}}\leq\beta\|v\|^{s};\,P_{j}v=0,j>i\\}.$ ###### Lemma 3.12. For every $\delta,\beta,s>0$, $(Df_{0})^{-1}(C^{i}_{\delta}(\beta,s))\subset C^{i}_{\xi\delta}(\rho_{i}\beta,s).$ where $\xi=\max\\{1,\|(Df_{0})^{-1}\|\\}$ and $\rho_{i}=(\lambda_{i}-\rho)^{-{\sigma_{i}^{-1}}}(\lambda_{i}+\rho)^{s}$. ###### Proof. Let $v\in(Df_{0})^{-1}(C^{i}_{\delta}(\beta,s))$. Then $\displaystyle(\lambda_{i}-\rho)^{\sigma_{i}^{-1}}\|P_{i}v\|^{\sigma_{i}^{-1}}\leq\beta\left(\sum_{j\leq i}(\lambda_{j}+\rho)\|P_{j}v\|\right)^{s}\leq\beta(\lambda_{i}+\rho)^{s}\|v\|^{s}.$ This implies the lemma. ∎ ###### Proof of Theorem 3.10. We start by setting up notation for the Jordan form of $Df_{0}$ for $\lambda_{i}=1$. It follows from our choice of the norm that there exist linear maps $Q_{1},\ldots,Q_{j_{0}}$ such that (16) $\|(Df_{0}^{k})v\|=\left\|\sum_{j=0}^{j_{0}}k^{j}Q_{j}v\right\|\quad\hbox{for $k\geq 0$ and $v\in\hbox{im}(P_{i})$.}$ Let $s>0$ be as in Lemma 3.11. Recall that $s\to\infty$ as $\nu\to 1^{+}$ and $\rho\to 0^{+}$. We choose $\rho>0$ and $\nu>1$ so that $\displaystyle s-\sigma_{i}^{-1}>0\quad\hbox{when $\lambda_{i}=1$},$ $\displaystyle\rho_{i}:=(\lambda_{i}-\rho)^{-{\sigma_{i}^{-1}}}(\lambda_{i}+\rho)^{s}<1\quad\quad\hbox{when $\lambda_{i}<1$.}$ Let $\xi\geq 1$ be as in Lemma 3.12 and $\beta,\epsilon>0$ as in Lemma 3.11. Take $\delta\in(0,1)$ such that for $\|v\|<\xi\delta$, the exponential coordinates $v\mapsto\exp(v)z$, $z\in X$, are one-to-one, and (17) $\|Q_{j}P_{i}v\|<\beta^{-(s-\sigma_{i}^{-1})^{-1}}$ when $\lambda_{i}=1$ and $j=0,\ldots,j_{0}$. In addition, we assume that $\epsilon$ is sufficiently small so that $B_{\epsilon}(x)\subset\exp(\\{\|v\|<\delta\\})x\quad\hbox{for all $x\in X$.}$ Then by Lemma 3.11, (18) $\mathcal{L}_{\epsilon}(x)\subset\exp(C(\beta,s)\cap\\{\|v\|<\delta\\})x\quad\hbox{for every $x\in X$.}$ In particular, (19) $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{d}_{\delta}(\beta,s))x.$ If $\lambda_{d}\leq 1$, we argue as in the following paragraph. Otherwise, we observe that since $\delta<1$, we have $C^{d}_{\delta}(\beta,s_{1})\subset C^{d}_{\delta}(\beta,s_{2})\quad\hbox{for $s_{1}>s_{2}$},$ and hence inclusion (18) also holds for $s>0$ such that $\rho_{d}=(\lambda_{d}-\rho)^{-{\sigma_{d}^{-1}}}(\lambda_{d}+\rho)^{s}<1$. Applying $f_{0}^{-1}$ to (19), we deduce from Lemma 3.12 that (20) $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{d}_{\xi\delta}(\rho_{d}\beta,s))x$ for every $x\in X$. Using that the exponential coordinates are one-to-one, we obtain from (20) and (18) that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{d}_{\delta}(\rho_{d}\beta,s))x.$ Repeating this argument, we conclude that $\mathcal{L}_{\epsilon}(x)\subset\bigcap_{k\geq 1}\exp(C^{d}_{\delta}(\rho^{k}_{d}\beta,s))x=\exp(C^{d}_{\delta}(0,s))x.$ Now (18) implies that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{d-1}_{\delta}(\beta,s))x.$ Applying the same reasoning inductively on $i$, we deduce that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(0,s))x$ provided that $\lambda_{i}>1$. It follows from (18) that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i-1}_{\delta}(\beta,s))x$. Suppose $\lambda_{i}=1$ and $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(\beta,s))x$ for some $\beta>0$. We will show that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(0,s))x.$ Applying $f_{0}^{-1}$, we deduce that for $y=\exp(v)x\in\mathcal{L}_{\epsilon}(x)$, $\|v\|<\delta$, and $k\geq 0$, we have $\|(Df_{0}^{k})P_{i}v\|^{\sigma_{i}^{-1}}\leq\beta\left(\sum_{j<i}(\lambda_{j}+\rho)^{k}\|P_{j}v\|+\|(Df_{0}^{k})P_{i}v\|\right)^{s}.$ Using that $\lambda_{j}+\rho<1$ for $j<i$ and taking $k\to\infty$, we deduce from (16) that $\|Q_{j_{0}}P_{i}v\|^{\sigma_{i}^{-1}}\leq\beta\|Q_{j_{0}}P_{i}v\|^{s}.$ By the choice of $\delta$ (see (17)), $\|Q_{j_{0}}P_{i}v\|=0$. Similar arguments imply that $\|Q_{j}P_{i}v\|=0$ for all $j=0,\ldots,j_{0}$. Hence, $P_{i}v=0$ and $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(0,s))x$. Combining this estimate with (18), we deduce that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i-1}_{\delta}(\beta,s))x$. Now we consider the case when $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(\beta,s))$ for some $i$ such that $\lambda_{i}<1$ and $\beta>0$. Applying $f_{0}^{-1}$, it follows from Lemma 3.12 that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\xi\delta}(\rho_{i}\beta,s))x\quad\hbox{for every $x\in X$.}$ Then it follows from (18) that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{i}_{\delta}(\rho_{i}\beta,s))x,$ and repeating this argument, we deduce that $\mathcal{L}_{\epsilon}(x)\subset\bigcap_{k\geq 1}\exp(C^{i}_{\delta}(\rho^{k}_{i}\beta,s))x=\exp(C^{i}_{\delta}(0,s))x.$ Since the above argument can be applied inductively on $i$, and we conclude that $\mathcal{L}_{\epsilon}(x)\subset\exp(C^{2}_{\delta}(0,s))x$. This completes the proof. ∎ ###### Proof of Theorem 3.5. The first claim of Theorem 3.5 follows from Theorem 3.10 with $\mathcal{L}_{\epsilon}(x)=\mathcal{S}_{\epsilon^{\prime}}(x)$ where $\mathcal{S}_{\epsilon^{\prime}}(x)$ is as in (6) with sufficiently small $\epsilon^{\prime}>0$. Note that $\alpha_{0}(\Gamma)$ contains an essential subset by Lemma 3.9, and (11) follows from (9) where the parameter $\nu$ is close to one if $f$ and $f_{0}$ are $C^{1}$-close. It remains to show that $\Phi$ is bi-Hölder with respect to the metrics $d^{fs}$. There exists $\epsilon>0$ such that for every $x\in X$, any points $z,w\in X$ lying on the same local leaf of $W^{fs}_{f}$ in $B(x,\epsilon)$ satisfy (4). Let $\delta>0$ be such that $\Phi(B_{\delta}(y))\subset B_{\epsilon}(\Phi(y))$ for every $y\in X$. Consider points $z_{0},w_{0}\in X$ lying on the same leaf of $W^{fs}_{f_{0}}$ such that $d^{fs}(z_{0},w_{0})<\delta$. Let $\ell$ be a curve from $z_{0}$ to $w_{0}$ contained in $W^{fs}_{f_{0}}(z_{0})$ such that $L(\ell)=d^{fs}(z_{0},w_{0})$. Then $\Phi(\ell)$ is contained in $B_{\epsilon}(\Phi(z_{0}))\cap W^{fs}_{f}(\Phi(z_{0}))$. Moreover, since $\Phi(\ell)$ is connected, $\Phi(\ell)$ is contained in a single local leaf of $W^{fs}_{f}$ in $B_{\epsilon}(\Phi(z_{0}))$. Hence, $d^{fs}(\Phi(z_{0}),\Phi(w_{0}))\ll d(\Phi(z_{0}),\Phi(w_{0})).$ Since $\Phi$ is Hölder with respect to $d$, this implies that $\Phi$ is Hölder with respect to $d^{fs}$ as well. The proof that $\Phi^{-1}$ is Hölder with respect to $d^{fs}$ is similar. ∎ ### 3.3. Convergence of the sequences $f^{-n}gf^{n}$ In this section, we study convergence of the sequence of maps $f^{-n}gf^{n}$ as $n\to\infty$. First, we consider the algebraic setting: ###### Proposition 3.13. Let $f_{0},g_{0}\in\hbox{\rm Aff}(X)$ be such that $Df_{0}:W_{f_{0}}^{min}\to W_{f_{0}}^{min}$ is semisimple. Then 1. (1) Given a sequence $\\{m_{i}\\}$ such that $(f_{0}^{-m_{i}}g_{0}f_{0}^{m_{i}})(x)\to y\quad\hbox{as $i\to\infty$}$ for some $x,y\in X$, the sequence of maps $f_{0}^{-m_{i}}g_{0}f_{0}^{m_{i}}:{W^{fs}_{f_{0}}(x)}\to X$ is precompact in the $C^{0}$-topology. 2. (2) There exist a sequence $\\{n_{i}\\}$ and a linear map $A:W_{f_{0}}^{min}\to W_{f_{0}}^{min}$ such that if for some $x,y\in X$ and a subsequence $\\{n_{i_{j}}\\}$, $(f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}})(x)\to y\quad\hbox{as $j\to\infty$,}$ then uniformly on $v\in W^{min}_{f}$ in compact sets, $(f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}})\exp(v)x\to\exp(Av)y\quad\hbox{as $j\to\infty$}.$ The map $A$ is nondegenerate provided that $P_{f_{0}}^{min}Dg_{0}:W_{f_{0}}^{min}\to W_{f_{0}}^{min}$ is nondegenerate. ###### Remark 3.14. If $\dim W_{f_{0}}^{min}=1$, one can take $n_{i}=i$ and $A=P_{f_{0}}^{min}Dg_{0}$. In general, $A=\lim_{i\to\infty}\omega^{-n_{i}}P^{min}_{f_{0}}(Dg_{0})\omega^{n_{i}}$ for some $\omega\in\hbox{Isom}(W^{min}_{f_{0}})$. ###### Proof. We have $(f_{0}^{-n}g_{0}f_{0}^{n})\exp(v)x=\exp(D(f_{0}^{-n}g_{0}f_{0}^{n})v)(f_{0}^{-n}g_{0}f_{0}^{n})x.$ It follows from the assumption on $f_{0}$ that $Df_{0}|_{W_{f_{0}}^{min}}=\lambda\cdot\omega$ where $\lambda>0$ and $\omega$ is an isometry of $W_{f_{0}}^{min}$. Then $\displaystyle D(f_{0}^{-n}g_{0}f_{0}^{n})v=(\omega^{-n}P^{min}_{f_{0}}(Dg_{0})\omega^{n})v+(Df_{0})^{-n}P^{max}_{f_{0}}(Dg_{0})\lambda^{n}\omega^{n}v$ where $P^{min}_{f_{0}}$ denotes the projection on $W^{min}_{f_{0}}$ and $P^{max}_{f_{0}}$ denotes the projection on the sum of eigenspaces complimentary to $W^{min}_{f_{0}}$. Since $\omega$ is an isometry, and $(Df_{0})^{-n}P^{max}_{f_{0}}(Dg_{0})\lambda^{n}\omega^{n}v\to 0,$ it is clear that the sequence of maps $v\mapsto D(f_{0}^{-n}g_{0}f_{0}^{n})v$ is precompact in $C^{0}$-topology. This implies that the sequence $f_{0}^{-m_{i}}g_{0}f_{0}^{m_{i}}|_{W^{fs}_{f_{0}}(x)}$ is precompact in $C^{0}$-topology as well. To prove (2), it suffices to choose the sequence $\\{n_{i}\\}$ so that $\\{\omega^{n_{i}}\\}$ converges. This proves the proposition. ∎ We show that the convergence of $f_{0}^{-n}g_{0}f_{0}^{n}|_{W^{fs}_{f_{0}}(x)}$ persists under small perturbations: ###### Theorem 3.15. Let $f_{0},g_{0}\in\hbox{\rm Aff}(X)$ satisfy 1. (i) The map $f_{0}$ is partially hyperbolic, 2. (ii) The map $Df_{0}:W^{min}_{f_{0}}\to W^{min}_{f_{0}}$ is semisimple. Let $f,g\in\hbox{\rm Diff}(X)$ be $C^{1}$-small perturbations of $f_{0}$ and $g_{0}$ and $\Phi:X\to X$ a Hölder isomorphism such that $\Phi\circ f_{0}=f\circ\Phi\quad\hbox{and}\quad\Phi\circ g_{0}=g\circ\Phi$ and $\Phi(W^{fs}_{f_{0}}(x))=W^{fs}_{f}(\Phi(x))\quad\hbox{for every $x\in X$.}$ Then for every $x\in X$ and a sequence $\\{m_{i}\\}$ as in Proposition 3.13(1), the sequence of maps $f^{-m_{i}}gf^{m_{i}}:W^{fs}_{f}(x)\to X,\quad i\geq 0,$ is precompact in the $C^{1}$-topology. Throughout this section, we assume that $X$ is a submanifold of $\mathbb{R}^{N}$, which allows us to identify tangent spaces at different points. We have a Hölder continuous decomposition (cf. (2)) (21) $T_{x}X=E^{-}_{x}\oplus E^{+}_{x},\quad x\in X,$ where $E^{-}_{x}=T_{x}W^{fs}_{f}(x)$. Let $P_{x}:T_{x}X\to E^{-}_{x}\quad\hbox{and}\quad P^{+}_{x}:T_{x}X\to E^{+}_{x}$ denote the corresponding projections. The following proposition is the main ingredient of the proof of Theorem 3.15. ###### Proposition 3.16. Let $r>0$. Then under the assumptions of Theorem 3.15, for every $x,y\in X$ satisfying $y\in W^{fs}_{f}(x)$ and $d^{fs}(x,y)\leq r$, $\|D(f^{-n}gf^{n})_{x}P_{x}-D(f^{-n}gf^{n})_{y}P_{y}\|\ll d^{fs}(x,y)^{\vartheta}\|D(f^{-n}gf^{n})_{x}P_{x}\|+\delta_{n}$ where $\vartheta>0$ and $\delta_{n}\to 0$. ###### Proof. Note that $\Phi$ and $\Phi^{-1}$ are also Hölder with respect to the metrics $d^{fs}$ on the fast stable leaves of $f_{0}$ and $f$ (see proof of Theorem 3.5). By Proposition 3.6, $\displaystyle d(f_{0}^{-k}g_{0}f_{0}^{n}(\Phi^{-1}(x)),f_{0}^{-k}g_{0}f_{0}^{n}(\Phi^{-1}(y)))$ $\displaystyle\ll\lambda_{0}^{n-k}d^{fs}(\Phi^{-1}(x),\Phi^{-1}(y))$ $\displaystyle\ll\lambda_{0}^{n-k}d^{fs}(x,y)^{\omega_{0}}$ where $\omega_{0}>0$ is the Hölder exponent of $\Phi^{-1}$ with respect to $d^{fs}$. Then it follows that we have the estimate (22) $d(f^{-k}gf^{n}(x),f^{-k}gf^{n}(y))\ll\lambda_{0}^{\omega(n-k)}d^{fs}(x,y)^{\omega_{0}\omega}$ where $\omega>0$ is the Hölder exponent of $\Phi$ with respect to $d$. Since the decomposition (21) is $f$-invariant, we have $P_{f(x)}D(f)_{x}P_{x}=D(f)_{x}P_{x}\quad\hbox{and}\quad P_{f^{-1}(x)}D(f^{-1})_{x}P_{x}=D(f^{-1})_{x}P_{x}.$ By (3), there exist $\lambda\in(0,1)$ and $\mu>\lambda$ such that (23) $\|D(f^{n})_{x}P_{x}\|\ll\lambda^{n}\quad\hbox{and}\quad\|D(f^{-n})_{x}P^{+}_{x}\|\ll\mu^{-n}$ uniformly on $x\in X$ and $n\geq 0$. It is crucial for the proof that the map $D(f)_{x}P_{x}$ is approximately conformal (cf. assumption (ii) on $f_{0}$). Namely, for some small $\epsilon>0$, (24) $\|D(f^{-n})_{x}P_{x}\|\ll(\lambda-\epsilon)^{-n}$ uniformly on $x\in X$ and $n\geq 0$. We also recall that for $\rho>\lambda$ and $x,y\in X$ such that $y\in W^{fs}_{f}(x)$, (25) $d^{fs}(f^{n}(x),f^{n}(y))\ll\rho^{n}d^{fs}(x,y).$ Note that the parameter $\epsilon$ in (24) satisfies $\epsilon\to 0$ as $d_{C^{1}}(f_{0},f)\to 0$. We assume $f$ is sufficiently close to $f_{0}$ so that $\zeta:=(\lambda-\epsilon)^{-1}\lambda\rho^{\theta}<1\quad\hbox{and}\quad\nu:=(\lambda-\epsilon)^{-1}\lambda\lambda_{0}^{\omega}<1$ where $\theta$ is the Hölder exponent of the map $x\mapsto P_{x}$. We have $\displaystyle D(f^{-n}gf^{n})_{x}P_{x}=$ $\displaystyle D(f^{-n})_{gf^{n}(x)}P_{gf^{n}(x)}D(g)_{f^{n}(x)}D(f^{n})_{x}P_{x}$ $\displaystyle+D(f^{-n})_{gf^{n}(x)}P^{+}_{gf^{n}(x)}D(g)_{f^{n}(x)}D(f^{n})_{x}P_{x}.$ It follows from (23) that $\|D(f^{-n})_{gf^{n}(x)}P^{+}_{gf^{n}(x)}D(g)_{f^{n}(x)}D(f^{n})_{x}P_{x}\|\ll\lambda^{n}\mu^{-n}\to 0.$ Hence, to prove the theorem, it suffices to show that for $A_{n}(x):=\left(\prod_{i=n-1}^{0}D(f^{-1})_{f^{-i}gf^{n}(x)}\right)P_{gf^{n}(x)}D(g)_{f^{n}(x)}\left(\prod_{i=n-1}^{0}D(f)_{f^{i}(x)}\right)P_{x},$ we have $\|A_{n}(x)-A_{n}(y)\|\ll d^{fs}(x,y)^{\kappa}\|A_{n}(x)\|.$ We consider the operators $\displaystyle A_{n,k}(x,y):=$ $\displaystyle\left(\prod_{i=n-1}^{0}D(f^{-1})_{f^{-i}gf^{n}(x)}\right)P_{gf^{n}(x)}D(g)_{f^{n}(x)}$ $\displaystyle\times\left(\prod_{i=n-1}^{k+1}D(f)_{f^{i}(x)}\right)P_{f^{k+1}(x)}\left(\prod_{i=k}^{0}D(f)_{f^{i}(y)}\right)P_{y}.$ Note that (26) $\displaystyle\|A_{n}(x)-A_{n,-1}(x,y)\|\leq\|A_{n}(x)\|\cdot\|P_{x}-P_{x}P_{y}\|\ll\|A_{n}(x)\|d(x,y)^{\theta}.$ Now we estimate $\|A_{n,n-1}(x,y)-A_{n,-1}(x,y)\|$. We use that $\displaystyle A_{n,k}(x,y)-A_{n,k-1}(x,y)=A_{n}(x)B_{n,k}(x,y)$ where $\displaystyle B_{n,k}(x,y):=$ $\displaystyle\left(\prod_{i=0}^{k}D(f)^{-1}_{f^{i}(x)}\right)P_{f^{k+1}(x)}\left(D(f)_{f^{k}(y)}P_{f^{k}(y)}-D(f)_{f^{k}(x)}P_{f^{k}(x)}\right)$ $\displaystyle\times\left(\prod_{i=k-1}^{0}D(f)_{f^{i}(y)}\right)P_{y}.$ By (25), we have $\|D(f)_{f^{k}(y)}P_{f^{k}(y)}-D(f)_{f^{k}(x)}P_{f^{k}(x)}\|\ll d(f^{k}(x),f^{k}(y))^{\theta}\ll\rho^{\theta k}d^{fs}(x,y)^{\theta},$ and by (23) and (24), $\displaystyle\left\|\left(\prod_{i=k-1}^{0}D(f)_{f^{i}(y)}\right)P_{y}\right\|$ $\displaystyle\ll\lambda^{k},$ $\displaystyle\left\|\left(\prod_{i=0}^{k}D(f)^{-1}_{f^{i}(x)}\right)P_{f^{k+1}(x)}\right\|$ $\displaystyle\ll(\lambda-\epsilon)^{-k-1}.$ Hence, $\|B_{n,k}(x,y)\|\ll\zeta^{k}d^{fs}(x,y)^{\theta}$ Since $\zeta<1$, it follows that (27) $\displaystyle\|A_{n,n-1}(x,y)-A_{n,-1}(x,y)\|$ $\displaystyle\leq\sum_{k=0}^{n-1}\|A_{n,k}(x,y)-A_{n,k-1}(x,y)\|$ $\displaystyle\ll\|A_{n}(x)\|d^{fs}(x,y)^{\theta}.$ We claim that for some $c>0$ and all $k=-1,\ldots,n-1$, (28) $\|A_{n,k}(x,y)\|\ll(1+c\,d^{fs}(x,y)^{\theta})\cdot\|A_{n}(x)\|.$ Setting $C_{k}(x,y):=\left(\prod_{i=0}^{k}D(f)^{-1}_{f^{i}(x)}\right)P_{f^{k+1}(x)}\left(\prod_{i=k}^{0}D(f)_{f^{i}(y)}\right)P_{y},$ we have $A_{n,k}(x,y)=A_{n}(x)C_{k}(x,y).$ Now equation (28) will follow from the estimate $\|C_{k}(x,y)\|\ll 1+c\,d^{fs}(x,y)^{\theta}.$ In fact, we will show that (29) $\|C_{k}(x,y)-P_{x}P_{y}\|\ll d^{fs}(x,y)^{\theta}.$ Using (23) and (24), we deduce that $\displaystyle\|C_{k}(x,y)-C_{k-1}(x,y)\|$ $\displaystyle=$ $\displaystyle\left\|\left(\prod_{i=0}^{k-1}D(f)^{-1}_{f^{i}(x)}\right)P_{f^{k}(x)}\left(D(f)^{-1}_{f^{k}(x)}D(f)_{f^{k}(y)}-id\right)\right.$ $\displaystyle\quad\quad\left.\times\left(\prod_{i=k-1}^{0}D(f)_{f^{i}(y)}\right)P_{y}\right\|$ $\displaystyle\ll$ $\displaystyle(\lambda-\epsilon)^{-k}d(f^{k}(x),f^{k}(y))^{\theta}\lambda^{k}\ll\zeta^{k}d^{fs}(x,y)^{\theta}.$ Since $C_{-1}(x,y)=P_{x}P_{y}$ and $\zeta<1$, the last estimate implies (29) and (28). Next, we consider the operators $\displaystyle D_{n,k}(x,y):=$ $\displaystyle\left(\prod_{i=n-1}^{k}D(f^{-1})_{f^{-i}gf^{n}(y)}\right)P_{f^{-k}gf^{n}(y)}\left(\prod_{i=k-1}^{0}D(f^{-1})_{f^{-i}gf^{n}(x)}\right)$ $\displaystyle\times P_{gf^{n}(x)}D(g)_{f^{n}(x)}P_{f^{n}(x)}\left(\prod_{i=n-1}^{0}D(f)_{f^{i}(y)}\right)P_{y}.$ Using (22), we deduce that $\displaystyle\|A_{n,n-1}(x,y)-D_{n,n}(x,y)\|$ $\displaystyle\leq\|P_{f^{-n}gf^{n}(x)}-P_{f^{-n}gf^{n}(y)}P_{f^{-n}gf^{n}(x)}\|\cdot\|A_{n,n-1}(x,y)\|$ $\displaystyle\ll d(f^{-n}gf^{n}(x),f^{-n}gf^{n}(y))^{\theta}\|A_{n,n-1}(x,y)\|$ $\displaystyle\ll d^{fs}(x,y)^{\theta\omega_{0}\omega}\|A_{n,n-1}(x,y)\|$ $\displaystyle\ll d^{fs}(x,y)^{\theta\omega_{0}\omega}\|A_{n}(x)\|.$ To estimate $\|D_{n,n}(x,y)-D_{n,0}(x,y)\|$, we use the argument similar to the proof of (27). We have $D_{n,k}(x,y)-D_{n,k-1}(x,y)=E_{n,k}(x,y)A_{n,n-1}(x,y)$ where $\displaystyle E_{n,k}(x,y):=$ $\displaystyle\left(\prod_{i=n-1}^{k}D(f^{-1})_{f^{-i}gf^{n}(y)}\right)P_{f^{-k}gf^{n}(y)}$ $\displaystyle\times\left(D(f^{-1})_{f^{-(k-1)}gf^{n}(x)}P_{f^{-(k-1)}gf^{n}(x)}-D(f^{-1})_{f^{-(k-1)}gf^{n}(y)}P_{f^{-(k-1)}gf^{n}(y)}\right)$ $\displaystyle\times\left(\prod_{i=k-1}^{n-1}D(f^{-1})^{-1}_{f^{-i}gf^{n}(x)}\right)P_{f^{-n}gf^{n}(x)}$ Applying (24), (22), and (23), we deduce that $\|E_{n,k}(x,y)\|\ll\nu^{n-k}d^{fs}(x,y)^{\theta\omega_{0}\omega}.$ Since $\nu<1$, it follows that (30) $\displaystyle\|D_{n,n}(x,y)-D_{n,0}(x,y)\|$ $\displaystyle\leq\sum^{n}_{k=1}\|D_{n,k}(x,y)-D_{n,k-1}(x,y)\|$ (31) $\displaystyle\ll d^{fs}(x,y)^{\theta\omega_{0}\omega}\|A_{n,n-1}(x,y)\|\ll d^{fs}(x,y)^{\theta\omega_{0}\omega}\|A_{n}(x)\|.$ Next, we compare the maps $A_{n}(y)$ and $D_{n,0}(x,y)$: $\displaystyle\|A_{n}(y)-D_{n,0}(x,y)\|=$ $\displaystyle\left\|\left(\prod_{i=n-1}^{0}D(f^{-1})_{f^{-i}gf^{n}(y)}\right)P_{gf^{n}(y)}\right.$ $\displaystyle\quad\quad\times\left(P_{gf^{n}(y)}D(g)_{f^{n}(y)}P_{f^{n}(y)}-P_{gf^{n}(x)}D(g)_{f^{n}(x)}P_{f^{n}(x)}\right)$ $\displaystyle\left.\quad\quad\times\left(\prod_{i=n-1}^{0}D(f)_{f^{i}(y)}\right)P_{y}\right\|.$ We have $\displaystyle\left\|P_{gf^{n}(y)}D(g)_{f^{n}(y)}P_{f^{n}(y)}-P_{gf^{n}(x)}D(g)_{f^{n}(x)}P_{f^{n}(x)}\right\|$ $\displaystyle\ll d(f^{n}(x),f^{n}(y))^{\theta}$ $\displaystyle\ll\rho^{\theta n}d^{fs}(x,y)^{\theta}.$ Combining this estimate with (23) and (24), we deduce that $\|A_{n}(y)-D_{n,0}(x,y)\|\ll\zeta^{n}d^{fs}(x,y)^{\theta}.$ Finally, the proposition follows from the estimate $\displaystyle\|A_{n}(x)-A_{n}(y)\|\leq$ $\displaystyle\|A_{n}(x)-A_{n,-1}(x,y)\|+\|A_{n,-1}(x,y)-A_{n,n-1}(x,y)\|$ $\displaystyle+\|A_{n,n-1}(x,y)-D_{n,n}(x,y)\|+\|D_{n,n}(x,y)-D_{n,0}(x,y)\|$ $\displaystyle+\|D_{n,0}(x,y)-A_{n}(y)\|.$ This completes the proof. ∎ ###### Proposition 3.17. Let $x_{0}\in X$ and $r>0$. Then under the assumptions of Theorem 3.15, $\sup\\{\|D(f^{-n}gf^{n})_{x}P_{x}\|:\,\,x\in W^{fs}_{f}(x_{0}),\,d^{fs}(x,x_{0})\leq r,\,\,n\in\mathbb{N}\\}<\infty.$ ###### Proof. Suppose that the claim fails, i.e., there exist sequences $x_{i}\in W^{fs}_{f}(x_{0})$, $d^{fs}(x_{i},x_{0})\leq r$, and $n_{i}\in\mathbb{N}$, $n_{i}\to\infty$, such that $\|D(f^{-n_{i}}gf^{n_{i}})_{x_{i}}P_{x_{i}}\|\to\infty.$ Passing to a subsequence, we may assume that $x_{i}\to x_{\infty}$ for some $x_{\infty}\in W^{fs}_{f}(x_{0})$ such that $d^{fs}(x_{\infty},x_{0})\leq r$. It follows from Proposition 3.16 that $\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}P_{x_{\infty}}\|\geq(1-c\cdot d^{fs}(x_{i},x_{\infty})^{\kappa})\|D(f^{-n_{i}}gf^{n_{i}})_{x_{i}}P_{x_{i}}\|-\delta_{n_{i}}\to\infty.$ Let $v_{i}\in T_{x_{\infty}}(W^{fs}_{f}(x_{\infty}))$ with $\|v_{i}\|=1$ be such that $\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}P_{x_{\infty}}\|=\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}v_{i}\|.$ Passing to a subsequence, we may assume that $v_{i}\to v_{\infty}$. We have $\displaystyle\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}v_{\infty}\|$ $\displaystyle\geq\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}v_{i}\|-\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}P_{x_{\infty}}(v_{\infty}-v_{i})\|$ $\displaystyle\geq\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}P_{x_{\infty}}\|\cdot(1-\|v_{\infty}-v_{i}\|).$ Hence, for sufficiently large $i$, we have $\displaystyle\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}v_{\infty}\|\geq\frac{1}{2}\|D(f^{-n_{i}}gf^{n_{i}})_{x_{\infty}}P_{x_{\infty}}\|.$ Let $\alpha_{n}=\|D(f^{-n}gf^{n})_{x_{\infty}}v_{\infty}\|$. Note that $\alpha_{n_{i}}\to\infty$. Fix small $\epsilon>0$. Let $x\in W^{fs}_{f}(x_{\infty})$ be such that $d^{fs}(x,x_{\infty})<\epsilon$ and $v\in T_{x}W^{fs}_{f}(x)$ such that $\|v-v_{\infty}\|<\epsilon$. We have $\displaystyle\|D(f^{-n}gf^{n})_{x}v-D(f^{-n}gf^{n})_{x_{\infty}}v_{\infty}\|\leq$ $\displaystyle\|D(f^{-n}gf^{n})_{x}P_{x}v-D(f^{-n}gf^{n})_{x_{\infty}}P_{x_{\infty}}v\|$ $\displaystyle+\|D(f^{-n}gf^{n})_{x_{\infty}}P_{x_{\infty}}v-D(f^{-n}gf^{n})_{x_{\infty}}P_{x_{\infty}}v_{\infty}\|$ $\displaystyle\ll$ $\displaystyle d^{fs}(x,x_{\infty})^{\kappa}\|D(f^{-n}gf^{n})_{x_{\infty}}P_{x_{\infty}}\|+\delta_{n}$ $\displaystyle+\|D(f^{-n}gf^{n})_{x_{\infty}}P_{x_{\infty}}\|\cdot\|v-v_{\infty}\|$ $\displaystyle\ll$ $\displaystyle(\epsilon^{\kappa}\alpha_{n}+\delta_{n})+\epsilon\alpha_{n}.$ For some $\rho=\rho(\epsilon)>0$, there exists a smooth curve $\ell:[0,\rho]\to W^{fs}_{f}(x_{\infty})$ such that $\displaystyle\ell(0)=x_{\infty},\quad\ell^{\prime}(0)=v_{\infty},\quad\ell^{\prime}(t)\in T_{\ell(t)}W^{fs}_{f}(\ell(t)),$ $\displaystyle\hbox{diam}(\ell([0,\rho]))<\epsilon,\quad\quad\|\ell^{\prime}(t)-\ell^{\prime}(0)\|<\epsilon.$ We consider the sequence of curves $\ell_{n}=(f^{-n}gf^{n})\ell$. Note that $\|\ell_{n}^{\prime}(0)\|=\alpha_{n}$, and it follows from the previous computation that, choosing $\epsilon$ sufficiently small, $\|\ell_{n_{i}}^{\prime}(t)-\ell_{n_{i}}^{\prime}(0)\|\leq\frac{1}{3}\|\ell_{n_{i}}^{\prime}(0)\|$ for all $t\in[0,\rho]$ and sufficiently large $i$. Since $\|\ell^{\prime}_{n_{i}}(0)\|\to\infty$, it follows that the distance between $\ell_{n_{i}}(0)$ and $\ell_{n_{i}}(\rho)$ in the ambient Eucledean space goes to infinity as $i\to\infty$. This contradiction proves the proposition. ∎ ###### Proof of Theorem 3.15. By Proposition 3.13(1), the maps $\\{f^{-m_{i}}gf^{m_{i}}|_{W^{fs}_{f}(x)}\\}$ are precompact in $C^{0}$-topology. Then it follows from Proposition 3.17 that the maps $\\{f^{-m_{i}}gf^{m_{i}}|_{W^{fs}_{f}(x)}\\}$ are uniformly bounded in the $C^{1}$-topology. Also, combining Proposition 3.16 and Proposition 3.17, we obtain that for every $z$ and $w$ in a compact neighborhood of $x$ in $W^{fs}_{f}(x)$, $\|D(f^{-m_{i}}gf^{m_{i}})_{z}P_{z}-D(f^{-m_{i}}gf^{m_{i}})_{w}P_{w}\|\ll d^{fs}(z,w)^{\kappa}+\delta_{m_{i}}.$ Since $\delta_{m_{i}}\to 0$, it follows that the maps $\\{f^{-m_{i}}gf^{m_{i}}|_{W^{fs}_{f}(x)}\\}$ are equicontinuous in the $C^{1}$-topology. This implies the theorem. ∎ ### 3.4. Hölder implies $C^{1}$ along fast stable manifolds ###### Theorem 3.18. Let $f_{0},g_{0}\in\hbox{\rm Aff}(X)$ be a good pair, $f,g\in\hbox{\rm Diff}(X)$ $C^{1}$-small perturbations of $f_{0},g_{0}$, and $\Phi:X\to X$ a Hölder isomorphism such that $\Phi\circ f_{0}=f\circ\Phi\quad\hbox{and}\quad\Phi\circ g_{0}=g\circ\Phi,$ and $\Phi(W^{fs}_{f_{0}}(x))=W^{fs}_{f}(\Phi(x))\quad\hbox{for all $x\in X$.}$ Then for a.e. $x\in X$, the maps $\Phi|_{W^{fs}_{f_{0}}(x)}$ and $\Phi^{-1}|_{W^{fs}_{f}(\Phi(x))}$ are $C^{1}$-diffeomorphisms. ###### Proof. Fix a sequence $\\{n_{i}\\}$ and $A\in\hbox{GL}(W^{min}_{f_{0}})$ as in Proposition 3.13(2). For a set of $x\in X$ of full measure, the sequence $\\{f_{0}^{-n_{i}}g_{0}f_{0}^{n_{i}}(x)\\}$ is dense in $X$. In particular, for a.e. $x\in X$ and every $y\in W^{fs}_{f_{0}}(x)$, there exists a subsequence $\\{n_{i_{j}}\\}$ such that $f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}}(x)\to y$. Then by Proposition 3.13(2), for every $v\in W^{min}_{f_{0}}$, (32) $f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}}(\exp(v)x)\to\exp(Av)y$ uniformly on compact sets. For $k\in\mathbb{N}$ and $y\in W^{fs}_{f_{0}}(x)$, we consider maps $\displaystyle\rho^{0}_{k,y}:W^{fs}_{f_{0}}(x)\to W^{fs}_{f_{0}}(x):\exp(v)x\mapsto\exp(A^{k}v)y,$ $\displaystyle\rho^{1}_{k,y}:W^{fs}_{f}(\Phi(x))\to W^{fs}_{f}(\Phi(x)):\Phi(\exp(v)x)\mapsto\Phi\left(\exp(A^{k}v)y\right),$ where $v\in W^{min}_{f_{0}}$. Note that (33) $\rho^{0}_{k,y}=\Phi^{-1}\circ\rho^{1}_{k,y}\circ\Phi.$ In particular, it follows that $\rho^{1}_{k,y}$ is a homeomorphism, and by (32), $\rho^{1}_{1,y}=\lim_{j\to\infty}(f^{-n_{i_{j}}}gf^{n_{i_{j}}})|_{W^{fs}_{f}(\Phi(x))}.$ in the $C^{0}$-topology. By Theorem 3.15, the sequence of maps $(f^{-n_{i_{j}}}gf^{n_{i_{j}}})|_{W^{fs}_{f}(\Phi(x))}$ is precompact in the $C^{1}$-topology. Hence, there exists a subsequence which converges in the $C^{1}$-topology, and $\rho^{1}_{1,y}$ is a $C^{1}$-map for every $y\in W^{fs}_{f_{0}}(x)$. Since each map $\rho_{k,y}^{1}$, $k\geq 1$, is a composition of maps $\rho_{1,z}^{1}$, it is also $C^{1}$. Next, we show that (34) $D(\rho^{1}_{1,y})_{z}\neq 0\quad\hbox{for every $y\in W^{fs}_{f_{0}}(x)$ and $z\in W^{fs}_{f}(\Phi(x))$.}$ Suppose that, to the contrary, $D(\rho^{1}_{1,y_{0}})_{z_{0}}=0$ for some $y_{0}\in W^{fs}_{f_{0}}(x)$ and $z_{0}\in W^{fs}_{f}(\Phi(x))$. We claim that for every $y\in W^{fs}_{f_{0}}(x)$, there exists $y_{1}\in W^{fs}_{f_{0}}(x)$ such that (35) $\rho^{0}_{2,y}=\rho^{0}_{1,y_{1}}\rho^{0}_{1,y_{0}}.$ Indeed, if we write $y=\exp(v)x$, $y_{1}=\exp(v_{1})x$, $y_{0}=\exp(v_{0})x$ for some $v,v_{1},v_{0}\in W^{min}_{f_{0}}$, then (35) is equivalent to $A^{2}w+v=A(Aw+v_{0})+v_{1},\quad w\in W^{min}_{f_{0}},$ and we can take $v_{1}=v-Av_{0}$. Now by (33) and (35), $\rho^{1}_{2,y}=\rho^{1}_{1,y_{1}}\rho^{1}_{1,y_{0}}.$ Hence, $D(\rho^{1}_{2,y})_{z_{0}}=0$ for every $y\in W^{fs}_{f_{0}}(x)$. Similarly, using (33), we deduce that for every $z\in W^{fs}_{f}(\Phi(x))$, there exists $y_{z}\in W^{fs}_{f_{0}}(x)$ such that $\rho^{1}_{1,y_{z}}(z)=z_{0}$. If we fix $y_{2}\in W^{fs}_{f_{0}}(x)$, there exists $y_{z}^{\prime}\in W^{fs}_{f_{0}}(x)$ such that $\rho^{1}_{3,y_{2}}=\rho^{1}_{2,y_{z}^{\prime}}\rho^{1}_{1,y_{z}}.$ Then we have $D(\rho^{1}_{3,y_{2}})_{z}=0\quad\hbox{for every $z\in W^{fs}_{f}(\Phi(x))$.}$ This contradicts the map $\rho^{1}_{3,y_{2}}$ being a homeomorphism, and (34) follows. We have proved that $\rho^{1}_{1,y}$ is a $C^{1}$-diffeomorphism for every $y\in W^{fs}_{f_{0}}(x)$. This implies that the map $\rho^{1}_{0,y}$, which can be represented as a composition of $\rho^{1}_{1,z_{1}}$ and $(\rho^{1}_{1,z_{2}})^{-1}$, is also a $C^{1}$-diffeomorphism for every $y\in W^{fs}_{f_{0}}(x)$. We have a free transitive action of $W^{min}_{f_{0}}$ on $W^{fs}_{f}(\Phi(x))$ defined by (36) $s(v,\Phi(\exp(w)x))=\Phi(\exp(v+w)x)$ where $v,w\in W^{min}_{f_{0}}$. Note that the action $s:W^{min}_{f_{0}}\times W^{fs}_{f}(\Phi(x))\to W^{fs}_{f}(\Phi(x))$ is continuous, and since $s(v,\Phi(\exp(w)x))=\rho^{1}_{0,\exp(v)x}(\Phi(\exp(w)x)),$ the map $s(v,\cdot)$ is a $C^{1}$-diffeomorphism for every $v\in W^{min}_{f_{0}}$. Hence, by the Bochner–Montgomery theorem [3], the map $s$ is $C^{1}$. Now it follows from (36) that the map $\Phi_{x}(v):=\Phi(\exp(v)x)$ is $C^{1}$. Suppose that for some $v\in W^{min}_{f_{0}}$, we have $\phi^{\prime}(0)=0$ where $\phi(t)=\Phi(\exp(tv)x)$. Then since $\phi(t_{1}+t)=\rho^{1}_{0,\exp(tv)x}(\Phi(\exp(t_{1}v)x))$, it follows that $\phi^{\prime}(t)=0$ for every $t\in\mathbb{R}$. This contradicts the action $s$ being free. Hence, we conclude that $D(\Phi_{x})_{0}$ is nondegenerate, and because $\Phi(\exp(v+w)x)=\rho^{1}_{0,\exp(v)x}(\Phi(\exp(w)x)),$ $D(\Phi_{x})_{v}$ is nondegenerate for every $v\in W^{min}_{f_{0}}$. This shows that $\Phi|_{W^{fs}_{f_{0}}(x)}$ is a $C^{1}$-diffeomorphism for a.e. $x\in X$. ∎ ### 3.5. Completion of the proof of the main theorem Let $\\{f_{0},g_{0}\\}\subset\alpha_{0}(\Gamma)$ be a good pair and $\\{f,g\\}\subset\alpha_{1}(\Gamma)$ its conjugate under $\Phi$. We use notation $A$, $\\{n_{i}\\}$, $\omega$ as in Proposition 3.13 and Remark 3.14. Recall that $A=\lim_{i\to\infty}\omega^{-n_{i}}P^{min}_{f_{0}}(Dg_{0})\omega^{n_{i}}.$ Hence, replacing the pair $\\{f_{0},g_{0}\\}$ by the pair $\\{f_{0},f_{0}^{l}g_{0}\\}$ for some $l\geq 1$, we can get a good pair with $A$ satisfying $\|A\|<1$, which we now assume. By Theorem 3.5, $\Phi(W^{fs}_{f_{0}}(x))=W^{fs}_{f}(\Phi(x))$ for all $x\in X$, so we consider the maps $\displaystyle\alpha^{0}_{x}:W^{fs}_{f_{0}}(x)\to W^{fs}_{f_{0}}(x):\exp(v)x\mapsto\exp(Av)x,$ $\displaystyle\alpha_{x}:W^{fs}_{f}(\Phi(x))\to W^{fs}_{f}(\Phi(x)):\Phi(\exp(v)x)\mapsto\Phi\left(\exp(Av)x\right),$ where $v\in W^{min}_{f_{0}}$. Note that (37) $\Phi\circ\alpha^{0}_{x}=\alpha_{x}\circ\Phi.$ In particular, it follows that $\alpha_{x}$ is a homeomorphism. For a.e. $x\in X$, there exists a subsequence $\\{n_{i_{j}}\\}$ such that $f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}}(x)\to x$ as $j\to\infty$. Then by Proposition 3.13, $\alpha^{0}_{x}=\lim_{j\to\infty}(f_{0}^{-n_{i_{j}}}g_{0}f_{0}^{n_{i_{j}}})|_{W^{fs}_{f_{0}}(x)},$ and by (37), $\alpha_{x}=\lim_{j\to\infty}(f^{-n_{i_{j}}}gf^{n_{i_{j}}})|_{W^{fs}_{f}(\Phi(x))}$ in the $C^{0}$-topology. It follows from Theorem 3.15 that the sequence of maps $(f^{-n_{i_{j}}}gf^{n_{i_{j}}})|_{W^{fs}_{f}(\Phi(x))}$ is precompact in the $C^{1}$-topology. Hence, it contains a subsequence which converges in the $C^{1}$-topology, and $\alpha_{x}$ is a $C^{1}$-map for a.e. $x\in X$. Let $W^{min}_{f_{0},g_{0}}$ be the sum of eigenspaces of $A$ with eigenvalues of minimal modulus. Our aim is to show that the map $\Phi$ restricted to the leaves $\exp(W^{min}_{f_{0},g_{0}})x$ is linear in suitable $C^{\infty}$-coordinate systems which depend continuously on $x$. The first step is to show that the maps $\alpha_{x}$ are linear in suitable coordinates on the fast stable leaves. Consider the measurable function $\sigma(x)=\sup\\{\|D(f^{-n}gf^{n})_{\Phi(x)}P_{\Phi(x)}\|:\,\,n\in\mathbb{N}\\},$ which is well defined by Proposition 3.17. For $c>0$, let $X(c)$ be the subset of $x\in X$ such that $\sigma(x)\leq c$ and the sequence $\\{f_{0}^{-n_{i}}g_{0}f_{0}^{n_{i}}(x)\\}$ has $x$ as an accumulation point. By property (iv) of good pair and Proposition 3.17, the set $\cup_{c>0}X(c)$ has full measure in $X$. Let $Df_{0}|_{W^{min}_{f_{0}}}=\lambda\cdot\omega$ where $\lambda\in\mathbb{R}$, $|\lambda|<1$, and $\omega\in\hbox{Isom}(W^{min}_{f_{0}})$. Using the Poincare recurrence theorem, for a.e. $(x,\omega^{\prime})\in X(c)\times\hbox{Isom}(W^{min}_{f_{0}})$, one can construct a sequence $\\{k_{j}\\}$, $k_{0}=0$, such that $\displaystyle f_{0}^{k_{j}}(x)\in X(c)\quad\hbox{for every $j\geq 1$}\quad\hbox{and}\quad\omega^{k_{j}}\omega^{\prime}\to\omega^{\prime}\quad\hbox{as $j\to\infty$.}$ Then $\omega^{k_{j}}\to id$. Hence, by the Fubini theorem, for a.e. $x\in X(c)$, there exists a sequence $\\{k_{j}\\}$, $k_{0}=0$, such that (38) $f_{0}^{k_{j}}(x)\in X(c)\quad\hbox{for every $j\geq 1$}\quad\hbox{and}\quad\omega^{k_{j}}\to id\quad\hbox{as $j\to\infty$.}$ Now we assume that $x\in X(c)$ satisfies (38). Let $\\{n^{(j)}_{i}\\}$ be a subsequence such that $(f_{0}^{-n^{(j)}_{i}}g_{0}f_{0}^{n^{(j)}_{i}})f_{0}^{k_{j}}(x)\to f_{0}^{k_{j}}(x)\quad\hbox{as $i\to\infty$,}$ Then by Proposition 3.13(2), (39) $(f_{0}^{-n^{(j)}_{i}}g_{0}f_{0}^{n^{(j)}_{i}})|_{W^{fs}_{f_{0}}(f_{0}^{k_{j}}(x))}\to\alpha^{0}_{f_{0}^{k_{j}}(x)}\quad\hbox{as $i\to\infty$}$ in the $C^{0}$-topology. A direct computation shows that $\alpha_{x,j}^{0}=f_{0}^{-k_{j}}\circ\alpha^{0}_{f_{0}^{k_{j}}(x)}\circ f_{0}^{k_{j}}$ where $\alpha^{0}_{x,j}:W^{fs}_{f_{0}}(x)\to W^{fs}_{f_{0}}(x):\exp(v)x\mapsto\exp((\omega^{-k_{j}}A\omega^{k_{j}})v)x,\quad v\in W^{min}_{f}.$ Clearly, $\alpha^{0}_{x,j}\to\alpha^{0}_{x}$ as $j\to\infty$ in the $C^{0}$-topology. It follows from (37) that (40) $\alpha_{x,j}=f^{-k_{j}}\circ\alpha_{f_{0}^{k_{j}}(x)}\circ f^{k_{j}}$ where $\alpha_{x,j}=\Phi\circ\alpha^{0}_{x,j}\circ\Phi^{-1}\to\alpha_{x}\quad\hbox{as $j\to\infty$}$ in the $C^{0}$-topology. Since $f$ is $C^{1}$-close to the algebraic map $f_{0}$, its Mather spectrum on fast stable leaves is contained in a small interval, and by the nonstationary Sternberg linearization [12, 11], $f|_{W^{fs}_{f}(z)}$ is linear in suitable coordinate systems. Namely, there exists a family of $C^{\infty}$-diffeomorphisms $L_{z}:W^{min}_{f_{0}}\to W^{fs}_{f}(z),\quad z\in X,$ such that the map $z\mapsto L_{z}$ is continuous in the $C^{\infty}$-topology, $L_{z}(0)=z$, $D(L_{z})_{0}=id$, and (41) $(L_{f(z)}^{-1}\circ f\circ L_{z})(v)=\rho(z)v,\quad v\in W^{min}_{f_{0}},$ with $\rho(z)\in\hbox{GL}(W^{min}_{f_{0}})$, $\|\rho(z)\|<1$. Consider the sequence of maps $G_{k}=L_{f^{k}(\Phi(x))}^{-1}\circ\alpha_{f_{0}^{k}(x)}\circ L_{f^{k}(\Phi(x))}:W^{min}_{f_{0}}\to W^{min}_{f_{0}}.$ We claim that the sequence of maps $G_{k_{j}}$ restricted to compact sets is uniformly bounded and equicontinuous in the $C^{1}$-topology. This is equivalent to the sequence $\\{\alpha_{f_{0}^{k_{j}}(x)}\\}$ being uniformly bounded and equicontinuous in the $C^{1}$-topology. It follows from (37), (39), and (40) that $F_{i,j}:=(f^{-k_{j}-n^{(0)}_{i}}gf^{n^{(0)}_{i}+k_{j}})|_{W^{fs}_{f}(\Phi(x))}\to\alpha_{x,j}\quad\hbox{as $i\to\infty$}$ in the $C^{0}$-topology, and by Theorem 3.15, we may assume, after passing to a subsequence, that convergence also holds in the $C^{1}$-topology. By (40), $\displaystyle\alpha_{f_{0}^{k_{j}}(x)}=$ $\displaystyle(f^{k_{j}}\circ\alpha_{x,j}\circ f^{-k_{j}})|_{W^{fs}_{f}(f^{k_{j}}(\Phi(x)))}.$ We observe that $(f^{k_{j}}\circ F_{i,j}\circ f^{-k_{j}})|_{W^{fs}_{f}(f^{k_{j}}(\Phi(x)))}$ converges in the $C^{1}$-topology to $\alpha_{f_{0}^{k_{j}}(x)}$ as $i\to\infty$, and since $f_{0}^{k_{j}}(x)\in X(c)$ for all $j$, the derivative of $(f^{k_{j}}\circ F_{i,j}\circ f^{-k_{j}})|_{W^{fs}_{f}(f^{k_{j}}(\Phi(x)))}=(f^{-n_{i}^{(0)}}gf^{n_{i}^{(0)}})|_{W^{fs}_{f}(f^{k_{j}}(\Phi(x)))}$ is uniformly bounded over compact sets and $i\in\mathbb{N}$. This implies that the sequence $\\{\alpha_{f_{0}^{k_{j}}(x)}\\}$ is uniformly bounded in the $C^{1}$-topology. To prove equicontinuity, we observe that for $z,w\in W^{fs}_{f}(f^{k_{j}}(\Phi(x)))$, $\displaystyle\left\|D(\alpha_{f_{0}^{k_{j}}(x)})_{z}P_{z}-D(\alpha_{f_{0}^{k_{j}}(x)})_{w}P_{w}\right\|$ $\displaystyle\leq\|D(f^{k_{j}}\circ(\alpha_{x,j}-F_{i,j})\circ f^{-k_{j}})_{z}P_{z}\|$ $\displaystyle+\|D(f^{k_{j}}F_{i,j}f^{-k_{j}})_{z}P_{z}-D(f^{k_{j}}F_{i,j}f^{-k_{j}})_{w}P_{w}\|$ $\displaystyle+\|D(f^{k_{j}}\circ(F_{i,j}-\alpha_{x,j})\circ f^{-k_{j}})_{w}P_{w}\|.$ Since $F_{i,j}\to\alpha_{x,j}$ as $i\to\infty$ in the $C^{1}$-topology, taking $i=i(j)$ sufficiently large, we can make the first and the last terms arbitrary small. To estimate the middle term, we use that $f_{0}^{k_{j}}(x)\in X(c)$ for all $j$ and Proposition 3.16. We get $\|D(f^{k_{j}}F_{i,j}f^{-k_{j}})_{z}P_{z}-D(f^{k_{j}}F_{i,j}f^{-k_{j}})_{w}P_{w}\|\ll d^{fs}(z,w)^{\kappa}+\delta_{n_{i}^{(0)}},$ where $\delta_{n}\to 0$ as $n\to\infty$. This proves equicontinuity, and in fact, the stronger conclusion: (42) $\left\|D(\alpha_{f_{0}^{k_{j}}(x)})_{z}P_{z}-D(\alpha_{f_{0}^{k_{j}}(x)})_{w}P_{w}\right\|\ll d^{fs}(z,w)^{\kappa}.$ Let $\rho_{k}=\prod_{s=k-1}^{0}\rho(f^{s}(\Phi(x)))$ with $\rho$ defined as in (41). Since $f$ is $C^{1}$-close to the map $f_{0}$, which is conformal on the fast stable leaves, it follows that for some $\lambda<1$ and small $\epsilon>0$, we have (43) $\|\rho_{k}(x)\|\ll(\lambda+\epsilon)^{k}\quad\hbox{and}\quad\|\rho_{k}(x)^{-1}\|\ll(\lambda-\epsilon)^{-k}$ uniformly on $x\in X$ and $k\in\mathbb{N}$. We deduce from (40) and (41) that (44) $G_{0,j}=\rho_{k_{j}}^{-1}G_{k_{j}}(\rho_{k_{j}}v),\quad v\in W^{min}_{f_{0}},$ where $G_{0,j}=L_{x}^{-1}\circ\alpha_{x,j}\circ L_{x}\to G_{0}$ as $j\to\infty$. Fix a basis $\\{e_{\ell}\\}$ of $W^{min}_{f_{0}}$ and write $G_{k}(v)=\sum_{\ell}G_{k,\ell}(v)e_{\ell}.$ Applying the mean value theorem to the functions $t\mapsto G_{k_{j},\ell}(t\rho_{k_{j}}v)$, $t\in[0,1]$, we deduce that (45) $G_{k_{j},\ell}(\rho_{k_{j}}v)=\sum_{s}\frac{\partial G_{k_{j},\ell}}{\partial x_{s}}(t_{j,\ell}(v)\rho_{k_{j}}v)(\rho_{k_{j}}v)_{s}$ for some $t_{j,\ell}(v)\in[0,1]$. Hence, by (44), (46) $G_{0,j}(v)=(\rho_{k_{j}}^{-1}B_{j}(v)\rho_{k_{j}})v$ where $B_{j}(v)$ is the linear map of $W^{min}_{f_{0}}$ with coefficients coming from (45). Let $B_{j}=D(G_{k_{j}})_{0}$. From (42), we deduce that $\|B_{j}(v)-B_{j}\|\ll\|\rho_{k_{j}}v\|^{\kappa},$ and it follows from (43) that (47) $\|\rho_{k_{j}}^{-1}B_{j}(v)\rho_{k_{j}}-\rho_{k_{j}}^{-1}B_{j}\rho_{k_{j}}\|\to 0\quad\hbox{as $j\to\infty$.}$ Let $B=B(x)$ be a limit point of the sequence of maps $\rho_{k_{j}}^{-1}B_{j}(v)\rho_{k_{j}}$. The crucial point of our argument is that $B(x)$ is independent of $v$ because of the equicontinuity estimate. Taking $j\to\infty$, we deduce from (46) that $G_{0}(v)=B\,v$, and by the definition of $G_{0}$, (48) $L_{x}^{-1}(\Phi(\exp(Av)x))=B(x)\,L_{x}^{-1}(\Phi(\exp(v)x))).$ This equality holds for a.e. $x\in X(c)$ with $c>0$ and hence, for a.e. $x\in X$. Note that since $\Phi$ is a homeomorphism, the linear map $B(x)$ is nondegenerate. Although $B(x)$ is defined only for a.e. $x\in X$, it follows from (48) that it can be extended continuously to the whole space so that (48) holds everywhere. Now we consider the maps $\Phi_{x}(v)=L_{\Phi(x)}^{-1}(\Phi(\exp(v))x),\quad x\in X,\;v\in W^{min}_{f_{0}},$ which satisfy the equivariance relation $\Phi_{x}\circ A=B(x)\circ\Phi_{x}$. Recall that by Theorem 3.18, $\Phi_{x}$ is a $C^{1}$-diffeomorphism for a.e. $x\in X$. Hence, we have $D(\Phi_{x})_{0}A=B(x)D(\Phi_{x})_{0}$, and it follows that the map $\Psi_{x}:=D(\Phi_{x})_{0}^{-1}\circ\Phi_{x}$ commutes with the contraction $A$. We write $A|_{W^{min}_{f_{0},g_{0}}}=\rho\cdot\theta$ where $\rho\in\mathbb{R}$, $|\rho|<1$, and $\theta\in\hbox{Isom}(W^{min}_{f_{0},g_{0}})$. Since $W^{min}_{f_{0},g_{0}}=\\{v\in W^{min}_{f_{0}}:\,\|A^{n}v\|=O(\rho^{n})\quad\hbox{as $n\to\infty$}\\},$ and $\Psi_{x}$ is a $C^{1}$-map, we deduce that $\Psi_{x}(W^{min}_{f_{0},g_{0}})\subset W^{min}_{f_{0},g_{0}}.$ We claim that $\Psi_{x}|_{W^{min}_{f_{0},g_{0}}}$ is linear. We fix a basis $\\{e_{\ell}\\}$ of $W^{min}_{f_{0},g_{0}}$ and write $\Psi:=\Psi_{x}$ as $\Psi(v)=\sum_{\ell}\Psi_{\ell}(v)e_{\ell},\quad v\in W^{min}_{f_{0},g_{0}}.$ By the mean value theorem, $\Psi_{\ell}(A^{n}v)=\sum_{s}\frac{\partial\Psi_{\ell}}{\partial x_{s}}(t_{\ell}(v)A^{n}v)(A^{n}v)_{s}$ for some $t_{\ell}(v)\in[0,1]$. Hence, (49) $\Psi(v)=A^{-n}\Psi(A^{n}v)=(A^{-n}C_{n}(v)A^{n})v=(\theta^{-n}C_{n}(v)\theta^{n})v$ where $C_{n}(v)$ is the matrix with coefficients $\frac{\partial\Psi_{\ell}}{\partial x_{s}}(t_{\ell}(v)A^{n}v)$. Since $\Psi$ is a $C^{1}$-map and $\|A\|<1$, it follows that $C_{n}(v)\to D(\Psi)_{0}$ as $n\to\infty$. Passing to a subsequence, we may assume that the sequence of isometries $\theta^{n}$ also converges. Then it follows from (49) that $\Psi$ is linear. We conclude that for a.e. $x\in X$, there exists a linear map $C(x):W^{min}_{f_{0},g_{0}}\to W^{min}_{f_{0},g_{0}}$ such that (50) $\Phi(\exp(v)x)=L_{\Phi(x)}(C(x)v),\quad v\in W^{min}_{f_{0},g_{0}}.$ It follows from this relation that $C(x)$ is nondegenerate. Moreover, by continuity, we may assume that (50) holds for all $x\in X$, and $C(x)$ depends continuously on $x$. Now relation (50) also implies that $\Phi$ is a $C^{\infty}$-diffeomorphism along the leaves $\exp(W^{min}_{f_{0},g_{0}})x$, and the partial derivatives along this leaves depend continuously on $x\in X$. Note that if $\\{f_{0},g_{0}\\}$ is a good pair, then $\\{h^{-1}f_{0}h,h^{-1}f_{0}h\\}$ is good as well for every $h\in\alpha_{0}(\Gamma)$, and we have $W^{min}_{h^{-1}f_{0}h,h^{-1}g_{0}h}=(Dh)^{-1}W^{min}_{f_{0},g_{0}}$. Hence, it follows from the irreducibility of the $\Gamma$-action on $\hbox{Lie}(G)$ that (51) $\sum_{\\{f_{0},g_{0}\\}\subset\alpha_{0}(\Gamma)\hbox{\tiny- good}}W^{min}_{f_{0},g_{0}}=\hbox{\rm Lie}(G).$ Now we consider the elliptic differential operator $\mathcal{D}^{s}=\sum_{i}\frac{\partial^{2s}}{\partial x_{i}^{2s}}$ where the partial derivatives $\frac{\partial}{\partial x_{i}}$ span the tangent space and are taken in directions of $W^{min}_{f_{0},g_{0}}$ for some choice of good pairs $\\{f_{0},g_{0}\\}$. It follows from the previous paragraph that $\mathcal{D}^{s}\Psi$ is continuous for every $s\geq 2$. Hence, by the regularity of solutions of elliptic PDE, $\Psi$ is $C^{\infty}$. Since $D(\Phi)_{x}$ is onto when restricted to fast stable distributions of good $f_{0}$ and its conjugate $f$, it follows that $D(\Phi)_{x}$ is onto as well. Thus, $\Psi^{-1}$ is $C^{\infty}$ by the inverse function theorem. ## 4\. Existence of good pairs ### 4.1. Tori In this section, we set $X=\mathbb{T}^{d}$, $d\geq 2$, and prove ###### Proposition 4.1. Let $\Gamma$ be a subgroup of $\hbox{\rm Aff}(X)$ such that the Zariski closure of $D\Gamma$ contains $\hbox{\rm SL}_{d}$. Then $\Gamma$ contains a good pair. We will use the following lemma, which is easy to prove using Fourier analysis (see, for example, [2, Corollary 1.6 and Remark 1.8]). Let $\phi$ be the Euler totient function. ###### Lemma 4.2. Let $f_{1},f_{2}\in\hbox{\rm Aff}(X)$ be such that for every $l\geq 1$ satisfying $\phi(l)\leq d^{2}$, the map $Df_{1}^{-l}Df_{2}^{l}$ does not have eigenvalue 1. Then for every $\phi_{1},\phi_{2}\in L^{2}(X)$, $\int_{X}\phi_{1}(f_{1}^{n}(x))\phi_{2}(f_{2}^{n}(x))d\mu(x)\to\left(\int_{X}\phi_{1}\,d\mu\right)\left(\int_{X}\phi_{2}\,d\mu\right)\quad\hbox{as $n\to\infty$.}$ If the conclusion of Lemma 4.2 holds, then we call the pair $\\{f_{1},f_{2}\\}$ mixing. Mixing pairs can be used to construct affine maps satisfying property (iv) of good pairs. ###### Lemma 4.3. Let $f,g\in\hbox{\rm Aff}(X)$ and suppose the pair $\\{f^{-1},gf^{-1}g^{-1}\\}$ is mixing. Then for every subsequence $\\{n_{i}\\}$ and for a.e. $x\in X$, the sequence $\\{f^{-n_{i}}gf^{n_{i}}(x)\\}_{n\geq 0}$ is dense in $X$. ###### Proof. We have $\int_{X}\phi_{1}(gf^{-n}g^{-1}(x))\phi_{2}(f^{-n}(x))d\mu(x)\to\left(\int_{X}\phi_{1}\,d\mu\right)\left(\int_{X}\phi_{2}\,d\mu\right)\quad\hbox{as $n\to\infty$}$ for every $\phi_{1},\phi_{2}\in L^{2}(X)$. By invariance of the measure, this also implies that $\int_{X}\phi_{1}(x)\phi_{2}(f^{-n}gf^{n}(x))d\mu(x)\to\left(\int_{X}\phi_{1}\,d\mu\right)\left(\int_{X}\phi_{2}\,d\mu\right)\quad\hbox{as $n\to\infty$}$ for every $\phi_{1},\phi_{2}\in L^{2}(X)$. Now we show that for $\delta_{n}=f^{-n}gf^{n}$, the sequence $\\{\delta_{n_{i}}x\\}$ is dense in $X$ for a.e. $x\in X$. Let $U$ be a nonempty open subset of $X$ and $A=\cup_{i\geq 0}\delta_{n_{i}}^{-1}(U)$. We have $0=\int_{X}\chi_{U}(\delta_{n_{i}}(x))\chi_{A^{c}}(x)\,d\mu(x)\to\mu(U)\mu(A^{c}).$ This implies that $\mu(A^{c})=0$, i.e. for a.e. $x\in X$, $\\{\delta_{n_{i}}x\\}_{i\geq 0}\cap U\neq\emptyset.$ Since $X$ has countable base of topology, this proves the lemma. ∎ ###### Proof of Proposition 4.1. Since $D\Gamma$ is Zariski dense, there is $f\in\Gamma$ such that $Df$ is $\mathbb{R}$-regular (see [1, 21]). In particular, $Df$ is semisimple and hyperbolic. Because of Lemmas 4.2 and 4.3, it suffices to find $g\in\Gamma$ such that $Dg$ belongs to the set $\displaystyle\left\\{X\in\hbox{SL}_{d}:\;\;\det(P^{min}_{f}X|_{W^{min}_{f}})\neq 0,\quad\det([Df^{l},X]-id)\neq 0\;\;\hbox{for $\phi(l)\leq d^{2}$}\right\\}.$ One can check that this is a nonempty Zariski open subset of $\hbox{SL}_{d}$. Hence, existence of such $g\in\Gamma$ follows from Zariski density. ∎ ### 4.2. Semisimple groups Let $G$ be a connected semisimple Lie groups with no compact factors, $\Lambda$ a lattice in $G$, and $X=G/\Lambda$. ###### Proposition 4.4. Let $\Gamma$ be a subgroup of $\hbox{\rm Aff}(X)$ such that the Zariski closure of $D\Gamma$ contains $\hbox{\rm Ad}(G)$. Then $\Gamma$ contains a good pair. ###### Proof. Since $D\Gamma$ contains a finite index subgroup consisting of inner automorphisms, we may assume without loss of generality that $D\Gamma$ is a subgroup of $\hbox{Ad}(G)$. It follows from Zariski density [1, 21] that $\Gamma$ contains an element $f$ such that $Df$ is $\mathbb{R}$-regular. In particular, it is partially hyperbolic and semisimple, and hence it satisfies properties (i)–(ii) of the definition of a good pair. If we choose $g\in\Gamma$ so that the pair $\\{f^{-1},gf^{-1}g^{-1}\\}$ is mixing, then by Lemma 4.3, $f$ and $g$ will satisfy property (iv) of the definition of a good pair. By the Howe–Moore theorem, the pair $\\{f^{-1},gf^{-1}g^{-1}\\}$ is mixing provided that for all projections $\pi_{i}:\hbox{Ad}(G)\to\hbox{Ad}(G_{i})$ on simple factors of $\hbox{Ad}(G)$, the sequence $\\{\pi_{i}(Dg(Df)^{-n}(Dg)^{-1}(Df)^{n})\\}$ is divergent. Since $\pi_{i}(Df)$ is also $\mathbb{R}$-regular, $P_{i}=\\{g\in G_{i}:\,\pi_{i}(Df)^{-n}\cdot g\cdot\pi_{i}(Df)^{n}\hbox{ is nondivergent}\\}$ is a proper parabolic subgroup of $G_{i}$. By Zariski density, there exists $g\in\Gamma$ such that $\pi_{i}(Dg)\notin P_{i}$ for all $i$, and $P_{f}^{min}(Dg):W^{min}_{f}\to W^{min}_{f}$ is nondegenerate. Such $f$ and $g$ provide a good pair. ∎ ## References * [1] Y. Benoist and F. Labourie, Sur les difféomorphismes d’Anosov affines à feuilletages stable et instable différentiables. Invent. Math. 111 (1993), no. 2, 285–-308. * [2] V. Bergelson and A. Gorodnik, Ergodicity and mixing of non-commuting epimorphisms. Proc. Lond. Math. Soc. (3) 95 (2007), no. 2, 329–359. * [3] S. Bochner and D. Montgomery, Groups of differentiable and real or complex analytic transformations. Ann. of Math. (2) 46, (1945). 685–694. * [4] E. Cawley, The Teichmüller space of the standard action of $\hbox{\rm SL}(2,\mathbf{Z})$ on $T^{2}$ is trivial. Internat. Math. Res. Notices 1992, no. 7, 135–141. * [5] D. Fisher, Local Rigidity: Past, Present, Future. in Dynamics, Ergodic Theory and Geometry (Mathematical Sciences Research Institute Publications), 45–98, Cambridge University Press, 2007. * [6] D. Fisher, Bending group actions and cohomology of arithmetic groups, in preparation. * [7] D. Fisher, Deformations of group actions. Trans. Amer. Math. Soc. 360 (2008), no. 1, 491–505. * [8] D. Fisher and T. J. Hitchman, Cocycle superrigidity and harmonic maps with infinite-dimensional targets. Int. Math. Res. Not. 2006, Art. ID 72405, 19 pp. * [9] D. Fisher and T. J. Hitchman, Harmonic Maps into Infinite Dimensional Manifolds and Cocycle Superrigidity, in preparation. * [10] D. Fisher and G. Margulis, Local rigidity for affine actions of higher rank Lie groups and their lattices, preprint. * [11] M. Guysinsky, The theory of non-stationary normal forms. Ergodic Theory Dynam. Systems 22 (2002), no. 3, 845–862. * [12] M. Guysinsky and A. Katok, Normal forms and invariant geometric structures for dynamical systems with invariant contracting foliations. Math. Res. Lett. 5 (1998), no. 1-2, 149–163. * [13] M. Hirsch; C. Pugh; M. Shub, Invariant manifolds. Lecture Notes in Mathematics, Vol. 583. Springer-Verlag, Berlin-New York, 1977. * [14] T. J. Hitchman, Deformations and smooth rigidity for toral actions of lattices in rank one groups, Ph.D. thesis, University of Michigan, 2003. * [15] A. Katok and B. Hasselblatt, Introduction to the Modern Theory of Dynamical Systems. Cambridge University Press, Cambridge, 1995. * [16] A. Katok and J. Lewis, Local rigidity for certain groups of toral automorphisms. Israel J. Math. 75 (1991), no. 2-3, 203–241. * [17] A. Katok and R. Spatzier, Differential rigidity of Anosov actions of higher rank Abelian groups and applications to rigidity, Proc. Steklov Inst. Math. 216 (1997) 292-319. * [18] W. Kryszewski and S. Plaskacz, Topological methods for the local controllability of nonlinear systems, SIAM J. Control and Optimization 32:1 (1994), 213–223. * [19] G. Margulis and N. Qian, Rigidity of weakly hyperbolic actions of higher real rank semisimple Lie groups and their lattices. Ergodic Theory Dynam. Systems 21 (2001), no. 1, 121–164. * [20] Y. Pesin, Lectures on partial hyperbolicity and stable ergodicity. Zurich Lectures in Advanced Mathematics. European Mathematical Society (EMS), Zürich, 2004. * [21] G. Prasad, $\mathbb{R}$-regular elements in Zariski-dense subgroups. Quart. J. Math. Oxford Ser. (2) 45 (1994), no. 180, 541–-545. * [22] B. Schmidt, Weakly hyperbolic actions of Kazhdan groups on tori. Geom. Funct. Anal. 16 (2006), no. 5, 1139–1156.

arxiv-papers

2024-09-04T02:48:54.594151

0803.4224

# A New two-dimensional Second Order Non-oscillatory Central Scheme Applied to multiphase flows in heterogeneous porous media F. Furtado2, F. Pereira1 and S. Ribeiro1 1 State University of Rio de Janeiro, Nova Friburgo, RJ 28630-050, Brazil. 2 University of Wyoming, Laramie, WY 82071-3036, U.S.A. ## Abstract We are concerned with central differencing schemes for solving scalar hyperbolic conservation laws. We compare the Kurganov-Tadmor (KT) two- dimensional (Kurganov and Tadmor, 2000) second order semi-discrete central scheme in dimension by dimension formulation with a new two-dimensional approach introduced here and applied in numerical simulations for two-phase, two-dimensional flows in heterogeneous formations. This semi-discrete central scheme is based on the ideas of Rusanov’s method (Rusanov (1961)) using a more precise information about the local speeds of wave propagation computed at each Riemann Problem in two-space dimensions. We find the KT dimension by dimension has a much simpler mathematical description than the genuinely two- dimensional one with a little more numerical diffusion, particularly in the presence of viscous fingers. Unfortunately, as one can see in Abreu et al. (2007), the KT with the dimension by dimension approach might produce incorrect boundary behavior in a typical geometry used in the study of porous media flows: the quarter of a five spot. This problem has been corrected by the authors with the new semi-discrete scheme proposed here. We conclude with numerical examples of two-dimensional, two-phase flow associated with two distinct flooding problems: a two-dimensional flow in a rectangular heterogeneous reservoir (called slab geometry) and a two-dimensional flow in a 5-spot geometry homogeneous reservoir. ## 1\. INTRODUCTION We consider a model for two-phase flow, immiscible and incompressible displacement in heterogeneous porous media. The highly nonlinear governing equations are of very practical importance in petroleum engineering (Peaceman, 1977; Chavent and Jaffré, 1986) (see also (Furtado and Pereira, 2003) and the references therein for recent studies for the scale-up problem for such equations). The conventional theoretical description of two-phase flow in a porous medium, in the limit of vanishing capillary pressure, is via Darcy’s law coupled to the Buckley-Leverett equation. The two phases will be referred to as water and oil, and indicated by the subscripts $w$ and $o$, respectively. We also assume that the two fluid phases saturate the pores. With no sources or sinks, and neglecting the effects of gravity these equations are (see Peaceman (Peaceman, 1977)): $\nabla\cdot{\bf v}=0,\quad{\bf v}=-\lambda(s)K({\bf x})\nabla p,$ (1) $\frac{\partial s}{\partial t}+\nabla\cdot(f(s){\bf v})=0.$ (2) Here, ${\bf v}$ is the total seepage velocity, $s$ is the water saturation, $K({\bf x})$ is the absolute permeability, and $p$ is the pressure. The constant porosity has been scaled out by a change of the time variable. The constitutive functions $\lambda(s)$ and $f(s)$ represent the total mobility and the fractional flow of water, respectively. We are concerned with numerical schemes for solving scalar hyperbolic conservation laws arising in the simulation of multiphase flows in multidimensional heterogeneous porous media. These schemes are non-oscillatory and enjoy the main advantage of Godunov-type central schemes: simplicity, i.e., they employ neither characteristic decomposition nor approximate Riemann solvers. This makes them universal methods that can be applied to a wide variety of physical problems, including hyperbolic systems of conservation laws. The two main classes of Godunov methods are upwind and central schemes. The Lax-Friedrichs (LxF) scheme (Lax, 1954) is the canonical first order central scheme, which is the forerunner of all central differencing schemes. It is a non-oscillatory scheme based on piecewise constant approximate solution and it also enjoys simplicity. Unfortunately the excessive numerical dissipation in the LxF recipe (of order ${\mathcal{O}}(\Delta X^{2}/\Delta t)$) yields a poor resolution, which seems to have delayed the development of high resolution central schemes when compared with the earlier developments of the high resolution upwind methods. Only in 1990 a second order generalization to the LxF scheme was introduced in (Nessyahu and Tadmor, 1990). They used a staggered form of the LxF scheme and replaced the first order piecewise constant solution with a van Leer’s MUSCL-type piecewise linear second order approximation (Van Leer, 1979). The numerical dissipation present in this new central scheme has an amplitude of order ${\mathcal{O}}({\Delta X}^{4}/\Delta t)$. (see (Abreu et al., 2004c), (Abreu et al., 2004b), (Abreu et al., 2004a), (Abreu et al., 2006), for recent studies in three phase flows using the Nessyahu and Tadmor (NT) central scheme). In spite of the good resolution obtained by the Nessyahu and Tadmor scheme, much higher than in the first order LxF scheme, there are still some difficulties with small time steps which arise, e.g. in multiphase flows in highly heterogeneous petroleum reservoirs or aquifers. To overcome this difficulty, we can use a semi- discrete formulation coupled with an appropriate ODE solver retaining simplicity and high resolution with lower numerical viscosity, proportional to the vanishing size of the time step $\Delta t$. Both LxF and NT schemes do not admit a semi-discrete form; see (Kurganov and Tadmor, 2000) for a detailed description of the one-dimensional Kurganov and Tadmor central scheme which is the first fully discrete Godunov-Type central scheme admitting a semi-discrete form. We compare the Kurganov-Tadmor (KT) two-dimensional (Kurganov and Tadmor, 2000) second order semi-discrete central scheme in dimension by dimension formulation with a genuinely two-dimensional approach applied in numerical simulations for two-phase, two-dimensional flows in heterogeneous formations. We find the KT dimension by dimension has a much simpler mathematical description than the genuinely two-dimensional one adding only a little more diffusion, particularly in the presence of viscous fingers. Unfortunately, the KT with the dimension by dimension approach might produce incorrect boundary behavior in a typical geometry used in the study of porous media flows: the quarter of a five spot. These results are presented in (Abreu et al., 2007). This problem motivated the authors to develop a genuinely two-dimensional formulation which is then presented in section (2.2). Although a similar two- dimensional formulation was available in a early work (Kurganov and Petrova, 2001), ours was developed independently to deal with two-phase flows, immiscible and incompressible displacement in heterogeneous porous media. It shares the same general ideas with the work of Kurganov-Petrovna but differs in many technical details. This paper is organized as follows. In Section 2 we introduce the model for two-phase flows, immiscible and incompressible displacement in heterogeneous porous media. In Section 2.2 we discuss the mathematical formulation for the KT central scheme in dimension by dimension approach and in a genuinely two- dimensional one. In Section 3 we will present some numerical results when we apply the KT central differencing scheme with both approaches mentioned above to porous media flows. ## 2\. NUMERICAL APPROXIMATION OF TWO-PHASE FLOW ### 2.1. Operator splitting for two-phase flow. An operator splitting technique is employed for the computational solution of the saturation equation (2) and the pressure equation (1) which are solved sequentially with distinct time steps. This method has proved to be computationally efficient in producing accurate numerical solutions for two- phase flow (Douglas et al., 1997). Typically, for computational efficiency, larger time steps are used to compute the pressure (1). The splitting allows time steps used in the solution of the pressure-velocity equation that are longer than those allowed in the convection calculation (2). Thus, we introduce two time steps: $\Delta t_{c}$ used to compute the solution of the hyperbolic problem, and $\Delta t_{p}$ used in the pressure-velocity calculation such that $\Delta t_{p}\geq\Delta t_{c}$. We remark that in practice, variable time steps are always useful, especially for the convection micro-steps subject dynamically to a $CFL$ condition. For the global pressure solution (the pertinent elliptic equation), we use a (locally conservative) hybridized mixed finite element discretization equivalent to cell-centered finite differences (Douglas et al., 1997), which effectively treats the rapidly changing permeabilities that arise from stochastic geology and produces accurate velocity fields. The pressure and the Darcy velocity are approximated at times $t^{m}=m\Delta t_{p}$, $m=0,1,2,\dots$. The algebraic system resulting from the discretized equations can be solved by a preconditioned conjugate gradient procedure (PCG) or by a multi-grid procedure ((Douglas et al., 1997)). We use high resolution numerical central scheme (see (Kurganov and Tadmor, 2000)) for solving the scalar hyperbolic conservation laws arising in the convection of the fluid phases in heterogeneous porous media for two-phase flows - we will discuss the application of these schemes for two-phase flows in Section 3. Theses methods can accurately resolve sharp fronts in the fluid saturations without introducing spurious oscillations or excessive numerical diffusion. The saturation equation is approximated at times $t^{\kappa}_{m}=t^{m}+{\kappa}\Delta t_{c}$ for $t^{m}<{t^{\kappa}_{m}}\leq t^{m+1}$ that take into account the advective transport of water. We will write $t^{\kappa}$ to indicate the time step ${t^{\kappa}_{m}}$ and $t^{\kappa+1}$ to indicate ${t^{\kappa}_{m}}+\Delta t_{c}$. A detailed description of the numerical method that we employ for the solution of Eqs. (1)-(2) can be found in Douglas et al. (1997) and references therein (see also Abreu et al. (2006, 2004c) and Abreu et al. (2008) for applications of the operator splitting technique for three phase flows that takes into account capillary pressure (diffusive effects)). _R_ emark: To simplify notation, we denote: * • NT1d for one-dimensional NT scheme; * • NT2d for two-dimensional NT scheme; * • KT1d for one-dimensional KT scheme; * • KTdxd for the KT scheme with dimension by dimension approach and * • SD2-2D for our two-dimensional approach. ### 2.2. TWO SPATIAL DIMENSIONS SECOND ORDER SEMI-DISCRETE CENTRAL SCHEME In this section, we will develop a two-spatial dimension second order semi- discrete central scheme (SD2-2D) based on the ideas of Lax (1954); Rusanov (1961); Nessyahu and Tadmor (1990); Kurganov and Tadmor (2000) and Jiang and Tadmor (1998) which are then applied in numerical approximation of the scalar hyperbolic conservation law modeling the convective transport of the fluid phases in two-phase flows and its coupling with lowest order Raviart-Thomas (Raviart and Thomas, 1977) locally conservative mixed finite elements for the associated elliptic (velocity-pressure part) problem (See Raviart and Thomas (1977)). We summarize below the basic ideas of the construction of SD2$-$2D numerical scheme: * • The Lax-Friedrichs method in two-spatial dimensions LxF2D written in the REA algorithm setup (See Jiang and Tadmor (1998)) will be used to obtain the two dimensional Rusanov’s method SD1-2D. We follow the same procedures presented in Kurganov and Tadmor (2000) in one spatial dimension. * • The new SD2-2D numerical scheme will then be obtained as a second order generalization of the SD1-2D. This second order precision is achieved approximating the solution with piecewise linear functions. ### 2.3. The staggered non-uniform grid of the SD2-2D central scheme We begin then extending the LxF2D to obtain the SD1-2D following the same procedures for one dimensional problems. First, we define the non-staggered and the staggered grids of retangular cells used in the LxF2D. The points $(x_{j},y_{k},t^{\kappa})$ are defined as follows. $\displaystyle x_{j}$ $\displaystyle=$ $\displaystyle j\cdot\Delta x,\qquad j=\ldots,-1,0,1,\ldots$ $\displaystyle y_{k}$ $\displaystyle=$ $\displaystyle k\cdot\Delta y,\qquad k=\ldots,-1,0,1,\ldots$ $\displaystyle t^{\kappa}$ $\displaystyle=$ $\displaystyle t^{m}+\kappa\cdot\Delta t_{c},\qquad\kappa=0,\ldots,i,$ We denote the cells of the non-staggered grid by $I_{j,k}:=(x_{j-1/2},x_{j+1/2})\times(y_{k-1/2},y_{k+1/2})$. Its area $A_{j,k}\equiv\Delta x\cdot\Delta y=(x_{j+1/2}-x_{j-1/2})\cdot(y_{k+1/2}-y_{k-1/2})$. The time step of the convective equation (2) is $\Delta t_{c}=t^{\kappa+1}-t^{\kappa}$. The staggered grid is obtained moving the cells $\Delta x/2$ to the right and $\Delta y/2$ upward These staggered cells will be denoted by $I_{j+1/2,k+1/2}:=(x_{j},x_{j+1})\times(y_{k},y_{k+1})$. Its center is the point $(x_{j+1/2},y_{k+1/2})$, where $x_{j+1/2}=x_{j}+\Delta x/2$ and $y_{k+1/2}=y_{j}+\Delta y/2$. The Figure 1 illustrates the non-staggered and the staggered grids showing the cells $I_{j,k}$ e $I_{j+3/2,k+1/2}$ as examples of non-staggered and staggered cells, respectively. Figure 1. The LxF2D uniform grid. The cells $I_{j,k}$ of the original non- staggered grid are limited by the solid lines and the cells $I_{j+3/2,k+1/2}$ of the staggered grid are limited by the dashed lines. The scalar hyperbolic conservation law (2) can be written as $\frac{\partial s}{\partial t}+\frac{\partial}{\partial x}({{}^{x}\\!v}f(s))+\frac{\partial}{\partial y}({{}^{y}\\!v}f(s))=0,$ (3) where ${{}^{x}\\!v}\equiv{{}^{x}\\!v}(x,y,t)$ and ${{}^{y}\\!v}\equiv{{}^{y}\\!v}(x,y,t)$ denote the $x$ and $y$ components of the velocity field ${\bf v}$. The cell averages at time $t^{\kappa}$ are $\displaystyle\overline{S}^{\kappa}_{j,k}:=\overline{S}_{j,k}(t^{\kappa})\equiv\frac{1}{\Delta X\\!\Delta Y}\displaystyle\int_{x_{j-\frac{1}{2}}}^{x_{j+\frac{1}{2}}}\displaystyle\int_{y_{k-\frac{1}{2}}}^{y_{k+\frac{1}{2}}}s(x,y,t^{\kappa})\,dxdy.$ (4) The solution $s(x,y,t^{\kappa})$ of the problem (2) at time $t^{\kappa}$ is approximated using piecewise-linear MUSCL-type interpolants (See Van Leer (1979)). $\widetilde{S}^{\kappa}_{j,k}(x,y)=\overline{S}^{\kappa}_{j,k}+(S_{x})^{\kappa}_{j,k}\cdot(x-x_{j})+(S_{y})^{\kappa}_{j,k}\cdot(y-y_{k}),$ (5) where $x_{j-1/2}\leq x\leq x_{j+1/2}$ e $y_{k-1/2}\leq y\leq y_{k+1/2}$. The second-order resolution is guaranteed if the so-called vector of numerical derivative $(S_{x})^{\kappa}_{j,k}$ and $(S_{y})^{\kappa}_{j,k}$ satisfy $\displaystyle(S_{x})^{\kappa}_{j,k}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial s}{\partial x}\,\right|_{x=x_{j},y=y_{k},t=t^{\kappa}}+\mathcal{O}(\Delta X);$ (6a) $\displaystyle(S_{y})^{\kappa}_{j,k}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial s}{\partial y}\right|_{x=x_{j},y=y_{k},t=t^{\kappa}}+\mathcal{O}(\Delta Y),$ (6b) These numerical derivatives are computed using the MinMod limiter $\displaystyle(S_{x})^{\kappa}_{j,k}$ $\displaystyle=$ $\displaystyle\text{MM}\theta\frac{1}{\Delta x}\left\\{\overline{S}^{\kappa}_{j-1,k},\overline{S}^{\kappa}_{j,k},\overline{S}^{\kappa}_{j+1,k}\right\\}$ (7a) $\displaystyle:=$ $\displaystyle\text{MM}\left(\theta\frac{\Delta S^{\kappa}_{j+1/2,k}}{\Delta x},\frac{\Delta S^{\kappa}_{j-1/2,k}-\Delta S^{\kappa}_{j+1/2,k}}{2\Delta x},\theta\frac{\Delta S^{\kappa}_{j-1/2,k}}{\Delta x}\right);$ $\displaystyle(S_{y})^{\kappa}_{j,k}$ $\displaystyle=$ $\displaystyle\text{MM}\theta\frac{1}{\Delta y}\left\\{\overline{S}^{\kappa}_{j,k-1},\overline{S}^{\kappa}_{j,k},\overline{S}^{\kappa}_{j,k+1}\right\\}$ (7b) $\displaystyle:=$ $\displaystyle\text{MM}\left(\theta\frac{\Delta S^{\kappa}_{j,k+1/2}}{\Delta y},\frac{\Delta S^{\kappa}_{j,k-1/2}-\Delta S^{\kappa}_{j,k+1/2}}{2\Delta y},\theta\frac{\Delta S^{\kappa}_{j,k-1/2}}{\Delta y}\right).$ Here $\Delta$ denotes the centered difference, $\Delta S^{\kappa}_{j+1/2,k}=\overline{S}^{\kappa}_{j+1,k}-\overline{S}^{\kappa}_{j,k}$ and the paramter $\theta\in[1,1.8]$ has been chosen in the optimal way in every numerical example with $\theta=1.8$ beeing the less dissipative limiter. The minmod limiter (7) guarantees the nonoscillatory property and the second- order accuracy. The reconstruction given by Equations (5)-(7) also retains conversation, i.e., $\int_{I_{j,k}}\widetilde{S}^{\kappa}_{j,k}(x,y)\,dx\,dy=\overline{S}^{\kappa}_{j,k}.$ (8) Remark: We notice that if $(S_{x})^{\kappa}_{j,k}$ and $(S_{y})^{\kappa}_{j,k}$ are equal to zero, then we will get the first-order two-dimensional semi-discrete scheme SD1-2D. Otherwise, we will obtain the second-order two spatial dimensions semi-discrete central scheme SD2-2D. We consider the model of hyperbolic conservation laws given by Equation (2) with cell averages as in (4) and the two-dimensional piecewise linear reconstruction defined in (5) and (6) with the conservative property (8). Our goal is to compute an approximated solution $S_{j,k}({t^{\kappa}_{m}}+\Delta t_{c})$ in the original grid at the next time step. To this end, we apply the Godunov’s algorithm REA. To solve this problem, we integrate the conservation law over some control volumes that we need to specify. Constructing the staggered nonuniform grid: Kurganov and Tadmor developed the KT1D scheme along the lines of NT1D (See (Kurganov and Tadmor, 2000)). The nonuniform staggered grid in the KT1D was constructed directly from the staggered uniform grid of NT1D with additional information on the local speeds of wave propagation. In a similar way the nonuniform staggered grid in the SD2-2D scheme is defined from the uniform staggered grid of the NT2D scheme as follows. $(1)$ We set $C_{j,k}=[x_{j-1/2},x_{j+1/2}]\times[y_{k-1/2},y_{k+1/2}]$ to denote the cells of the original non-staggered grid; $C_{j+1/2,k+1/2}=[x_{j},x_{j+1}]\times[y_{k},y_{k+1}]$ to denote the cells of the uniform staggered grid. We start with a piecewise constant approximated solution $\bar{S}^{\kappa}_{j,k}$ over the original cells $C_{j,k}$. $(2)$ Next, we move to the staggered uniform grid with cells given by $C_{j+1/2,k+1/2}$. $(3)$ The NT2D scheme computes four averaged solutions at time step ${t^{\kappa}_{m}}+\Delta t_{c}$ (See the hachured regions in Figure 2 corresponding to the cells $C_{j-1/2,k-1/2}$, $C_{j+1/2,k-1/2}$, $C_{j-1/2,k+1/2}$ and $C_{j+1/2,k+1/2}$ on the staggered grid). These averaged staggered solution are then properly projected back onto the original non- staggered grid to obtain the desired solution (See Jiang and Tadmor (1998)). Repeating the same ideias presented by Rusanov in his modification of Lax- Friedrichs’ method, we compute the local speed of propragation at each Riemann Problem. These local speeds define new non-uniform cells where the evolution step will take place. Computing the local speed of propagation: we begin with the cell $C_{j-1/2,k-1/2}$ to find the local speeds at the following Riemann Problems: 1. (1) Y direction: 1. (a) $\left\\{\begin{array}[]{ll}\overline{S}^{\kappa}_{j-1,k-1},&x_{j-3/2}\leq x\leq x_{j-1/2},\quad y_{k-3/2}\leq y\leq y_{k-1/2}\\\ \overline{S}^{\kappa}_{j-1,k},&x_{j-3/2}\leq x\leq x_{j-1/2},\quad y_{k-1/2}\leq y\leq y_{k+1/2}.\end{array}\right.$ The local speed $a^{y}_{j-1,k-1/2}$ defines the following points: $\displaystyle p_{1}=(x_{j-1},\,y_{k-1/2}-a^{y}_{j-1,k-1/2}{\Delta t_{c}}),$ $\displaystyle p_{2}=(x_{j-1},\,y_{k-1/2}+a^{y}_{j-1,k-1/2}{\Delta t_{c}})$ sketched in Figure 2. We denote the distance between them by $\Delta y_{j-1,k-1/2}:=2a^{y}_{j-1,k-1/2}{\Delta t_{c}}$. 2. (b) $\left\\{\begin{array}[]{ll}\overline{S}^{\kappa}_{j,k-1},&x_{j-1/2}\leq x\leq x_{j+1/2},\quad y_{k-3/2}\leq y\leq y_{k-1/2}\\\ \overline{S}^{\kappa}_{j,k},&x_{j-1/2}\leq x\leq x_{j+1/2},\quad y_{k-1/2}\leq y\leq y_{k+1/2}.\end{array}\right.$ The local speed $a^{y}_{j,k-1/2}$ defines the points $\displaystyle p_{3}=(x_{j},\,y_{k-1/2}-a^{y}_{j,k-1/2}{\Delta t_{c}}),$ $\displaystyle p_{4}=(x_{j},\,y_{k-1/2}+a^{y}_{j,k-1/2}{\Delta t_{c}})$ shown in Figure 2 and the distance between them is $\Delta y_{j,k-1/2}=2a^{y}_{j,k-1/2}{\Delta t_{c}}$. 2. (2) X direction: 1. (a) $\left\\{\begin{array}[]{ll}\overline{S}^{\kappa}_{j-1,k-1},&x_{j-3/2}\leq x\leq x_{j-1/2},\quad y_{k-3/2}\leq y\leq y_{k-1/2}\\\ \overline{S}^{\kappa}_{j,k-1},&x_{j-1/2}\leq x\leq x_{j+1/2},\quad y_{k-3/2}\leq y\leq y_{k-1/2}.\end{array}\right.$ The local speed $a^{x}_{j-1/2,k-1}$ defines the points $\displaystyle p_{5}=(x_{j-1/2}-a^{x}_{j-1/2,k-1}{\Delta t_{c}},\,y_{k-1}),$ $\displaystyle p_{6}=(x_{j-1/2}+a^{x}_{j-1/2,k-1}{\Delta t_{c}},\,y_{k-1})$ also shown in 2 and $\Delta x_{j-1/2,k-1}=2a^{x}_{j-1/2,k-1}{\Delta t_{c}}$ is the distance between them. 2. (b) $\left\\{\begin{array}[]{ll}\overline{S}^{\kappa}_{j-1,k},&x_{j-3/2}\leq x\leq x_{j-1/2},\quad y_{k-1/2}\leq y\leq y_{k+1/2}\\\ \overline{S}^{\kappa}_{j,k},&x_{j-1/2}\leq x\leq x_{j+1/2},\quad y_{k-1/2}\leq y\leq y_{k+1/2}.\end{array}\right.$ The local speed $a^{x}_{j-1/2,k}$ defines the points $\displaystyle p_{7}=(x_{j-1/2}-a^{x}_{j-1/2,k}{\Delta t_{c}},\,y_{k}),$ $\displaystyle p_{8}=(x_{j-1/2}+a^{x}_{j-1/2,k}{\Delta t_{c}},y_{k})$ shown in Figure 2 and $\Delta X_{j-1/2,k}=2a^{x}_{j-1/2,k}{\Delta t_{c}}$ denotes the distance between them. Given these four local speed of wave propagation $a^{y}_{j-1,k-1/2}$, $a^{y}_{j,k-1/2}$, $a^{x}_{j-1/2,k-1}$ e $a^{x}_{j-1/2,k}$, we can define the Region I (also represented by $R_{j-1/2,k-1/2}$) as follows: * $\rhd$ Region I: $\displaystyle R_{j-1/2,k-1/2}$ $\displaystyle:=$ $\displaystyle[x_{j-1/2}-b^{x}_{j-1/2,k-1/2}{\Delta t_{c}},\,x_{j-1/2}+b^{x}_{j-1/2,k-1/2}{\Delta t_{c}}]\times$ $\displaystyle[y_{k-1/2}-b^{y}_{j-1/2,k-1/2}{\Delta t_{c}},\,y_{k-1/2}+b^{y}_{j-1/2,k-1/2}{\Delta t_{c}}]$ where $\displaystyle b^{x}_{j-1/2,k-1/2}:=\max\\{a^{x}_{j-1/2,k},a^{x}_{j-1/2,k-1}\\}\hskip 8.5359pt\mbox{e}\hskip 8.5359pt$ $\displaystyle b^{y}_{j-1/2,k-1/2}:=\max\\{a^{y}_{j,k-1/2},a^{y}_{j-1,k-1/2}\\}.$ Figure 3 shows the new cell $R_{j-1/2,k-1/2}$ of the new staggered non-uniform grid. Figure 2. SD2-2D: the construction of the two-dimensional grid We repeat the same procedures with the cell $C_{j-1/2,k+1/2}$ to define the Region III in terms of the local speed of propagation $a^{y}_{j-1,k+1/2}$, $a^{y}_{j,k+1/2}$, $a^{x}_{j-1/2,k+1}$ e $a^{x}_{j-1/2,k}$. These local speeds determine analogously the points $q_{1}$ a $q_{8}$ sketched in Figure 3 and define the new Region III: * $\rhd$ Region III: $\displaystyle R_{j-1/2,k+1/2}$ $\displaystyle:=$ $\displaystyle[x_{j-1/2}-b^{x}_{j-1/2,k+1/2}{\Delta t_{c}},\,x_{j-1/2}+b^{x}_{j-1/2,k+1/2}{\Delta t_{c}}]\times$ $\displaystyle[y_{k+1/2}-b^{y}_{j-1/2,k+1/2}{\Delta t_{c}},\,y_{k+1/2}+b^{y}_{j-1/2,k+1/2}{\Delta t_{c}}]$ where $\displaystyle b^{x}_{j-1/2,k+1/2}:=\max\\{a^{x}_{j-1/2,k},a^{x}_{j-1/2,k+1}\\}\hskip 8.5359pt\mbox{and}\hskip 8.5359pt$ $\displaystyle b^{y}_{j-1/2,k+1/2}:=\max\\{a^{y}_{j,k+1/2},a^{y}_{j-1,k+1/2}\\}.$ Figure 3. Regions I and III Following these same procedures with the staggered cells $C_{j+1/2,k-1/2}$ and $C_{j+1/2,k+1/2}$ we define Regions VII and IX shown in Figure 4. We will denote by Group A the set of Regions I, III, VII and IX. Figure 4. Regions I, III, VII and IX Finally, to finish the construction of the new non-uniform staggered grid, we need to define: * • Four cells which lie in the empty spaces between the regions of Group A. This new set will be denoted by Group B. * • A central region where the solution is smooth. The definitions of the cells of Group B will determine the central region. This central region will not be a rectangle, but a set of retangles. There are many ways to define the cells of Group B. Our definition will be as follows (See Figura 5): Figure 5. SD2-2D: The two dimensional semi-discrete central scheme SD2-2D: construction of the non-uniform staggered grid. * $\rhd$ Region II: $\displaystyle R_{j-1/2,k}$ $\displaystyle:=$ $\displaystyle[x_{j-1/2}-c^{x}_{j-1/2,k}{\Delta t_{c}},\,x_{j-1/2}+c^{x}_{j-1/2,k}{\Delta t_{c}}]\times$ $\displaystyle[y_{k-1/2}+b^{y}_{j-1/2,k-1/2}{\Delta t_{c}},\,y_{k+1/2}-b^{y}_{j-1/2,k+1/2}{\Delta t_{c}}]$ $\displaystyle\mbox{where}\quad c^{x}_{j-1/2,k}:=\max\\{b^{x}_{j-1/2,k-1/2},b^{x}_{j-1/2,k+1/2}\\}$ The Regions VI and VIII can be obtained analogously As soon as the cells of Group A and B are determined, the central region is automatically defined: * $\rhd$ Region V: $\displaystyle R_{j,k}$ $\displaystyle:=$ $\displaystyle[x_{j-1/2}+c^{x}_{j-1/2,k}{\Delta t_{c}},\,x_{j+1/2}-c^{x}_{j+1/2,k}{\Delta t_{c}}]\times$ $\displaystyle[y_{k-1/2}+d^{y}_{j,k-1/2}{\Delta t_{c}},\,y_{k+1/2}-d^{y}_{j,k+1/2}{\Delta t_{c}}].$ We also would like to emphasize that our choice for these regions does not introduce more numerical diffusion. We will call BR the black rectangle that can be seen in Figure 5 and we notice that, by construction, its area is proportional to $(\Delta t_{c})^{2}$. ### 2.4. The new SD2-2D central scheme using Algorithm REA After defining the new control volumes performed in the section above, we are now able to develop our new SD2-2D central scheme following the REA algorithm. Reconstruction step: We suppose that we know an approximated solution constant in each cell at time step $t^{\kappa}$ as in Equation (4). This approximated solution is then reconstructed as a piecewise bilinear polinomial as defined in Equations (5) and (6). Evolution step: Let $D$ represents one of the nine regions defined above $R_{j\pm 1/2,k\pm 1/2}$, $R_{j,k\pm 1/2}$, $R_{j\pm 1/2,k}$, the central region $R_{j,k}$ or the black rectangle BR. We will denote by $D^{+}$ the part of region $D$ inside the non-staggered cell $I_{j,k}$ and by $D^{-}$, the part of region $D$ outside the cell $I_{j,k}$. We integrate the conservation law (2) in the control volumes $D\times[{t^{\kappa}_{m}},{t^{\kappa}_{m}}+{\Delta t_{c}}]$ to obtain an approximate averaged solution $\bar{w}^{\kappa+1}(D)$ at the next time step, in each cell $D$ of the staggered non-uniform grid. Projection step: These averaged solutions $\overline{w}^{\kappa+1}(D)$ are then reconstructed as piecewise bilinear polynomials $\widetilde{w}^{\kappa+1}(x,y)$ in each of the ten regions $D$. These new reconstructions are then projected back onto the original grid of uniform non- staggered cells, $\overline{S}^{\kappa+1}_{j,k}:=\frac{1}{\Delta x\,\Delta y}\int_{\bigcup D}\widetilde{w}^{\kappa+1}(x,y)\,dx\,dy.$ (9) The new reconstructions $\widetilde{w}^{\kappa+1}(x,y)$ are defined analogously as in Equation (5). For instance, for Region $R_{j+1/2,k}$, $\displaystyle{\widetilde{w}^{\kappa+1}_{j+1/2,k}}(x,y)=\overline{w}^{\kappa+1}_{j+1/2,k}+(w_{x})^{\kappa+1}_{j+1/2,k}(x-x_{j+1/2})+$ $\displaystyle(w_{y})^{\kappa+1}_{j+1/2,k}(y-y_{k}),$ (10) $\displaystyle(x,y)\in R_{j+1/2,k}.$ The numerical derivatives $(w_{x})^{\kappa+1}_{j+1/2,k}$ and $(w_{y})^{\kappa+1}_{j+1/2,k}$ satisfy the conditions $\displaystyle(w_{x})^{\kappa+1}_{j+1/2,k}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial w}{\partial x}\,\right|_{(x_{j+1/2},y_{k},t^{\kappa+1})}+\mathcal{O}(\Delta x);$ (11) $\displaystyle(w_{y})^{\kappa+1}_{j+1/2,k}$ $\displaystyle=$ $\displaystyle\left.\frac{\partial w}{\partial y}\,\right|_{(x_{j+1/2},y_{k},t^{\kappa+1})}+\mathcal{O}(\Delta y);$ (12) in order to guarantee the second order approximation. Also, the reconstruction $\widetilde{w}^{\kappa+1}_{j+1/2,k}(x,y)$ retains the conservation property (8). We remark that this is a theoretical step and it will not be necessary to compute these numerical derivatives in the final semi-discrete formulation. This completes the construction of our totally discrete central scheme in a rectangular grid. It is very laborious to write a totally discrete version of this central scheme. Instead, we will proceed directly to our semi-discrete formulation. In order to to this, we compute the following limit when ${\Delta t_{c}}\rightarrow 0$, $\displaystyle\lim_{{\Delta t_{c}}{\rightarrow}0}$ $\displaystyle\frac{\overline{S}_{j,k}(t+{\Delta t_{c}})-\overline{S}_{j,k}(t)}{{\Delta t_{c}}}=\frac{d}{dt}\overline{S}_{j,k}(t)=$ $\displaystyle=\lim_{{\Delta t_{c}}{\rightarrow}0}\frac{1}{{\Delta t_{c}}}\cdot\frac{1}{\Delta x\,\Delta y}\left\\{\sum_{p=j\pm 1/2}\int_{R^{+}_{p,k+1/2}}\widetilde{w}^{\kappa+1}_{p,k+1/2}(x,y)\right.$ $\displaystyle\left.+\sum_{p=j\pm 1/2}\int_{R^{+}_{p,k-1/2}}\widetilde{w}^{\kappa+1}_{p,k-1/2}(x,y)+\sum_{p=j\pm 1/2}\int_{R^{+}_{p,k}}\widetilde{w}^{\kappa+1}_{p,k}(x,y)\,dx\,dy\right.$ $\displaystyle\left.+\sum_{q=k\pm 1/2}\int_{R^{+}_{j,q}}\widetilde{w}^{\kappa+1}_{j,q}(x,y)\,dx\,dy+\int_{R_{j,k}}\widetilde{w}^{\kappa+1}_{j,k}(x,y)\,dx\,dy\right.$ $\displaystyle\left.+\int_{RP}\widetilde{w}^{\kappa+1}_{j+1/2,k-1/2}(x,y)dxdy-(\Delta x\,\Delta y)\overline{S}_{j,k}(t)\right\\}$ (13) The conservation property of the reconstructions $\widetilde{w}^{\kappa+1}_{j,k}$ in he regions $D$ results $\int_{D}\widetilde{w}^{\kappa+1}(x,y)\,dx\,dy=|D|\cdot\overline{w}^{\kappa+1}(D),$ (14) where $\overline{w}^{\kappa+1}(D)$ is the averaged solution in region $D$. Note that, by reconstruction, the area of regions I, III, VII and IX are proportional to $({\Delta t_{c}})^{2}$, that is, $\begin{array}[]{ll}\mbox{Region I: }&|R_{j-1/2,k-1/2}|=\mathcal{O}\big{(}({\Delta t_{c}})^{2}\big{)}\\\ \mbox{Region III: }&|R_{j-1/2,k+1/2}|=\mathcal{O}\big{(}({\Delta t_{c}})^{2}\big{)}\\\ \mbox{Region VII: }&|R_{j+1/2,k-1/2}|=\mathcal{O}\big{(}({\Delta t_{c}})^{2}\big{)}\\\ \mbox{Region IX: }&|R_{j+1/2,k+1/2}|=\mathcal{O}\big{(}({\Delta t_{c}})^{2}\big{)}\\\ \mbox{Region RP: }&|RP|=\mathcal{O}\big{(}({\Delta t_{c}})^{2}\big{)}\\\ \end{array}$ For example, considering Region IX, we conclude $\displaystyle\int_{R_{j+1/2,k+1/2}}\widetilde{w}^{\kappa+1}_{j+1/2,k+1/2}(x,y)\,dx\,dy=\mathcal{O}({\Delta t_{c}})^{2}$ $\displaystyle\hskip 85.35826pt\Rightarrow\lim_{{\Delta t_{c}}{\rightarrow}0}\frac{1}{{\Delta t_{c}}}\int_{R_{j+1/2,k+1/2}}\widetilde{w}^{\kappa+1}_{j+1/2,k+1/2}(x,y)\,dx\,dy=0;$ (15) Our goal is to obtain the semi-discrete formulation of this new central cheme. Therefore, the Equation (2.4) show that we do not need to computed the averaged soltions over the Regions I, III, VII and IX. Note that, by simetry, we only need to compute the solutions over the Regions VI, VIII, V and the Black Rectangle. For the Region $R_{j+1/2,k}$, we obtain: $\displaystyle\int_{R^{+}_{j+1/2,k}}\widetilde{w}^{\kappa+1}_{j+1/2,k}(x,y)dxdy=$ $\displaystyle=\int_{R^{+}_{j+1/2,k}}\Big{[}\bigg{(}\overline{w}^{\kappa+1}_{j+1/2,k}+(w_{x})^{\kappa+1}_{j+1/2,k}\cdot(x-x_{j+1/2})+(w_{y})^{\kappa+1}_{j+1/2,k}\cdot(y-y_{k})\bigg{)}\Big{]}dxdy$ $\displaystyle=\overline{w}^{\kappa+1}_{j+1/2,k}\cdot|R^{+}_{j+1/2,k}|+\mathcal{O}(({\Delta t_{c}})^{2}).$ (16) Analogously, we compute the averaged solution over Region $R^{+}_{j,k+1/2}$: $\displaystyle\int_{R^{+}_{j,k+1/2}}\widetilde{w}^{\kappa+1}_{j,k+1/2}(x,y)dxdy=\overline{w}^{\kappa+1}_{j,k+1/2}\cdot|R^{+}_{j,k+1/2}|+\mathcal{O}(({\Delta t_{c}})^{2}).$ (17) Note that the solution has no discontinuities inside $R_{j,k}$. So, it isn’t necessary to reconstruction as a piecewise bilinear polynomials. The averaged solution is $\displaystyle\int_{R_{j,k}}\widetilde{w}^{\kappa+1}_{j,k}(x,y)dxdy=\overline{w}^{\kappa+1}_{j,k}\cdot|R_{j,k}|.$ (18) Substituting the Equations (2.4), (2.4), (17) and (18) in Equation (2.4), we obtain: $\displaystyle\frac{d}{dt}\overline{S}_{j,k}(t)=$ $\displaystyle\lim_{{\Delta t_{c}}{\rightarrow}0}\left\\{\frac{c^{x}_{j-1/2,k}}{\Delta x}\overline{w}_{j-1/2,k}(t+{\Delta t_{c}})+\frac{c^{x}_{j+1/2,k}}{\Delta x}\overline{w}_{j+1/2,k}(t+{\Delta t_{c}})\right.$ $\displaystyle\left.+\frac{d^{y}_{j,k-1/2}}{\Delta y}\overline{w}_{j,k-1/2}(t+{\Delta t_{c}})+\frac{d^{y}_{j,k+1/2}}{\Delta Y}\overline{w}_{j,k+1/2}(t+{\Delta t_{c}})\right.$ $\displaystyle\left.-\left(\frac{d^{y}_{j,k+1/2}+d^{y}_{j,k-1/2}}{\Delta y}+\frac{c^{x}_{j+1/2,k}+c^{x}_{j-1/2,k}}{\Delta x}\right)\cdot\overline{w}_{j,k}(t+{\Delta t_{c}})\right.$ $\displaystyle\left.+\left(\frac{1}{{\Delta t_{c}}}\overline{w}_{j,k}(t+{\Delta t_{c}})-\frac{1}{{\Delta t_{c}}}\overline{S}_{j,k}(t+{\Delta t_{c}})\right)\right\\}.$ (19) For the final formulation, we have to compute the averaged solution over the non-uniform staggered grid. $\overline{w}^{\kappa+1}_{j+1/2,k},\overline{w}^{\kappa+1}_{j,k+1/2}\mbox{ e }\overline{w}^{\kappa+1}_{j,k},$ To this end, we integrate the conservation law (LABEL:eq:sist2_cont) over the control volumes $R_{j+1/2,k}\times[{t^{\kappa}_{m}},{t^{\kappa}_{m}}+{\Delta t_{c}}],\quad R_{j,k+1/2}\times[{t^{\kappa}_{m}},{t^{\kappa}_{m}}+{\Delta t_{c}}]\quad\mbox{e}\quad R_{j,k}\times[{t^{\kappa}_{m}},{t^{\kappa}_{m}}+{\Delta t_{c}}],$ respectively. Therefore, $\displaystyle w^{\kappa+1}_{j+1/2,k}$ $\displaystyle=\frac{1}{|R_{j+1/2,k}|}\int_{R_{j+1/2,k}}s(x,y,t^{\kappa+1})\,dx\,dy$ $\displaystyle=\frac{1}{|R_{j+1/2,k}|}\int_{R_{j+1/2,k}}\widetilde{S}^{\kappa}(x,y)\,dx\,dy$ $\displaystyle-\frac{1}{|R_{j+1/2,k}|}\int_{R_{j+1/2,k}}\int_{t^{\kappa}}^{t^{\kappa+1}}\bigg{[}\frac{\partial}{\partial x}\bigg{(}{{}^{x}\\!v}(x,y,\tau)\cdot f(s(x,y,\tau))\bigg{)}$ (20) $\displaystyle\hskip 85.35826pt+\frac{\partial}{\partial y}\bigg{(}{{}^{y}\\!v}(x,y,\tau)\cdot f(s(x,y,\tau))\bigg{)}\bigg{]}\,dx\,dyd\tau.$ Let us denote the double integral by Int${}_{\widetilde{S}}$ and the flux integral by Intf. The integral Int${}_{\widetilde{S}}$ is computed analytically. $\displaystyle\mbox{Int}_{\widetilde{S}}=$ $\displaystyle\frac{1}{2}(\overline{S}^{\kappa}_{j,k}+\overline{S}^{\kappa}_{j+1,k})$ $\displaystyle+\frac{1}{4}\left[\Delta x-c^{x}_{j+1/2,k}{\Delta t_{c}}\right]\cdot\left[(S_{x})^{\kappa}_{j,k}-(S_{x})^{\kappa}_{j+1,k}\right]$ (21) $\displaystyle+\frac{{\Delta t_{c}}}{4}\left[b^{y}_{j+1/2,k-1/2}-b^{y}_{j+1/2,k+1/2}\right]\cdot\left[(S_{y})^{\kappa}_{j,k}-(S_{y})^{\kappa}_{j+1,k}\right].$ To compute the flux integral, Intf, we first denote the limits of region $R_{j+1/2,k}$ as follows: $\displaystyle a:=x_{j+1/2}-c^{x}_{j+1/2,k}{\Delta t_{c}}$ $\displaystyle b:=x_{j+1/2}+c^{x}_{j+1/2,k}{\Delta t_{c}}$ $\displaystyle c:=y_{k+1/2}+b^{y}_{j+1/2,k-1/2}{\Delta t_{c}}$ $\displaystyle d:=y_{k+1/2}-b^{y}_{j+1/2,k+1/2}{\Delta t_{c}}$ Using the Calculus Fundamental Theorem together with the trapezoid rule, we obtain $\displaystyle\mbox{Int}_{f}$ $\displaystyle=$ $\displaystyle\frac{1}{2\Delta x_{j+1/2,k}}\int_{t^{\kappa}}^{t^{\kappa+1}}\Big{[}{{}^{x}\\!v}(b,d,\tau)\,f(s(b,d,\tau))-{{}^{x}\\!v}(a,d,\tau)\,f(s(a,d,\tau))$ (22) $\displaystyle+{{}^{x}\\!v}(b,c,\tau)\,f(s(b,c,\tau))-{{}^{x}\\!v}(a,c,\tau)\,f(s(a,c,\tau))\Big{]}\,d\tau$ $\displaystyle+\frac{1}{2\Delta y_{j+1/2,k}}\int_{t^{\kappa}}^{t^{\kappa+1}}\Big{[}{{}^{y}\\!v}(b,d,\tau)\,f(s(b,d,\tau))-{{}^{y}\\!v}(b,c,\tau)\,f(s(b,c,\tau))$ $\displaystyle+{{}^{y}\\!v}(a,d,\tau)\,f(s(a,d,\tau))-{{}^{y}\\!v}(a,c,\tau)\,f(s(a,c,\tau))\Big{]}\,d\tau$ If the CFL condition $\max\left(\frac{{\Delta t_{c}}}{\Delta x}\max_{S}|{{}^{x}\\!v}\,f^{\prime}(s)|,\frac{{\Delta t_{c}}}{\Delta y}\max_{S}|{{}^{y}\\!v}\,f^{\prime}(s)|\right)<\frac{1}{2}$ (23) holds and since the functions ${{}^{x}\\!v}\,f(s(x,y,\tau))$ and ${{}^{y}\\!v}\,f(s(x,y,\tau))$ are computed away from the discontinuities then they are differential functions of $\tau$ and therefore, the time integral can be approximated using the middle point rule. Denoting $t^{\kappa+1/2}:=t+{\Delta t_{c}}/2$, we obtain: $\displaystyle\mbox{Int}_{f}$ $\displaystyle=$ $\displaystyle\displaystyle{\frac{1}{4c^{x}_{j+1/2,k}}\Big{[}{{}^{x}\\!v}(b,d,t^{\kappa+1/2})\,f(S(b,d,t^{\kappa+1/2}))-{{}^{x}\\!v}(a,d,t^{\kappa+1/2})\,f(S(a,d,t^{\kappa+1/2}))}$ (24) $\displaystyle\displaystyle{+{{}^{x}\\!v}(b,c,t^{\kappa+1/2})\,f(S(b,c,t^{\kappa+1/2}))-{{}^{x}\\!v}(a,c,t^{\kappa+1/2})\,f(S(a,c,t^{\kappa+1/2}))\Big{]}\,d\tau}$ $\displaystyle\displaystyle{+\frac{\alpha_{Y}}{2-2\alpha_{Y}(b^{y}_{j+1/2,k+1/2}-b^{y}_{j+1/2,k-1/2})}}\cdot$ $\displaystyle\displaystyle{\Big{[}{{}^{y}\\!v}(b,d,t^{\kappa+1/2})\,f(S(b,d,t^{\kappa+1/2}))-{{}^{y}\\!v}(b,c,t^{\kappa+1/2})\,f(S(b,c,t^{\kappa+1/2}))}$ $\displaystyle\displaystyle{+{{}^{y}\\!v}(a,d,t^{\kappa+1/2})\,f(S(a,d,t^{\kappa+1/2}))-{{}^{y}\\!v}(a,c,t^{\kappa+1/2})\,f(S(a,c,t^{\kappa+1/2}))\Big{]}\,d\tau}$ where $\alpha_{X}={\Delta t_{c}}/\Delta X$ and $\alpha_{Y}={\Delta t_{c}}/\Delta Y$. The midpoint values are computed using the Taylor expansions and the conservation law (2). For instance, $\left\\{\begin{array}[]{l}S(a,d,t^{\kappa+1/2}):=S(a,d,t)\\\ \hskip 85.35826pt\displaystyle{-\frac{{\Delta t_{c}}}{2}\Big{(}{{}^{x}\\!v}(a,d,t)\,f(S(a,d,t))\Big{)}_{x}}\displaystyle{-\frac{{\Delta t_{c}}}{2}\Big{(}{{}^{y}\\!v}(a,d,t)\,f(S(a,d,t))\Big{)}_{y}}\\\ \\\ \displaystyle{S(a,d,t):=\overline{S}^{\kappa}_{j+1,k}-\Delta x\,(S_{x})^{\kappa}_{j+1,k}\Big{(}\frac{1}{2}-\alpha_{X}c^{x}_{j+1/2,k}\Big{)}}\displaystyle{-\Delta y\,(S_{y})^{\kappa}_{j+1,k}\Big{(}\frac{1}{2}-\alpha_{Y}b^{y}_{j+1/2,k+1/2}\Big{)}}.\end{array}\right.$ As the time step ${\Delta t_{c}}$ goes to zero, the limit $\displaystyle\lim_{{\Delta t_{c}}{\rightarrow}0}S(a,d,t^{\kappa+1/2})=\overline{S}^{\kappa}_{j+1,k}-\frac{\Delta X}{2}\,(S_{x})^{\kappa}_{j+1,k}-\frac{\Delta Y}{2}\,(S_{y})^{\kappa}_{j+1,k}:=S^{++}_{j+1/2,k-1/2}.$ (25) These are called the intermediate values and their general form is $\displaystyle S^{\pm\pm}_{j+1/2,k+1/2}$ $\displaystyle=$ $\displaystyle\overline{S}^{\kappa}_{j+1/2\pm 1/2,k+1/2\pm 1/2}\pm\frac{\Delta X}{2}\,(S_{x})^{\kappa}_{j+1/2\pm 1/2,k+1/2\pm 1/2}(x_{j+1/2}-x_{j+1/2\pm 1/2})$ (26) $\displaystyle\pm\frac{\Delta Y}{2}\,(S_{y})^{\kappa}_{j+1/2\pm 1/2,k+1/2\pm 1/2}(y_{k+1/2}-y_{k+1/2\pm 1/2})$ We notice that the cell averages $w^{\kappa+1}_{j,k+1/2}$ and $w^{\kappa+1}_{j,k}$ are obtained analogously to (2.4). And also, $w^{\kappa+1}_{j-1/2,k}=w^{\kappa+1}_{j+1/2-1,k}$ e $w^{\kappa+1}_{j,k-1/2}=w^{\kappa+1}_{j,k+1/2-1}$. Substituting all these cell averages in time step ${t^{\kappa}_{m}}+{\Delta t_{c}}$ into Equation (2.4) and computing the limit when ${\Delta t_{c}}{\rightarrow}0$, we obtain the second order central scheme in semi- discrete formulation: $\frac{d}{dt}\overline{S}_{jk}(t)=-\frac{H^{x}_{j+1/2,k}(t)-H^{x}_{j-1/2,k}(t)}{\Delta X}-\frac{H^{y}_{j,k+1/2}(t)-H^{y}_{j,k-1/2}(t)}{\Delta Y},$ (27) where the numerical fluxes are $\displaystyle H^{x}_{j+1/2,k}(t)=$ $\displaystyle\frac{1}{4}\Big{\\{}{{}^{x}\\!v}_{j+1/2,k+1/2}(t)\Big{[}f(S^{+-}_{j+1/2,k+1/2}(t))+f(S^{--}_{j+1/2,k+1/2}(t))\Big{]}$ $\displaystyle+{{}^{x}\\!v}_{j+1/2,k-1/2}(t)\Big{[}f(S^{++}_{j+1/2,k-1/2}(t))+f(S^{-+}_{j+1/2,k-1/2}(t))\Big{]}\Big{\\}}$ $\displaystyle-\frac{c^{x}_{j+1/2,k}}{2}\Big{[}S^{+}_{j+1/2,k}(t)-S^{-}_{j+1/2,k}(t)\Big{]};$ (28a) $\displaystyle H^{y}_{j,k+1/2}(t)$ $\displaystyle=\frac{1}{4}\Big{\\{}{{}^{y}\\!v}_{j+1/2,k+1/2}(t)\Big{[}f(S^{-+}_{j+1/2,k+1/2}(t))+f(S^{--}_{j+1/2,k+1/2}(t))\Big{]}$ $\displaystyle+{{}^{y}\\!v}_{j-1/2,k+1/2}(t)\Big{[}f(S^{++}_{j-1/2,k+1/2}(t))+f(S^{+-}_{j-1/2,k+1/2}(t))\Big{]}\Big{\\}}$ $\displaystyle-\frac{d^{y}_{j,k+1/2}}{2}\Big{[}S^{+}_{j,k+1/2}(t)-S^{-}_{j,k+1/2}(t)\Big{]}.$ (28b) If the numerical derivatives are equal to zero then we obtain the two- dimensional Rusanov’s central scheme in semi-discrete formulation. $\displaystyle\mbox{{Rusa}}^{x}_{j+1/2,k}(t)=$ $\displaystyle\frac{1}{4}\Big{\\{}{{}^{x}\\!v}_{j+1/2,k+1/2}(t)\Big{[}f(S_{j+1,k}(t))+f(S_{j,k}(t))\Big{]}$ $\displaystyle+{{}^{x}\\!v}_{j+1/2,k-1/2}(t)\Big{[}f(S_{j+1,k}(t))+f(S_{j,k}(t))\Big{]}\Big{\\}}$ $\displaystyle-\frac{c^{x}_{j+1/2,k}}{2}\Big{[}S_{j+1,k}(t)-S_{j,k}(t)\Big{]};$ (29a) $\displaystyle\mbox{{Rusa}}^{y}_{j,k+1/2}(t)=$ $\displaystyle\frac{1}{4}\Big{\\{}{{}^{y}\\!v}_{j+1/2,k+1/2}(t)\Big{[}f(S_{j,k+1}(t))+f(S_{j,k}(t))\Big{]}$ $\displaystyle+{{}^{y}\\!v}_{j-1/2,k+1/2}(t)\Big{[}f(S_{j,k+1}(t))+f(S_{j,k}(t))\Big{]}\Big{\\}}$ $\displaystyle-\frac{d^{y}_{j,k+1/2}}{2}\Big{[}S_{j,k+1}(t)-S_{j,k}(t)\Big{]}.$ (29b) This new two-dimensional semi-discrete central scheme with the numerical fluxes given by (28) or (29) comprises a system of ordinary differential equations. To solve this system, we use the explicit second order Runge-Kutta method. ### 2.5. The velocity field Finally, to complete the description of the genuinely two-dimensional KT scheme, we have to define the velocity field. The velocity is defined at the vertices of the cells. We can not use directly the velocity field from the Raviart-Thomas space as we did in the dimension by dimension approach. Instead we will use a bilinear interpolation of it preserving the null divergence necessary for the incompressible flows. For instance, to compute the value of ${{}^{x}\\!v}_{j+1/2,k+1/2}$ on the vertex $(x_{j+1/2},y_{k+1/2})$ at some time step $t^{m}$, we have to use all the four cells which share this vertex, $\displaystyle{{}^{x}\\!v}_{j+1/2,k+1/2}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\Big{(}{{}^{x}\\!v}_{j,k+1/2}+{{}^{x}\\!v}_{j+1,k+1/2}\Big{)}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\Big{[}\frac{1}{2}\Big{(}{{}^{x}\\!v}_{j,k}+{{}^{x}\\!v}_{j,k+1}\Big{)}+\frac{1}{2}\Big{(}{{}^{x}\\!v}_{j+1,k}+{{}^{x}\\!v}_{j+1,k+1}\Big{)}\Big{]}$ $\displaystyle=$ $\displaystyle+\frac{1}{2}\Big{[}\frac{1}{2}\Big{(}\frac{1}{2}(\vec{v_{r}}_{j+1/2,k}-\vec{v_{l}}_{j-1/2,k})$ $\displaystyle+\frac{1}{2}(\vec{v_{r}}_{j+1/2,k+1}-\vec{v_{l}}_{j-1/2,k+1}\Big{)}+\frac{1}{2}\Big{(}\frac{1}{2}(\vec{v_{r}}_{j+3/2,k}-\vec{v_{l}}_{j+1/2,k})+$ $\displaystyle\frac{1}{2}(\vec{v_{r}}_{j+3/2,k+1}-\vec{v_{l}}_{j+1/2,k+1})\Big{)}$ $\displaystyle=$ $\displaystyle\frac{1}{8}\Big{(}\vec{v_{r}}_{j+1/2,k}-\vec{v_{l}}_{j-1/2,k}+\vec{v_{r}}_{j+1/2,k+1}-\vec{v_{l}}_{j-1/2,k+1}$ $\displaystyle+\vec{v_{r}}_{j+3/2,k}-\vec{v_{l}}_{j+1/2,k}+\vec{v_{r}}_{j+3/2,k+1}-\vec{v_{l}}_{j+1/2,k+1}\Big{)}$ ## 3\. TWO-DIMENSIONAL NUMERICAL EXPERIMENTS We present and compare the results for numerical simulations of two- dimensional, two-phase flow associated with two distinct flooding problems using the KT scheme with the dimension by dimension approach (KT dxd) and the genuinely two-dimensional formulation (SD2-2D). The first problem is a two- dimensional flow in a rectangular heterogeneous reservoir (called slab geometry) with $256$ m $\times$ $64$ m in size, and the second is a two- dimensional flow in a 5-spot geometry homogeneous reservoir having $64$ m $\times$ $64$ m. In the 5-spot geometry homogeneous reservoir simulation we used two distinct uniform five-spot well configurations intended to illustrate different flow patterns, with parallel and diagonal grid orientations, and boundary behavior. In all simulations the reservoir contains initially $79\%$ of oil and $21\%$ of water. Water is injected at a constant rate of $0.2$ pore volumes every year. The fractional volumetric flow, the total mobility, and the relative permeabilities are assumed to be: $f(s)=\frac{k_{rw}(s)/\mu_{w}}{\lambda(s)},\quad\lambda(s)=\frac{k_{rw}(s)}{\mu_{w}}+\frac{k_{ro}(s)}{\mu_{o}},$ and $k_{ro}(s)=(1-(1-s_{ro})^{-1}s)^{2},\quad k_{rw}(s)=(1-s_{rw})^{-2}(s-s_{rw})^{2},$ where $s_{ro}=0.15$ and $s_{rw}=0.2$ are the residual oil and water saturations, respectively. The viscosity of oil and water used in all simulations are $\mu_{o}=10.0\,cP$ and $\mu_{o}=0.05\,cP$. For the heterogeneous reservoir studies we consider a scalar absolute permeability field $K({\bf x})$ taken to be log-normal (a fractal field, see (Glimm et al., 1993) and (Furtado and Pereira, 2003) for more details) with moderately large heterogeneity strength. The spatially variable permeability field is defined on a 256 $\times$ 64 grid with three different values of the coefficient of variation $C_{v}$ (standard deviation)/mean: $0.5,1.0,$ and $2.4$. The boundary conditions, injection and production specifications for two-phase flow equations (1)-(2)) are as follows. For the horizontal slab geometry, injection is made uniformly along the left edge of the reservoir and the (total) production rate is taken to be uniform along the right edge; no flow is allowed along the edges appearing at the top and bottom of the reservoir in the graphics (Figures 6, 7, and 8). In the case of a five-spot flood with diagonal grid (Figure 9), injection takes place at one corner and production at the diametrically opposite corner; no flow is allowed across the entirety of the boundary. In the case of a five- spot flood with the parallel grid, injection takes place at two corners (diametrically opposite), say left down and right up, and production in the diametrically ’off corners’, say right down and left up. Figure 6. Water saturation surface plots for two-phase flow in a two- dimensional heterogeneous reservoir having $256$ m $\times$ $64$ m, with the coefficient of variation $C_{v}=0.5$ and viscous ratio 20. From top to bottom:$1)$ (KT dxd) scheme with $256\times 64$ grid; $2)$ (KT dxd) scheme with $512\times 128$ grid; $3)$ (KT two) scheme with $256\times 64$ grid. Figure 7. Water saturation surface plots for two-phase flow in a two- dimensional heterogeneous reservoir having $256$ m $\times$ $64$ m, with the coefficient of variation $CV=1.2$ and viscous ratio 20. From top to bottom:$1)$ (KT dxd) scheme with $256\times 64$ grid; $2)$ (KT dxd) scheme with $512\times 128$ grid; $3)$ (KT two) scheme with $256\times 64$ grid. Figure 8. Water saturation surface plots for two-phase flow in a two- dimensional heterogeneous reservoir having $256$ m $\times$ $64$ m, with the coefficient of variation $CV=2.2$ and viscous ratio 20. From top to bottom:$1)$ (KT dxd) scheme with $256\times 64$ grid; $2)$ (KT dxd) scheme with $512\times 128$ grid; $3)$ (KT two) scheme with $256\times 64$ grid. (a) NT2D: $64\\!\times\\!64$ c lulas (b) KTdxd: $64\\!\times\\!64$ c lulas (c) NT2D: $128\\!\\!\times\\!\\!128$ c lulas (d) KTdxd: $128\\!\times\\!128$ c lulas Figure 9. Water saturation level curves for two-phase flow in a five-spot well configuration. The SD2-2D scheme was used in pictures in the left column and the KTdxd was used in pictures in the right column. ### 3.1. Analyzing the Numerical Results. Conclusions The Figures 6, 7 and 8 refer to a comparative study for the KT dimension by dimension and a genuinely two-dimensional KT schemes showing the water saturation surface plots after $350$ days of simulation for three different values for the strength of the heterogeneity of the fractal permeability field, $CV=0.5,1.2$ and $2.2$. Note that the genuinely two-dimensional KT scheme gives a more accurate solution than the solutions computed by the KT dimension by dimension scheme for the same grid. In fact we observe that the KT dxd is only comparable in accuracy with one degree of refinement (see Figures 6, 7 and 8). The better accuracy of the genuinely two-dimensional approach is due to a more precise computation of the genuinely two-dimensional numerical fluxes, with respect to the one dimensional numerical fluxes in the dimension by dimension approach. In the case of a five-spot geometry homogeneous reservoir, Figure 9 (diagonal grid) shows the saturation level curves after $260$ days of simulation obtained with KTdxd and SD2-2D schemes for two levels of spatial discretization. In this figure 9, the pictures on the left column are the results obtained with the SD2-2D scheme and the ones on the right were computed with the KTdxd scheme. In these Figures, the grid become finer from top to bottom, having $64\,\times\,64$ and $128\,\times\,128$ cells, respectively. It is clear that the KTdxd scheme (right column pictures in Figures 9 is producing incorrect boundary behavior. Moreover as the computational grid is refined (right column and bottom picture in Figure 9) this problem seems to be emphasized indicating that the solution is not convergent. The KTdxd scheme uses numerical fluxes in the $x$ and $y$ directions which can be viewed as generalizations of one-dimensional numerical fluxes. We state that this type of approximation for the fluxes leads to the incorrect boundary behavior discussed above. This incorrect boundary behavior led us to develop a new genuinely two-dimensional KT scheme. The results obtained with this new scheme can be seen in the left column of Figure 9. It is clear that we have corrected the boundary behavior just changing the approach from dimension by dimension to our two-dimensional approach. This fact indicates (but does not prove) our idea that computing two-dimensional numerical fluxes using straight generalizations of one-dimensional numerical fluxes may produce incorrect numerical solutions. ## References * Abreu et al. (2004a) E. Abreu, F. Furtado, D. Marchesin, and F. Pereira. Transitional waves in three-phase flows in heterogeneous formations. _Computational Methods for Water Resources_ , I:609–620, 2004a. * Abreu et al. (2004b) E. Abreu, F. Furtado, D. Marchesin, and F. Pereira. Three-phase flow in petroleum reservoirs. _Proceedings of the Conference on Analysis, Modeling and Computation of PDEs and Multiphase Flow, in Celebration of James Glimm’s 70th Birthday, SUNY at Stony Brook. To appear_ , 2004b. * Abreu et al. (2004c) E. Abreu, F. Furtado, and F. Pereira. On the numerical simulation of three-phase reservoir transport problems. _Transport Theory Statist. Phys._ , 33(5-7):503–526, 2004c. * Abreu et al. (2006) E. Abreu, F. Furtado, and F. Pereira. Three-phase immiscible displacement in heterogeneous petroleum reservoirs. _Mathematics and computers in simulation_ , 73:2–20, 2006\. * Abreu et al. (2007) E. Abreu, Furtado F., F. Pereira, and S. Ribeiro. Central schemes for porous media flow. _(Paper to be submited.)_ , 1:6, 2007. * Abreu et al. (2008) E. Abreu, Jim Douglas Jr., F. Furtado, and F. Pereira. Operator splitting for three-phase flow in heterogeneous porous media, 2008. _To appear in: International Journal of Computational Science_ , 2008\. * Chavent and Jaffré (1986) G. Chavent and J. Jaffré. _Mathematical Methods and Finite Elements for Reservoir Simulation_ , volume 17. Studies in Mathematics and its Applications, North-Holland, Amsterdam, studies in mathematics and its applications edition, 1986. * Douglas et al. (1997) J. Douglas, Jr., F. Furtado, and F. Pereira. On the numerical simulation of waterflooding of heterogeneous petroleum reservoirs. _Comput. Geosci._ , 1(2):155–190, 1997. * Furtado and Pereira (2003) F. Furtado and F. Pereira. Crossover from nonlinearity controlled to heterogeneity controlled mixing in two-phase porous media flows. _Comput. Geosci._ , 7(2):115–135, 2003. * Glimm et al. (1993) J. Glimm, B. Lindquist, F. Pereira, and Q. Zhang. A theory of macrodispersion for the scale up problem. _Transport in Porous Media_ , 13:97–122, 1993. * Jiang and Tadmor (1998) G.-S. Jiang and E. Tadmor. Non-oscillatory central schemes for multidimensional hyperbolic conservation laws. _SIAM Journal on Scientific Computing_ , 19:1892–1917, 1998\. * Kurganov and Petrova (2001) A. Kurganov and G. Petrova. A third-order semi-discrete genuinely multidimensional central scheme for hyperbolic conservation laws and related problems. _Numer. Math._ , 88(4):683–729, 2001. * Kurganov and Tadmor (2000) A. Kurganov and E. Tadmor. New high-resolution central schemes for nonlinear conservation laws and convection-diffusion equations. _J. Comput. Phys._ , 160(1):241–282, 2000. * Lax (1954) P.D. Lax. Weak solutions of non-linear hyperbolic equations and their numerical computation. _Comm. Pure Appl. Math._ , 7:159–193, 1954. * Nessyahu and Tadmor (1990) H. Nessyahu and E. Tadmor. Nonoscillatory central differencing for hyperbolic conservation laws. _J. Comput. Phys._ , 87(2):408–463, 1990. * Peaceman (1977) D. W. Peaceman. _Fundamentals of Numerical Reservoir Simulation_. Elsevier, New York, 1977. * Raviart and Thomas (1977) P.-A. Raviart and J. M. Thomas. A mixed finite element method for 2nd order elliptic problems. In _Mathematical aspects of finite element methods (Proc. Conf., Consiglio Naz. delle Ricerche (C.N.R.), Rome, 1975)_ , pages 292–315. Lecture Notes in Math., Vol. 606. Springer, Berlin, 1977. * Rusanov (1961) V. V. Rusanov. The calculation of the interaction of non-stationary shock waves with barriers. _Ž. Vyčisl. Mat. i Mat. Fiz._ , 1:267–279, 1961. * Van Leer (1979) B. Van Leer. Towards the ultimate conservative difference scheme. v. a second order sequel to godunov’s metthod. _J. Comp. Phys._ , 32:101–136, 1979.

arxiv-papers

2024-09-04T02:48:54.603832

0803.4256

# Melting and freezing of argon in a granular packing of linear mesopore arrays Christof Schaefer, Tommy Hofmann Dirk Wallacher Patrick Huber p.huber@physik.uni-saarland.de Klaus Knorr knorr@mx.uni-saarland.de Faculty of Physics and Mechatronics Engineering, Saarland University, D-66041 Saarbrücken, Germany ###### Abstract Freezing and melting of Ar condensed in a granular packing of template-grown arrays of linear mesopores (SBA-15, mean pore diameter $8~{}\mathrm{n}\mathrm{m}$) has been studied by specific heat measurements $C$ as a function of fractional filling of the pores. While interfacial melting leads to a single melting peak in $C$, homogeneous and heterogeneous freezing along with a delayering transition for partial fillings of the pores result in a complex freezing mechanism explainable only by a consideration of regular adsorption sites (in the cylindrical mesopores) and irregular adsorption sites (in niches of the rough external surfaces of the grains, and at points of mutual contact of the powder grains). The tensile pressure release upon reaching bulk liquid/vapor coexistence quantitatively accounts for an upward shift of the melting/freeezing temperature observed while overfilling the mesopores. ###### pacs: 64.70.Nd, 65.80.+n, 65.40.Ba The arguably most conspicuous effect of pore confinement on molecular condensates is the shift of phase transitions with respect to the non-confined bulk state. Thus condensation (the vapor-liquid transition) does not occur at the saturated vapor pressure $p_{0}$ of the bulk liquid but in a wide range of reduced vapor pressures $P=p/p_{0}$, $P<1$, starting with the condensation of an adsorbed layer on the pore walls and culminating in pore filling via capillary condensationCole and Saam (1974). In a similar way melting and freezing of the material in the pores occurs at temperatures $T$ well below the bulk triple point temperature $T_{\rm 3}$ Christenson (2001). These reductions are usually interpreted in terms of the Kelvin- and the Gibbs- Kelvin equation, resp. Here interfacial energies are introduced and because of their competition with volume free energies, the shift scales with the inverse of the characteristic linear dimension $L$ of the geometric configuration. In case of tubular pores this is the pore radius $R$. Here we present an experimental calorimetric study on the melting and freezing of one of the most simple condensates imaginable, the rare gas Ar, in a template grown mesoporous silica, known as SBA-15 and considered one of the most ideal mesoporous substrates available Zhao et al. (1998). Figure 1: Ar sorption isotherm in SBA-15, measured at $86~{}\mathrm{K}$. Inset: Scanning electron micrographs of a SBA-15 grain taken with different spatial resolutions as indicated in the figure. The preparation has been described elsewhere Hofmann et al. (2005). The material obtained is a powder of grains, about $1$ micron in size, which have a relatively rough surface (Fig. 1). The pores within the grains are linear, non-ramified, parallel, have a relatively uniform cross section, and form a periodic 2D hexagonal array. Hence network effects and the blocking of the solidification front in bottle necks Khokhlov et al. (2007) are not expected to play a major role. The Bragg angles of the diffraction pattern of the empty pore lattice give a lattice parameter (= pore-pore-distance) of $10.7~{}\mathrm{n}\mathrm{m}$ Hofmann et al. (2005). A fit of a radial electron density $\rho(r)$ profile to the Bragg intensities (with $\rho=0$ for $r<R_{0}-d$ , $\rho=\text{constant}$ for $r>R_{\rm 0}+d$, and with a linear function in the ”corona” in between yields a pore radius $R_{\rm 0}$ of $3.8~{}\mathrm{n}\mathrm{m}$, corresponding to a porosity of $0.49$, and a corona thickness $2d$ of $2.2~{}\mathrm{n}\mathrm{m}$. A volumetric Ar sorption isotherm (normalized uptake $f=n/n_{0}$ as function of the reduced vapour pressure $P$, where $n$ is the adsorbed amount of Ar and $n_{0}=30$ mmol is the amount of Ar for complete filling of the pores), recorded in the liquid regime of the Ar pore filling at $86~{}\mathrm{K}$, is shown in Fig. 1. The initial reversible part is due to the condensation of Ar onto the pore walls, the hysteretic part to the filling of the pores by capillary condensation on $p$-increase (starting at about $f_{\rm A}=0.65$) and to the evaporation of the capillary condensate on $p$-decrease (terminating at about $f_{\rm D}=0.42$). The branches of the hysteresis loop are relatively steep compared to what is observed for other mesoporous substrates, suggesting a narrow distribution of the pore radius and the absence of pore network effects such as pore blocking Yortsos (1999). The final slope in the post-filling regime $f>1$ (absent e.g. for monolithic Vycor substrates) demonstrates that there are further sites available for condensation, with characteristic linear dimensions larger than $R_{\rm 0}$, but still finite. We think of tapered pore mouths, niches of the rough external surface of the grains, and the points at which the powder grains are in mutual contact. Figure 2: (color online). Specific heat curves of cooling (solid line) and heating (dashed line) cycles for selected filling fractions $f$ indicated in the figure. The calorimeter consists of the sample cell, heat shield, and the outer jacket, with two separate vacuum spaces in between. The sample cell and the heat shield are equipped with thermometers ($1000~{}\Omega$ Pt-resistors) and resistance heaters. The heat shield is in thermal contact with the cold plate of a closed cycle He refrigerator. The cell contains $1.3~{}\mathrm{g}$ SBA-15 and a coil of Ag wire (that is meant to reduce the thermal time constant within the powder). $12$ fractional fillings have been investigated that have been prepared by introducing suitable amounts of Ar into the cell via a fill line at $86~{}\mathrm{K}$. We employ a scanning experiment the calorimetric signal is proportional to $dQ/dT$,the amount of heat per unit $T$-interval that goes into or comes out of the sample. $dQ/dT=K\Delta T/r$, $\Delta T$ and $K$ are the temperature difference and the heat leak between cell and shield, $r$ is the cooling/heating rate of the shield. The heat leak has been realized by a He gas pressure of a few mbar in the vacuum space between cell and shield, $\Delta T$ was typically $1~{}\mathrm{K}$, the heating/cooling rate was held constant at a value of $0.05~{}\mathrm{K}\mathrm{/}\mathrm{min}$. Quite elaborate thermometry allowed us to work with such low rates which are more than one order of magnitude smaller than what is standard with commercial equipment. The samples have been subject to a complete freezing/heating cycle starting at $86~{}\mathrm{K}$ in the liquid regime, taking the sample to a minimum temperature of $50~{}\mathrm{K}$ deep in the solid regime, and bringing it back to $86~{}\mathrm{K}$. The results for a selected set of fillings are shown separately for cooling and heating in Fig. 2. Samples with $f\geq 0.45$ do show heating and freezing anomalies, samples with $f\leq 0.38$ do not. Analogous observations have been made for Ar in Vycor Wallacher and Knorr (2001). Thus a phase-transition like change from the liquid to the crystalline state is basically reserved to the capillary condensed part of the pore filling. The initial ”dead” fraction of the condensate up to about $f=0.4$, adsorbed on the pore walls and filling the niches of the corona, does not participate in a collective freezing melting process. Melting of the ”mobile” part occurs in a relatively narrow $T$-interval of $2~{}\mathrm{K}$, which for $f<1$ is centred at $74.5~{}\mathrm{K}$, whereas freezing - say for $f=0.92$ \- extends over $7~{}\mathrm{K}$ from $71.5~{}\mathrm{K}$ down to $64.5~{}\mathrm{K}$ with the freezing anomaly having a rather complex shape (Fig. 2). The entropy of fusion, $S_{\rm 0}$, is obtained by integrating the calorimetric signal above the smooth background, $S_{\rm 0}=c\smallint(dQ^{\prime}/dT)/TdT$. The coefficient $c$ has not been determined, but is the same for all heating and cooling runs. A plot of $S_{\rm 0}$ as function of $f$ is shown in Fig. 3. The values of $S_{\rm 0}$ obtained from cooling and heating runs are identical within experimental error. This is not only a consistency check but also means that all parts of the freezing anomalies are indeed due to no other phase transition but solidification. For $f<1$, $S_{\rm 0}$ is a linear function of $f$ that extrapolates to zero at about $f=0.39$, close to $f_{\rm D}$ of the $86~{}\mathrm{K}$-sorption isotherm. Thus the part of the condensate that supports the collective freezing/melting process comprises not only what is formed by capillary condensation but also the fraction which grows initially in form of a metastabel film at the pore walls ($f_{\rm A}-f_{\rm D}$). For ease of discussion we divide the freezing anomaly into the parts I,II,III which are meant to pertain to certain fractions of the pore filling. Part II and III are the peaks at $67.5~{}\mathrm{K}$ and $65.5~{}\mathrm{K}$, respectively, part I the remainder at higher $T$. The $f$-dependence of the three parts is also shown in Fig. 3. The fact that the pore filling solidifies sequentially, but melts practically in a single step strongly suggests that solidification leads to a reorganization of the capillary condensate upon which the different fractions I, II, III of pore material loose their identity. The melting anomalies have an asymmetric shape. For $f<1$, the low-$T$ wings of the anomalies for different filling fractions superimpose. We have interpreted the analogous observations on Ar in Vycor in terms of interfacial melting in a cylindrical pore starting at the boundary between the dead and the mobile part of the pore filling rather than by referring to a pore size distribution which had to be rather special in order to reproduce the peculiar shape of the anomalies Wallacher and Knorr (2001). This interfacial melting scenario is similar to surface melting at semi-confined, planar surfaces Dash (1986). Calorimetry can of course not prove whether the melting front propagates along the pores or radially inward, but the mere fact that the asymmetric shape of the anomaly is recovered in the highly homogeneous pores of the present substrate is a strong argument in favor of interfacial i.e. radial melting. As for freezing, part III of the freezing anomaly represents the component of the filling that is the last to solidify on cooling. $S_{\rm 0}$(III) shows the peculiar type of $f$-dependence , $S_{\rm 0}$(III) scales with the vapor filled fraction of pore space. $S_{\rm 0}$(III) is due to the solidification of the mobile part of the liquid film on the pore walls in otherwise empty pore segments. Upon solidification the film delayers and the material involved joins the already solidified capillary condensate (and melts as such on subsequent heating). The same observations have been made for Ar in Vycor Wallacher and Knorr (2001). Still referring to the underfilled situation, $f<1$, this reasoning leaves parts I and II for the freezing of the capillary condensate in the regular pores. We recall that the interfacial melting model not only explains the lowering of the equilibrium freezing/melting temperature $T_{\rm 0}$ in pore confinement but also predicts metastable states, liquid ones down to the lower spinodal temperature $T^{\rm-}$ and solid ones up to the upper spinodal temperature $T^{\rm+}$. We propose to identify peak II with the freezing of the capillary condensate via homogeneous nucleation of the solid phase at $T^{\rm-}$. Peak II is sharp because of the narrow pore size distribution of SBA-15. Part I of the freezing anomaly on the other hand stems from the freezing via heterogeneous nucleation within the rather broad ($T_{\rm 0}$, $T^{\rm-}$)-interval, triggered by the presence of various types of small nuclei residing at irregular sites. The chance that a parcel of liquid is in contact with such nuclei increases with $f$. That is why part II dominates (in coexistence with part III) at low $f$ and goes through a maximum slightly below $f=1$ whereas part I gradually develops with increasing $f$ in a somewhat delayed fashion (Figs. 2 and 3). Figure 3: Entropy plots of the different anomaly parts as a function of filling fraction $f$. The dashed lines are guides for the eye. The data on the overfilled sample, $1.04<f<1.19$, shows that the freezing/melting of different parts of the Ar condensate are not independent from each other, an observation that supports the idea of heterogeneous freezing. Quite in contrast to the results on a monolithic Vycor sample, the melting and freezing of the excess material, $f-1$, does not lead to a $\delta$-peak at the triple point of the bulk system (at $83.8~{}\mathrm{K}$), but in first respect to some extra calorimetric signal at the high-$T$ end of the anomalies, both for freezing and melting. Obviously most of the excess material resides in sites, say in niches of the external surface of the powder grains, with dimensions that are still in the nm-range just slightly larger than the diameter of the regular pores. The presence of this material has an effect on the freezing and melting of the material in the pores. The melting peak for $f=0.92$ for example is centered at $T_{\rm m}=74.8~{}\mathrm{K}$, but adding excess material the melting peak eventually shifts to $76.1~{}\mathrm{K}$ and even reduces the calorimetric signal at the former $T_{\rm m}$. Thus the presence of material outside stabilizes the solid state in the pores. The pore filling experiences a negative hydrostatic pressure $p_{\rm h}$ of $70\pm 10$ bar for $f<1$ Kanda (2004). This value is dictated by the Laplace equation while assuming semi-spherical, concave menisci with negative curvature radii of the condensate-vapor interface on the order of the pore radius Huber and Knorr (1999). Upon post-filling of the pores the negative curvature of these interfaces gradually vanishes; hence the tensile pressure in the pore condensate disappears. Based on the $p_{\rm h}$ dependency of the melting line of Ar Michels (1962), an increase of the melting temperature of $1.7\pm 0.3$ K is expected for a $70\pm 10$ pressure release, in good agreement with the $\sim$ 1.5 K $T$-shift of the melting peak documented in Fig. 2. The upward shift of the onset of the freezing anomaly for $f>1$ (part I) can be explained in an analogous way. An opposite, downward shift of $T_{\rm m}$ upon approaching $f=1$, observed for water in mesopores Findenegg (2001), is consistent with our tensile pressure hypothesis due to water’s anomalous $T_{\rm m}(p_{\rm h})$-curve. Figure 4: (color online). Incomplete (partial) cooling-heating cycles at $f=0.66$ (solid lines) in comparison to heating, cooling curves after complete solidification, melting, resp. at $f=0.62$ (dotted lines). For $f=0.66$ partial cooling/heating cycles have been investigated (Fig. 4). In the first type of such cycles a part of the condensate is solidified by cooling to some temperature within the $T$-range of the freezing (Fig. 4a) and the melting of this part is then studied on subsequent heating (Fig. 4b). In the first run the freezing of part I is initiated by cooling down to $68~{}\mathrm{K}$, in the second one parts I and II are solidified by cooling down to $66.5~{}\mathrm{K}$. As far as position and shape are concerned, the resulting melting anomalies are identical to what is obtained after complete solidification (see Fig. 2 for data on $f=0.62$ and $0.72$), it is just that $S_{\rm 0}$ is scaled down in proportion to the amount of material solidified on cooling. Parts of the condensate with largely different freezing temperatures melt in the same narrow $T$-interval. The second type of cycles starts at $60~{}\mathrm{K}$ deep in the solid regime and a part of the solid is melted by heating up to temperatures within the $T$-range of melting (Fig. 4c). The re-solidification is then studied by subsequent cooling (Fig. 4d). In such partial cycles the calorimetric signal right at the onset of freezing is enhanced. Obviously the fraction of material that has been melted in the heating cycle is still in contact with the solid remaining and can therefore resolidify directly without having to overcome a nucleation barrier. Peak III is recovered in the partial cycles. Whenever some part of the pore filling is liquefied, the mobile part of film coating on the pore walls is re-established. As for peak II, the completed freezing process of the first cooling run down to 60 K leads to a re-arrangement of the pore filling in pore space by which the isolated parcels are eliminated, the freezing of which (via homogeneous nucleation) gives rise to peak II. These parcels are not re-established in case melting along the heating run is incomplete. This is why peak II is absent in the lower trace of Fig. 4d. Even in case of complete melting by heating up to a temperature of 75.5 K slightly above the high-$T$ cutoff of the melting anomaly, the isolated parcels are not fully recovered compared to the situation of the virgin sample right after preparation by vapor condensation at 86 K, see the dotted line in Fig. 4. This explains the reduced size of peak II in the upper trace of Fig. 4d. After a freezing-melting cycle the pore liquid keeps some memory of the spatial distribution of the pore solid Soprunyuk et al. (2003). The freezing/melting phenomenology encountered in SBA-15 challenges the common notion which underlies pore size spectroscopy techniques (”thermoporometry”) Brun et al. (1977) relying on simple geometric relations between pore diameter and specific heat anomalies. As demonstrated here for a prototype mesoporous system, neither a resorting to the specific heat anomaly in freezing nor in melting will help in a granular packing of mesoporous grains. Freezing is demonstrated to be significantly affected by heterogeneous nucleation, not to mention the delayering transition for partial fillings, whereas melting is governed by interfacial melting processes. Freezing and melting in such a granular packing of comparably simple linear mesopores is clearly a complicated process. Metastable states, an interplay of homogeneous and heterogeneous nucleation processes and the coupling between different regions of pore space lead to a remarkable complex phenomenology. ###### Acknowledgements. This work has been supported by the SFB 277 of the Deutsche Forschungsgemeinschaft. ## References * Cole and Saam (1974) M. W. Cole and W. F. Saam, Phys. Rev. Lett. 32, 18 (1974). * Christenson (2001) H. K. Christenson, J. Phys.: Condens. Matter 13, R95 (2001); C. Alba-Simionescoet. al., J. Phys. Cond. Matt. 18, R15 (2006); K. Knorr, P. Huber, and D. Wallacher, Z. Phys. Chem. 222, 257 (2008). * Zhao et al. (1998) D. Zhao et. al., Science 279, 548 (1998). * Hofmann et al. (2005) T. Hofmann et. al., Phys. Rev. B 72, 064122 (2005); G. A. Zickler et. al., ibid. 73, 184109 (2006). * Khokhlov et al. (2007) A. Khokhlov et. al., New J. Phys. 9, 272 (2007). * Yortsos (1999) Y. C. Yortsos, _Methods in the Physics of Porous Media_ (Academic Press, 1999). * Wallacher and Knorr (2001) D. Wallacher and K. Knorr, Phys. Rev. B 63, 104202 (2001). * Dash (1986) D. Zhu and J.G. Dash, Phys. Rev. Lett. 57, 2959 (1986). * Kanda (2004) H. Kanda, M. Miyahara, and K. Higashitani, J. Chem. Phys. 120, 6173 (2004); H. Kanda and M. Miyahara, Adsorption 13, 191 (2007). * Huber and Knorr (1999) P. Huber and K. Knorr, Phys. Rev. B 60, 12657 (1999); J. Hoffmann and P. Nielaba, Phys. Rev. E 67, 036115 (2003). * Michels (1962) A. Michels and C. Prins, Physica 28, 101 (1962). * Findenegg (2001) A. Schreiber, I. Ketelsen, and G.H. Findenegg, Phys. Chem. Chem. Phys. 3, 1185 (2001). * Soprunyuk et al. (2003) V.P. Soprunyuk, D. Wallacher, P. Huber, and K. Knorr, and A.V. Kityk, Phys. Rev. B 67, 144105 (2003). * Brun et al. (1977) M. Brun, A. Lallemand, J. F. Quinson, and C. Eyraud, Thermochim. Acta 21, 59 (1977).

arxiv-papers

2024-09-04T02:48:54.609545

0803.4310

Corresponding author:]kchang@red.semi.ac.cn # Anomalous Rashba spin-orbit interaction in InAs/GaSb quantum wells Jun Li Kai Chang [ SKLSM, Institute of Semiconductors, Chinese Academy of Sciences, P. O. Box 912, Beijing 100083, China G. Q. Hai Instituto de Física de Säo Carlos, Universidade de Säo Paulo, 13560-970 Säo Carlos, Säo Paulo, Brazil K. S. Chan Department of Physics and Materials Science, City University of Hong Kong, Hong Kong, China ###### Abstract We investigate theoretically the Rashba spin-orbit interaction in InAs/GaSb quantum wells (QWs). We find that the Rashba spin-splitting (RSS) depends sensitively on the thickness of the InAs layer. The RSS exhibits nonlinear behavior for narrow InAs/GaSb QWs and the oscillating feature for wide InAs/GaSb QWs. The nonlinear and oscillating behaviors arise from the weakened and enhanced interband coupling. The RSS also show asymmetric features respect to the direction of the external electric field. ###### pacs: 78.40.Ri, 42.70.Qs, 42.79.Fm InAs/GaSb superlattices (SLs) and quantum wells (QWs) have attracted intensive attention in the past decades due to their potential application in nanoelectronics and remarkable electronic properties, e.g., the infrared detector and laser diode, as well as interband tunneling diodes and transistorsEsaki ; McGill ; Ting1 ; Houng . An interesting feature of this broken-gap structure is that the top of the valence band of GaSb lies above the bottom of the conduction band of InAs. A two-dimensional electron gas (2DEG) in the InAs layer can coexist with a two-dimensional hole gas in the GaSb layer since the electron can move across the InAs/GaSb interface from the valence band of GaSb to the conduction band of InAs, consequently leading to a semimetallic phaseEsaki ; JLuo . The energy spectrum exhibits an anticrossing behavior between the top valence subband and the lowest conduction subband at finite in-plane momentum when the lowest conduction subband in InAs layer lies below the top valence subband in GaSb layerBandStructure . The hybridized gap caused by the anticrossing was observed experimentallyGapExperiment , and leads to the semiconducting behavior of system. Recently, by utilizing the unique characteristics of the InAs/GaSb/AlSb system, e.g., the strong spin- orbit interaction (SOI) in InAs and GaSb and the high electron mobility of InAs, it may be possible to realize high-speed spintronic devices, e.g., Rashba spin filtersVosk ; Koga ; Ting2 , a spin field effect transistorDatta and a high-frequency optical modulator utilizing spin precessionHallstein . The SOI also has a significant influence on the spin relaxation of electrons in semiconductors and could be used to generate the spin current. In this Letter, we investigate theoretically the Rashba spin-splitting (RSS) in undoped InAs/GaSb quantum wells sandwiched by AlSb barriers by solving the eight-band Kane HamiltonianBurt ; RSS and Poisson equation self-consistently. We find a spontaneous RSS that arises from the interface contribution in the absence of external electric field, and RSS is asymmetric with respect to the direction of external electric field. It is interesting to note that RSS exhibits distinct behavior, i.e., nonlinear and oscillating feature as a function of the in-plane momentum, at small and large thicknesses of InAs layers due to the interband coupling and hybridization between the conduction band and the valence band. Figure 1: (a) Band profile and the probability density distribution of E1(+) (blue), HH1(+) (red) subband in InAs/GaSb ASQW at $k_{\parallel}=0$ (solid line) and $k_{\parallel}=0.4~{}nm^{-1}$ (dashed line), ($L_{InAs}=10~{}nm$,$L_{GaSb}=10~{}nm$). (b) The four band-edge state components, $|S\rangle$ (blue), $|HH\rangle$ (red), $|LH\rangle$ (green) and $|SO\rangle$ (cyan) as a function of in-plane wavevector of E1 band. The solid and dashed line denote the spin up (E1(+)) and spin down(E1(-)) state, respectively. The inset shows the calculated band structure. (c) and (d), the same as (a) and (b), but for a GaSb/InAs/GaSb SQW. We consider an undoped InAs/GaSb quantum well (grown on the [001] plane) shown schematically in Fig. 1(a). If an external electric field $F$ is applied along the direction perpendicular to the QW plane, an external electric field term $V_{E}\left(z\right)=eFz$ should be added to the total Hamiltonian. When the layer thickness or the external electric field is sufficiently large so that the bottom of the lowest conduction subband in InAs layer falls below the top of the highest valence subband in GaSb layer, electrons could transfer from GaSb layer to InAs layer, thus induced an internal electrostatic potential $V_{in}\left(z\right)$ in InAs and GaSb layersBastard . The total Hamiltonian writes as $H=H_{k}+V_{E}\left(z\right)+V_{in}\left(z\right)$ including the external and internal electrostatic potential. A self-consistent iteration procedure is performed until $V_{in}\left(z\right)$ is stable. The Kane parameters of materials used in our calculation are obtained from Ref. Parameters, . Figure 2: (a)Rashba spin-splitting of E1 band as a function of the in-plane momentum in the absence of external electric field for different thicknesses of InAs layer. (b) The same as (a), but for different external electric fields. The insets depict the band structure near the band gap. Fig. 1 (a) shows schematically the self-consistent band profile of a 10 nm-10 nm InAs/GaSb asymmetric chemical multilayer QW (ASQW) at $k_{\parallel}=0$. The energy dispersion of the ASQW obtained from the self-consistent calculation is plotted in the inset of Fig. 1(b). From the energy dispersion, we can find that the energy dispersion of the lowest (highest) conduction (valence) subband shows a minimum (maximum) at a finite $k_{\parallel}^{a}$, i.e., anticrossing behavior, since the bottom of the lowest electron subband in the InAs layer lies below the top of the highest heavy-hole subband in the GaSb layer. A spin-dependent hybridized gap($\sim$3 meV) at the anticrossing point forms due to the strong mixing effect of the InAs electron state and the GaSb hole state. In this QW, the concentration of electrons transferred from GaSb layer to InAs layer is found to be at the order of 1.9 $\times$ 10-11 cm-2, which induces a 7 meV bending down (up) in the profile of InAs’s conduction band (GaSb’s valence band) near the interface. The small band- bending only change the results slightly and shifts $k_{\parallel}^{a}$ to a smaller value. In Fig. 1(b), we plot the four components of the E1 state as a function of $k_{\parallel}$. The dominant component of E1 states (E1(+) and E1(-)) experiences a crossover from $|HH\rangle$ to $|S\rangle$ when the in- plane momentum $k_{\parallel}$ sweeps across the anticrossing point $k_{\parallel}^{a}$. Meanwhile, the dominant component of the HH1 band varies from electron-like to hole-like feature. This feature can also be demonstrated in Fig. 1(a) in which we also plot the density distribution of the E1(+) and HH1(+) at $k_{\parallel}=0$ and $0.4~{}nm^{-1}$. At $k_{\parallel}=0$, the E1 (HH1) state is mostly heavy-hole-like (electron-like) and therefore localizes in the GaSb (InAs) layer. At $k_{\parallel}>k_{\parallel}^{a}$, the two anticrossing subbands E1 and HH1 exchange their main characteristics, so the density distribution of E1(HH1) state localizes in InAs(GaSb) layer. Fig. 1 (c) and (d) plot the situation of a symmetric chemical multilayer QW (SQW). In this symmetric structure, the internal electrostatic potential is symmetric respect to the center of InAs layer. Thus there is no RSS existing, and the components of the two spin branches of E1 are identical in Fig. 1(d). In Fig. 2 (a) we plot the Rashba spin-splitting (RSS) of the lowest conduction subband (E1) as a function of the in-plane momentum at fixed thicknesses of the InAs and GaSb layers. As shown in the insets, the energy bands of narrow QW, e.g., 8 nm InAs layer with 10 nm GaSb layer, manifest a normal semiconductor phase since the strong quantum confinement pushes the lowest conduction subband to a higher energy. The RSS in this QW is a nonlinear function of in-plane wave vector, just as the RSS in biased narrow band gap semiconductor QWs we reported elsewhere before RSS . Interesting difference between this work and the previous work RSS is that there exist a spontaneous RSS in the InAs/GaSb QW in the absence of external electric fields, since it arises mainly from the asymmetric potential profile of InAs/GaSb QW, i.e., the asymmetric interband coupling at the left and right interfaces for InAs and GaSb layers. This is clearly demonstrated by the disappearing of the spontaneous RSS in the GaSb/InAs/GaSb SQW (see Fig. 1(c) and (d)). Besides the nonlinear RSS, one can also see an oscillating RSS of the E1 band in a wider QW, e.g., the QW of 10 nm InAs layer with 10 nm GaSb layer, in which an anticrossing occurs in the energy spectrum (see the inset of Fig. 2 (a)). A sharp drop of RSS corresponding to the anticrossing point $k_{\parallel}^{a}$ appears. It is interesting to see that the RSS changed its sign near the anticrossing point. Obviously the sign change of RSS corresponds to the cross of the two spin branches. In addition, the RSS is anisotropic with respect to different $k_{\parallel}$ direction, e.g., [100] and [110] direction (see Fig. 2). The anisotropy of RSS comes from the anisotropy of valence bands of InAs/GaSb QW through interband coupling. Figure 3: (a) Contour plot of self-consistently calculated RSS of E1 band as a function of the in-plane momentum and the thickness of the InAs layer without external electric field. (b) The same as (a), but with an external electric field $F=40~{}kV/cm$. In each panel the thickness of GaSb layer is fixed at 8 nm. Fig. 2 (b) displays the RSS of an InAs/GaSb quantum well as a function of the in-plane momentum for different external perpendicular electric fields. This figure shows that the RSS is heavily controlled by the external electric field. This behavior of RSS can be explained by the interplay between the asymmetric interface contribution and the interband coupling induced by the external electric field. In Fig. 2 (b), if an external electric field is applied parallel to the positive direction of the $z$ axis, the energy of the conduction subbands decreases while the energy of the valence subbands increases. Thus, the anticrossing behavior occurs or is enhanced even for QW with a narrow width InAs layer that exhibits a normal-semiconductor phase in zero electric fieldEfieldEffect . A valley of RSS appears once an anticrossing happens. The energy difference between the E1 and HH1 subbands is increased when the external electric field is applied antiparallel to the z axis. Therefore the anticrossing behavior is weakened and even smeared out. Compared to the conventional type-I QW, RSS in the type-II InAs/GaSb broken-gap QW exhibits unique features, i.e., nonlinear and oscillating behaviors which can be tuned by the external electric field. For the wide ASQW case (see Figs. 2 (a) and (b)), the internal electrostatic potential induced by charge transfer tends to weaken the coupling between the conduction subbands in InAs layer and the valence subbands in GaSb layer, i.e., the internal electric field compensates partly the external electric field, and shifts the anticrossing point to a smaller $k_{\parallel}$. In normal semiconductor phases of undoped InAs/GaSb QW, the charge transfer process doesn’t happen, therefore the internal electrostatic potential disappears. Figs. 3 (a) and (b) describe the RSS as function of the in-plane momentum and the thickness of the InAs layer in zero and finite electric field. From this contour plot one can see more clearly that the RSS shows a nonlinear feature for the narrow InAs layer, and an oscillating behavior at large thickness of InAs layer. The nonlinear behavior arises from the interface contribution which also depends on the interband couplingRSS . The oscillation of RSS is caused by the strong mixing between the conduction subband E1 and the heavy- hole subband HH1. Interestingly, this oscillating behavior of RSS can be enhanced by an electric field, e.g., $F=40~{}kV/cm$, (see Fig. 3 (b)). Note that there is a critical thickness of the InAs layer $L_{c}$, the RSS exhibits oscillating features corresponding to the different phases when $L>L_{c}$. Fig.4 (a) displays the phase diagram of InAs/GaSb QWs for different external electric fields. This figure indicates that the critical thickness of the InAs layer decreases as the thickness of GaSb layer increases, and tends to saturate at different thicknesses determined by the external electric fields (see Fig. 4 (a)). Positive external electric fields decrease the critical thickness, while negative external electric fields increase the critical thickness. Fig. 4 (b) gives the critical thickness of InAs layer and the critical electric field for different thicknesses of the GaSb layers. The critical electric fields saturate at different valyes for different thicknesses of the GaSb layers. Figure 4: (a) The phase diagram of InAs/GaSb QWs as function of the thickness of InAs and GaSb layers for different electric fields. (b) The Same as (a), but as function of the thickness of InAs layer and the external electric field for different thicknesses of the GaSb layers. In summary, we theoretically investigated Rashba spin-orbit interaction in InAs/GaSb asymmetric chemical multilayer QWs. We found a spontaneous RSS that arises from the interface contribution induced by the asymmetric structure of the QW. The RSS exhibits distinct behaviors, i.e., nonlinear and oscillating behavior, depending on the thickness of the QW and the external electric field. The oscillating RSS comes from the strong interband mixing between the lowest InAs conduction and the highest valence GaSb subbands. This crossover between two distinct behaviors can be tuned by the thicknesses of the InAs or GaSb layers and the external electric field. The unique features of RSS in InAs/GaSb QWs could provide us an interesting way to manipulate the electron spin and construct spintronic devices. ###### Acknowledgements. This work was supported by the NSFC Grant No. 60525405 and the knowledge innovation project from CAS, , City University of Hong Kong Strategic Research Grant (project no. 7002029). GQH was supported by FAPESP and CNPq (Brazil). ## References * (1) G. A. Sai-Halasz, R. Tsu, and L. Esaki, Appl. Phys. Lett. 30, 651 (1977); G. A. Sai-Halasz, L. Esaki, and W. A. Harrison, Phys. Rev. B 18, 2812 (1978). * (2) J. R. Södeström, D. H. Chow and T. C. McGill, Appl. Phys. Lett. 55, 1094 (1989); L. F. Luo, R. Beresford and W. I. Wang, Appl. Phys. Lett. 55, 2023 (1989). * (3) D. Z-Y. Ting, D. A. Collins, E. T. Yu, D. H. Chow and T. C. McGill, Appl. Phys. Lett. 57, 1257 (1990). * (4) M. P. Houng, Y. H. Wang, S. L. Shen, J. F. Chen and A. Y. Cho, Appl. Phys. Lett. 60, 713 (1992). * (5) J. Luo, H. Munekata, F. F. Fang, and P. J. Stiles, Phys. Rev. B 41, 7685 (1990). * (6) Y.-C. Chang and J. N. Schulman, Phys. Rev. B 31, 2069 (1985); J.-C. Chiang, S.-F. Tsay, Z. M. Chau, and I. Lo, Phy. Rev. Lett 77, 2053 (1996); S. de-Leon, L. D. Shvartsman, and B. Laikhtman, Phy. Rev. B 60, 1861 (1999); A. Zakharova, S. T. Yen, and K. A. Chao, Phy. Rev. B 66, 085312 (2002). * (7) M. J. Yang, C. H. Yang, B. R. Bennett, and B. V. Shanabrook, Phy. Rev. Lett. 78, 4613 (1997); M. Lakrimi, S. Khym, R. J. Nicholas, D. M. Symons, F. M. Peeters, N. J. Mason, and P. J. Walker, Phy. Rev. Lett. 79, 3034 (1997); L. J. Cooper, N. K. Patel, V. Drouot, E. H. Linfield, D. A. Ritchie, and M. Pepper, Phy. Rev. B 57, 11915 (1998); T. P. Marlow, L. J. Cooper, D. D. Arnone, N. K. Patel, D. M. Whittaker, E. H. Linfield, D. A. Ritchie, and M. Pepper, Phy. Rev. Lett. 82, 2362 (1999). * (8) A. Voskoboynikov, S. S. Lin, C. P. Lee, and O. Tretyak, J. Appl. Phys. 87, 387 (2000). * (9) T. Koga, J. Nitta, H. Takayanagi, and S. Datta, Phys. Rev. Lett. 88, 126601 (2002). * (10) D. Z. -Y. Ting and X. Cartoixa, Appl. Phys. Lett. 81, 4198 (2002). * (11) S. Datta and B. Das, Appl. Phys. Lett. 56, 665 (1990). * (12) S. Hallstein, J. D. Berger, M. Hilpert, H.C. Schneider, W. W. Rühle, F. Jahnke, S.W. Koch, H. M Gibbs, G. Khitrova, and M. Oestreich, Phys. Rev. B 56, 7076 (1997). * (13) M. G. Burt, J. Phys.: Condens. Matter 4, 6651 (1992); B. A. Foreman, Phys. Rev. B 56, R12748 (1997); T. Darnhofer and U. Rossler, ibid. 47, 16020 (1993). * (14) W. Yang and Kai Chang, Phys. Rev. B, 73, 113303 (2006); W. Yang and Kai Chang, ibid. 74, 193314 (2006). * (15) G. Bastard, E. E. Mendez, L. L. Chang, and L. Esaki, J. Vac. Sci. Technol. 21, 531 (1982). * (16) I. Vurgaftmana, J. R. Meyer and L. R. Ram-Mohan, J. Appl. Phys. 89, 5815 (2001). * (17) Y. Naveh and B. Laikhtman, Appl. Phys. Lett. 66, 713 (1995). * (18) H. B. de Carvalho, et al., Phys. Rev. B 74, 041305 (2006).

arxiv-papers

2024-09-04T02:48:54.616117

0803.4355

# Grammar-Based Random Walkers in Semantic Networks111Rodriguez, M.A., ”Grammar-Based Random Walkers in Semantic Networks”, Knowledge-Based Systems, volume 21, issue 7, pages 727-739, ISSN: 0950-7051, Elsevier, doi:10.1016/j.knosys.2008.03.030, LA-UR-06-7791, October 2008. Marko A. Rodriguez Digital Library Research and Prototyping Team Los Alamos National Laboratory Los Alamos, New Mexico 87545 ###### Abstract Semantic networks qualify the meaning of an edge relating any two vertices. Determining which vertices are most “central” in a semantic network is difficult because one relationship type may be deemed subjectively more important than another. For this reason, research into semantic network metrics has focused primarily on context-based rankings (i.e. user prescribed contexts). Moreover, many of the current semantic network metrics rank semantic associations (i.e. directed paths between two vertices) and not the vertices themselves. This article presents a framework for calculating semantically meaningful primary eigenvector-based metrics such as eigenvector centrality and PageRank in semantic networks using a modified version of the random walker model of Markov chain analysis. Random walkers, in the context of this article, are constrained by a grammar, where the grammar is a user defined data structure that determines the meaning of the final vertex ranking. The ideas in this article are presented within the context of the Resource Description Framework (RDF) of the Semantic Web initiative. ††preprint: LAUR-06-7791 ## I Introduction There exists a large collection of centrality metrics that have been used extensively to rank vertices in single-relational (or unlabeled) networks. Any metric for determining the centrality of a vertex in a single-relational network can be generally defined by the function $f:G\rightarrow\mathbb{R}^{|V|}$, where a single-relational network is denoted $G^{1}=(V=\\{i,\ldots,j\\},E\subseteq V\times V)$ and the range of $f$ is the rank vector representing the centrality value assigned to each vertex in $V$ 222The superscript $1$ on $G^{1}$ denotes that the network is a single- relational network as opposed to a semantic network which will be denoted as $G^{n}$.. The work in Wasserman and Faust (1994); Brandes and Erlebach (2005); Getoor and Diehl (2005) provide reviews of the many popular centrality measures that are currently used today to analyze single-relational networks. Of particular importance to this article are those metrics that use the primary eigenvector of the network to rank the vertices in $V$ (namely eigenvector centrality Bonacich (1987) and PageRank Page _et al._ (1998)). If $\mathbf{A}\in\mathbb{R}^{|V|\times|V|}$ is the adjacency matrix representation of $G^{1}$, then the primary eigenvector of $\mathbf{A}$ is $\mathbf{\pi}$ when $\mathbf{A}\mathbf{\pi}=\lambda\mathbf{\pi}$, where $\lambda$ is the greatest eigenvalue of all eigenvectors of $\mathbf{A}$ and $\mathbf{\pi}\in\mathbb{R}^{|V|}$ Trefethen and Bau (1997). The primary eigenvector has been applied extensively to ranking vertices in all types of networks such as social networks Bonacich (1987), scholarly networks of articles Chen _et al._ (2007) and journals Bollen _et al._ (2006), and technological networks such as the web citation network Page _et al._ (1998). In single-relational networks, determining the primary eigenvector of the network can be computed using the power method which simulates the behavior of a collection of random walkers traversing the network Brandes and Erlebach (2005). Those vertices that have a higher probability of being traversed by a random walker are the most “central” or “important” vertices. For aperiodic, strongly connected networks, $\mathbf{\pi}$ is the eigenvector centrality ranking Bonacich (1987). For networks that are not strongly connected or are periodic, the network’s topology can be altered such that a “teleportation” network can be overlaid with $G^{1}$ to produce an irreducible and aperiodic network for which the power method will yield a real valued $\mathbf{\pi}$. This is the method that was introduced by Brin and Page and is popularly known as the random web-surfer model of the PageRank algorithm Page _et al._ (1998). The PageRank algorithm is one of the primary reasons for the (subjectively) successful rankings of web pages from the Google search engine Langville and Meyer (2006). In a single-relational social network, for example, the network data structure can only represent a single type of relationship such as friendship. However, in a semantic network (or multi-relational network), the vertices can be connected to each other by a heterogeneous set of relationships such as friendship, kinship, collaboration, communication, etc. For a semantic network instance, there usually exists an ontology (or schema) which specifies how vertex types are related to one another. For example, an ontology may say that a vertex of type human can have another vertex of type human as a friend, but a human cannot have a vertex of type animal as a friend. An ontology is nearly analogous to the object-specifications of object-oriented programming minus the method declarations Rodriguez (2007a) and loosely related to the schema definitions of relational databases. The Resource Description Framework (RDF) is a popular data model for explicitly representing semantic networks for the distribution and use amongst computers Miller (1998); Klyne and Carroll (2004); Manola and Miller (2004). The Resource Description Framework Schema (RDFS) is a popular ontology language for RDF Brickley and Guha (2004). An RDF network can be represented as a triple list $G^{n}\subseteq(V\times\Omega\times V)$, where $\Omega$ is a set of edge labels denoting the semantic (or meaning) of the relationship between the vertices in $V$ and any ordered triple $\langle i,\omega,j\rangle\in G^{n}$ states that vertex $i$ is related to vertex $j$ by the semantic $\omega$. The use of labeled edges complicates the meaning of the rank vector returned by single-relational centrality measures because some vertices may be deemed more central than others with respect to one edge label, but not with respect to another. For example, the relationship isFriendOf may be considered more relevant than livesInSameCityAs. Therefore, due to the number of ways by which two adjacent vertices can be related and the focus on the semantics of such relations, the aim of recent semantic network metrics have been on ranking semantic associations Rada _et al._ (1989); Lin (2004); Sheth _et al._ (2005); Aleman-Meza _et al._ (2005), not the vertices themselves. A semantic association between vertices $i$ and $j$ is defined by the ordered multi-set path $q$, where $q=(i,\omega_{a},\ldots,\omega_{b},j)$, $i,j\in V$, and $\omega_{a},\omega_{b}\in\Omega$ Anyanwu and Sheth (2003). If $Q_{i,j}$ is the set of all possible semantic associations between vertices $i$ and $j$ in $G^{n}$, then a path metric function is generally defined as $f:Q_{V,V}\rightarrow\mathbb{R}^{|Q_{V,V}|}$, where the range of $f$ denotes the ranking of each path in $Q_{i,j}$. This article focuses on vertex ranking, not path ranking. Moreover, this article is primarily interested in eigenvector-based metrics such as eigenvector centrality Bonacich (1987) and PageRank Page _et al._ (1998). While eigenvector-based metrics on semantic networks have been proposed to rank vertices, the algorithms rely on prescribed semantic network ontologies and therefore, have not been generalized to handle any semantic network instance Zhuge and Zheng (2003); Mihalcea _et al._ (2004); Rodriguez (2007b). This article presents a method for applying eigenvector-based centrality metrics to semantic networks such that the semantic network’s ontology is respected. The proposed method extends the random walker model of Markov chain analysis Häggström (2002) to support its application to semantic network vertex ranking without altering the original data set or isolating subsets of the data set for analysis. This method is called the grammar-based random walker method. While the random walker’s of Markov chain analysis are memoryless, grammar-based random walkers of semantic networks utilize a user- defined grammar (or program) that instructs the grammar-based random walker to take particular ontological paths through the semantic network instance. Moreover, a grammar-based random walker maintains a memory of its path in the network and in the grammar in order for it to execute simple logic along its path. This simple logic allows the grammar-based random walker to generate semantically complex eigenvector rankings. For example, given a scholarly semantic network and the grammar-based method, it is possible to calculate $\mathbf{\pi}$ over all author vertices such that the authors indexed by $\mathbf{\pi}$ are located at some institution and they wrote an article that cites another article of a different author of the same institution. The next section provides an overview of the class of eigenvector-based metrics for single-relational networks that use the random walker model and then proposes a method for meaningfully applying such metrics to semantic networks. The result is a vertex valuing function generally defined as $f:G\times\Psi\rightarrow\mathbb{R}^{|\subseteq V|}$ where $\Psi$ is a user defined grammar and $\mathbf{\pi}\in\mathbb{R}^{|\subseteq V|}$. ## II Random Walkers in Single-Relational Networks The random walker model comes from the field of Markov chain analysis. Markov chains are used to model the dynamics of a stochastic system by explicitly representing the states of the system and the probability of transition between those states Mitrani (1998); Ching and Ng (2006). A Markov chain can be represented by a directed weighted network $G^{1}=(V,E,w)$ where the set of vertices in $V$ are system states, $E\subseteq V\times V$ are the set of directed edges representing the transitions between states, and $w:E\rightarrow[0,1]$ is the function that maps each edge to a real weight value that represents the state transition probability 333Note that while the weight function $w$ does in fact label edges in $E$, the meaning of the edges are homogenous and thus, $\omega$ simply denotes the extent to which the meaning is applied. Therefore, with respects to this article, a weighted Markov chain is considered a single-relational network, not a semantic network.. The outgoing edge weights of any state in the Markov chain form a probability distribution such that $\sum_{e\in\Gamma^{+}(i)}w(i)=1\;:\;|\Gamma^{+}(i)|\geq 1$, where $\Gamma^{+}(i)\subseteq E$ is the set of outgoing edges of vertex $i$. The future state of the system at time $n+1$ is based solely on the current state of the system at time $n$ and its respective outgoing edges. Given that a Markov chain can be represented by a weighted directed network, one can envision a random walker moving from vertex to vertex (i.e. state to state). A random walker moves through the Markov chain by choosing a new vertex according to the transition probabilities outgoing from its current vertex. This process continues indefinitely where the long run behavior, or stationary distribution denoted $\mathbf{\pi}$, of the random walker makes explicit the probability of the random walker being located at any one vertex at some random time in the future. However, only aperiodic, irreducible, and recurrent Markov chains can be used to generate a $\mathbf{\pi}$ that is the stationary distribution of the chain Brandes and Erlebach (2005). If the Markov chain is aperiodic then the random walker does not return to some previous vertex in a periodic manner. A Markov chain is considered recurrent and irreducible if there exists a path from any vertex to any other vertex. In the language of graph theory, the weighted directed network representing the Markov chain must be strongly connected. If $\mathbf{A}\in\mathbb{R}^{|V|\times|V|}$ is the weighted adjacency matrix representation of $G^{1}$ and there exists a vertex vector $\mathbf{\pi}\in\mathbb{R}^{|V|}$ where $\sum_{i\in V}\mathbf{\pi}_{i}=1$ and $\mathbf{A}\mathbf{\pi}=\lambda\mathbf{\pi}$, where $\lambda$ is the greatest eigenvalue of all eigenvectors of $\mathbf{A}$, then $\mathbf{\pi}$ is the stationary distribution of $G^{1}$ as well as the primary eigenvector of $\mathbf{A}$ Parzen (1962). The vector $\mathbf{\pi}$ represents the eigenvector centrality values for all vertices in $V$ Bonacich (1987). In the real world, periodicity is highly unlikely in most natural networks Brandes and Erlebach (2005). However, a strongly connected network is not always guaranteed. If the network is not strongly connected, then the problem of rank sinks and subset cycles is introduced and $\mathbf{\pi}$ is not a real valued vector. Therefore, many networks require some manipulation to ensure strong connectivity. For example, the web citation network, represented as $G^{1}=(V,E)$, is not strongly connected Broder _et al._ (2000) and therefore, in order to calculate $\mathbf{\pi}$ for the web citation network, it is necessary to transform $G^{1}$ into a strongly connected network. One such method was introduced in Brin and Page (1998); Page _et al._ (1998) where a probabilistic web citation network is overlaid with a fully connected web citation network. In matrix form, the probabilistic adjacency matrix of the web citation network, $\mathbf{A}\in\mathbb{R}^{|V|\times|V|}$, is created, where $\mathbf{A}_{i,j}=\begin{cases}\frac{1}{|\Gamma^{+}(i)|}&\text{if }(i,j)\in E\\\ \frac{1}{|V|}&\text{if }|\Gamma^{+}(i)|=0.\end{cases}$ In $\mathbf{A}$, all rank sinks (i.e. vertices with no out degree, absorbing vertices) connect to every other vertex in $V$ with equal probability. Next, the matrix $\mathbf{B}$ is created such that $\mathbf{B}\in\mathbb{R}^{|V|\times|V|}$ and $\mathbf{B}_{i,j}=\frac{1}{|V|}$ for all $i$ and $j$ in $V$. $\mathbf{B}$ denotes a fully connected network (i.e. a complete network) where every vertex is connected to every other vertex with equal probability. The composite adjacency matrix $\mathbf{C}=\delta\mathbf{A}+(1-\delta)\mathbf{B}$, where $\delta\in(0,1]$ is a parameter weighting the contribution of each adjacency matrix, guarantees that there is some finite probability that each vertex in $V$ is reachable by every other vertex in $V$. Therefore, the network denoted by $\mathbf{C}$ is strongly connected and there exists a unique stationary distribution $\mathbf{\pi}$ such that $\mathbf{C}\mathbf{\pi}=\lambda\mathbf{\pi}$. This method of inducing strong connectivity is called PageRank and has been used extensively to rank vertices in a unlabeled, single-relational networks Langville and Meyer (2006). The primary contribution of this article is that it ports the eigenvector- based algorithms of single-relational networks over to the semantic network domain. This article presents a method for calculating a semantically meaningful stationary distribution within some subset of a semantic network (called grammar-based eigenvector centrality) as well as how to implicitly induce strong connectivity irrespective of the network’s topology (called grammar-based PageRank). This general method is called the grammar-based random walker model because a random walker does not blindly move from vertex to vertex, but instead is constrained by a grammar that ensures that the stationary distribution is calculated in a “grammatically correct” subset of $G^{n}$. Before discussing the grammar-based random walker method, the next section provides a brief review of semantic networks, ontologies, and current standards for their representation. ## III Semantic Networks A semantic network is also known as a multi-relational network or directed labeled network. In a semantic network, there exists a heterogeneous set of vertex types and a heterogeneous set of edge types such that any two vertices in the network can be connected by zero or more edges. In order to make a distinction between two edges connecting the same vertices, a label denotes the meaning, or semantic, of the relationship. A semantic network can be represented by the triple list $G^{n}\subseteq(V\times\Omega\times V)$. A vertex to vertex relationship is called a triple because there exists the relationship $\langle i,\omega,j\rangle$ where $i\in V$ is called the subject, $\omega\in\Omega$ is called the predicate, and $j\in V$ is called the object. Perhaps the most popular standard for representing semantic networks is the Resource Description Framework (RDF) of the Semantic Web initiative Manola and Miller (2004); Klyne and Carroll (2004). There currently exists many applications to support the creation, query, and manipulation of RDF-based semantic networks. High-end, modern day triple-stores (RDF databases) can reasonably support on the order of $10^{9}$ triples Aasman (2006). For this reason, and due to the fact that RDF is becoming a common data model for various disciplines including digital libraries Bax (2004), bioinformatics Ruttenberg _et al._ (2007), and computer science Rodriguez and Bollen (2007), all of the constructs of the grammar-based random walker model will be presented according RDF and its ontology modeling language RDFS. RDF identifies vertices in a semantic network by Uniform Resource Identifiers (URI) Berners-Lee _et al._ (2005), literals, or blank nodes (also called anonymous nodes) and edge labels are represented by URIs. An example RDF triple where all components are URIs is $\langle\texttt{lanl:marko},\texttt{lanl:hasFriend},\texttt{lanl:johan}\rangle.$ In this triple, lanl is a namespace prefix that represents http://www.lanl.gov. This prefix convention is used throughout the article to ensure brevity of text and diagram clarity. Figure 1 is a graphic representation of the previous triple. Figure 1: A example triple in RDF. Another example of a triple where the object is a literal is $\langle\texttt{lanl:marko},\texttt{lanl:hasFirstName},\texttt{"Marko"${}^{\wedge}$${}^{\wedge}$xsd:string}\rangle.$ In this triple, the literal "Marko"∧∧xsd:string is an XML schema datatype string (xsd) Biron and Malhotra (2004). While a semantic network instance is represented in pure RDF, a semantic network ontology is represented in RDFS (a language represented in RDF). ### III.1 Ontologies Due the heterogeneous nature of the vertices and edges in a semantic network, an ontology is usually defined as way of specifying the range of possible interactions between the vertices in the network. Ontologies articulate the relation between abstract concepts and make no explicit reference to the instances of those classes Sowa (1987). For example, the ontology for the web citation network can be defined by a single class representing the abstract concept of a web page and the single semantic relationship representing a web link or citation (i.e. href). This simple ontology states that the network representing the semantic model of the web is constrained to only instances of one class (a web page) and one relationship (a web link). Given the previous single triple represented in Figure 1, the semantic network ontology could be represented as diagramed in Figure 2, where the lanl:hasFriend property must have a domain of lanl:Human and a range of lanl:Human, where lanl:marko and lanl:johan are both lanl:Humans. Figure 2: A example of the relationship between an ontology and its instance. Note that ontological diagrams can be abbreviated by assuming that the tail of an edge is the rdfs:domain and the head of the edge is the rdfs:range. This abbreviated form is diagrammed in Figure 3. Figure 3: An abbreviation of the diagramed in Figure 2. In general, the relationship between an ontology and its corresponding semantic network instantiation is depicted in Figure 4 where the rdf:type property denotes that the vertices in $V$ are an instance of some abstract class in the ontology. Figure 4: The relationship between a semantic network instance and its ontology. RDFS does not provide a large enough vocabulary to describe many of the types of relations needed for modeling class interactions Lacy (2005). For this reason, other modeling languages, based on RDFS, have been developed such as the Web Ontology Language (OWL) McGuinness and van Harmelen (2004); Lacy (2005). OWL allows a modeler to represent restrictions on properties (e.g. cardinality) and provides a broader range of property types (e.g. inverse relationships, functional relationships). Even though RDFS is limited in its expressiveness it will be used as the modeling language for describing the grammar-based random walker ontology. Note that it is trivial to map the presented concepts over to other modeling languages such as OWL. For a more in-depth review of ontology modeling languages, their history, and their application, please refer to Lacy (2005) and Gasevic _et al._ (2006). The next section brings together the concepts of random walkers, semantic networks, and ontologies in order to formalize this article’s proposed grammar-based random walker model. ## IV Grammar-Based Random Walkers A grammar-based random walker moves through a semantic network in a manner that respects the labels of the edges connecting the network’s vertices. The purpose of the grammar-based random walker is to identify the stationary distribution of some subset of the full semantic network (i.e. the primary eigenvector of a sub-network of the network). Unlike the random walkers of Markov chain analysis, a grammar-based random walker does not take any outgoing edge from its current vertex, but instead, depending on the user defined grammar, traverses particular edges types to particular vertex types. Any designed grammar uses the constructs and algorithms defined by the grammar ontology (prefixed as rwr). The grammar ontology defines rule classes, attribute classes, data structures, and properties that are intended to be combined with instances and classes of $G^{n}$ to create a $G^{n}$ specific grammar denoted $\Psi$. The rules of the grammar ultimately determine which vertices in $V$ are indexed by the returned rank vector $\mathbf{\pi}$. The rank vector $\mathbf{\pi}$ is created by a set of grammar-based random walkers $P$ traversing through $G^{n}$ and obeying $\Psi$. Figure 5 diagrams the relationship between $\Psi$, $P$, $G^{n}$, and their respective ontologies. Note that $\Psi$, $\Psi$’s ontology, $G^{n}$, and $G^{n}$’s ontology are all semantic networks and thus, can be represented by the same semantic network data structure. However, in order to make the separation between the components clear, each data structure will be discussed as a separate semantic network. Figure 5: The grammar-based random walker architecture. The meaning of the vertex rank vector $\mathbf{\pi}$ of the grammar-based model, both semantically and theoretically, depends primarily on the grammar used. Some $\Psi$s will generate a $\mathbf{\pi}$ that is the stationary probability distribution of some subset of $G^{n}$, while others will be more representative of a discrete form of the spreading activation models, where calculating the long run behavior of the random walker is undesirable Cohen and Kjeldsen (1987); Savoy (1992); Crestani (1997); Crestani and Lee (2000). In practice, determining whether $\mathbf{\pi}$ is a stationary distribution of the analyzed subset of $V$ is a matter of determining whether the subset of $G^{n}$ that is traversed by $P$ is strongly connected and the normalized $\mathbf{\pi}$ has converged to a stable set of values. Any grammar-based random walker implementation is a function generally defined as $f:G\times\Psi\rightarrow\mathbb{R}^{|\subseteq V|}$. It is noted that there exists two related ontologies for modeling the distribution of discrete entities in a semantic network. These ontologies were inspirational to the ideas presented in this article. The marker passing Petri net ontology of Gasevic and Devedzic (2006) and the particle swarm ontology of Rodriguez (2007b). However, both ontologies were designed for a different application space. The first is for Petri net algorithms while the latter was defined specifically for collective decision making systems. Finally, the grammar-based model presented in Rodriguez and Watkins (2007) for calculating geodesics in a semantic network combined with the grammar-based model presented in this article form a unified framework for porting many of the popular single-relational network analysis algorithms over to the semantic network domain (more specifically, the RDF and Semantic Web domain). ### IV.1 The Grammar-Based Random Walker Ontology The complete grammar ontology is graphically represented in Figure 6, where squares are rdfs:Classes and edge labels are rdf:Property types. The tail of each edge is the rdfs:domain of the rdf:Property and the head is the rdfs:range. For the purpose of diagram clarity, the dashed edges denote a relationship of rdfs:subClassOf. Finally, note that the two dashed squares should be instances or classes that are in $G^{n}$ or its ontology, respectively. Figure 6: The complete grammar-based random walker ontology. The grammar ontology follows a convention similar to most object oriented programming languages Sebesta (2005) in that a rwr:Context (i.e. class) has a set of attributes (i.e. fields) and rules (i.e. methods). The general idea is that any grammar instance $\Psi$ is a collection of rwr:Context objects connected to one another by rwr:Traverse rules. rwr:Contexts and their rwr:Traverse rules are an abstract model of what triples a grammar-based random walker can traverse in $G^{n}$. The rwr:Is and rwr:Not attributes further constrain the types of vertices that can be traversed by the random walker and are used for path “bookkeeping” and path logic. The rwr:IncrCount and rwr:SubmitCounts rules determine which vertices in $V$ should be indexed by $\mathbf{\pi}$. Finally, the rwr:Reresolve rule is the means by which the random walker is able to “teleport” to other regions of $G^{n}$. The rwr:Reresolve rule is used to model the PageRank algorithm and therefore, is a mechanism for guaranteeing that the subset of $G^{n}$ that is traversed is strongly connected and $\mathbf{\pi}$ is a stationary distribution. ### IV.2 High-Level Overview of the Grammar-Based Model This section will provide a high-level overview of the components of the grammar diagrammed in Figure 6. $\Psi$ is a user defined data structure that is created specifically for $G^{n}$ and $G^{n}$’s respective ontology. Any $\Psi$ must obey the constraints defined by the grammar ontology diagrammed in Figure 6. A single grammar-based random walker (denoted $p\in P$) “walks” both $G^{n}$ and $\Psi$ in order to dynamically generate a vertex rank vector denoted $\mathbf{\pi}$. If the $p$-traversed subset of $G^{n}$ is strongly connected, then only a single random walker is needed to compute $\mathbf{\pi}$ Häggström (2002). When random walker $p\in P$ is at some rwr:Context in $\Psi$, the rwr:Context is “resolved” to a particular vertex in $V$. This is the relationship between $\Psi$ and $G^{n}$. For example, if $p$ is at some rwr:Context in $\Psi$ that is rwr:forResource lanl:Human, then $p$ must also be at some vertex in $V$ that is of rdf:type lanl:Human. Thus, $\Psi$ is an abstract representation of the legal vertices that $p$ can traverse in $V$. When $p$ is at a rwr:Context, $p$ will execute the rwr:Context’s collection of rwr:Rules, while at the same time respecting rwr:Context rwr:Attributes. The collection of rwr:Rules is an ordered rdf:Seq Brickley and Guha (2004). This means that $p$ must execute the rules in their specified sequence. This is represented as the set of properties rdf:_1, rdf:_2, rdf:_3, etc. (i.e. rdfs:subPropertyOf rdfs:ContainerMembershipProperty). Any grammar-based random walker $p$ has three local variables: * • a reference to its path history in $G^{n}$ (denoted $g^{p}$) * • a reference to its path history in $\Psi$ (denoted $\psi^{p}$) * • a local vertex vector (denoted $\mathbf{\pi}^{p}\in\mathbb{N}^{|\subseteq V|}$) and a reference to a single global variable: * • a global vertex vector (denoted $\mathbf{\pi}\in\mathbb{N}^{|\subseteq V|}$) The path history $g^{p}$ is an ordered multi-set of vertices, edge labels, and edge directionalities. If the random walker $p$ traversed the path diagrammed in Figure 1 from left to right, then $g^{p}=\\{\texttt{lanl:marko},\texttt{lanl:hasFriend},+,\texttt{lanl:johan}\\}$. Note that $g^{p}_{0}=\texttt{lanl:marko}$, $g^{p}_{1^{\prime}}=\texttt{lanl:hasFriend}$, $g^{p}_{1^{\prime\prime}}=+$, and $g^{p}_{1}=\texttt{lanl:johan}$, where $n^{\prime}$ denotes the edge label used to get to the vertex at time $n$ and $n^{\prime\prime}$ denotes the direction that $p$ traversed over that edge. In the grammar-based random walker model, a random walker can, if stated in $\Psi$, oppose an edge’s directionality. For example, if $p$ had traversed the edge diagrammed in Figure 1 from right to left, then $g^{p}=\\{\texttt{lanl:johan},\texttt{lanl:hasFriend},-,\texttt{lanl:marko}\\}$. A similar convention holds for $p$’s $\Psi$-history $\psi^{p}$. However, in $\psi^{p}$ the vertices are rwr:Contexts, the edge labels are the rdf:Property of the rwr:Edge chosen, and the directionalities are determined by whether an rwr:OutEdge or rwr:InEdge was traversed. The “walking” aspect of $p$ for both $\Psi$ and $G^{n}$ is governed by the rwr:Traverse rule. When $p$ executes a rwr:Traverse rule in $\Psi$, it selects a particular rwr:Edge to traverse. For rwr:OutEdges, a triple in $G^{n}$ is selected with the subject being its current location $g^{p}_{n}$, and predicate and objects are instances of the respective resource specified by the rwr:OutEdge (rwr:hasPredicate and rwr:hasObject). For rwr:InEdges, a triple in $G^{n}$ is selected where $g^{p}_{n}$ is the object of the triple and the subject and predicate are instances of the resource specified by the rwr:InEdge (rwr:hasPredicate and rwr:hasSubject). The rwr:Context chosen is $\psi^{p}_{n+1}$ and the rdfs:Resource of the triple $\langle\psi^{p}_{n+1},\texttt{rwr:forResource},?x\rangle\in\Psi$ determines $g^{p}_{n+1}$, where $?x$ is any class in $G^{n}$’s ontology or instance in $G^{n}$’s vertex set $V$. The newly chosen $g^{p}_{n+1}$ is called the resolution of $\psi^{p}_{n+1}$. The rwr:IncrCount and rwr:SubmitCounts rules effect the random walker’s local vertex vector $\mathbf{\pi}^{p}$ and the global vertex vector $\mathbf{\pi}$, respectively. The distinction between $\mathbf{\pi}^{p}$ and $\mathbf{\pi}$ is that $\mathbf{\pi}^{p}$ is a temporary counter that is not submitted to the global counter $\mathbf{\pi}$ until the rwr:SubmitCounts rule has been executed. The walker $p$ does not submit its vertex counts until it has determined that it is in a $\Psi$-correct subset of $G^{n}$. The process of moving $p$ through a semantic network and allowing it to increment a counter for specific vertices continues until the ratio between the values of the global $\mathbf{\pi}$ converge. Note that $\mathbf{\pi}$ does not provide a probability distribution, $\sum_{i\in\mathbf{\pi}}\mathbf{\pi}_{i}\neq 1$. Instead, $\mathbf{\pi}$ represents the number of times an indexed vertex of $\mathbf{\pi}$ has been counted by a grammar-based random walker. Therefore, to determine the probability of being at any one vertex that is indexed by $\mathbf{\pi}$, $\mathbf{\pi}$ can be normalized to generate a new vector denoted $\mathbf{\pi}^{\prime}\in\mathbb{R}^{|\subseteq V|}$, where $\mathbf{\pi}^{\prime}_{i}=\frac{\mathbf{\pi}_{i}}{\sum_{j\in\mathbf{\pi}}\mathbf{\pi}_{j}}$. If $\mathbf{\pi}^{\prime}$ is the normalization of $\mathbf{\pi}$ then, when $\mathbf{\pi}^{\prime}$ no longer changes with successive executions of the rwr:SubmitCounts rule, the process is complete. More formally, if $\epsilon\in\mathbb{R}$ is an argument specifying the smallest change accepted for convergence consideration, then the grammar-based random walker algorithm is complete when ${||\mathbf{\pi}^{\prime}_{{}_{n}}-\mathbf{\pi}^{\prime}_{{}_{m}}||}_{2}<\epsilon$, where $n$ and $m$ are the time steps of consecutive calls to rwr:SubmitCounts. However, like Markov chains, this convergence will only occur if the subset of $G^{n}$ that is traversed is strongly connected and aperiodic. If the traversed subset of $G^{n}$ is not strongly connected or is periodic, then the rwr:Reresolve rule can be used to simulate grammar-based random walker “teleportation”. With the inclusion of the rwr:Reresolve rule, a grammar-based PageRank can be executed on $G^{n}$. The next section will formalized each of the rwr:Rules and rwr:Attributes of the grammar ontology. ## V The Rules and Attributes of the Grammar Ontology The following rwr:Rules and rwr:Attributes are presented in a set theoretic form that borrows much of its structure from semantic query languages such as SPARQL Prud’hommeaux and Seaborne (2004). The query triple $\langle?x,\texttt{rdf:type},\texttt{lanl:Author}\rangle\in G^{n}$ will bind $?x$ to any lanl:Author in the semantic network $G^{n}$. The $?x$ notation represents that $?x$ is a variable that is bound to any vertex (i.e. URI) that matches the query pattern. The same query can return many resources that bind to $?x$. In such cases, the results are returned as a set. Thus $X=\\{?x\;|\;\langle?x,\texttt{rdf:type},\texttt{lanl:Author}\rangle\in G^{n}\\}$ denotes the set of all vertices in $V$ that are of rdf:type lanl:Author. The following subsections present each of the rwr:Rules and rwr:Attributes that a grammar-based random walker must execute and respect during its journey through both $\Psi$ and $G^{n}$. ### V.1 Entering $\Psi$ and $G^{n}$ Every random walker “walks” both $\Psi$ and $G^{n}$ in parallel. However, before a walker can walk either data structure, it must enter both $\Psi$ and $G^{n}$. The entry points of $\Psi$ are rwr:EntryContexts and are represented by the set $s(\Phi)$, where $\displaystyle s(\Phi)=$ $\displaystyle\\{?x\;|\;\langle?x,\texttt{rdf:type},\texttt{rwr:EntryContext}\rangle\in\Psi\\}.$ The starting location $\phi\in s(\Phi)$ of $p$ is chosen with probability $\frac{1}{|s(\Phi)|}$. Once some $\phi$ is chosen, $\psi^{p}_{0}=\phi$ (time $n$ starts at $0$). An entry location into $V$ can be determined by randomly selecting some vertex $i\in s(V\;|\;\phi)$, where $s(V\;|\;\phi)$ is the set of all $i\in V$ given that $i$ is a proper resolution of the rwr:EntryContext $\phi$. Thus, $\displaystyle s(V\;|\;\phi)=$ $\displaystyle\\{?i\;|\;\langle\phi,\texttt{rwr:forResource},?z\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?i,\texttt{rdf:type},?z\rangle\in\Psi\;\vee\;?i=?z)\\},$ where type inheritance is strictly followed. For instance, if $i$ is an rdf:type of $z$ then $i$ is an instance of $z$ or an instance of $u$ where $u$ is a rdfs:subClassOf $z$. This is subsumption in RDFS reasoning and will be used repeatedly throughout the remainder of this article. Given the set $s(V\;|\;\phi)$, the probability of $p$ choosing some $i\in s(V\;|\;\phi)$ is $\frac{1}{|s(V\;|\;\phi)|}$. The chosen vertex $i$ becomes the starting location of $p$ in $G^{n}$ and thus, $g^{p}_{0}=i$. Note that $g^{p}_{0^{\prime}}=\emptyset$, $\psi^{p}_{0^{\prime}}=\emptyset$, $g^{p}_{0^{\prime\prime}}=\emptyset$ and $\psi^{p}_{0^{\prime\prime}}=\emptyset$ since a random walker enters both $\Psi$ and $G^{n}$ at a vertex without using an intervening edge label or directionality. Figure 7 depicts how rwr:EntryContexts in $\Psi$ are related to vertices in $G^{n}$. Figure 7: The relationship between rwr:EntryContexts in $\Psi$, $G^{n}$, and $G^{n}$’s ontology. ### V.2 The rwr:Not Attribute Before presenting the rwr:Traverse rule, it is important to discuss the two attributes that constrain the rwr:Traverse rule: namely, rwr:Not and rwr:Is. This subsection will discuss the rwr:Not attribute. The next section will discuss the rwr:Is attribute. The rwr:Not atttribute ensures that the random walker $p$ does not traverse an edge to a particular, previously seen vertex in $g^{p}$. Any rwr:Not attribute is the subject of a triple with a predicate rwr:steps and literal $m\in\mathbb{N}$. The literal $m$ denotes which vertex from $m$-steps ago $p$ must avoid. In other words, $p$ must not have a $g^{p}_{n+1}$ that equals $g^{p}_{n-m}$. Thus, the rwr:Context $\psi^{p}_{n+1}$ cannot resolve to $g^{p}_{n-m}$. If $\displaystyle M=$ $\displaystyle\\{?m\;|\;\langle\psi^{p}_{n+1},\texttt{rwr:hasAttributes},?x\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?x,\texttt{rwr:hasAttribute},?y\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{rwr:Not}\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rwr:steps},?m\rangle\in\Psi\\},$ then $\displaystyle X(p)_{n+1}=\bigcup_{m\in M}g^{p}_{n-m},$ where $X(p)_{n+1}\subseteq V$ and $X(p)_{n+1}\cap g^{p}_{n+1}=\emptyset$. The set $X(p)_{n+1}$ is the set of vertices in $V$ that $g^{p}_{n+1}$ must not equal. The rwr:Not attribute is useful when $p$ must not return to a vertex in $V$ that has been previously visited. Imagine that $p$ is determining whether or not a particular article has at least two authors (or must traverse an implicit coauthorship network). Such an example is depicted in Figure 8, where the numbered circles are the location of $p$ at particular time steps and author vertices are only connected to their authored articles. If, at $n=1$, $p$ is located at lanl:marko then $p$ will traverse the lanl:wrote predicate to the lanl:DDD article. If $p$ is checking for another author that is not lanl:marko then $p$ can only take the lanl:wrote predicate to lanl:dsteinbock. If lanl:DDD only had one author, then $p$ would be stuck (i.e. halt) at lanl:DDD since no legal lanl:wrote predicate could be traversed. At which point, it is apparent that the article has only one author. Moreover, by traversing to lanl:dstreinbock and not back to lanl:marko at $n=3$, a coauthorship network is implicitly traversed. Figure 8: An example situation for the rwr:Not attribute ### V.3 The rwr:Is Attribute Unlike the rwr:Not attribute, the rwr:Is atttribute is used to ensure that the random walker $p$ does, in fact, traverse an edge to a previously visited vertex in $V$. Any rwr:Is attribute is the subject of a triple with a predicate rwr:steps and literal $m\in\mathbb{N}$. The literal $m$ denotes which vertex from $m$-steps ago $p$ must traverse to. If this set of vertices returned by the rwr:Is attribute is greater than $1$, then $p$ must traverses to one of the vertices from the set. Thus, the random walker $p$ must have vertex $g^{p}_{n+1}$ equal some $g^{p}_{n-m}$. In other words, the rwr:Context $\psi^{p}_{n+1}$ must resolve to some $g^{p}_{n-m}$. If $\displaystyle M=$ $\displaystyle\\{?m\;|\;\langle\psi^{p}_{n+1},\texttt{rwr:hasAttributes},?x\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?x,\texttt{rwr:hasAttribute},?y\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{rwr:Is}\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rwr:steps},?m\rangle\in\Psi\\},$ then $\displaystyle O(p)_{n+1}=\bigcup_{m\in M}g^{p}_{n-m}\;,$ where $O(p)_{n+1}\subseteq V$ and $g^{p}_{n+1}\in O(p)_{n+1}$. Again, unless $O(p)_{n+1}=\emptyset$, one of the vertices in $O(p)_{n+1}$ must be $p$’s location in $G^{n}$ at $n+1$. The rwr:Is attribute is useful when $p$ must search particular properties of a vertex and later return to the original vertex. For instance, imagine the triple $\langle\texttt{lanl:LANL},\texttt{rdf:type},\texttt{lanl:Laboratory}\rangle\in G^{n}$ as depicted in Figure 9, where the numbered circles represent the $p$’s location at particular time steps $n$. Assume that $p$ is at the lanl:LANL vertex at $n=1$ and $p$ must check to determine if lanl:LANL is, in fact, a lanl:Laboratory. In order to do so, $p$ must traverse the rdf:type predicate to arrive at lanl:Laboratory at $n=2$. At $n=3$, $p$ should return to the original lanl:LANL vertex. Without the rwr:Is attribute, $p$ has the potential for choosing some other lanl:Laboratory, such as lanl:PNNL. Once back at lanl:LANL, it is apparent that lanl:LANL is a lanl:Laboratory and $p$ can move to some other vertex at $n=4$. Figure 9: An example situation for the rwr:Is attribute ### V.4 The rwr:Traversal Rule The rwr:Travere rule allows the random walker $p$ to traverse to a new rwr:Context in $\Psi$ and a new vertex in $V$. If there exists some rwr:Context $\phi$ with the rwr:Traverse rule $t$, then when $g^{p}_{n}=a$ and $\psi^{p}_{n}=\phi$, the probability of $p$ traversing some outgoing triple from $a$ or some incoming triple to $a$ is $\frac{1}{|\Gamma(a,p)|}$, where if $\displaystyle Y_{\text{out}}=$ $\displaystyle\\{?y\;|\;\langle t,\texttt{rdfs:hasEdge},?y\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{rwr:OutEdge}\rangle\in\Psi\\},$ $\displaystyle Y_{\text{in}}=$ $\displaystyle\\{?y\;|\;\langle t,\texttt{rdfs:hasEdge},?y\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{rwr:InEdge}\rangle\in\Psi\\},$ $\displaystyle\Gamma^{+}(a,p)=\bigcup_{y\in Y_{\text{out}}}$ $\displaystyle\\{\langle a,?\omega,?b\rangle\;|\;\langle a,?\omega,?b\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle y,\texttt{rwr:hasPredicate},?w\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?\omega,\texttt{rdfs:subPropertyOf},?w\rangle\in G^{n}$ $\displaystyle\;\;\;\;\;\;\vee\;?\omega=?w)$ $\displaystyle\;\wedge\;\langle y,\texttt{rwr:hasObject},?x\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?x,\texttt{rdf:forResource},?z\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?b,\texttt{rdf:type},?z\rangle\in G^{n}\;\vee\;?b=?z)$ $\displaystyle\;\wedge\;(O(p)_{n+1}=\emptyset\;\vee\;?b\in O(p)_{n+1})$ $\displaystyle\;\;\wedge\;?b\not\in X(p)_{n+1}\\},$ $\displaystyle\Gamma^{-}(a,p)=\bigcup_{y\in Y_{\text{in}}}$ $\displaystyle\\{\langle?b,?\omega,a\rangle\;|\;\langle?b,?\omega,a\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle y,\texttt{rwr:hasPredicate},?w\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?\omega,\texttt{rdfs:subPropertyOf},?w\rangle\in G^{n}$ $\displaystyle\;\;\;\;\;\;\vee\;?\omega=?w)$ $\displaystyle\;\wedge\;\langle y,\texttt{rwr:hasSubject},?x\rangle\in\Psi$ $\displaystyle\;\wedge\;\langle?x,\texttt{rdf:forResource},?z\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?b,\texttt{rdf:type},?z\rangle\in G^{n}\;\vee\;?b=?z)$ $\displaystyle\;\wedge\;(O(p)_{n+1}=\emptyset\;\vee\;?b\in O(p)_{n+1})$ $\displaystyle\;\;\wedge\;?b\not\in X(p)_{n+1}\\},$ then $\Gamma(a,p)=\Gamma^{+}(a,p)\cup\Gamma^{-}(a,p).$ At the completion of the traversal, $g^{p}_{n+1}=b$, $g^{p}_{n+1^{\prime}}=\omega$, $\psi^{p}_{n+1}=x$, and $\psi^{p}_{n+1^{\prime}}=w$. If the edge was chosen from $\Gamma^{+}(a,p)$ then $g^{p}_{n+1^{\prime\prime}}=+$ and $\psi^{p}_{n+1^{\prime\prime}}=+$. If the edge was chosen from $\Gamma^{-}(a,p)$ then $g^{p}_{n+1^{\prime\prime}}=-$ and $\psi^{p}_{n+1^{\prime\prime}}=-$. It is always the case that $\forall n:\psi^{p}_{n^{\prime\prime}}=g^{p}_{n^{\prime\prime}}$. Note the relationship between $G^{n}$ and $\Psi$ in the definition of both $\Gamma^{-}(a,p)$ and $\Gamma^{+}(a,p)$. It is necessary that the rwr:hasPredicate $?w$ and the rwr:forResource $?z$ as defined in $\Psi$ also exist in $G^{n}$. It is through the rwr:Traverse rule that the relationship between $\Psi$ and $G^{n}$ is made explicit and demonstrates how $\Psi$ constrains the path that $p$ can traverse in $G^{n}$. Figure 10 depicts an example of a traversal. In Figure 10, $\Gamma_{\psi}^{-}(a,p)=\\{\langle j,\omega,a\rangle\\}$ and $\Gamma_{\psi}^{+}(a,p)=\\{\langle a,\omega,e\rangle,\langle a,\omega,f\rangle\\}$, where $\Gamma_{\psi}(i)=\\{\langle j,\omega,a\rangle,\langle a,\omega,e\rangle,\langle a,\omega,f\rangle\\}$, and any one triple is selected with $\frac{1}{3}$ probability. Figure 10: An example of the set of edges allowed for traversal by $p$ when $g^{p}_{n}=a$. ### V.5 The rwr:IncrCount and rwr:SubmitCounts Rules The purpose of the rwr:IncrCount and rwr:SubmitCounts rules is to increment the local vertex rank vector $\mathbf{\pi}^{p}$ and global vertex rank vector $\mathbf{\pi}$, respectively. While $\mathbf{\pi}^{p}$ is a local variable of $p$, only $\mathbf{\pi}$ is returned at the completion of the grammar-based random walker algorithm. The reason for $\mathbf{\pi}^{p}$ is to ensure that prior to incrementing $\mathbf{\pi}$, the vertices indexed by $\mathbf{\pi}^{p}$ are in a grammatically correct region of $G^{n}$ as determined by the grammar $\Psi$. For example, if $p$ is to index a particular lanl:Human, it will do so in $\mathbf{\pi}^{p}$. However, before that lanl:Human is considered legal according to $\Psi$, $p$ may have to check to see if the lanl:Human is lanl:locatedAt the same lanl:University of some previously encountered lanl:Human. Thus, when $p$ has submitted its $\mathbf{\pi}^{p}$ to $\mathbf{\pi}$, it will have guaranteed that all the appropriate aspects of its incremented vertices in $\mathbf{\pi}^{p}$ have been validated by $\Psi$. This concept will be made more salient in the example to follow in the next section. Formally, if $\langle\phi,\texttt{rdf:type},\texttt{rwr:Context}\rangle\in\Psi$, $\psi^{p}_{n}=\phi$, $g^{p}_{n}=i$, and $\phi$ has the rwr:IncrCount rule, then ${\mathbf{\pi}^{p}_{i}}_{(n+1)}={\mathbf{\pi}^{p}_{i}}_{(n)}+1.$ Next, if $g^{p}=i$, $\psi^{p}=\phi$, $\langle\phi,\texttt{rdf:type},\texttt{rwr:Context}\rangle\in\Psi$, and $\phi$ has the rwr:SubmitCounts rule, then ${\mathbf{\pi}_{i}}_{(n+1)}={\mathbf{\pi}_{i}}_{(n)}+{\mathbf{\pi}^{p}_{i}}_{(n)}\;:\;\forall i\in\mathbf{\pi}^{p}$ and ${\mathbf{\pi}^{p}_{i}}_{(n+1)}=0\;:\;\forall i\in\mathbf{\pi}^{p}.$ As stated above, once $\mathbf{\pi}^{p}$ has been submitted to $\mathbf{\pi}$, the values of $\mathbf{\pi}^{p}$ are set to $0$. ### V.6 The rwr:Reresolve Rule The rwr:Reresolve rule is a way to “teleport” the random walker to some random vertex in $V$ and is perhaps the most complicated rule of the grammar-based random walker ontology. If there exists the rwr:Context $\phi$, $\psi^{p}_{n}=\phi$, $\phi$ has the rwr:Reresolve rule $u$, $\langle u,\texttt{rwr:probability},?d\rangle\in\Psi$, and $\langle u,\texttt{rwr:steps},?m\rangle\in\Psi$, then $p$ will have a $(d\cdot 100)$% chance of re-resolving its path from $m$ steps ago to the current step $n$, where $d=0.15$ in most PageRank implementations. If the random walker re- resolves, then the path from $g^{p}_{n-m}$ to $g^{p}_{n}$ is recalculated. In other words, a new path in $G^{n}$ is determined with respects to the rwr:Contexts $\psi^{p}_{n-m}$ to $\psi^{p}_{n}$ such that no rules are executed and only those attributes specified by the rwr:obeys property are respected. For example, suppose $\psi^{p}_{(n-m)\rightarrow n}=(\phi_{(n-m)},\omega_{(n-m)+1^{\prime}},\pm_{(n-m)+1^{\prime\prime}},\ldots,\omega_{n^{\prime}},\pm_{n^{\prime\prime}},\phi_{n})$ and context $\psi^{p}_{n}$ has a rwr:Reresolve rule, where $\psi^{p}_{n}=\phi_{n}$. If the rwr:Reresolve rule rwr:obeys both the rwr:Is and rwr:Not attributes, then the grammar-based random walker $p$ will re- resolve its history in $G^{n}$. Thus it will recalculate $g^{p}_{n-m}$ to $g^{p}_{n}$. The set of legal re-resolved paths from $n-m$ steps ago to $n$ is denoted $Q_{(n-m),n}$. Given that the probability $d$ is met, $\displaystyle Q_{(n-m),n}=$ $\displaystyle\\{(?i,?\omega_{(n-m)+1^{\prime}},\pm_{(n-m)+1^{\prime\prime}},?a,\ldots,$ $\displaystyle\;\;\;\;\;\;\;\;\;\;?b,?\omega_{n^{\prime}},\pm_{n^{\prime\prime}},?j)\;|$ $\displaystyle\;\;\;\;\;\;\langle\psi^{p}_{(n-m)},\texttt{rwr:forResource},?x\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?i,\texttt{rdf:type},?x\rangle\in G^{n}\;\vee\;?i=?x)$ $\displaystyle\;\wedge\;(O(p)_{(n-m)}=\emptyset\;\vee\;?i\in O(p)_{(n-m)})$ $\displaystyle\;\;\wedge\;?i\not\in X(p)_{(n-m)}$ $\displaystyle\;\wedge\;(\langle?\omega_{n^{\prime}},\texttt{rdfs:subPropertyOf},\psi^{p}_{n^{\prime}}\rangle\in G^{n}\;$ $\displaystyle\;\;\;\;\;\;\;\vee\;\;?\omega_{(n-m)+1^{\prime}}=\psi^{p}_{(n-m)+1^{\prime}})$ $\displaystyle\;\wedge\;((\pm_{(n-m)+1^{\prime\prime}}=+$ $\displaystyle\;\;\;\;\;\;\;\;\;\wedge\;(?i,?\omega_{(n-m)+1^{\prime}},?a)\in G^{n})$ $\displaystyle\;\;\;\;\;\;\vee\;(\pm_{(n-m)+1^{\prime\prime}}=-$ $\displaystyle\;\;\;\;\;\;\;\;\;\wedge\;(?a,?\omega_{(n-m)+1^{\prime}},?i)\in G^{n}))$ $\displaystyle\;\wedge\;\ldots$ $\displaystyle\;\wedge\;((\pm_{n^{\prime\prime}}=+\;\wedge\;(?b,?\omega_{n^{\prime}},?j)\in G^{n})$ $\displaystyle\;\;\;\;\;\;\vee\;(\pm_{n^{\prime\prime}}=-\;\wedge\;(?j,?\omega_{n^{\prime}},?b)\in G^{n}))$ $\displaystyle\;\wedge\;(\langle?\omega_{n^{\prime}},\texttt{rdfs:subPropertyOf},\psi^{p}_{n^{\prime}}\rangle\in G^{n}\;$ $\displaystyle\;\;\;\;\;\;\;\vee\;?\omega_{n^{\prime}}=\psi^{p}_{n^{\prime}})$ $\displaystyle\;\wedge\;\langle\psi^{p}_{n},\texttt{rwr:forResource},?y\rangle\in\Psi$ $\displaystyle\;\wedge\;(\langle?j,\texttt{rdf:type},?y\rangle\in G^{n}\;\vee\;?j=?y)$ $\displaystyle\;\wedge\;(O(p)_{n}=\emptyset\;\vee\;?j\in O(p)_{n})$ $\displaystyle\;\;\wedge\;?j\not\in X(p)_{n}\\}.$ The probability of $p$ choosing some re-resolved path $q\in Q_{(n-m),n}$ is $\frac{1}{Q_{(n-m),n}}$, where $g^{p}_{k^{\prime\prime}}=q_{k^{\prime\prime}}$, $g^{p}_{k^{\prime}}=q_{k^{\prime}}$, and $g^{p}_{k}=q_{k}$ for all $k$ such that $m\leq k\leq n$. While the above equation is perhaps notationally tricky, it has a relatively simple meaning. In short, $p$ must recalculate (or re-resolve) its path from $m$ step ago to the present step $n$. This recalculation must follow the exact same grammar path denoted in $\psi^{p}$. Thus, if from $m$ to $n$, $p$ had ensured that its current vertex is a lanl:Human that is lanl:locatedAt lanl:Laboratory then when $p$ “teleports”, the new vertex at $n$ will be guaranteed to also be a lanl:Human that is lanl:locatedAt a lanl:Laboratory. If there are no rank sinks, this rule guarantees a strongly connected network; any vertex can be reached by any other vertex in the grammatically correct region of $G^{n}$. However, note that rank sinks are remedied by the next rule. ### V.7 The Empty Rule Random walker halting occurs when $p$ arrives at some rwr:Context where no rule exists or there are no more rules to execute (e.g. when a rwr:Traverse rule does not provide any transition edges – $\Gamma(a,p)=\emptyset$). At halt points, a new random walker with an empty $\mathbf{\pi}^{p}$ and no $G^{n}$ or $\Psi$ history (i.e. $|g^{p}|=0$ and $|\psi^{p}|=0$), enters $G^{n}$ at some rwr:EntryContext $\phi$ in $\Psi$ and some $i\in s(V\;|\;\phi)$. The new random walker executes the grammar. Note that the global rank vector $\mathbf{\pi}$ remains unchanged. The combination of the empty rule and the rwr:Reresolve rule are necessary to ensure that $\mathbf{\pi}$ is a stationary distribution. Both rules are used in conjunction to support grammar-based PageRank calculations. In order to demonstrate the aforementioned ideas, the next section presents a particular grammar instance developed for a scholarly network ontology and instance. ## VI A Scholarly Network Example This section will demonstrate the application of grammar-based random walkers to a scholarly semantic network denoted $G^{n}$. Figure 11 diagrams the ontology of $G^{n}$ where the tail of the edge is the rdfs:domain and the head of the edge is the rdfs:range. The dashed lines represent the rdfs:subClassOf relationship. This ontology represents the relationships between lanl:Institutions, lanl:Researchers, lanl:Articles, and their respective children classes. Figure 11: An example scholarly ontology The first example calculates the stationary distribution of the subset of $G^{n}$ that is semantically equivalent to the coauthorship network resulting from lanl:ConferenceArticles written by lanl:Researchers that are lanl:locatedAt a lanl:University only. The second example presents a grammar for calculating the stationary distribution over all vertices in a semantic network irrespective of the edge labels (i.e. an unconstrained grammar). The second example is equivalent to running the single-relational implementation of PageRank on a semantic network. ### VI.1 Conference Article Co-Authorship Grammar $\Psi_{\text{coaut}}$ Let $\Psi_{\text{coaut}}$ denote the grammar for generating a $\mathbf{\pi}$ for the subset of $G^{n}$ that is semantically equivalent to the coauthorship network resulting from lanl:ConferenceArticles for all lanl:Researchers from a lanl:University. $\Psi_{\text{coaut}}$ is diagrammed in Figure 12 where, for the sake of convenience, the context names, without the _#, denote the rdfs:Resource pointed to by the rwr:forResource property of the respective rwr:Context. The bolded $+$ or $-$ on the edges denotes whether the rwr:Edge is an rwr:OutEdge or rwr:InEdge, respectively. The dashed square represents an rwr:EntryContext. The stack of rules for each rwr:Context denotes the rdf:Seq of rules ordered from top to bottom and rwr:Context attributes are also stacked (in no particular order) with their respective rwr:Context. Figure 12: A grammar to calculate eigenvector centrality on a conference article coauthorship network of university researchers. A single grammar-based random walker $p\in P$ will begin its journey in $G^{n}$ at some vertex $i\in s(V\;|\;\texttt{lanl:University\\_0})$, where $\displaystyle s(V\;|\;$ $\displaystyle\texttt{lanl:University\\_0})=$ $\displaystyle\\{?i\;|\;\langle?i,\texttt{rdf:type},\texttt{lanl:University}\rangle\in G^{n}\\}$ and the $i\in s(V\;|\;\texttt{lanl:University\\_0})$ is chosen with probability $\frac{1}{|s(V\;|\;\texttt{lanl:University\\_0})|}$. After a vertex in $s(V\;|\;\texttt{lanl:University\\_0})$ is chosen, $g^{p}_{0}=i$ and $\psi^{p}_{0}=\texttt{lanl:University\\_0}$. There are $2$ sequentially ordered rules at University_0: rwr:SubmitCounts_0 and rwr:Traverse_0. The first rule has no effect on $\mathbf{\pi}$ or $\mathbf{\pi}^{p}$ because for all $i$ ${\mathbf{\pi}^{p}_{i}}_{(0)}=0$. The rwr:SubmitCounts_0 rule is important on the next time around $\Psi_{\text{coaut}}$. With the rwr:Traverse_0 rule, $p$ randomly chooses a single vertex $w$ in $\displaystyle W=$ $\displaystyle\\{?w\;|\;\langle?w,\texttt{lanl:locatedAt},i\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?w,\texttt{rdf:type},\texttt{lanl:Researcher}\rangle\in G^{n}\\},$ where rwr:Is_1 requires that $g^{p}_{1}=g^{p}_{-1}$ and $g^{p}_{-1}=\emptyset$ (i.e. $O(p)_{1}=\emptyset$). The rwr:Is_1 attribute is important the second time around $\Psi_{\text{coaut}}$. At time step $1$, $g^{p}_{1}=w$ and $\psi^{p}_{1}=\texttt{lanl:Researcher\\_1}$. Researcher_1 has the rwr:IncrCount_1 rule and thus, ${\mathbf{\pi}^{p}_{w}}_{(1)}=1$. After the rwr:IncrCount_1 rule is executed, $p$ will execute the rwr:Traverse_1 rule. The random walker $p$ will randomly choose some $x$ in $\displaystyle X=$ $\displaystyle\\{?x\;|\;\langle w,\texttt{lanl:wrote},?x\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?x,\texttt{rdf:type},\texttt{lanl:ConferenceArticle}\rangle\in G^{n}\\}.$ If $x$ is properly resolved, then $g^{p}_{2}=x$ and $\psi^{p}_{2}=\texttt{lanl:ConferenceArticle\\_2}$. However, if $w$ has not written a lanl:ConferenceArticle, then $x=\emptyset$. At which point, the rwr:Traverse_1 rule fails and $(i,\texttt{lanl:locatedAt},-,w)$ is an ungrammatical path in $G^{n}$ according to $\Psi_{\text{coaut}}$. If $x=\emptyset$, a new random walker (i.e. a $p$ with no history and zero $\mathbf{\pi}^{p}$) randomly chooses some entry point into $\Psi_{\text{coaut}}$ and $G^{n}$ and the process begins again. If, on the other hand, $w$ has written some lanl:ConferenceArticle $x$, then $p$ will randomly select a $y$ in $\displaystyle Y=$ $\displaystyle\\{?y\;|\;\langle?y,\texttt{lanl:wrote},x\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{lanl:Researcher}\rangle\in G^{n}$ $\displaystyle\;\;\wedge\;?y\;\neq\;w\\}.$ Note the role of the rwr:Not_3 property in Researcher_3. rwr:Not_3 guarantees that the $x$ lanl:ConfereneArticle was written by two or more lanl:Researchers and that only those lanl:Researchers that are not $w$ are selected since $X(p)_{3}=\\{w\\}$. Semantically, this ensures that the subset of $G^{n}$ that is traversed is a coauthorship network. If $y=\emptyset$, then $(i,\texttt{lanl:locatedAt},-,w,\texttt{lanl:wrote},+,x)$ is an ungrammatical path with respects to $\Psi_{\text{coaut}}$. If $y\neq\emptyset$, then $g^{p}_{3}=y$, $\psi^{p}_{3}=\texttt{Researcher\\_3}$, and ${\mathbf{\pi}^{p}_{y}}_{(3)}=1$. Finally, because of the rwr:Traverse_3 rule, $p$ randomly selects some $z$ in $\displaystyle Z=$ $\displaystyle\\{?z\;|\;\langle y,\texttt{lanl:locatedAt},?z\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?z,\texttt{rdf:type},\texttt{lanl:University}\rangle\in G^{n}\\}.$ Thus, $g^{p}_{4}=z$ and $\psi^{p}_{4}=\texttt{University\\_0}$. At this point in time, $g^{p}=(i,$ lanl:locatedAt, $-,w$, lanl:wrote, $+,x$, lanl:wrote, $-,y$, lanl:locatedAt, $+,z)$ and $g^{p}$ is a $\Psi_{\text{coaut}}$-correct and $w$ and $y$ are indexed by $\mathbf{\pi}$. The rwr:SubmitCounts_0 rule ensures that ${\mathbf{\pi}_{w}}_{(4)}={\mathbf{\pi}^{p}_{w}}_{(4)}$ and ${\mathbf{\pi}_{y}}_{(4)}={\mathbf{\pi}^{p}_{y}}_{(4)}$. Finally, when rwr:SubmitCounts_0 has completed, ${\mathbf{\pi}^{p}_{w}}_{(4)}={\mathbf{\pi}^{p}_{y}}_{(4)}=0$. This process continues until the ratio between the counts in $\mathbf{\pi}$ converge. At $n=5$, the rwr:Is_1 rule is important to ensure that, after checking if the $y$ rwr:Researcher is rwr:locatedAt a rwr:University, $p$ return to $y$ before locating a rwr:ConferenceArticle written by $y$ and continuing its traversal through the implicit coauthorship network in $G^{n}$ as defined by $\Psi_{\text{coaut}}$. What is provided by $\mathbf{\pi}$ is the number of times a particular vertex in $V$ has been visited over a given number of time steps $n$. If vertex $i\in V$ was visited $\mathbf{\pi}_{i}$ times then the probability of observing a random walker at $i$ is $\frac{n}{\mathbf{\pi}_{i}}$. However, given that $\sum_{i\in V}\mathbf{\pi}_{i}\leq n$ because other vertices not indexed by $\mathbf{\pi}$ exist on a $\Psi_{\text{coaut}}$-correct path of $G^{n}$, the probability of the random walker being at vertex $i$ when observing only those vertices indexed by $\mathbf{\pi}$ is $\mathbf{\pi}_{i}^{\prime}=\frac{\mathbf{\pi}_{i}}{\sum_{j\in V}\mathbf{\pi}_{j}}\;:\;i\in V.$ Thus, $\sum_{i\in V}\mathbf{\pi}_{i}^{\prime}=1.$ This step is called the normalization of $\mathbf{\pi}$ and is necessary for transforming the number of times a vertex in $V$ is visited into the probability that the vertex is being visited at any one time step. When ${||\mathbf{\pi}^{\prime}_{(n)}-\mathbf{\pi}^{\prime}_{(m)}||}_{2}\leq\epsilon$, where $m<n$ and $m$ and $n$ are consecutive $\mathbf{\pi}$ update steps (i.e. consecutive rwr:SubmitCounts), $\mathbf{\pi}$ has converged to a range acceptable by the $\epsilon\in\mathbb{R}$ provided argument. However, $\mathbf{\pi}$ may never converge if the $p$-traversed subset of $G^{n}$ is not strongly connected. For instance, let the triple list $A^{n}$ be defined as $\displaystyle A^{n}=$ $\displaystyle\\{\langle?i,\texttt{lanl:coauthor},?y\rangle\;|$ $\displaystyle\;\wedge\;\langle?w,\texttt{rdf:type},\texttt{lanl:University}\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?i,\texttt{lanl:locatedAt},?w\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?i,\texttt{lanl:wrote},?x\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?x,\texttt{rdf:type},\texttt{lanl:ConferenceArticle}\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?y,\texttt{lanl:wrote},?x\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?y,\texttt{lanl:locatedAt},?z\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?z,\texttt{rdf:type},\texttt{lanl:University}\rangle\in G^{n}$ $\displaystyle\;\wedge\;?i\neq?y\\}.$ Furthermore, let $V^{*}$ denote the set of unique lanl:Researcher vertices in $A^{n}$ and $\mathbf{A}\in\mathbb{R}^{|V^{*}|\times|V^{*}|}$ be a weighted adjacency matrix where $\mathbf{A}_{i,y}=\begin{cases}\frac{1}{|\Gamma^{+}(i)|}&\text{if }\langle i,\texttt{lanl:coauthor},y\rangle\in A^{n}\\\ \frac{1}{|V^{*}|}&\text{if }|\Gamma^{+}(i)|=0.\end{cases}$ If $\mathbf{A}\mathbf{\pi}^{\prime}=\lambda\mathbf{\pi}^{\prime}$ where $\lambda$ is the largest eigenvalue of the eigenvectors of $\mathbf{A}$, then $\mathbf{\pi}^{\prime}$ is the stationary distribution of $\mathbf{A}$ and thus, the $p$-traversed subset of $G^{n}$ given $\Psi_{\text{coaut}}$ is strongly connected. However, most coauthorship networks are not strongly connected Liu _et al._ (2006) and therefore, $\mathbf{\pi}^{\prime}$ may not be a stationary distribution. For example, there may exists some lanl:University denoted $R$ and lanl:locatedAt $R$ are only two lanl:Researchers, $x$ and $y$, that have a coauthor relationship with respects to a particular lanl:ConferenceArticle. If the random walker $p$ happens to enter $G^{n}$ at $x$, then the random walker will never leave the $x/y$ component. However, some new lanl:Researcher, and therefore some new lanl:University, can be introduced into the problem by re-resolving the lanl:ConferenceArticle uniting $x$ and $y$ such that $p$ teleports to some new researcher $w$ at some other lanl:University $S$. This example is depicted in Figure 13, where the dashed line represents a teleportation by $p$. This teleportation introduces the artificial relationship that $x$ coauthored with $w$. Thus, when there exists a non-zero probability of teleportation at every vertex in $V^{*}$, the coauthorship network becomes strongly connected. Figure 13: Teleportation required for connecting isolated components. In order to guarantee a strongly connected network, it is possible to simulate the behavior of randomly choosing some new entry point with probability $\delta\in(0,1]\;$ as an analogy to the method of inducing strong connectivity in Page _et al._ (1998). The rwr:Reresolve rule is introduced to $\Psi_{\text{coaut}}$ at ConferenceArticle_2 where rwr:Reresolve_2 has a $\delta=0.15$, a rwr:steps of $m=2$, and does not rwr:obey any rwr:Context attributes. $\Psi_{\text{coaut'}}$ is diagrammed in Figure 14, where the "0.15" literal is the object of the triple $\langle\texttt{rwr:Reresolve\\_2},\texttt{rwr:probability},\texttt{"0.15"}\rangle\in\Psi_{\text{coaut'}}$ and the "2" literal is the object of triple $\langle\texttt{rwr:Reresolve\\_2},\texttt{rwr:steps},\texttt{"2"}\rangle\in\Psi_{\text{coaut'}}$. Figure 14: A grammar to calculate PageRank on a conference article coauthorship network of university researchers. With respects to $G^{n}$, every time random walker $p$ encounters the rwr:ConferenceArticle_2 context, it has a 15% chance of teleporting to some new lanl:ConferenceArticle $i$ in $V$ such that $\displaystyle Q_{n-2,n}=$ $\displaystyle\\{(?w,?x,-,?y,?z,+,?i)\;|$ $\displaystyle\;\wedge\;\langle?w,\texttt{rdf:type},\texttt{lanl:University}\rangle\in G^{n}$ $\displaystyle\;\;\wedge\;?x=\texttt{lanl:locatedAt}$ $\displaystyle\;\wedge\;\langle?y,\texttt{rdf:type},\texttt{Researcher}\rangle\in G^{n}$ $\displaystyle\;\;\wedge\;?z=\texttt{lanl:wrote}$ $\displaystyle\;\wedge\;\langle?i,\texttt{rdf:type},$ $\displaystyle\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\texttt{lanl:ConferenceArticle}\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?y,?x,?w\rangle\in G^{n}$ $\displaystyle\;\wedge\;\langle?y,?z,?i\rangle\in G^{n}\\}.$ and a new path $q\in Q_{n-2,n}$ is chosen with probability $\frac{1}{|Q_{n-2,n}|}$. If $q=(w,x,-,y,z,+,i)$, $g^{p}_{(n-2)\rightarrow n}=\begin{cases}g^{p}_{(n-2)\rightarrow n}&\text{with probability }1-d\\\ q_{0\rightarrow 2}&\text{with probability }d.\end{cases}$ The rwr:Reresolve rule guarantees that any conference publishing researcher is reachable by any other conference publishing university researcher and thus, the coauthorship network of conference publications by university researchers is strongly connected. Theoretically, the rwr:Reresolve_2 rule ensures that there exists some hypothetical triple list $B^{n}$, such that $B^{n}=\\{\langle?i,\texttt{lanl:teleport},?j\rangle\;|\;?i,?j\in V^{*}\\},$ where $V^{*}$ is the set of lanl:Researchers from $A^{n}$. Let $\mathbf{B}\in\mathbb{R}^{|V^{*}|\times|V^{*}|}$ be a weighted adjacency matrix where for any entry in $\mathbf{B}$, $\mathbf{B}_{i,j}=\frac{1}{|V^{*}|}$. $\Psi_{\text{coaut'}}$ is equivalent to computing $\mathbf{\pi}^{\prime}$ for $\mathbf{C}$ where $\mathbf{C}=\delta\mathbf{A}+(1-\delta)\mathbf{B}$ and $\delta=0.85$. Therefore, $\mathbf{\pi}^{\prime}$ generated from $\Psi_{\text{coaut'}}$ is a stationary distribution. The eigenvector centrality or PageRank of the network could have been calculated by extracting the appropriate lanl:Researcher vertices from $V$ and generating the implicit lanl:ConferenceArticle coauthorship edge between them. This was done with the network $A^{n}$ and its “teleporation” network $B^{n}$, where $\mathbf{A}$ and $\mathbf{B}$ are the respective adjacency matrices representations of these networks. In this sense, the single-relational eigenvector centrality or PageRank algorithm would generate the same results. However, the grammar-based random walker algorithm is different than the “isolation-based” method. In the grammar-based method, there is no need to generate (i.e. make explicit) the implicit single-relational subset of $G^{n}$ and thus, create another data structure; the same $G^{n}$ can be used for different eigenvector calculations without altering it. Thus, multiple different grammars can be running in parallel on the same data set (on the same triple-store). For more complex grammars that involve rwr:Is and rwr:Not constraints over multiple cycles of a grammar, the query to isolate the sub- network becomes increasingly long as recursions cannot be expressed in the standard RDF query language SPARQL Prud’hommeaux and Seaborne (2004). ### VI.2 Simulating Single-Relational PageRank on a Semantic Network The grammar depicted in Figure 15 is denoted $\Psi_{\emptyset}$ and is the grammar that calculates $\mathbf{\pi}$ on any semantic network without consideration for edge directionality nor edge labels. Thus, this grammar is not constrained to the ontology of the semantic network and can be applied to any $G^{n}$ instance. Furthermore, the rwr:Reresolve rule guarantees that all vertices are reachable from all other vertices. Note that this grammar ensures that all vertices in $V$ are $\Psi_{\emptyset}$-correct. The presented grammar is equivalent to executing PageRank on an undirected single-relational representation of a semantic network. Figure 15: A grammar to calculate an undirected single-relational network PageRank on a semantic network. ## VII Analysis What has been presented thus far is an ontology for instantiating an $G^{n}$ specific grammar, the formalization of the rules and attributes that must be respected by a grammar-based random walker, and an eigenvector centrality and PageRank example involving a semantic scholarly network. This section will briefly discuss the various permutations of $G^{n}$ that are traversed by a grammar-based random walker. As stated previously, only a subset of the complete semantic network $G^{n}$ is traversed by any $p\in P$. Let $G^{\psi}\subseteq G^{n}$ denote the graph traversed by $p$ according to $\Psi$. It is noted that only a subset of $G^{\psi}$ is considered $\Psi$-correct (i.e. grammatically correct according to $\Psi$). If $p$ is unable to submit its $\mathbf{\pi}^{p}$ to the global vertex vector $\mathbf{\pi}$, then $p$ has taken an ungrammatical semantic path in $G^{n}$. On the other hand, if $p$ contributes its $\mathbf{\pi}^{p}$ to $\mathbf{\pi}$, $p$ has taken a grammatical semantic path (i.e. a $\Psi$-correct path). Let $G^{\psi+}\subseteq G^{\psi}$ denote the subset of $G^{\psi}$ that is grammatically correct according to $\Psi$. ###### Definition 1 (The $\Psi$-Correct Paths of $G^{\psi+}$) The path $g^{p}_{m\rightarrow n}$ in $G^{n}$ is considered grammatically correct with respects to $\Psi$ if and only if $\psi^{p}_{m}$ is an rwr:EntryContext or an rwr:Context with an rwr:SubmitCounts rule, $\psi^{p}_{n}$ is an rwr:Context with an rwr:SubmitCounts rule, and there exist some rwr:IncrCount rule at time $k$, such that $m\leq k\leq n$. The set of all grammtically correct paths form the semantic network $G^{\psi+}$, where $G^{\psi+}\subseteq G^{\psi}\subseteq G^{n}$. The grammatically correct path $g^{p}_{m\rightarrow n}$ ensures that some vertex in $g^{p}_{m\rightarrow n}$ was validated by the grammar $\Psi$ and indexed by $\mathbf{\pi}$. Figure 16 demonstrates a subset of $G^{n}$ that is traversed by $P$ to generate $G^{\psi+}$, where the bold labeled vertices are those indexed by $\mathbf{\pi}$. Figure 16: $G^{\psi+}$ as the $\Psi$-correct subset of $G^{n}$. Note that the vertices indexed by $\mathbf{\pi}$ are not necessarily all of the vertices encountered by the random walkers in $G^{\psi+}$. Similar to the coauthor example presented previous, while a $p\in P$ traverses vertices of type lanl:Article, lanl:University, and lanl:Researcher, only lanl:Resercher vertices are index by $\mathbf{\pi}$. Thus, those vertices indexed by $\mathbf{\pi}$ form an “implied” network. The $G^{\psi+}$ represented in Figure 16 has the implied network $G^{\mathbf{\pi}}$ as diagrammed in Figure 17. The probabilities on the edges are given by branches between the respective vertices in $G^{\psi+}$. Figure 17: $G^{\mathbf{\pi}}$ as the implied network of $G^{\psi+}$. ###### Theorem 1 If $G^{\psi+}$ is strongly connected and aperiodic, then $\mathbf{\pi}^{\prime}$ is a stationary probability distribution. _Proof._ If $G^{\psi+}$ is strongly connected, then every vertex in $G^{\psi+}$ is reachable from any other vertex. Given that $\mathbf{\pi}$ indexes a subset of the vertices in $G^{\psi+}$ and the vertices in $G^{\mathbf{\pi}}$ are reachable by means of the edges in $G^{\psi+}$, then the vertices indexed by $\mathbf{\pi}$ are strongly connected. Thus, the normalization of $\mathbf{\pi}$, $\mathbf{\pi}^{\prime}$, is a stationary probability distribution. $\Box$ Note that the above does not generalize to $G^{n}$. If $G^{n}$ is strongly connected, that does not guarantee that the grammar will permit the grammar- based random walker $p$ to traverse a subset of $G^{n}$ that is strongly connected. For example, imagine the network $G^{n}$ depicted in Figure 18. Even if $G^{n}$ is a strongly connected network, the $\Psi$-correct subgraph of $G^{n}$ traversed may not be. Figure 18: A strongly connected $G^{n}$ does not guarantee a strongly connected $G^{\psi+}$. Finally, if the path distance between the vertices in $G^{\mathbf{\pi}}$ is equal in $G^{\psi+}$, then $\mathbf{\pi}^{\prime}$ is the primary eigenvector of the $G^{\mathbf{\pi}}$. However, this is not always the case. Figure 19 demonstrates that the timing between indexing the different vertices in the network diagrammed in Figure 16 is different for different paths chosen by $p$. Figure 19: The variability of index delay times for $G^{\mathbf{\pi}}$. ###### Theorem 2 If the paths in $G^{\psi+}$ between the vertices indexed by $\mathbf{\pi}$ are of equal length, then $\mathbf{\pi}^{\prime}$ is the primary eigenvector of $G^{\mathbf{\pi}}$. _Proof._ If the path lengths in $G^{\psi+}$ between the vertices index by $\mathbf{\pi}$ are of equal length, then the intervening non-$\mathbf{\pi}$ vertices in $G^{\psi+}$ can be removed without interfering with the relative timing of respective increments to the vertices in $\mathbf{\pi}$. Given this network manipulation, a single-relational eigenvector centrality algorithm on the single-relational network $G^{\mathbf{\pi}}$ would yield $\mathbf{\pi}^{\prime}$. Thus, $\mathbf{\pi}^{\prime}$ is the primary eigenvector of the network $G^{\mathbf{\pi}}$. $\Box$ ## VIII Conclusion There is much disagreement to the high-level meaning of the primary eigenvector of a network. $\mathbf{\pi}$ has been associated with concepts such as “prestige”’, “value”, “importance”, etc. For Markov chain analysis, when vertices represent states of a system, the meaning is clear; $\mathbf{\pi}$ defines the probability that at some random time $n$, the system $G^{1}$ will be at some particular state in $V$, where more “central” states (i.e. those with a higher $\mathbf{\pi}$ probability) are more likely to been seen. However, the application of $\mathbf{\pi}$ to more abstract concepts of centrality such as “value” has been applied in the area of the web citation network. If the web is represented as a Markov chain, then $\mathbf{\pi}_{i}$ defines the probability that some random web surfer will be at a particular web page $i$ at some random time $n$. Does this phenomena denote that web pages with a higher $\mathbf{\pi}$ probability are more “valuable” than those with lower $\mathbf{\pi}$ probabilities? For the many of us who use Google daily, it does Brin and Page (1998). However, for other artifact networks, $\mathbf{\pi}$ can have a completely different meaning. In journal usage networks, $\mathbf{\pi}$ tends to be a component which makes a distinction between applied and theoretical journals, not “value” or “prestige” Bollen and Van de Sompel (2006). On the other hand, the $\mathbf{\pi}$ calculated for a journal citation network does provide us with the notion of “prestige” Bollen _et al._ (2006). This demonstrates that $\mathbf{\pi}$ has a different meaning depending on the semantics of the edges traversed. In other words, different grammars provide different interpretations of $\mathbf{\pi}$. Whether $\mathbf{\pi}$ represents “value” or some other dimension of distinction, this article has provided a method for calculating various $\mathbf{\pi}$ vectors in subsets of the semantic network $G^{n}$ by means of a random walker algorithm constrained to a grammar. For researchers with nework-based data sets containing heterogeneous entity types and heterogeneous relationship types, this article may provide a more intuitive way of studying the various $\mathbf{\pi}$s of $G^{n}$. ## Acknowledgments This work was funded by a grant from the Andrew W. Mellon Foundation and executed by the MESUR project (http://www.mesur.org). ## References * Aasman (2006) Aasman, J., 2006, _Allegro Graph_ , Technical Report 1, Franz Incorporated, URL www.franz.com/products/allegrograph/allegrograph.datasheet.pd%f. * Aleman-Meza _et al._ (2005) Aleman-Meza, B., C. Halaschek-Wiener, I. B. Arpinar, C. Ramakrishnan, and A. P. Sheth, 2005, IEEE Internet Computing 9(3), 37, ISSN 1089-7801. * Anyanwu and Sheth (2003) Anyanwu, K., and A. Sheth, 2003, in _Proceedings of the Twelfth International World-Wide Web Conference_ (Budapest, Hungary). * Bax (2004) Bax, M., 2004, in _International Conference on Electronic Publishing (ICCC2004)_ (Brasília, Brazil). * Berners-Lee _et al._ (2005) Berners-Lee, T., , R. Fielding, D. Software, L. Masinter, and A. Systems, 2005, Uniform Resource Identifier (URI): Generic Syntax. * Biron and Malhotra (2004) Biron, P. V., and A. Malhotra, 2004, _XML Schema Part 2: Datatypes Second Edition_ , Technical Report, World Wide Web Consortium, URL http://www.w3.org/TR/xmlschema-2/. * Bollen _et al._ (2006) Bollen, J., M. A. Rodriguez, and H. Van de Sompel, 2006, Scientometrics 69(3). * Bollen and Van de Sompel (2006) Bollen, J., and H. Van de Sompel, 2006, Scientometrics 69(2). * Bonacich (1987) Bonacich, P., 1987, American Journal of Sociology 92(5), 1170. * Brandes and Erlebach (2005) Brandes, U., and T. Erlebach (eds.), 2005, _Network Analysis: Methodolgical Foundations_ (Springer, Berling, DE). * Brickley and Guha (2004) Brickley, D., and R. Guha, 2004, _RDF Vocabulary Description Language 1.0: RDF schema_ , Technical Report, World Wide Web Consortium, URL http://www.w3.org/TR/rdf-schema/. * Brin and Page (1998) Brin, S., and L. Page, 1998, Computer Networks and ISDN Systems 30(1–7), 107. * Broder _et al._ (2000) Broder, A., R. Kumar, F. Maghoul, P. Raghavan, S. Rajagopalan, R. Stata, A. Tomkins, and J. Wiener, 2000, in _Proceedings of the 9th International World Wide Web Conference_ (Amsterdam, Netherlands). * Chen _et al._ (2007) Chen, P., H. Xie, S. Maslov, and S. Redner, 2007, Journal of Informetrics 1(1), 8, URL http://arxiv.org/abs/physics/0604130. * Ching and Ng (2006) Ching, W.-K., and M. K. Ng, 2006, _Markov Chains: Models, Algorithms, and Applications_ (Springer, New York, NY). * Cohen and Kjeldsen (1987) Cohen, P. R., and R. Kjeldsen, 1987, Information Processing and Management 23(4), 255\. * Crestani (1997) Crestani, F., 1997, Artificial Intelligence Review 11(6), 453. * Crestani and Lee (2000) Crestani, F., and P. L. Lee, 2000, Information Processing and Management 36(4), 585\. * Gasevic and Devedzic (2006) Gasevic, D., and V. Devedzic, 2006, Knowledge-Based Systems 19, 220. * Gasevic _et al._ (2006) Gasevic, D., D. Djuric, and V. Devedzic, 2006, _Model Driven Architecture and Ontology Development_ (Spring-Verlag, Berlin, DE). * Getoor and Diehl (2005) Getoor, L., and C. P. Diehl, 2005, SIGKDD Explorations Newsletter 7(2), 3, ISSN 1931-0145. * Häggström (2002) Häggström, O., 2002, _Finite Markov Chains and Algorithmic Applications_ (Cambridge University Press). * Klyne and Carroll (2004) Klyne, G., and J. J. Carroll, 2004, _Resource Description Framework (RDF): Concepts and Abstract Syntax_ , Technical Report, World Wide Web Consortium, URL http://www.w3.org/TR/rdf-concepts/. * Lacy (2005) Lacy, L. W., 2005, _OWL: Representing Information Using the Web Ontology Language_ (Trafford Publishing). * Langville and Meyer (2006) Langville, A. N., and C. D. Meyer, 2006, _Google’s PageRank and Beyond: The Science of Search Engine Rankings_ (Princeton University Press, Princeton, New Jersey). * Lin (2004) Lin, S., 2004, in _Sixteenth Conference on Innovative Applications of Artificial Intelligence_ , edited by D. L. McGuinness and G. Ferguson (MIT Press), pp. 991–992. * Liu _et al._ (2006) Liu, X., J. Bollen, M. L. Nelson, and H. Van de Sompel, 2006, Information Processing and Management 41(6), 1462, URL http://arxiv.org/abs/cs.DL/0502056. * Manola and Miller (2004) Manola, F., and E. Miller, 2004, RDF primer: W3C recommendation, URL http://www.w3.org/TR/rdf-primer/. * McGuinness and van Harmelen (2004) McGuinness, D. L., and F. van Harmelen, 2004, OWL web ontology language overview, URL http://www.w3.org/TR/owl-features/. * Mihalcea _et al._ (2004) Mihalcea, R., P. Tarau, and E. Figa, 2004, in _Proceedings of the 20th International Conference on Computational Linguistics (COLING 2004)_ (Switzerland, Geneva). * Miller (1998) Miller, E., 1998, D-Lib Magazine URL http://dx.doi.org/hdl:cnri.dlib/may98-miller. * Mitrani (1998) Mitrani, I., 1998, _Probabilistic Modeling_ (Cambridge University Press, Cambridge, UK). * Page _et al._ (1998) Page, L., S. Brin, R. Motwani, and T. Winograd, 1998, _The PageRank Citation Ranking: Bringing Order to the Web_ , Technical Report, Stanford Digital Library Technologies Project. * Parzen (1962) Parzen, E., 1962, _Stochastic Processes_ (Holden-Day, Inc.). * Prud’hommeaux and Seaborne (2004) Prud’hommeaux, E., and A. Seaborne, 2004, _SPARQL Query Language for RDF_ , Technical Report, World Wide Web Consortium, URL http://www.w3.org/TR/2004/WD-rdf-sparql-query-20041012/. * Rada _et al._ (1989) Rada, R., H. Mili, E. Bicknell, and M. Blettner, 1989, IEEE Transactions on Systems, Man and Cybernetics 19(1), 17\. * Rodriguez (2007a) Rodriguez, M. A., 2007a, _General-Purpose Computing on a Semantic Network Substrate_ , Technical Report LA-UR-07-2885, Los Alamos National Laboratory, URL http://arxiv.org/abs/0704.3395. * Rodriguez (2007b) Rodriguez, M. A., 2007b, in _40th Annual Hawaii International Conference on Systems Science (HICSS’07)_ (Waikoloa, Hawaii), URL http://dx.doi.org/10.1109/HICSS.2007.487. * Rodriguez and Bollen (2007) Rodriguez, M. A., and J. Bollen, 2007, _Modeling Computations in a Semantic Network Substrate_ , Technical Report LA-UR-07-3678, Los Alamos National Laboratory, URL http://arxiv.org/abs/0706.0022. * Rodriguez and Watkins (2007) Rodriguez, M. A., and J. H. Watkins, 2007, _Grammar-Based Geodesics in Semantic Networks_ , Technical Report LA-UR-07-4042, Los Alamos National Laboratory. * Ruttenberg _et al._ (2007) Ruttenberg, A., T. Clark, W. Bug, M. Samwald, O. Bodenreider, H. Chen, D. Doherty, K. Forsberg, Y. Gao, V. Kashyap, J. Kinoshita, J. Luciano, _et al._ , 2007, BMC Bioinformatics 8(3), S2, ISSN 1471-2105, URL http://www.biomedcentral.com/1471-2105/8/S3/S2. * Savoy (1992) Savoy, J., 1992, Information Processing and Management 28(3), 389. * Sebesta (2005) Sebesta, R. W., 2005, _Concepts of Programming Languages_ (Addison-Wesley). * Sheth _et al._ (2005) Sheth, A. P., I. B. Arpinar, C. Halaschek, C. Ramakrishnan, C. Bertram, Y. Warke, D. Avant, F. S. Arpinar, K. Anyanwu, and K. Kochut, 2005, Journal of Database Management 16(1), 33. * Sowa (1987) Sowa, J. F., 1987, _Encyclopedia of Artificial Intelligence_ (Wiley), chapter Semantic Networks. * Trefethen and Bau (1997) Trefethen, L. N., and D. Bau, 1997, _Numerical Linear Algebra_ (Society for Industrial and Applied Mathematics). * Wasserman and Faust (1994) Wasserman, S., and K. Faust, 1994, _Social Network Analysis: Methods and Applications_ (Cambridge University Press, Cambridge, UK). * Zhuge and Zheng (2003) Zhuge, H., and L. Zheng, 2003, in _Proceedings of the Twelfth International World Wide Web Conference (WWW03)_ (Budapest, Hungary).

arxiv-papers

2024-09-04T02:48:54.621814

0803.4429

A simple conceptual model of abrupt glacial climate events. A conceptual model of Dansgaard-Oeschger events H. Braun et al. H. Braun H. Braun Heidelberg Academy of Sciences and Humanities, c/o Institute of Environmental Physics, University of Heidelberg, Im Neuenheimer Feld 229, 69120 Heidelberg, Germany. A. Ganopolski Potsdam Institute for Climate Impact Research, P.O. Box 601203, 14412 Potsdam, Germany. M. Christl PSI/ETH Laboratory for Ion Beam Physics, c/o Institute of Particle Physics, ETH Zurich, 8093 Zurich, Switzerland. D. R. Chialvo Department of Physiology, Feinberg Medical School, Northwestern University, 303 East Chicago Ave. Chicago, IL 60611, USA. Here we use a very simple conceptual model in an attempt to reduce essential parts of the complex nonlinearity of abrupt glacial climate changes (the so-called Dansgaard-Oeschger events) to a few simple principles, namely (i) the existence of two different climate states, (ii) a threshold process and (iii) an overshooting in the stability of the system at the start and the end of the events, which is followed by a millennial-scale relaxation. By comparison with a so-called Earth system model of intermediate complexity (CLIMBER-2), in which the events represent oscillations between two climate states corresponding to two fundamentally different modes of deep-water formation in the North Atlantic, we demonstrate that the conceptual model captures fundamental aspects of the nonlinearity of the events in that model. We use the conceptual model in order to reproduce and reanalyse nonlinear resonance mechanisms that were already suggested in order to explain the characteristic time scale of Dansgaard-Oeschger events. In doing so we identify a new form of stochastic resonance (i.e. an overshooting stochastic resonance) and provide the first explicitly reported manifestation of ghost resonance in a geosystem, i.e. of a mechanism which could be relevant for other systems with thresholds and with multiple states of operation. Our work enables us to explicitly simulate realistic probability measures of Dansgaard-Oeschger events (e.g. waiting time distributions, which are a prerequisite for statistical analyses on the regularity of the events by means of Monte-Carlo simulations). We thus think that our study is an important advance in order to develop more adequate methods to test the statistical significance and the origin of the proposed glacial 1470-year climate cycle. DO events as seen in the GISP2 ice-core from Greenland [Grootes et al., 1993, Grootes and Stuiver, 1997]. The figure shows Greenland temperature changes over the interval between 10000 and about 40000 years before present. DO events (0-10) manifest themselves as saw-tooth shaped warm intervals. Dashed lines are spaced by 1470 years. Time series of North Atlantic atmospheric / sea surface temperatures during the last ice age reveal the existence of repeated large-scale warming events, the so-called Dansgaard-Oeschger (DO) events [Dansgaard et al., 1982, Grootes et al., 1993]. In climate records from the North Atlantic region the events have a characteristic saw-tooth shape (Fig. <ref>): They typically start with a warming by up to 10-15 K [Severinghaus and Brook, 1999, Leuenberger et al., 1999] over only a few years/decades. Temperatures remain high for centuries/millennia until they drop back to pre-events values over a century or so. A prominent feature of DO events is their millennial time scale: During Marine Isotope Stages (MIS) 2 and 3, successive events in the GISP2 ice core were reported to be often spaced by about 1470 years or multiples thereof [Alley et al., 2001, Schulz, 2002, Rahmstorf, 2003], compare Fig. <ref>. A leading spectral peak corresponding to the 1470-year period was found [Grootes and Stuiver, 1997, Yiou et al., 1997], and this spectral component was reported to be significant at least over a certain time interval [Schulz, 2002]. We note, however, that the statistical significance of this pattern is still under debate [Ditlevsen et al., 2007], in particular because of the lack of adequate nonlinear analysis methods. It was proposed that DO events represent rapid transitions between two fundamentally different modes of the thermohaline ocean circulation (THC) [Oeschger et al., 1984, Broecker et al., 1985], most likely corresponding to different modes of deep-water formation [Alley and Clark, 1999, Ganopolski and Rahmstorf, 2001]. The origin of these transitions is also under debate: In principle they could have been caused by factors from outside of the Earth system [Keeling and Whorf, 2000, Rial, 2004, Clemens, 2005, Braun et al., 2005], but they could also represent internal oscillations of the climate system [Broecker et al., 1990, Sakai and Peltier, 1997, van Kreveld et al., 2000]. Several nonlinear resonance mechanisms have been suggested in order to explain the characteristic timing of DO events, including coherence resonance [Ganopolski and Rahmstorf, 2002, Timmermann et al., 2003, Ditlevsen et al., 2005] and stochastic resonance [Alley et al., 2001, Alley et al., 2001, Ganopolski and Rahmstorf, 2002, Rahmstorf and Alley, 2002]. § SPECTRUM OF MODELS DO events have already been simulated by a variety of models, ranging from simple conceptual ones to Earth system models of intermediate complexity (EMICs). Conceptual models are most suitable to perform large numbers of long-term investigations because they require very little computational cost. However, they are often based on ad-hoc assumptions and only consider processes in a highly simplified way. In addition to that, the number of adjustable parameters is typically large compared to the degrees of freedom in those models. This implies that seemingly good results can often be obtained merely by excessive tuning. Nevertheless, conceptual models can provide important help for the interpretation of complex climatic The gap between conceptual models and the most comprehensive general circulation models (GCMs), which are not yet applicable for millennial-scale simulations because of their large computational cost, is bridged by EMICs [Claussen et al., 2002]. EMICs include most of the processes described in comprehensive models (in a more reduced form), and interactions between different components of the Earth system (atmosphere, hydrosphere, cryosphere, biosphere, etc.) are simulated. The number of degrees of freedom typically exceeds the number of adjustable parameters by orders of magnitude. Since many EMICs are fast enough for studies on the multi-millennial time scale, they are the most adequate tools for the simulation of DO events. § THE CONCEPTUAL MODEL The simple conceptual model which we use here is an extended version of the model described by Braun et al., 2005 (in the Supplementary Material of that publication). Here we use the model to demonstrate and analyse two apparently counterintuitive resonance phenomena (stochastic resonance and ghost resonance) that can exist in a large class of highly nonlinear systems. Due to the complexity of many of those systems it is often impossible to precisely identify the reasons for the occurrence of these resonance phenomena. Our conceptual model, in contrast, has a very clear forcing-response relation as well as a very low computational cost and thus provides a powerful tool to explore these phenomena and to test their robustness. Furthermore, we describe and discuss the applicability of the model for improved statistical analyses (i.e. Monte-Carlo simulations) on the regularity of DO events. In the following the key assumptions of the conceptual model are first discussed. In the Supplementary Material we then compare the model performance under a number of systematic forcing scenarios with the performance of a more comprehensive model (the EMIC CLIMBER-2), compare Supplementary Information File and Supplementary Figs. 1-6. In the framework of the conceptual model we finally demonstrate and interpret two hypotheses that were previously suggested in order to explain the recurrence time of DO events, and we discuss how these hypotheses could be tested. §.§ Model description Our conceptual model is based on three key assumptions: * DO events represent repeated transitions between two different climate states, corresponding to warm and cold conditions in the North Atlantic region. * These transitions are rapid compared to the characteristic life-time of the two climate states (i.e. in first order approximation they occur instantaneously) and take place each time a certain threshold is crossed. * With every transition between the two states the threshold overshoots and afterwards approaches equilibrium following a millennial-scale relaxation process. This implies that the conditions for a switch between both states ameliorate with increasing duration of the cold and warm Our three assumptions are supported by paleoclimatic evidence and/or by simulations with a climate model: * Since long, DO events have been regarded as repeated oscillations between two different climate states [Dansgaard et al., 1982]. It has been suggested that these states are linked with different modes of operation of the THC [Oeschger et al., 1984, Broecker et al., 1985]. This seminal interpretation has since then influenced numerous studies and is now generally accepted [Rahmstorf, 2002]. Indirect data indicate that the glacial THC indeed switched between different modes of operation [Sarnthein et al., 1994, Alley and Clark, 1999] which, according to their occurrence during cold and warm intervals in the North Atlantic, were labelled stadial and interstadial modes. A third mode named Heinrich mode (because of its presence during the so-called Heinrich events) is not relevant here. * High-resolution paleoclimatic data show that transitions from cold conditions in the North Atlantic region to warm ones often happened very quickly, i.e. on the decadal-scale or even faster [Taylor et al., 1997, Severinghaus and Brook, 1999]. The opposite transitions were slower, i.e. on the century-scale [Schulz, 2002], but nevertheless still faster that the characteristic life-time of the cold and warm intervals (which is on the centennial to multi-millennial time scale, compare Fig. 1). The abruptness of the shifts from cold conditions to warm ones has commonly been interpreted as evidence for the existence of a critical threshold in the climate system that needs to be crossed in order to trigger a shift between stadial and interstadial conditions [Alley et al., 2003]. Such a threshold could be provided by the THC (more precisely, by the process of deep-water formation): When warm and salty surface water from lower latitudes cools on its way to the North Atlantic, its density increases. If the density increase is large enough (i.e. if the surface gets denser than the deeper ocean water), surface water starts to sink. Otherwise, surface water can freeze instead of sinking. The onset of deep-water formation can thus hinder sea-ice formation and facilitate sea-ice melting (due to the vertical heat transfer between the surface and the deeper ocean). A switch between two fundamentally different modes of deep-water formation can thus dramatically change sea ice cover and can cause large-scale climate shifts. Such nonlinear, threshold-like transitions between different modes of deep-water formation are at present considered as the most likely explanation for DO events [Alley et al., 1999, Ganopolski and Rahmstorf, 2001]. * The time-evolution of Greenland temperature during the warm phase of DO events has a characteristic saw-tooth shape (Fig. <ref>). Highest temperatures typically occur during the first decades of the events. These temperature maxima are followed by a gradual cooling trend over centuries/millennia, before the system returns to cold conditions at the end of the events. This asymmetry supports the idea that the system overshoots in some way during the abrupt warming at the beginning of the events and that the subsequent cooling trend represents a millennial-scale relaxation towards a new equilibrium [Schulz et al., 2002, Centurelli et al., 2006]. We note that the time-evolution of Greenland temperature provides no clear evidence for an overshooting during the opposite transitions (i.e. from the warm state back to the cold one). This, however, is not necessarily in contradiction to our assumption: This lack of an overshooting in the temperature fields does not necessarily mean that the ocean-atmosphere system did not overshoot, since Greenland temperature evolution in the stadial state might have been dominated by factors other than the THC (respectively its stability), e.g. by Greenland ice accumulation, which would mask the signature of the THC in the ice core We will show later (in Sect. 3.3.) that the assumption of an overshooting in the stability of the system is in fact strengthened by the analysis of model results obtained with the coupled model CLIMBER-2. In that model the overshooting results from the dynamics of the transitions between the two climate states: In the stadial state deep convection occurs south of the Greenland-Scotland ridge (i.e. at about 50 $^\circ$N). In the interstadial state, however, deep convection takes place north of the ridge (i.e. at about 65 $^\circ$N). The onset of deep convection north of the Greenland-Scotland ridge, which releases accumulated energy to the atmosphere (i.e. heat that is stored mainly in the deep ocean), in first place starts DO events in the model. This heat release leads to a reduction of sea ice, which in turn further enhances sea surface densities between 50 $^\circ$N and 65 $^\circ$N (e.g. by increased local evaporation and reduced sea ice transport into that area). As a result deep convection also starts between 50 $^\circ$N and 65 $^\circ$N, and much more heat can be released to the atmosphere. Without a further response of the THC the system would return quickly (within years or decades, i.e. with the convective time scale) to its original state. In CLIMBER-2, however, the changes in deep convection trigger a northward extension and also an intensification of the ocean circulation (i.e. an overshooting of the Atlantic meridional overturning circulation; compare Ganopolski and Rahmstorf, 2001), which maintains the interstadial climate state since it is accompanied by an increase in the salinity and heat flux to the new deep convection area (at about 65 $^\circ$N). In response to the overshooting of the overturning circulation, the system relaxes slowly (within about 1000 years, i.e. with the advective time scale) towards the stadial state. We note that the advective time scale corresponds to the millennial relaxation time in our conceptual model. The model CLIMBER-2 also supports the validity of our overshooting assumption during the opposite transition (from the warm state back to the cold one), as we will show in Sect. 3.3. We would like to stress that our interpretation of the processes during DO events is, of course, not necessarily true since we can only speculate that the underlying mechanism of the events is correctly captured by CLIMBER-2. §.§ Model formulation We implement the above assumptions in the following way (compare Fig. <ref>): First we define a discrete index s(t) that indicates the state of the system at time t (in years). Since we postulate the existence of two states, s can only take two values (s=1: warm state, s=0: cold state). We further define a threshold function T(t) that describes the stability of the system at time t (i.e. the stability of the current model Dynamics of our conceptual model. Shown is the time evolution of the model, in response to a forcing that is large enough to trigger switches between both model states. Top: Forcing f (black) and threshold function T (red). Bottom: Model state s (grey; s=0 corresponds to the cold state, s=1 to the warm one) and state variable S (green). At time $t''$ the forcing falls below the threshold function and a shift from the cold state into the warm one is triggered. With this transition, the threshold function switches to a non-equilibrium value (representing an overshooting of the system) and afterwards approaches equilibrium following a millennial-scale relaxation. At time $t'$ the forcing exceeds the threshold function, and a transition from the warm state back into the cold one is triggered. With this transition, the threshold function switches to another non-equilibrium value and approaches equilibrium following another millennial-scale relaxation, until the forcing again falls below the threshold function and the next switch into the warm state is triggered. Note that the state variable S is chosen to be identical to the threshold function T. For convenience, discontinuities in T and S are eliminated by linear interpolation. T, S and f are normalised in the figure. Second we define rules for the time evolution of the threshold function T. When the system shifts its state, we assume a discontinuity in the threshold function: With the switch from the warm state to the cold one (at time t' in Fig. <ref>) T takes the value A$_{0}$. Likewise, with the switch from the cold state into the warm one (at time t” in Fig. <ref>) T takes the value A$_{1}$. As long as the system does not change its state the evolution of T is assumed to be given by a relaxation process: \begin{equation} \frac{dT}{dt} = - \frac{(T-B_s)}{\tau_s} \end{equation} (s labels the current model state, $\tau_s$ denotes the relaxation time in that state, B$_{s}$ is a state-dependent constant that labels the equilibrium value of T in each model state). These assumptions result in the following expression for the threshold function T: \begin{equation} T(t) = (A_s-B_s) \cdot \exp(- \frac{t-\delta_s}{\tau_s})+B_s. \end{equation} Note that in the above expression the index s again denotes the current state of the model (i.e. s=0 stands for the cold state and s=1 for the warm one), $\delta_0$ labels the time of the last switch from the warm state into the cold one, and $\delta_1$ indicates the time of the last switch from the cold state into the warm one. Third we assume that transitions from one state to the other are triggered each time a given forcing function f(t) crosses the threshold function T. More precisely, we assume that when the system is in the cold state ($s[t''] = 0$) and the forcing is smaller than the threshold value ($f[t''+1] < T[t''+1]$) the system switches into the warm state ($s[t''+1] = 1$). This shift marks the start of a DO event. Likewise, when the system is in the warm ($s[t'] = 1$) and the forcing is larger than the threshold value ($f[t'+1] > T[t'+1]$) the system switches into the cold state ($s[t'+1] = 0$). That shift represents the termination of a DO event. If none of these conditions is fulfilled, the system remains in its present state (i.e. $s[t+1] = s[t]$). To simplify the comparison of the model output with paleoclimatic records we further define a state variable S, which represents anomalies in Greenland temperature during DO events. For simplicity we assume that the state variable is equal to the threshold function: $S(t) = T(t)$ (i.e. we assume that Greenland temperature evolution during DO events is closely related to the current state of the THC, respectively to its stability). We stress that this assumption is of course highly simplified, because Greenland temperature is certainly not only influenced by the THC but also by other processes such as changes in ice accumulation during DO oscillation. However, this assumption is not crucial for the dynamics of our model, since the timing of the switches between both model states is solely determined by the relation between the forcing function f and the threshold function T. This means that even if we included a more realistic relation between T and S, the timing of the simulated climate shifts would be unchanged and the model dynamics would thus essentially be invariant. Note that $\delta_0$ and $\delta_1$ are not adjustable; they rather represent internal time markers. Thus, six adjustable parameters exist in our model as described here, namely $A_{0}$, $A_{1}$, $B_{0}$, $B_{1}$, $\tau_{0}$ and $\tau_{1}$. Our choice for these parameters is shown in Table <ref>. With these parameter values the system is bistable (i.e. no transition is ever triggered in the absence of any forcing, since $B_{0} \le 0$ and $B_{1} \ge 0$) and almost symmetric. That means that the average duration of the simulated warm and cold intervals is almost equal. When compared with Greenland paleotemperature records this situation most likely corresponds to the time interval between about 27000 and 45000 years before present, during which the duration of the cold and warm intervals in DO oscillations was also comparable (Fig. 1). The model can, however, also represent an unstable (for $B_{0} > 0$ and $B_{1} < 0$) or a mono-stable system, in which the stable state is either the warm one (for $B_{0} > 0$ and $B_{1} \ge 0$) or the cold one (for $B_{0} \le 0$ and $B_{1} < 0$); when compared to the ice core data this situation is closer to the time interval between 15000 and 27000 years before present, since during that time the system was preferably in its cold state and the forcing apparently crossed the threshold only infrequently and during short periods of time. Parameters of the conceptual model. All parameters have the same values as in the publication of Braun et al., 2005. $A_{0}$, $A_{1}$, $B_{0}$ and $B_{1}$ are given in freshwater units (i.e. in mSv = milli-Sverdrup; $1~\unit{mSv} = 10^{3}~\unit{m^{3}/s}$), since the conceptual model was originally designed to mimic the response of the THC to an anomaly in the surface freshwater flux. Parameter Chosen value A$_{0}$ -27 mSv A$_{1}$ 27 mSv B$_{0}$ -9.7 mSv B$_{1}$ 11.2 mSv $\tau_{0}$ 1200 years $\tau_{1}$ 800 years §.§ Comparison with a coupled climate model In order to test our conceptual model we compare its performance under a number of systematic forcing scenarios with the performance of the far more comprehensive model CLIMBER-2 (a short description of that model is given in the Appendix; a detailed description exists in the publication of Petoukhov et al., 2000). Analogous to Braun et al., 2005, we investigate the response of both models to a forcing that consists of two century-scale sinusoidal cycles. In the conceptual model, the forcing is implemented as the forcing function f. In the EMIC, the forcing is added to the surface freshwater flux in the latitudinal belt 50-70 $^{\circ}$N, following Ganopolski and Rahmstorf, 2001 and Braun et al., 2005. This anomaly changes the vertical density gradient in the ocean and can thus trigger DO events. Switches from the cold state into the warm one are excited by sufficiently large (order of magnitude: a few centimetre per year in the surface freshwater flux into the relevant area of the North Atlantic) negative freshwater anomalies (i.e. by positive surface density anomalies that are strong enough to trigger buoyancy [deep] convection), and the opposite switches are triggered by sufficiently large positive freshwater anomalies (i.e. by negative surface density anomalies that are strong enough to stop buoyancy [deep] convection). This justifies our choice for the logical relations that govern the dynamics of the transitions in the conceptual model (i.e. $f(t) < T(t)$ as the condition for the switch from the cold state to the warm one, $f(t) > T(t)$ for the opposite switch). A detailed comparison between both model outputs is presented in the Supplementary Material. We here only summarise the main results: We find a general agreement between both models, which is robust when the forcing parameters are varied over some range (Supplementary Figs. 1-6). The conceptual model reproduces the existence of three different regimes (cold, warm, oscillatory) in the output of the EMIC and also their approximate position in the forcing parameter-space. By construction only the nonlinear component in the response of the EMIC to the forcing is reproduced by the conceptual model (this component represents the saw-tooth shape of DO events). A second, more linear component is not included in the conceptual model (this component represents small-amplitude temperature anomalies which are superimposed on the saw-tooth shaped events in the EMIC). In particular, the conceptual model very well reproduces the timing of the onset of DO events in the EMIC. The fact that our conceptual model, despite its simplicity, agrees in so many aspects with the much more detailed model CLIMBER-2 suggests that it indeed captures the key features in the dynamics of DO events in that We would like to stress that the output of the EMIC indeed supports our assumption of an overshooting in the stability of the system during the transitions between both climate states: When driven by a periodic forcing (with a period of 1470 years), the EMIC can show periodic oscillations during which it remains in either of its states for more than one forcing period (i.e. for considerably more than 1470 years, compare Supplementary Fig. 4a). This implies that (at least in the EMIC) the conditions for a return to the opposite state indeed ameliorate with increased duration of the cold or warm intervals. If the thresholds in the model were constant (or gradually increasing with increasing duration of the simulated cold / warm intervals), in contrast, the duration of the cold and warm intervals during the simulated oscillation could never be longer than 1470 years: If a periodic forcing does not cross a constant (or gradually increasing) threshold within its first period, it never crosses the threshold, due to the periodicity of the forcing. § NONLINEAR RESONANCE MECHANISMS IN THE MODEL Strongly nonlinear systems can show complex and apparently counterintuitive resonance phenomena that cannot occur in simple linear systems. In this section we use our conceptual model to demonstrate and to discuss two of these phenomena, i.e. stochastic resonance (SR) and ghost resonance (GR). Since the explanation of the 1470-year cycle (and in fact even its significance) is still an open question, we further discuss how future tests could distinguish between the proposed mechanisms. §.§ Stochastic resonance (SR) Stochastic resonance. The input consists of: 1. a sub-threshold sinusoidal signal with a period of 1470 years and an amplitude of 4.5 mSv (about 40 percent of the threshold value B$_{1}$ above which DO events occur in the model), 2. a random Gaussian-distributed signal with white noise power signature (standard deviation $\sigma$ = 8 mSv) and a cutoff frequency of 1/(50 years). The cutoff is used since no damping exists in the model and it thus shows an unrealistically large sensitivity to high-frequency (i.e. decadal-scale or faster) forcing. A: Total input (black), periodic input component (grey), model output (green). Dashed lines are spaced by 1470 years. B: Relative frequency to obtain a spacing of 1470 years $\pm$10% (triangles) respectively $\pm$20% (squares) between successive events, as a function of the noise level $\sigma$. In linear systems that are driven by a periodic input, the existence of noise generally reduces the regularity of the output (e.g. the coherence between the input and the output). This is not necessarily the case in nonlinear systems: Excitable or bistable nonlinear systems with a threshold and with noise, which are driven by a sinusoidal sub-threshold input, can show maximum coherence between the input and the output for an intermediate noise level, for which the leading output frequency of the system is close to the input frequency. This phenomenon is called stochastic resonance (SR) [Benzi et al., 1982, Gammaitoni et al., 1998]. SR has been suggested to explain the characteristic timing of DO events [Alley et al., 2001], i.e. the apparent tendency of the events to occur with a spacing of about 1470 years or integer multiples thereof. It has further been demonstrated that DO events in the model CLIMBER-2 can be subject to SR [Ganopolski and Rahmstorf, 2002]. Here we apply our conceptual model to reproduce these results and to reanalyse the underlying mechanism. We use an input that is composed of: (i) a sinusoidal signal with a period of 1470 years, (ii) additional white noise. Figures <ref> and <ref> show that for a suitable noise level the model can indeed show DO events with a preferred spacing of about 1470 years or integer multiples thereof. The reason for this pattern in the output is easily understandable in the context of the model dynamics: DO events in the model are triggered by pronounced minima of the total input (total input = periodic signal plus noise). These minima generally cluster around the minima of the sinusoidal signal, and the start of the simulated events thus has a tendency to coincide with minima of the sinusoidal signal (Fig. <ref>a). Some minima of the sinusoidal signal, however, are not able to trigger an event, because the magnitude of the noise around these minima is too small so that the threshold function is not reached by the total input. Consequently, a cycle is sometimes missed, and the spacing of successive events can change from about 1470 years to multiples thereof. Unlike the model CLIMBER-2 (which has a complex relationship between the input and the output and also a large computational cost) our conceptual model can be used for a detailed investigation of the SR, e.g. because the dynamics of the model is simple and precisely known and because probability measures (such as waiting time distributions) can be explicitly computed. In fact, the resonant pattern in the conceptual model (Fig. <ref>) is due to two time-scale matching conditions: The noise level is such that the average waiting time between successive noise-induced transitions is comparable to half of the period of the periodic forcing, and also comparable to the relaxation times $\tau_{0}$ and $\tau_{1}$ of the threshold function (compare Fig. <ref>b). This situation is different from the usual SR, in which thresholds (or potentials) are constant in time (apart from the influence of the periodic input). In the usual SR, only one time-scale matching condition exists [Gammaitoni et al., 1998], namely the one that the average waiting time between successive noise-induced transitions (i.e. the inverse of the so-called Kramers rate) is comparable to half of the period of the periodic forcing. Distribution of the spacing $\Delta$T between successive events. The input in a and b consists of noise only, with a standard deviation $\sigma$ of 8 mSv (as in Fig. <ref>a). In c and d, a sinusoidal input component (amplitude = 4.5 mSv, period = 1470 years; compare Fig. <ref>) is added to the noise. In a and c the threshold function is constant in each state (20 mSv in the warm state and -20 mSv in the cold one), while the overshooting relaxation assumption is used in b and d (with threshold parameters as shown in Table 1). Thus, c corresponds to the “usual” SR while d shows our “overshooting” SR. In order to investigate the implications of the second condition we simulate histograms for four different scenarios in the conceptual model (Fig. <ref>): 1. noise-only input, constant threshold (Fig. <ref>a); 2. noise-only input, overshooting threshold (Fig. <ref>b); 3. noise plus periodic input, constant threshold (Fig. <ref>c); 4. noise plus periodic input, overshooting threshold (Fig. <ref>d). We note that 3. corresponds to the usual SR, while 4. describes our overshooting stochastic resonance. As can be seen from the histograms, the existence of the millennial-scale relaxation process leads to a synchronisation in the sense that the waiting times between successive events are confined within a much smaller time interval (about 1000-4500 years with the overshooting, compared to about 0-10000 years without This confinement is plausible since the transition probability between both model states strongly depends on the magnitude of the threshold, which declines with increasing duration of the cold or warm intervals: When the standard deviation of the noise level is chosen such that the average waiting time between successive noise-induced transitions is comparable to the relaxation times $\tau_{0}$ and $\tau_{1}$, as in Fig. <ref>, the overshooting relaxation strongly reduces the transition probability for waiting times much smaller than the relaxation time (since the corresponding values of the threshold function are large) and increases the transition probability for waiting times of the order of the relaxation time or larger (since the corresponding values of the threshold function are considerably smaller). The probability to find an only century-scale spacing between successive events is thus small, because the corresponding transition probabilities are small. On the other hand, the probability to find a multi-millennial spacing is also small, because the states are already depopulated before (i.e. the probability to obtain lifetimes considerably larger than the relaxation time is almost zero). This explains why the possible values for the spacing between successive DO events are restricted to a much smaller range than in the usual SR (i.e. in the case with constant This synchronisation effect is indeed not unique to the conceptual model: The output of the coupled model CLIMBER-2 shows a similar pattern (with possible waiting times between successive DO events of e.g. about 1500-5000 years or about 1000-3000 years, depending on the noise level; compare Fig. 4a-d in the publication of Ganopolski and Rahmstorf, 2002). This similarity is of course not surprising, since the conceptual model is apparently able to mimic the events in the EMIC and since an overshooting in the stability of both states clearly also exists in CLIMBER-2 (compare Sect. 3.3). We note that in the GISP2 ice core data, DO events in the time interval 27000-45000 years before present (which, as discussed in Sect. 3.2, is the best analogue to the “background climate state” in our conceptional model, since the duration of the cold and warm intervals in the data is comparable in that interval) have spacings of about 1000-3000 years (compare Fig. 1) and were reported to cluster around values of either about 1470 years or about 2940 years [Schulz, 2002]. Because the SR mechanism could explain such a pattern (compare Fig. <ref>d) it has originally been proposed. However, this mechanism requires a sinusoidal input with a period of about 1470 years, which has so far not been detected. §.§ Ghost resonance (GR) In linear systems which are driven by a periodic input, the frequencies of the output are always identical to the input frequencies. This is not necessarily the case in nonlinear systems. For example, nonlinear excitable (or bistable) systems that are driven by an input with frequencies corresponding to harmonics of a fundamental frequency (which itself is not present in the input) can show a resonance at the fundamental frequency, i.e. at a frequency with zero input power. This phenomenon, which was first described in order to explain the pitch of complex sounds [Chialvo et al., 2002, Chialvo, 2003] and later observed experimentally e.g. in laser systems [Buldu et al., 2003], is called ghost resonance (GR). GR and SR can indeed occur in the same class of systems, e.g. in bistable or excitable systems with thresholds. However, unlike SR, GR requires a periodic driver with more than one frequency. Although many geophysical systems might be subject to GR (since the relevant processes often have thresholds), the occurrence of this mechanism has so far not expressly been demonstrated in geoscience. Ghost resonance. Top: Forcing (black) and model response (green). Middle: Amplitude spectrum of the forcing. Bottom: Amplitude spectrum of the model response. We use two sinusoidal forcing cycles, with frequencies of 7/(1470 years) and 17/(1470 years), respectively, and with equal amplitudes. These two cycles coincide every 1470 years and create peaks of particularly pronounced magnitude, spaced by exactly that period. Thus, despite the fact that there is no spectral power at the corresponding frequency (see middle panel), the forcing repeatedly crosses the threshold at those intervals. Consequently, the response of the conceptual model (i.e. the time evolution of the state variable S) shows strictly repetitive DO events with a period of 1470 years (as indicated by the dashed lines, which are spaced by 1470 years). Despite the lack of a 1470-year spectral component in the forcing, the output shows a very prominent peak at the corresponding frequency. Here we discuss a hypothesis that was recently proposed to explain the 1/(1470 years) leading frequency of DO events [Braun et al., 2005]. The underlying mechanism of the hypothesis is in fact the first reported manifestation of GR in a geophysical model system. According to that hypothesis the 1/(1470 years) frequency of DO events could represent the nonlinear climate response to forcing cycles with frequencies close to harmonics of 1/(1470 years). Our conceptual model illustrates the plausibility of this mechanism: We use a bi-sinusoidal input with frequencies of 7/(1470 years) and 17/(1470 years), i.e. with frequencies corresponding to the 7th and the 17th harmonic of a 1/(1470 years) fundamental frequency, and with equal amplitudes. A spectral component corresponding to the fundamental frequency is not explicitly present in the input. Since the two sinusoidal cycles correspond to harmonics of the missing fundamental, the input signal repeats with a period of 1470 years. For an appropriate range of input amplitudes, the output of the conceptual models shows periodic DO events with a period of 1470 years (Fig. <ref>). Unlike the input, the model output exhibits a pronounced frequency of 1/(1470 years), corresponding to the leading frequency of DO events and to the fundamental frequency that is absent in the input. This apparent paradox is explained by the fact that the two driving cycles enter in phase every 1470 years, thus creating pronounced peaks spaced by that period. Because the magnitude of these peaks results from constructive interference of the two driving cycles, it is indeed robust that a threshold process can be much more sensitive to these peaks than to the two original driving cycles [Chialvo, 2003]. The main strength of the GR mechanism is that – unlike the SR mechanism – it can relate the leading frequency of DO events to a main driver of natural (non-anthropogenic) climate variability, since proxies of solar activity suggest the existence of solar cycles with periods close to 1470/7 (=210) years (De Vries or Suess cycle) and 1470/17 ($\approx$86.5) years (Gleissberg cycle) [Stuiver and Braziunas, 1993, Wagner er al., 2001, Peristykh and Damon, 2003]. So far, however, no empirical evidence for this mechanism has been found [Muscheler and Beer, 2006], nor has it been shown yet that changes in solar activity over the solar cycles are sufficiently strong to actually trigger DO events. In order to investigate the stability of this mechanism we further add a stochastic component (i.e. white noise) to the forcing. In this case the events are – of course – not strictly periodic anymore. Similar to the SR case, an optimal (i.e. intermediate) noise level exists for which the waiting time distribution of the simulated events exhibits a maximum at a value of 1470 years, corresponding to the period of the fundamental frequency of the two input cycles (Fig. <ref>). In contrast to the SR case, in which a fairly simple waiting time distribution with a few broad maxima of century scale width is obtained (compare Fig. <ref>d), we now find a much more complex pattern with a large number of very sharp lines of only decadal scale width. Since the waiting time distributions of both mechanisms are considerably different, it could – at least in principle – be possible to distinguish between both mechanisms by analysing the distribution of the observed DO events. In practise, however, this approach is complicated by the fact that only about ten events appear to be sufficiently well dated for this kind of analysis, and even their spacing has an uncertainty of about 50 years [Ditlevsen et al., 2007], which is already of the same order as the width of the peaks in Fig. <ref>b. We note that the mechanism that is described in Fig. <ref> is known as ghost stochastic resonance (GSR), and its occurrence and robustness has already been reported before in other systems with thresholds and multiple states of operation [Chialvo, 2003]. At least in our system, however, this mechanism is even more complex than the other two types of resonance (SR, GR). It is beyond the scope of this paper to describe the GSR mechanism in more detail. Ghost stochastic resonance. The input consists of: 1. two sinusoidal forcing cycles with frequencies of 7/(1470 years) and 17/(1470 years), respectively, and with an amplitude of 8 mSv, 2. a random Gaussian-distributed signal with white noise power signature and a cutoff frequency of 1/(50 years), as in Fig. <ref>. In a, the relative frequency to obtain a spacing of 1470 years $\pm$1% (triangles) respectively $\pm$2% (squares) between successive events is shown as a function of the noise level $\sigma$. B shows the distribution of the spacing $\Delta$T between successive events (standard deviation of the noise: 5.5 mSv). §.§ Testing the proposed mechanisms The most direct way to test which of the proposed mechanisms – if any – provides the correct explanation for the timing of DO events would be to reconstruct decadal-scale density anomalies in the North Atlantic in connection with the events. This is not possible since even the most highly resolved oceanic records do not allow to reconstruct the variability of the glacial ocean on that time scale. Thus, only indirect tests can be performed. The identification of the postulated 1/(1470 years) forcing frequency, which has so far not been detected, would certainly give further support for the SR mechanism. And in order to support the GR mechanism, it remains crucial to demonstrate that century-scale solar irradiance variations are indeed of sufficiently large amplitude to trigger repeated transitions (with a preferred time scale of about 1470 years) between the two glacial climate states. This could be tested with climate models. An elegant and simple test is to make use of the observation that DO events in the Earth system model CLIMBER-2 represent the nonlinear response to the forcing, and that an additional (and much smaller) linear response is superimposed on the events. In the absence of any threshold crossing (e.g. in the Holocene, during which DO events did not occur) the response to the forcing, in contrast, does not show a strong nonlinear component. This suggests that Holocene climate archives from the North Atlantic region might be able to reveal what triggered glacial DO events. This approach has two major advantages: First, more reliable (e.g. better resolved and dated) records are available to solve this issue. Second, linear analysis methods can be used for that purpose, e.g. linear correlations. In the context of the GR mechanism, for example, the existence of a pronounced correlation between Holocene climate indices from the North Atlantic and solar activity proxies (reconstructed e.g. from $^{14}$C variations in precisely dated tree rings) would be expected. Up to now at least one study exists that supports this prediction of a linear relationship between century-scale solar forcing and North Atlantic climate variability throughout the Holocene: Proxies of drift ice anomalies in the North Atlantic show a persistent correlation and a statistically significant coherency with “rapid (100- to 200-year), conspicuously large-amplitude variations” in solar activity proxies [Bond et al., 2001], in accordance with the proposed GR mechanism. The most challenging test, however, is the direct analysis of the glacial climate records. We are convinced that one of the main difficulties in this approach is the high degree of nonlinearity of the events, which – according to our interpretation – has so far not been adequately addressed in many previous studies. For example, several attempts have already been made in order to investigate the 1470-year cycle by means of linear spectral analysis methods, and significance levels have commonly been calculated by assuming a red noise background. To us this assumption seems to be oversimplified, since the system responds at a preferred time scale even when driven by white noise (compare Fig. <ref>b). We thus suspect that the significance levels obtained by this method are unrealistically high. We further think that the reported lack of a clear phase relation between solar activity proxies and DO events [Muscheler and Beer, 2006] cannot rule out the idea that solar forcing synchronised DO events, since in the case of an additional stochastic forcing component (i.e. in the GSR case) the events are triggered by the combined effect of solar forcing and noise. Thus, the observed lack could also imply that only some of the events were in first place triggered by the Sun, whereas others were caused mainly by random variability (e.g. by noise). A new and promising approach, which is based on a Monte-Carlo method, has recently been proposed in order to test the significance of the glacial 1470-year climate cycle: Ditlevsen et al., 2007 define a certain measure in order to distinguish between different hypotheses for the timing of DO events, and they explicitly calculate the value of this measure for the series of events observed in the ice core. They then compare the calculated value with the values obtained by several hypothetic processes, e.g. by a random process (for which assumptions concerning the probability distribution of the recurrence times of the events have to be made). Significance levels are obtained from the (numerically estimated) probability distributions of the measure as generated by the considered process. Although we do not share their conclusions (because we think that more adequate measures can be chosen, which give considerably different results) we think that this approach is elegant because significance levels are not calculated based on linear theories. The method is thus also applicable to highly nonlinear time series. A major hurdle in this method is that for each considered process the probability distribution of the waiting times – which is unknown for almost all processes – somehow has to be specified. For example, Ditlevsen et al. use a simple mathematical (i.e. an exponential) distribution in order to mimic random DO events. In order to improve their novel approach, some method is thus needed to calculate waiting time distributions in response to any possible input. Comprehensive models are not applicable, due to their large computational cost. Our conceptual model, in contrast, is well designed for that purpose because it is combines the ability to mimic the complex nonlinearity of DO events as described by an accepted Earth system model with the extremely low computational cost of a very simple (zeroth order) model. We thus think that our work is an important step in order to develop improved statistical analysis methods which are able to cope with the extreme nonlinearity of DO events. [Discussion and conclusions] We here discussed DO events in the framework of a very simple conceptual model that is based on three key assumptions, namely (i) the existence of two different climate states, (ii) a threshold process and (iii) an overshooting in the stability of the system at the start and the end of the events, which is followed by a millennial-scale relaxation. These assumptions are in accordance with paleoclimatic records and/or with simulations performed with CLIMBER-2, a more complex Earth system model. In a couple of systematic tests we showed (in the Supplementary Material) that despite its simplicity, our model very well reproduces DO events as simulated with CLIMBER-2, whose dynamics is based on a (albeit reduced) description of the underlying hydro-/thermodynamic processes. The correspondence between both models thus strengthens our interpretation that the conceptual model can successfully mimic key features of DO events, and that these can be regarded as a new type of non-equilibrium oscillation (i.e. as an overshooting relaxation oscillation) between two states of a nonlinear system with a threshold. Although we discussed our model dynamics in the context of the (thermohaline) ocean circulation, our model does not explicitly assume that DO events are linked with changes in the ocean circulation: Threshold behaviour and multiple states exist in many compartments of the climate system (not only in the ocean, but e.g. also in the atmosphere and in the cryosphere). Our model thus cannot rule out a leading role of non-oceanic processes in DO oscillations. The millennial time scale of the events (which is represented in our model by the assumption of a millennial-scale relaxation), however, corresponds to the characteristic time scale of the thermohaline circulation and thus points to a key role of the ocean in DO oscillations. The main strength of our model is its simplicity: Due to the obvious relationship between forcing and response, the model can demonstrate why even a simple bistable (or excitable) system with a threshold can respond in a complex way to a simple forcing, which consists of only one or two sinusoidal inputs and noise. We applied our model to discuss two highly nonlinear and apparently counterintuitive resonance mechanisms, namely stochastic resonance and ghost resonance. In doing so we reported a new form of stochastic resonance (i.e. an overshooting stochastic resonance), in which the overshooting of the system leads to a further synchronisation effect compared to the usual stochastic resonance. Our study provides the first explicitly reported manifestation of ghost resonance in a geophysical (model) system. Since threshold behaviour and multiple equilibria are not unique to DO events but exist in many geophysical systems, we would indeed expect that ghost resonances could be inherent in many geosystems and not just in our model. In addition to its applicability to demonstrate and interpret nonlinear resonance mechanisms, and to test their stability, we further illustrated the ability of our conceptual model to simulate probability measures (e.g. waiting time distributions, which are required in order to test the significance and the cause of the proposed glacial 1470-year climate cycle by means of Monte-Carlo simulations). Because it combines the ability to reproduce essential aspects of DO events with the extremely low computational cost of a conceptual model (which is up to 10$^{7}$ times lower than in the Earth system model CLIMBER-2), we think that our model represents an important advance in order to develop adequate nonlinear methods for improved statistical analyses on DO events. § APPENDIX: DESCRIPTION OF CLIMBER-2 The Earth system model CLIMBER-2, which we used for our analysis, has dynamic components of the atmosphere, of the oceans (including sea ice) and the vegetation. Dynamic ice sheets were not included in our study. CLIMBER-2 is a global model with coarse resolution: For the atmosphere and the continents the spatial resolution is 10$^{\circ}$ in latitude, and 7 sectors are considered in longitude. The ocean is zonally averaged with a latitudinal resolution of 2.5$^{\circ}$ for the three large ocean basins. A detailed description of the model is given in the publication of Petoukhov et al., 2000. DO events in the model represent abrupt switches between two different climate states (stadial [i.e. cold] and interstadial [i.e. warm]), corresponding to two different modes of the THC: In the interstadial mode, North Atlantic deep water (NADW) forms at about 65 $^\circ$N and much of the North Atlantic is ice-free. In the stadial mode, NADW forms at about 50 $^\circ$N and a considerably larger area of the North Atlantic is ice-covered. We note that for the climatic background conditions of the Last Glacial Maximum (LGM) only the stadial mode is stable in the model whereas the interstadial mode is excitable but unstable [Ganopolski and Rahmstorf, 2001]. Moreover, the stability of both modes depends on the actual climate state (e.g. on the configuration of the Laurentide ice sheet and on the freshwater input into the North Atlantic), and the stability properties of the system change when the background conditions are modified (more precisely, the system can be bistable or mono-stable). Transitions between both modes can be triggered by anomalies in the density field of the North Atlantic, for example by variations in the surface freshwater flux (since the density of ocean water increases with increasing salinity). In our study we thus implement the forcing as a perturbation in the freshwater flux (in the latitudinal belt 50-70 $^{\circ}$N): We start the model with the climatic background conditions of the Last Glacial Maximum (LGM). Following earlier simulations [Braun et al., 2005] we then add a small constant offset of 17 mSv ($1~\unit{mSv} = 10^{3}~\unit{m^{3}/s}$) to the freshwater flux. For this climate state (which we label perturbed LGM) the THC is in fact bistable and DO events can be triggered more easily than for LGM conditions. This perturbed LGM state gives us the background conditions for the model simulations as presented in this paper. The authors thank R. Calov, A. Mangini, S. Rahmstorf, K. Roth and A. Witt for discussion, and P. Ditlevsen (in particular for observing the difference between the usual stochastic resonance and our overshooting stochastic resonance) and two anonymous reviewers for helpful comments. H. Braun was funded by Deutsche Forschungsgemeinschaft, DFG project number MA 821/33. [Alley and Clark, 1999] Alley, R. B. and Clark, P. U., The deglaciation of the northern hemisphere: A global perspective, Ann. Rev. Earth Planet. Sci., 27, 149–182, 1999. [Alley et al., 1999] Alley, R. B., Clark, P. U., Keigwin, L. D., and Webb, R. S., Making sense of millennial-scale climate change, in: Mechanisms of Global Climate Change at Millennial Time Scales, edited by: Clark, P. U., Webb, R. S., and Keigwin, L. D., pp. 385–394, AGU, Washington, DC, 1999. [Alley et al., 2001] Alley, R. B., Anandakrishnan, S., and Jung, P., Stochastic resonance in the North Atlantic, Paleoceanography, 16, 190–198, 2001a. [Alley et al., 2001] Alley, R. B., Anandakrishnan, S., Jung, P., and Clough, A., Stochastic resonance in the North Atlantic: Further insights, in: The Oceans and Rapid Climate Change: Past, Present and Future, edited by Seidov, D., Maslin, M., Haupt, B. J., pp. 57–68, AGU, Washington, DC, 2001b. [Alley et al., 2003] Alley, R. B., Marotzke, J., Nordhaus, W. D., Overpeck, J. T., Peteet, D. M., Pielke Jr., R. A., Pierrehumbert, R. T., Rhines, P. B., Stocker, T. F., Talley, L. D., and Wallace, J. M., Abrupt Climate Change, Science, 299, 2005–2010, [Benzi et al., 1982] Benzi, R., Parisi, G., Sutera, A., and Vulpiani, A., Stochastic resonance in climatic change. Tellus, 34, 10–16, 1982. [Bond et al., 2001] Bond, G., Kromer, B., Beer, J., Muscheler, R., Evans, M. N., Showers, W., Hoffmann, S., Lotti-Bond, R., Hajdas, I., and Bonani, G., Persistent Solar Influence on North Atlantic Climate During the Holocene. Science, 294, 2130–2136, 2001. [Braun et al., 2005] Braun, H., Christl, M., Rahmstorf, S., Ganopolski, A., Mangini, A., Kubatzki, C., Roth, K., and Kromer, B., Possible solar origin of the 1,470-year glacial climate cycle demonstrated in a coupled model, Nature, 2005, 438, 208–211. [Broecker et al., 1985] Broecker, W. S., Peteet, D. M., and Rind, D., Does the ocean-atmosphere system have more than one stable mode of operation? Nature, 315, 21–26, 1985. [Broecker et al., 1990] Broecker, W. S., Bond, G., Klas, M., Bonani, G., and Wolfli, W., A salt oscillator in the glacial Atlantic? 1. The concept, Paleoceanography, 5, 469–477, 1990. [Buldu et al., 2003] Buldu, J. M., Chialvo, D. R., Mirasso, C. R., Torrent, M. C., and Garcia-Ojalvo, J., Ghost resonance in a semiconductor laser with optical feedback, Europhysics Letters, 64, 178–184, 2003. [Centurelli et al., 2006] Centurelli, R., Musacchioa, S., Pasmanterc, R. A., and Vulpiani, A., Resemblances and differences in mechanisms of noise-induced resonance, Physica A, 360, 261–273, 2006. [Chialvo et al., 2002] Chialvo, D. R., Calvo, O., Gonzalez, D. L., Piro, O., and Savino, G. V., Subharmonic stochastic synchronization and resonance in neuronal systems, Physical Review E, 65, 050902, [Chialvo, 2003] Chialvo, D. R., How we hear what is not there: A neuronal mechanism for the missing fundamental illusion, Chaos, 13, 1226-1230, 2003. [Claussen et al., 2002] Claussen, M., Mysak, L. A., Weaver, A. J., Crucifix, M., Fichefet, T., Loutre, M.-F., Weber, S. L., Alcamo, J., Alexeev, V. A., Berger, A., Calov, R., Ganopolski, A., Goose, H., Lohmann, G., Lunkeit, F., Mokhov, I. I., Petoukhov, V., Stone, P., Wang, Z., Earth system models of intermediate complexity: closing the gap in the spectrum of climate system models. Clim. Dyn., 18, 579–586, 2002. [Clemens, 2005] Clemens, S. C., Millennial-band climate spectrum resolved and linked to centennial-scale solar cycles. Quat. Sci. Rev., 24, 521–531, 2005. [Dansgaard et al., 1982] Dansgaard, W., Clausen, H. B., Gundestrup, N., Hammer, C. U., Johnsen, S. F., Kristinsdottir, P. M., and Reeh, N., A New Greenland Deep Ice Core, Science, 218, 1273–1277, 1982. [Ditlevsen et al., 2005] Ditlevsen, P. D., Kristensen, M. S., Andersen, K. K., The recurrence time of Dansgaard-Oeschger events and possible causes, J. Clim., 18, 2594–2603, 2005. [Ditlevsen et al., 2007] Ditlevsen, P. D., Andersen, K. K., Svensson, A., The DO-climate events are probably noise induced: statistical investigation of the claimed 1470 years cycle, Clim. Past, 3, 129–134, 2007. [Gammaitoni et al., 1998] Gammaitoni, L., Hänggi, P., Jung, P., and Marchesoni, F., Stochastic resonance. Rev. Mod. Phys., 70, 223–288, 1998. [Ganopolski and Rahmstorf, 2001] Ganopolski, A. and Rahmstorf, S., Simulation of rapid glacial climate changes in a coupled climate model, Nature 409, 153–158, 2001. [Ganopolski and Rahmstorf, 2002] Ganopolski, A. and Rahmstorf, S., Abrupt glacial climate changes due to stochastic resonance, Phys. Rev. Lett., 88, 038501, 2002. [Grootes et al., 1993] Grootes, P. M., Stuiver, M., White, J. W. C., Johnsen, S., and Jouzel, J., Comparison of oxygen isotope records from the GISP2 and GRIP Greenland ice cores, Nature, 366, 552–554, 1993. [Grootes and Stuiver, 1997] Grootes, P. M. and Stuiver, M., Oxygen 18/16 variability in Greenland snow and ice with 10$^{3}$ to 10$^{5}$-year time resolution. J. Geophys. Res., 102, 26455–26470, 1997. [Keeling and Whorf, 2000] Keeling, C. D. and Whorf, T. P., The 1,800-year oceanic tidal cycle: A possible cause of rapid climate change. Proc. Natl. Acad. Sci. USA, 97, 3814–3819, 2000. [Leuenberger et al., 1999] Leuenberger, M., Schwander, J., and Johnsen, S., 16$^\circ$C Rapid Temperature Variations in Central Greenland 70,000 Years Ago, Science, 286, 934–937, 1999. [Muscheler and Beer, 2006] Muscheler, R. and Beer, J., Solar forced Dansgaard/Oeschger events? Geophys. Res. Lett., 33, L20706, 2006. [Oeschger et al., 1984] Oeschger, H., Beer, J., Siegenthaler, U., Stauffer, B., Dansgaard, W., Langway, Jr., C. C., Late glacial climate history from ice cores, in: Climate Processes and Climate Sensitivity, edited by Hansen, J. E. and Takahashi, T., pp. 299–306, AGU, Washington, DC, 1984. [Peristykh and Damon, 2003] Peristykh, A. N. and Damon, P. E., Persistence of the Gleissberg 88-yr solar cycle over the last 12,000 years: Evidence from cosmogenic isotopes. J. Geophys. Res., 108; 1003, 2003. [Petoukhov et al., 2000] Petoukhov, V., Ganopolski, A., Brovkin, V., Claussen, M., Eliseev, A., Kubatzki, C., and Rahmstorf, S., CLIMBER-2: A climate system model of intermediate complexity. Part I: Model description and performance for present climate, Clim. Dyn., 16, 1–17, 2000. [Rahmstorf, 2002] Rahmstorf, S., Ocean circulation and climate during the past 120,000 years, Nature, 419, 207–214, 2002. [Rahmstorf, 2003] Rahmstorf, S., Timing of abrupt climate change: a precise clock, Geophys. Res. Lett., 30, 1510–1514, 2003. [Rahmstorf and Alley, 2002] Rahmstorf, S. and Alley, R. B., Stochastic resonance in glacial climate, EOS, 83, 129–135, 2002. [Rial, 2004] Rial, J. A., Abrupt Climate Change: Chaos and Order at Orbital and Millennial Scales, Glob. Planet. Change, 41, 95–109, 2004. [Sakai and Peltier, 1997] Sakai, K. and Peltier, W. R., Dansgaard-Oeschger oscillations in a coupled atmosphere-ocean climate model, J. Clim., 10, 949–970, 1997. [Sarnthein et al., 1994] Sarnthein, M., Winn, K., Jung, S. J. A., Duplessy, J. C., Labeyrie, L., Erlenkeuser, H., and Ganssen, G., Changes in East Atlantic Deepwater Circulation over the Last 30,000 Years: Eight Time Slice Reconstructions, Paleoceanography, 9, 209–267, 1994. [Schulz, 2002] Schulz, M., On the 1470-year pacing of Dansgaard-Oeschger warm events. Paleoceanography, 17, 1014–1022, 2002. [Schulz et al., 2002] Schulz, M., Paul, A., and Timmermann, A., Relaxation oscillators in concert: A framework for climate change at millennial timescales during the late Pleistocene. Geophys. Res. Lett., 29, 2193–2197, 2002. [Severinghaus and Brook, 1999] Severinghaus, J. P. and Brook, E., Abrupt Climate Change at the End of the Last Glacial Period Inferred from Trapped Air in Polar Ice, Science, 286, 930–934, 1999. [Stuiver and Braziunas, 1993] Stuiver, M. and Braziunas, T. F., Sun, ocean, climate and atmospheric CO2: An evaluation of causal and spectral relationships, Holocene, 3, 289–305, 1993. [Taylor et al., 1997] Taylor, K. C., Mayewski, P. A., Alley, R. B., Brook, E. J., Gow, A. J., Grootes, P. M., Meese, D. A., Saltzman, E. S., Severinghaus, J. P., Twickler, E. S., White, J. W. C., Whitlow, S., and Zielinski, G. A., The Holocene-Younger Dryas Transition Recorded at Summit, Greenland, Science, 278, 825–827, 1997. [Timmermann et al., 2003] Timmermann, A., Gildor, H., Schulz, M., and Tziperman, E., Coherent Resonant Millennial-Scale Climate Oscillations Triggered by Massive Meltwater Pulses. J. Clim., 16, 2569–2585, 2003. [van Kreveld et al., 2000] van Kreveld, S. A., Sarnthein, M., Erlenkeuser, H., Grootes, P., Jung, S., Nadeau, M. J., Pflaumann, U., and Voelker, A., Potential links between surging ice sheets, circulation changes and the Dansgaard-Oeschger cycles in the Irminger Sea, 60-18 kyr, Paleoceanography, 15, 425–442, 2000. [Wagner er al., 2001] Wagner, G., Beer, J., Masarik, J., Muscheler, R., Kubik, P. W., Mende, W., Laj, C., Raisbeck, G. M., and Yiou, F., Presence of the solar de Vries cycle ($\approx$205 years) during the last ice age, Geophys. Res. Lett., 28, 303–306, 2001. [Yiou et al., 1997] Yiou, P., Fuhrer, K., Meeker, L. D., Jouzel, J., Johnsen, S., and Mayewski, P. A., Paleoclimatic variability inferred from the spectral analysis of Greenland and Antarctic ice-core data. J. Geophys. Res., 102, 26441–26454, 1997.

arxiv-papers

2024-09-04T02:48:54.630468

0803.4449

General relativistic plasma in higher dimensional space time D. Panigrahi 111Relativity and Cosmology Research Centre, Jadavpur University, Kolkata - 700032, India , e-mail: dibyendupanigrahi@yahoo.co.in , Permanent Address : Kandi Raj College, Kandi, Murshidabad 742137, India and S. Chatterjee222Relativity and Cosmology Research Centre, Jadavpur University, Kolkata - 700032, India, and also at NSOU, New Alipore College, Kolkata 700053, e-mail : chat_ sujit1@yahoo.com Correspondence to : S. Chatterjee ###### Abstract The well known (3+1) decomposition of Thorne and Macdonald is invoked to write down the Einstein-Maxwell equations generalised to (d+1) dimensions and also to formulate the plasma equations in a flat FRW like spacetime in higher dimensions (HD). Assuming an equation of state for the background metric we find solutions as also dispersion relations in different regimes of the universe in a unified manner both for magnetised(un) cold plasma. We find that for a free photon in expanding background we get maximum redshift in 4D spacetime, while for a particular dimension it is so in pre recombination era. Further wave propagation in magnetised plasma is possible for a restricted frequency range only, depending on the number of dimensions. Relevant to point out that unlike the special relativistic result this allowed range evolves with time. Interestingly the dielectric constant of the plasma media remains constant, not sharing the expansion of the background, which generalises a similar 4D result of Holcomb-Tajima in radiation background to the case of higher dimensions with cosmic matter obeying an equation of state . Further, analogous to the flat space static case we observe the phenomenon of Faraday rotation in higher dimensional case also. KEYWORDS : cosmology; higher dimensions; plasma PACS : 04.20, 04.50 +h ## 1\. Introduction While great strides have been made by general relativists to address the issues coming out of the recent observations in the field of astrophysics and cosmology and despite the fact that more than 90 percent of the cosmic stuff in stellar interior and intergalactic spaces is made up of matter in plasma state the much sought after union between the plasma dynamics and general relativity still remains elusive. Although we occasionally come across stray works like ‘plasma suppression of large scale structure formation in the universe’ [1] as also even the simulation techniques [2] where it is argued that the geometry of the spacetime can be purported to be structured by observing sound waves in primordial plasmas the fact remains that as the field equations both in general relativity and plasma dynamics are highly nonlinear it is very difficult to obtain closed form solutions in physics. So either numerical relativity or a linearised approximation of the plasma equations is preferred, before going in for more complicated non linear phenomena. If we briefly trace the thermal history of the universe theorists believe that from approximately $t=10^{-3}$ s to $t=1$ s at temperatures $T>10^{10}$ K there was an electron positron plasma at ultra relativistic temperatures. With cooling the plasma comprises mainly electron and hydrogen ions, may be with small amount of helium and other light elements in thermal equilibrium with photons. This period has come to be known as radiation dominated era because it is believed that the energy density of photons much exceed that of matter [3]. As the temperature decreased further with expansion at around $t\sim 10^{13}$ s recombination of ionized hydrogen atoms took place with consequent decoupling of matter and radiation and at a certain stage the universe becomes matter dominated. Consequently the temporal behaviour of the scale factor correspondingly changes with the evolution of matter at different time scales. Nevertheless, the studies of the effects of this expansion on the electromagnetic interactions of the matter, including the longitudinal and transverse modes have not, so far got the legitimate attention it deserves. Another area of current interest is the role of an external magnetic field in many astrophysical systems and cosmology and possible sources for the field in different scales [4] although the origin of the field continues to evade plausible physical explanations [5] so far. Granted that temperature (particularly at early universe) is a known enemy of magnetism there are no compelling reasons why magnetic fields should not have been present in the early Universe either. Indeed, the presence of large scales magnetic fields in our observed Universe is a well established experimental fact. Since their first evidence in diffuse astrophysical plasmas beyond the solar corona [4, 5] magnetic fields have been detected in our galaxy and in our local group through Zeeman splitting and through Faraday rotation measurements of linearly polarized radio waves. The Milky Way possesses a magnetic field whose strength is of the order of the microgauss corresponding to an energy density roughly comparable with the energy density today stored in the Cosmic Microwave Background Radiation (CMBR) energy spectrum peaked around a frequency of 30 GHz. Faraday rotation measurements of radio waves from extra-galactic sources also suggest that various spiral galaxies are endowed with magnetic fields whose intensities are of the same order as that of the Milky Way. The existence of magnetic fields at even larger scales (intergalactic scale, present horizon scale, etc.) cannot be excluded, but it is still quite debatable since, in principle, dispersion measurements (which estimate the electron density along the line of sight) cannot be applied in the intergalactic medium because of the absence of pulsar signals. Given the fact that a seed field does exist its amplification mechanism through the so called dynamo effect is relatively well understood [6] for an expanding cosmological model with the help of magnetohydrodynamical (MHD) equations [7]. On the other hand when discussing a primordial magnetic field one should also bear in mind that a large value would create significant anisotropy of the background geometry [8] with consequent impact on CMBR findings. As the large scale isotropy of the CMBR fairly excludes that possibility there should be efficient mechanism to rapidly damp that field. On the other hand there has been, of late, a resurgence of interests in physics in higher dimensional spacetime [9] in its attempts to unify all the forces in nature, to give a physical explanation of the current accelerating era of the universe without bringing in any hypothetical quintessencial type of scalar field [10] by hand, in the newly fashionable area of brane cosmology [11] where the gravity is supposed to act in the bulk while other forces in the physical 3D space. It has also received serious attention in the recently proposed induced matter theory pioneered by Wesson and others [12]. Most importantly both higher dimensional spacetime and cosmological plasma have one thing in common -_both are very relevant in the context of early universe_. While early universe may be loosely viewed as the history of evolution of matter in plasma state it can also be shown that starting from a higher dimensional phase the Einstein’s generalized field equations dictate results such that as time evolves the 3D space expands while the extra dimensions shrink till plackian length when some stabilizing mechanism ( for example, quantum gravity, casimir effect, a repulsive potential) halts the shrinkage down to a very small length, say planckian size as to be invisible with the low energy physics at the moment. So the world around us appears manifestly three dimensional. But it should be emphasized that the time scales for the spontaneous self compactification (SSC) and the onset of nucleosynthesis clearly differ with SSC occurring much earlier. So many of our findings in MHD lose much of their relevance when working in higher dimensional spacetime. While literature abounds with works on the effects of the expanding background on the matter distribution and vice versa as also on a large number of other physical processes scant attention has been paid so far to address the issues resulting from the expanding universe on the propagation of say, electromagnetic wave as also its interactions in a plasma media. To authors’ knowledge Holcomb and Tajima (HT) [13, 14], Banerjee et al [15] and later Dettmann et al [16] made important contributions in this regard. HT investigated the electromagnetic wave propagation in a radiation dominated and later in a matter dominated background both with or without any plasma material and also generalised it to the magnetized case both warm and cold. For the ultra relativistic case it is observed that all the modes redshift at the same rate i.e., photons are, so to say, self similar. Though not explicitly pointed out in their works we think this, however, may be a direct consequence of the conformal flatness of the FRW metric chosen by them. Later Banerjee et al discussed this in a little more general way. Dettmann et al got the similar results starting from a kinetic theory approach. HT’s findings for the matter dominated model, however, differ sharply from the first paper in the sense that while in the unmagnetised plasma case the photons redshift identically but here for the Alfven waves the frequency redshifts in a bizarre fashion unlike the case of free photons, depending on the magnitudes of the plasma density and also strength of the external magnetic field. On the other hand Dettmann shows that even for the unmagnetised plasma the plasma oscillations and the photons do not share the identical temporal dependence if they are decoupled. They, however, treated the whole situation from kinetic energy considerations. In the present work we have investigated the plasma dynamics in an expanding higher dimensional background in a very general way. Taking a (d+1) dimensional flat FRW type of metric as background, which one of us [17] derived earlier in a different context we first take the case of propagation of an electromagnetic wave in vacuum. To make things very general we assume an equation of state for the background as $p=\gamma\rho$. Taking $\gamma=\frac{1}{d}$, (as we are dealing with a (d+1) dimensional case) for the radiation and $\gamma=0$, for the dust case in the resulting solution we observe that solutions are very similar to the special relativistic form except that the frequencies red shift depending on $\gamma$ and the field amplitude is no longer a constant decaying with the expansion rate, reminiscent of the acoustic case where the damping occurs due to some form of dissipating mechanism. Here the expansion of the background, in a sense, takes the role of dissipation, causing this type of damping. It is observed that with number of dimensions the red shift decreases, being maximum in the usual 4D case. Moreover, red shift is less in the post recombination era compared to the early universe as expected. After briefly carrying out the so called $(3+1)$ decomposition of the Maxwell’s equations generalised to $(d+1)$ dimensional spacetime in section 2 we investigate the propagation of an electromagnetic wave in vacuum in section 3. In section 4 the electrostatic oscillations are very briefly discussed for a cold plasma and the dispersion relations obtained both for transverse and longitudinal modes. In section 5 we discuss in some detail the propagation of an electromagnetic wave in cold plasma in the presence of an ambient and homogeneous magnetic field both parallel and perpendicular to the wave propagator k. The presence of the magnetic field introduces newer and interesting modes of oscillation, creating Left and Right circularly polarized waves, resulting in the well known classical phenomenon of Faraday rotation. We also find that the wave propagation in the plasma is possible for certain range of frequencies only and this range critically depends on the number of dimensions. The paper ends with a short discussion in section 6. ## 2\. Field Equations and its Formalism We extend here the (3+1) decomposition of GR as formulated by Arnowitt, Deser and Misner (ADM) [18] to a higher dimensional space time of (d+1) dimensions. The ADM formalism was developed mainly to address the issue associated with numerical relativity as also quantization of gravity fields, specially when the space time has considerable symmetry. This work is connected with plasma physics in curved space time. To make use of the intuition from the known results of MHD in flat spacetime it is preferable to spilt the ordinary electromagnetic field tensor $F^{\mu\nu}$ into electric and magnetic fields $E$ and $B$ in terms of which the field equations are more familiar. As the split formalism has been extensively discussed and used in the literature [13, 14, 19] we shall very briefly touch upon its salient features as extended to higher dimensions. We define a set of observers ( Fiducial Observers or FIDO ) at rest in the space spanned by the hypersurfaces of constant universal time, having a d-velocity vector field, $n$ orthogonal to spatial slices. It is well known that for a rotation-free space time ( as we are dealing here ) $n_{i;j}=\sigma_{ij}+\frac{1}{d}~{}\theta\gamma_{ij}-a_{i}n_{j}$ (1) $(i,j=1,2,3,\ldots d)$ where $\sigma_{ij}$ is the shear of the Eulerian world lines given by $\sigma_{ij}\equiv\frac{1}{2}\left(n_{i;\mu}\gamma_{j}^{\mu}+n_{j;\mu}\gamma_{i}^{\mu}\right)-\frac{1}{d}~{}\theta\gamma_{ij}$ (2) Here $\gamma_{ij}$ is d-metric spatial tensors $a^{i}=n_{;j}^{i}n^{j}$ (3) and $\theta=n_{i}^{i}(=-K)$ (4) are the usual acceleration and expansion scalars and $K$ is the trace of the extrinsic curvature, $K_{;i}^{i}$. For our metric we take the ( d+1) dimensional generalized FRW space time as $ds^{2}=dt^{2}-A^{2}\left(dx^{2}+dy^{2}+dz^{2}+d\psi_{n}^{2}\right)$ (5) (n = 5, 6, 7, …, d ) where $A\equiv A(t)$ is the scale function. For this particular metric two relevant quantities $\alpha=\frac{d\tau}{dt}$ ( the lapse function - the rate of change of fiducial proper time to that of universal time ) and also the shift vector $\beta$ ( signifying how much spatial co-ordinates are shifted as one moves from one hypersurface to the other ) naturally reduce to $\alpha=1$ and $\beta=0$. In an earlier work [17] one of us extensively discussed the ( d+1) dimensional isotropic and homogeneous space time and assuming an equation of state, $P=\gamma\rho$ found the scale factor as ( $P$ = pressure, $\rho$ = energy density ) $A\sim t^{\frac{2}{d(1+\gamma)}}=t^{n},~{}~{}~{}n=\frac{2}{d(1+\gamma)}$ (6) With the extrinsic curvature scalar defined as $K=-\theta=-d\frac{\dot{A}}{A}=-d\frac{n}{t}$ (7) we finally write down the Maxwell’s equations [20, 21, 22] (see Mcdonald et al for ( 3+1 ) split for more details ) generalized to ( d+1) dimensions as $\displaystyle\nabla.E$ $\displaystyle=$ $\displaystyle 4\pi\rho_{e}$ (8) $\displaystyle\nabla.B$ $\displaystyle=$ $\displaystyle 0$ (9) $\displaystyle\frac{\partial E}{\partial t}$ $\displaystyle=$ $\displaystyle KE+cA^{-1}\nabla\times B-4\pi J$ (10) $\displaystyle\frac{\partial B}{\partial t}$ $\displaystyle=$ $\displaystyle KB-cA^{-1}\nabla\times E$ (11) $\displaystyle\frac{\partial\rho_{e}}{\partial t}$ $\displaystyle=$ $\displaystyle K\rho_{e}-\nabla.J~{}\textrm{(charge ~{} conservation)}$ (12) and finally the particle equation of motion in $(d+1)$ dimensional as $\frac{DA^{d-1}p}{D\tau}=A^{d-1}q\left(E+A\frac{v}{c}\times B\right)$ (13) or $\frac{Dp}{D\tau}=\frac{d-1}{d}Kp+q\left(E+A\frac{v}{c}\times B\right)$ (14) where $\frac{D}{D\tau}=\frac{1}{\alpha}\left(\partial_{t}+v.\nabla\right)$ (15) is the convective derivative and the d- momentum $p=m_{e}\Gamma v$ (16) ( $m_{e}$ is the rest mass, $\Gamma$ is the boost factor, and v, the d-velocity ). We thus see that for the simple metric given by equation(5)the decomposed ( d+1)- dimensional Maxwell equations closely mimic the flat space counterparts with some additional inputs from curved geometry( e.g., A and K terms). Here $\nabla\cdot$ and $\nabla\times$ are the ordinary Minkowskian divergence and curl in Cartesian co-ordinates. In what follows we shall consider, for simplicity, the small amplitude linear theory such that the convective derivative simply reduces to ordinary derivative, $\frac{d}{dt}$. ## 3\. Electromagnetic Waves in Vacuum With the set of equations split to ( d + 1) formalism we are now in a position to attempt applications in varied plasma phenomena. To start with we discuss briefly the propagation of an electromagnetic wave in free space in the expanding back ground. Using equations (9-11) we get via $\nabla\cdot E=0$ (for vacuum) the wave equation as $A^{2}\ddot{E}+\left(2d+1\right)A\dot{A}\dot{E}+\left(d^{2}\dot{A}^{2}+dA\ddot{A}\right)E=c^{2}\nabla^{2}E$ (17) Assuming separation of variables in electric field $E(\textbf{x},t)=E_{t}E_{r}$ (18) we finally get $\ddot{E_{t}}+(2d+1)\frac{\dot{A}}{A}\dot{E_{t}}+\left[d\left(\frac{\ddot{A}}{A}+d\frac{\dot{A}^{2}}{A^{2}}\right)+\frac{k_{i}^{2}c^{2}}{A^{2}}\right]E_{t}=0$ (19) which, through equation (6) finally reduces to $t^{2}\ddot{E_{t}}+(2d+1)nt\dot{E_{t}}+\left[dn\left(dn+n-1\right)+k_{i}^{2}c^{2}t^{2(1-n)}\right]E_{t}=0$ (20) (here $k_{i}$ is a separation constant) corresponding to some initial fiducial time $t_{i}$. We shall subsequently see that $k_{i}$ is also identified with the $d$\- dimensional wave vector. On the other hand, as we are dealing with a homogeneous world the spatial equation remains unchanged yielding a solution $e^{i(k_{i}\cdot~{}r)}$ as in special theory of relativity. A little algebra shows that the time equation is reducible to a Bessel equation of order $\frac{1}{2}$ as in the 4D case. Thus dimensionality or the equation of state has apparently no role in determining the order of the equation. Plugging everything together we get $\textbf{E}=E_{0}\hat{e}t^{\frac{1-(2d+1)n}{2}}H_{\frac{1}{2}}^{(2)}\left[\frac{t^{1-n}}{1-n}k_{i}c\right]e^{ik_{i}.r}$ (21) where $H_{\frac{1}{2}}^{(2)}$ is a Hankel function of order $\frac{1}{2}$. Replacing the asymptotic form of $H_{\frac{1}{2}}$ we get $E=E_{0}i\sqrt{\frac{2k_{i}cd(1+\gamma)}{\pi\\{d(1+\gamma)-2\\}}}t^{-\frac{2}{1+\gamma}}e^{-\frac{ik_{i}cd(1+\gamma)}{d(1+\gamma)-2}t^{\frac{d(1+\gamma)-2}{d(1+\gamma)}}}e^{ik_{i}.r}$ (22) It represents a d-dimensional ‘damped’ harmonic wave as one encounters in mechanical vibration. While in mechanical motion the damping occurs due to friction here the expansion of the universe seemingly causes some sort of damping. For pre recombination era in $(d+1)$ dimensional space time $\gamma=\frac{1}{d}$ and the damping factor is $A^{-d}$. Hence the damping of the wave amplitude apparently decreases with number of dimensions. This finding merits some explanation. We have remarked earlier that the amplitude decay is somewhat geometrical in nature caused by the expansion and curvature. In that case as with dimension the expansion rate decreases one expects that damping should be larger in 4D but a little inspection of the last relation shows that when we plug in the expression of ‘$A$’ (equation (6)) the last relation further reduces to $t^{-\frac{2d}{d+1}}$. So the damping actually increases in higher dimensional spacetime. On the other hand, for a fixed $d$ the damping factor is $t^{-2}$ for post recombination era ($\gamma=0$) (see figure 1). Alternatively for the case of a very large number of dimensions the damping asymptotically reaches $t^{-2}$, a form set for post recombination era. It has not escaped our notice that the scaling of $E$ or $B$ for $\gamma=0$ is independent of the number of dimensions unlike the radiation case. So the amplitude factor gets increasingly damped as the universe ages. The fact that damping of the $E$ or $B$ increases with the number of dimensions in the early universe has a number of interesting theoretical implications. Firstly we mention in the introduction that in order that the universe evolves isotropically according to the FRW model there should be efficient mechanisms to damp the primordial magnetic field as early as possible. In that respect the HD spacetime has some inherent advantage over the standard 4D in the sense that the damping is faster in HD. Secondly the magnetic field at small scales may influence the bigbang nucleosynthesis and change the primordial abundances of light elements by significantly changing the expansion rate of the universe at the corresponding time. The success of the standard BBN scenario can provide an interesting set of bounds on the intensity of the magnetic field at that epoch [5], indirectly constraining the number of dimensions of the spacetime. At this stage it may not be out of place to call attention to the fact that most of the above findings are of theoretical nature only and it is not feasible to relate them to current astrophysical data. Because multidimensional cosmological models lose much of their relevance well before the onset of bigbang nucleosynthesis and current observational findings can be explained for all practical purposes if the cosmological evolution be modelled along the standard four dimensional spacetime. If as usual we set $k_{i}c=\omega_{i}$ ( the angular frequency of the wave at some initial time $t=t_{i}$ ) then the above equation may be rewritten as Figure 1: _ $\mid E_{d}\mid\sim t$ graph_ $\displaystyle E$ $\displaystyle=$ $\displaystyle E_{0}i\sqrt{\frac{2k_{i}cd(1+\gamma)}{\pi\\{d(1+\gamma)-2\\}}}t^{-\frac{2}{1+\gamma}}e^{-i\omega_{i}t\frac{d(1+\gamma)}{d(1+\gamma)-2}t^{-\frac{2}{d(1+\gamma)}}}e^{ik_{i}.r}$ (23) $\displaystyle=$ $\displaystyle E_{0}i\sqrt{\frac{2k_{i}cd(1+\gamma)}{\pi\\{d(1+\gamma)-2\\}}}t^{-\frac{2}{1+\gamma}}e^{-i\frac{d(1+\gamma)}{d(1+\gamma)-2}\omega_{d}t}e^{ik_{i}.r}$ where $\omega_{d}=\omega_{i}t^{-\frac{2}{d(1+\gamma)}}$ (24) gives a measure of the red shift of the photon due to background expansion. For radiation dominated era $\gamma=\frac{1}{d}$ , $\omega_{d}=\omega_{i}t^{-\frac{2}{d+1}}$, so the rate at which the frequency decreases is maximum in 4D universe. Moreover damping is greater in radiation era (see figure 2). Figure 2: _ $\omega_{d}\sim t$ graph. As the number of dimension increases red shift decreases_ Returning again to the equation (22) we see that the horizon for our metric is given by $L_{d}=\int\frac{cdt}{t^{\frac{2}{d(1+\gamma)}}}=\frac{d(1+\gamma)}{d(1+\gamma)-2}ct^{\frac{d(1+\gamma)-2}{d(1+\gamma)}}$ (25) so the equation may be recast as $E=E_{0}(x,t)e^{i({k.r-k_{i}.L_{d}})}=E_{0}(x,t)e^{ik_{i}(\hat{k}x-L)}$ (26) exactly similar to the Newtonian result where the horizon is simply $L=ct$. ## 4\. Electrostatic Oscillation In this section we shall very briefly consider an electromagnetic wave in a 2-component plasma. For simplicity we assume a small amplitude wave such that, $v\times B=0$ because it is of second order in perturbed quantities. This, in turn, allows us to neglect the motion of ions. Skipping intermediate mathematical steps for space we get via equations (14) and (16) for our metric (5) $v=\frac{iqt^{\frac{2}{d(1+\gamma)}}}{m_{e}\Gamma\omega_{i}}E=-\frac{ie}{m_{e}\Gamma\omega_{d}}E$ (27) The last equation is very similar to the flat space case except that here, $\omega_{d}$ is not a constant but shares the background expansion. When one uses the last equation in Maxwell equation (10) through $J=n_{0}qv$ we arrive at Coulomb’s law $\nabla\times B=-\frac{i\omega_{i}}{c}\epsilon E$ (28) where $\epsilon(\omega_{d})=1-\frac{\omega_{pT}^{2}}{\omega_{d}^{2}}$ (29) is the dielectric constant of the plasma medium with the suffix ${}^{\prime}T^{\prime}$ signifying transverse mode. The $\omega_{pT}^{2}$ is related to the well known plasma frequency [23], $\displaystyle\omega_{p}^{2}$ $\displaystyle=$ $\displaystyle\frac{b_{d}n_{0}q^{2}}{m_{e}}~{},~{}~{}~{}\omega_{pT}^{2}\sim\frac{\omega_{p}^{2}}{\Gamma}$ $\displaystyle b_{d}$ $\displaystyle=$ $\displaystyle\frac{2^{\frac{d}{2}}\pi^{\frac{d}{2}}}{(d-1)!!}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}\mathrm{(d~{}~{}~{}even)},$ $\displaystyle=$ $\displaystyle\frac{2^{(\frac{d+1)}{2}}\pi^{\frac{(d-1)}{2}}}{(d-2)!!}~{}~{}~{}~{}~{}~{}~{}~{}~{}\mathrm{(d~{}~{}odd)}$ Skipping mathematical details it can also be shown that the transverse and the longitudinal modes give the dispersion relations as $\omega_{T}^{2}=\omega_{p}^{2}+c^{2}k^{2}$ (30) $\omega_{L}^{2}=\omega_{p}^{2}$ (31) These relations are pretty well known in the special relativistic case, excepting that here all the quantities depend on time as well as the total number of dimensions. Apparently the equation (29) has the same Newtonian form but both the frequencies depend on the scale factor, ‘$A$’ which, again, is a function of both the number of dimensions and the equation of state chosen.To end the section let us investigate the time dependence of the dielectric constant in equation(29). Now, $\omega_{pT}^{2}$ should share the time evolution of the background electron number density, $n_{0}$ ( the inverse of volume of the universe) i.e., $n_{0}\sim A^{-d}\sim t^{-\frac{2}{1+\gamma}}$. Again, from (27) we get, $v\Gamma\sim A^{n(1-d)}$. The equation (10) further dictates that $v\sim A^{-1}$, which gives $\Gamma\sim A^{(2-d)}$. So, $\omega_{pT}^{2}\sim\frac{n_{0}}{\Gamma}\sim A^{-2}\sim t^{-\frac{4}{d(1+\gamma)}}$. On the other hand the equation (24) implies that $\omega_{d}^{2}\sim A^{-2}$, hence $\epsilon$($\omega_{d}$) does not explicitly depend on time. This is a remarkable result in the sense that for a FRW type of metric the dielectric constant is a real constant irrespective of not only the total number of dimensions but also on the equation of state i.e., $\epsilon$($\omega_{d}$) continues to remain constant all through the evolution of the universe. ## 5\. Electromagnetic Oscillations in Cold Plasma In this section we investigate the situation where a plasma in thermodynamic equilibrium is slightly disturbed through the passage of an electromagnetic wave. We assume that an external ambient magnetic field is also present. We, however, assume the plasma medium to be cold so that the pressure can be neglected when considering the particle equation of motion. In stellar systems one often encounters situations where relaxation times are much larger than the age of the universe so that collisions ( hence pressure ) may be neglected. The effect of an electric field is not generally seriously considered because of the well known Debye shielding effect. The general problem of an electromagnetic wave propagating along an arbitrary direction with the external magnetic field is given by Appleton and Hartee in the Newtonian case when studying the propagation of radio waves in ionosphere. Holcomb [14] studied in FRW metric a specialised situation of the A-H equation in the dust case. Considering the fact that a general solution with arbitrary $\theta$ is very difficult to tackle in an expanding background with arbitrary number of dimension we shall restrict ourselves to the cases when the electromagnetic wave propagates parallel and perpendicular to the magnetic field. However the topic is of great importance in astrophysics and space science where electromagnetic wave propagation in magnetized plasma is very relevant. Case I ($\vec{B}\parallel\textbf{k}$) : we assume that the external, uniform magnetic field and the wave vector $\mathbf{k}$ are both aligned along the $i^{th}$ direction (say z direction with $i=3$) in the $d$\- dimensional space, such that $\textbf{k}=|k|\mathbf{e_{z}}$ and $\mathbf{B}=|B|\mathbf{e_{z}}$. As is customary in the analogous 3-dimensional static space we also assume that all the perturbed quantities have the same time dependence given by equation (23) such that the linearized equation of motion (13) takes the form $i\omega_{d}m_{e}v=e\left(E+A\frac{v}{c}\times B\right)$ (32) [ Here E is $\perp$r to $k$ and considering that we are dealing with a $(d+1)$ dimensional space time it has components $E_{1}$, $E_{2}$, $E_{3}$, $E_{4}$, …, $E_{j}$ . ] Replacing $\frac{\partial E}{\partial t}$ via equation (23) by $\frac{\partial E}{\partial t}=-\left[i\omega_{d}+\frac{2}{(1+\gamma)t}\right]E$ (33) which, when plugged in equation (10) gives, after a long but fairly straight forward calculation gives for j=1 $\displaystyle\left(\nabla\times B\right)_{1}$ $\displaystyle=$ $\displaystyle-\frac{i\omega_{d}}{c}A\left[\left(1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}}\right)E_{1}-\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}\frac{\omega_{c}^{2}}{\omega_{d}^{2}}E_{1}+i\sum_{j=2}^{d}\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}\frac{\omega_{c}}{\omega_{d}}E_{j}\right]$ $=-i\frac{\omega_{d}}{c}\left[\left(1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}\right)E_{1}+i\sum_{j=2,j\neq 1}^{d}\frac{\omega_{c}}{\omega_{d}}\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}E_{j}\right]$ (34) For the case $j=3$ (i.e., along the direction of the magnetic field ) it takes a simple form $\left(\nabla\times B\right)_{3}=-i\frac{\omega_{d}}{c}\left(1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}}\right)E_{3}$ (35) repeating the process for the remaining $(d-2)$ components we can write for the $\mu^{th}$ component a tensorial relation as $\left(\nabla\times B\right)_{\mu}=-i\frac{\omega_{d}}{c}\epsilon_{\mu\nu}E_{\nu}$ (36) ($\mu,\nu=1,2,3,\ldots,d$). where $\epsilon_{\mu\nu}$ is rank 2 skew symmetric tensor of order ‘d’. A little inspection shows that $\displaystyle\epsilon_{11}$ $\displaystyle=$ $\displaystyle\epsilon_{22}=\epsilon_{44}=\epsilon_{55}=\ldots=\epsilon_{dd}=1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}=p_{1}\textrm{(say)}$ (37) $\displaystyle\epsilon_{12}$ $\displaystyle=$ $\displaystyle\epsilon_{14}=\epsilon_{15}=\ldots=\epsilon_{1d}=\frac{\omega_{c}}{\omega_{d}}\frac{\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{c}^{2}}=p_{2}$ (38) $\displaystyle\epsilon_{31}$ $\displaystyle=$ $\displaystyle\epsilon_{32}=\epsilon_{34}=\ldots=\epsilon_{3d}=0$ (39) $\displaystyle\epsilon_{33}$ $\displaystyle=$ $\displaystyle 1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}}=p_{3}$ (40) so the $(d\times d)$ permitivity tensor comes out to be $\epsilon_{\mu\nu}=\left(\begin{array}[]{cccccc}p_{1}&ip_{2}&0&ip_{2}&.&ip_{d}\\\ -ip_{2}&p_{1}&0&ip_{2}&.&0\\\ 0&0&p_{3}&0&.&0\\\ -ip_{2}&-ip_{2}&0&p_{1}&.&0\\\ .&.&.&.&.&.\\\ -ip_{d}&.&.&.&.&p_{1}\\\ \end{array}\right)$ (41) Here $\omega_{p}$ is the plasma frequency given by $\omega_{p}^{2}=\frac{b_{d}n_{0}e^{2}}{m_{e}}$ (42) and the electron cyclotron frequency is given by Figure 3: _ $\omega_{c}\sim t$ graph for both radiation and matter dominated era. As the number of dimension increases $\omega_{c}$ decreases. Further the decay is sharper in dust case compared to radiation case. _ $\omega_{c}=\frac{eB}{m_{e}c}t^{\frac{2}{d(1+\gamma)}}\equiv\frac{e\hat{B}}{m_{e}c}$ (43) where $\hat{B}$, the orthogonal magnitude of the ambient magnetic field is given by $\hat{B}=|(B)_{z}(B)^{z}|^{1/2}=Bt^{\frac{2}{d(1+\gamma)}}$ for our system (see figure 3). If the magnetic field is switched off $(\omega_{c}=0)$ the equation (34) reduces to $\left(\nabla\times B\right)_{j}=-\frac{e\omega_{i}}{c}\epsilon E_{j}$ (44) exactly similar to the expression(28) of the section 4. Thus the introduction of the magnetic field generates varied modes transforming the dielectric constant scalar $\epsilon$ in equation (35) to a second rank tensor $\epsilon_{ij}$. Although the equations (30) - (40) exactly resemble the analogous expressions in Newtonian theory the fact remains that that all the frequencies now depend on time rather than being constant. Further the cyclotron frequency $\omega_{c}$ decays as $t^{-\frac{2(d-1)}{d(1+\gamma)}}$ exactly similar to the orthogonal component of the magnetic field. If we take the curl of the equation (36) and replace $\nabla$ by $ik\hat{e_{z}}$, after a long but fairly straight forward calculation we are led to the matrix form $\left(\begin{array}[]{cccccc}(1-\frac{p_{1}}{\mathbf{n}^{2}})&-i\frac{p_{2}}{\mathbf{n}^{2}}&0&0&.&0\\\ i\frac{p_{2}}{\mathbf{n}^{2}}&(1-\frac{p_{1}}{\mathbf{n}^{2}})&0&0&.&0\\\ 0&0&-\frac{p_{3}}{\mathbf{n}^{2}}&0&.&0\\\ .&.&.&.&.&.\\\ .&.&.&.&.&.\\\ 0&0&0&0&.&(1-\frac{p_{1}}{\mathbf{n}^{2}})\\\ \end{array}\right)\left(\begin{array}[]{c}E_{1}\\\ E_{2}\\\ E_{3}\\\ .\\\ .\\\ E_{d}\\\ \end{array}\right)=\left(\begin{array}[]{c}0\\\ 0\\\ 0\\\ .\\\ .\\\ 0\\\ \end{array}\right)$ (45) where $\mathbf{n}^{2}=\frac{c^{2}k^{2}}{\omega_{i}^{2}}$ (46) where $\mathbf{n}$ is the refractive index of the plasma medium. Three modes are possible. First the longitudinal mode characterized by $E_{1}=E_{2}=E_{4}=\ldots=E_{d}=0$, ($E_{3}\neq 0$ and $p_{3}=0$). Since the displacement is along $Z$ direction the magnetic field has no role to play and $\omega_{d}=\omega_{p}$. As we are more interested in the dynamics of the electromagnetic waves rather than the plasma oscillation as such we take $E_{3}=0$ in equation (45). Setting the determinant of the resulting $(d-1)\times(d-1)$ matrix in equation (45) to zero we get $\mathbf{n}^{2}=p_{1}\pm p_{2}$ (47) The plus sign gives via equation (36) $E_{\mu}-\emph{i}E_{\nu}=0~{}~{}~{}~{}~{}(\mathrm{\mu\neq\nu\neq 3}~{}~{}~{},~{}~{}~{}\emph{i}=\sqrt{-1})$ (48) such that $E_{l}=\left(\hat{e_{\mu}}-\emph{i}\hat{e_{\nu}}\right)e^{i(k_{L}z-\omega_{d}t)}$ (49) corresponding to left circularly polarized wave. The wave number $k_{L}$ can be found from equations (37, 38, 46, 47) as $k_{L}=\frac{\omega_{d}}{c}\left[1-\frac{\omega_{p}^{2}}{\omega_{d}(\omega_{d}+\omega_{c})}\right]^{1/2}$ (50) On the other hand for the minus sign in (47) we get $E_{R}=\left(\hat{e_{\mu}}+\hat{e_{\nu}}\right)e^{[\emph{i}(k_{R}z-\omega t)]^{1/2}}$ (51) representing RCP wave with $k_{R}=\frac{\omega_{d}}{c}\left[1-\frac{\omega_{p}^{2}}{\omega_{d}(\omega_{d}-\omega_{c})}\right]^{1/2}$ (52) It is clear that the two eigen modes have different phase and group velocities and unlike the former case the Right Circularly Polarized (RCP) wave has a resonance at $\omega=\omega_{f}=\omega_{c}$ where the phase velocities vanish. The expressions so far exactly resemble the ones found in the propagation of an electromagnetic wave with an ambient magnetic field in Newtonian mechanics. However, here all the quantities $\omega_{d}$, $\omega_{p}$ etc. depend on time, a dependence modelled by the form of line-element, the number of dimensions and also the equation of state. It should be noted that the time dependence of $\omega_{d}$ and $\omega_{c}$ are different being $\omega_{d}\sim t^{-\frac{2}{d(1+\gamma)}}$ and $\omega_{c}\sim t^{-\frac{2(d-1)}{d(1+\gamma)}}$. So there is no fixed resonant frequency as in Newtonian case but with time it changes. It also follows from equation (52) that for $\omega_{d}=\omega_{1}=\frac{1}{2}\left[\omega_{c}+\sqrt{\omega_{c}^{2}+4\omega_{p}^{2}}\right]$ (53) the wave vector $k$ vanishes and our analysis breaks down. So the wave propagates for $\omega_{d}<\omega_{c}$ and $\omega_{1}<\omega_{d}$, otherwise it becomes evanescent. Moreover, as the temporal dependence of $\omega_{c}$ and $\omega_{d}$ are different the magnitude of the allowed region changes. With dimensions $\omega_{c}$ decays more sharply than $\omega_{d}$. Thus the propagation of the electromagnetic wave is more restricted in higher dimensions than the usual 4D. Returning to the Left Circularly Polarized (LCP) wave we see that the wave vector vanishes for $\omega_{d}=\omega_{2}=\frac{1}{2}\left[-\omega_{c}+\sqrt{\omega_{c}^{2}+4\omega_{p}^{2}}\right]$ (54) and so the wave propagates for $\omega_{d}>\omega_{2}$ From what has been discussed above it is tempting to look for Faraday rotation (see ref. 4 for recent astrophysical data) analogous to the Newtonian case. Assuming that an electromagnetic wave traverses a distance $z$ in a plasma medium with a magnetic field subject to the restriction on frequencies discussed above the Faraday rotation is given by $\theta=\frac{k_{L}-k_{R}}{2}Z$ (55) It should be noted that one should revert to the physical co-ordinate rather than the co-moving one we are considering here. Accordingly $Z_{ph}=t^{-\frac{2}{d(1+\gamma)}}$, $Z_{cm}$ and $\theta$ finally comes out to be via equation (23) $\theta=\frac{k_{Li}-k_{Ri}}{2}Z_{ph}$ (56) with no dependence on time. so apparently the number of dimensions and the equation of state have no impact on this classical result. It may be relevant to mention that measurements of the radio waves from the extra galactic sources suggest that various spiral galaxies are endowed with magnetic fields whose intensities are of the same order of magnitude as that of Milky way [4] i.e., of the order of microgauss corresponding to an energy density stored today in CMBR energy spectrum peaked around a frequency of 30 GHz. Case II ($\vec{B}\bot\vec{k}$) : Let us very briefly consider the case of a plasma with a uniform magnetic field $\textbf{B}=B_{0}\hat{e}_{z}$, through which an electromagnetic wave is propagating with propagation vector $\vec{\textbf{k}}=k\hat{e_{x}}$, perpendicular to the magnetic field. Here two modes are possible. As the mathematical exercise closely resembles the case I we shall totally skip intermediate steps to write the final form as 1\. First mode (called ordinary wave) with displacements in z direction ( i.e. $\parallel B$) having the dispersion relation $\omega_{d}^{2}=\omega_{p}^{2}+k^{2}c^{2}$ (57) as the magnetic field has no influence for motion parallel to itself the equation (57) is exactly same as equation (30) for the electrostatic oscillation. 2\. Second mode (called extraordinary wave) with displacements in (d-1)-dimensional hypersurface ($\bot B$) having dispersion relation $\frac{c^{2}k^{2}}{\omega_{d}^{2}}=1-\frac{\omega_{p}^{2}}{\omega_{d}^{2}}\frac{\omega_{d}^{2}-\omega_{p}^{2}}{\omega_{d}^{2}-\omega_{p}^{2}-\omega_{c}^{2}}$ (58) Before ending the section a final remark may be in order. We know that pulsars are rotating neutron stars giving out pulses of radio waves periodically, which are affected by the interstellar medium during their propagation to reach us. If the interstellar medium has a component of magnetic field parallel to propagation direction then as shown earlier the plane of polarization will suffer Faraday rotation depending on frequency, having a spread in the rotation angle. This spread may have, _in principle at least_ , some imprint on the nature of expanding universe. ## 6\. Discussion With the help of (3+1) formalism the Einstein-Maxwell and the electrodynamical equations are written for a (d+1) dimensional FRW-like spacetime in presence of plasma and linearised equations are solved for different phases of the universe.The analysis essentially generalises to HD the well known results of Holcomb and Tajima. The salient features of our analysis may be summarised as: 1\. For a propagating wave in HD in vacuum the photons redshift most in 4D and for a fixed $d$ in radiation dominated model. 2\. Although the plasma is sharing the expansion of the background the dielectric constant remains a true constant. So the photons are in a sense self similar. This result was found earlier by Holcomb and Tajima. We here generalise this remarkable result to the case of extra dimensional spacetime and also for a fluid obeying a general equation of state. It may be tempting to suggest that the fact that the classical flat space result of the constancy of the dielectric constant is carried over to non static curved background and that too in higher dimensions may be due to the conformal flatness of the particular metric analysed here. So one should move with caution against any far fetched generalisation and in other complicated space time this result may not be true. 3\. In the presence of an external magnetic field many interesting oscillation modes manifest themselves. A simplified Appleton-Hartee type of solution generalised to higher dimensions is obtained in curved spacetime. Only a selected range of frequencies are available for propagation here. 4\. The well known phenomenon of Faraday rotation is obtained. To end a final remark may be in order. The present work suffers from two serious disqualifications. For sake of mathematical simplicity we work out everything assuming a linearised plasma theory. Conditions under which one may assume linearized plasma theory may well exist in Newtonnian theory, but we are not being able to clearly formulate those things for the case of a nonflat spacetime and that too when it is expanding. Secondly most observational evidences suggest that even if one starts with a higher dimensional phase the universe underwent the self compactification transition much earlier than the epoch when the big bang nucleosynthesis sets in. So although literature abounds with works (for example, higher dimensional black holes and its thermodynamics etc.) studying the standard electromagnetic as well as MHD laws in the framework of multidimensions becomes a sort of _suspect_. In that sense our analysis is more of a purely theoretical nature without much direct physical applications. In future work one should try to generalise these results in the realm of non linear plasma and also attempt to relate some of our findings to known astrophysical data. Acknowledgment : One of us(SC) acknowledges the financial support of UGC, New Delhi for carrying out this work. ## References * [1] Pisin Chen and Kwang- Chang Lai, 2007 _Phys. Rev. Lett._ 99 231302 * [2] G F Smoot, 2007 _Rev. Mod. Phys._ 79 1370 * [3] S Banerji and A Banerjee, 2007 _General Relativity and Cosmology_ , Elsevier * [4] P P Kronberg, 1994 _Rep.Prog.Phy_ 57 325 * [5] M Gasperini, M Giovanini and G Veneziano,1995 _Phy. Rev.Lett._ 75 3796 * [6] R Beck, A Brandenberg, D Moss, A A Shukurov and D. Sokoloff,1996 _Annu. Rev. Astron. Astrophys._ 34 155 * [7] C G Tsagas and J D Barrow , 1997 _Class. Quant. Grav._ 14 2539 * [8] Y B Zeldovich, A A Ruzmaikin and D D Sokoloff, 1983 _Magnetic Fields in Astrophysics_ , ( Mcgraw Hill), N Y * [9] D Hooper and S Profumo, ‘Dark Matter and Collider Phenomenology of Universal Extra Dimensions’, 2007 _Physics Reports_ 453 p 27 - 115 * [10] S Chatterjee, A Banerjee and Z H Zhang, 2006 _Int. J. Mod. Phys._ A21 4035 ; N Banerjee and S Das, 2006 _Mod. Phys. Lett._ A 21 2663 * [11] U Debnath, A Banerjee and S Chakraborty , 2004 _Class. Quant. Grav._ 21 5609 * [12] P S Wesson, 1999 _Space Time Matter_( World Scientific), Singapore * [13] K A Holcomb and T Tajima, 1989 _Phy. Rev._ D40 3809 * [14] K A Holcomb, 1990 _Astrophysical. J._ 362 381 * [15] A Banerjee, S Chatterjee, A Sil and N Banerjee, 1994 _Phy. Rev._ D50 1161 * [16] C P Dettmann, N E Frankel, 1993 _Phys. Rev._ D48 5655 * [17] S Chatterjee and B Bhui, 1990 _Mon. Not. R. Astron. Soc._ 247 57 * [18] R Arnowitt, S Deser and C W Misner, in Gravitation: 1962 _An Introduction to current Research_ , edited by L witten ( Wiley) New York * [19] X H Zhang, 1989 _Phy. Rev._ D39 2933 * [20] D A Macdonald and K Thorne, 1982 _Mon. Not. R. Astron. Soc._ 198 345 * [21] K Thorne and D A Macdonald, 1982 _Mon. Not. R. Astron. Soc._ 198 339 * [22] C R Evans and J F Hawley, 1988 _Astrophysical J._ 332 659 * [23] V M Emelyanov, Yu P Nikitin, I L Rozenbal and A V Berkov , 1996 _Phys.Rep._ 143 p 1-68

arxiv-papers

2024-09-04T02:48:54.637607

0803.4459

# Report on GRG18, Session A3 Mathematical Studies of the Field Equations Lars Andersson∗ laan@aei.mpg.de Albert Einstein Institute, Am Mühlenberg 1, D-14476 Potsdam, Germany and Department of Mathematics, University of Miami, Coral Gables, FL 33124, USA ###### Abstract. In this report I will give a summary of some of the main topics covered in Session A3, mathematical studies of the field equations, at GRG18, Sydney. Unfortunately, due to length constraints, some of the topics covered at the session will be very briefly mentioned or left out altogether. The summary is mainly based on extended abstracts submitted by the speakers and some of those who presented posters at the session. I would like to thank all participants for their contributions and help with this summary. ∗ Supported in part by the NSF, under contract no. DMS 0407732. ## 1\. The Buchdahl inequality The Buchdahl inequality, which is included in most textbooks on general relativity, states that for a static, self-gravitating body, (1) $2M/R\leq 8/9,$ where $M$ is the ADM mass and $R$ the area radius of the boundary of the static body. The proof by Buchdahl, cf. [16] of (1) assumed that the energy density is non-increasing outwards and that the pressure is isotropic. A bound on $2M/R$ has an immediate observational consequence: if $2M/R<8/9$ then the gravitational red shift is less than 2 but if $2M/R$ approaches 1 the red shift is unbounded. The assumptions used to derive the inequality are very restrictive, and as e.g. pointed out by Guven and Ó Murchhada [27] neither of them hold in a simple soap bubble and they do not approximate any known topologically stable field configuration. In addition to the restrictions implied by the hypotheses made by Buchdahl there are two other disadvantages associated with this inequality: it refers to the boundary of the body (i.e. the interior is excluded) and the solution which gives equality in the inequality is the Schwarzschild interior solution which has constant energy density and for which the pressure blows up at the centre. In particular it violates the dominant energy condition. Håkan Andréasson explained recent work [2], where all four restrictions described above are eliminated. He considered matter models for which the energy density $\rho$ and the radial pressure $p$ is non-negative and which satisfies $p+2p_{T}\leq\Omega\rho,$ where $\Omega$ is a non-negative constant and $p_{T}$ is the tangential pressure, and showed that (2) $\sup_{r>0}\frac{2m(r)}{r}\leq\frac{(1+2\Omega)^{2}-1}{(1+2\Omega)^{2}}.$ Here $m$ is the quasi-local mass so that $M=m(R).$ The case $\Omega=1$ gives the bound $2m/r\leq 8/9$. Among the matter models which satisfy the conditions stated are Vlasov matter and matter which satisfies the dominant energy condition and has non-negative pressure. Inequality (2) is sharp in the sense of measures and there are examples which come arbitrarily close to saturating the inequality. Andréasson has recently generalized the inequality to the case of charged spheres [3]. ## 2\. Initial-boundary value problems A long standing problem in the analytical and numerical study of the Einstein equations is the choice of boundary conditions for boundaries near “infinity”. As is well known, the Einstein equations in harmonic coordinates reduces to a quasilinear system of wave equations. It was proved in previous work by Kreiss and Winicour cf. [36] that the initial-boundary value problem is well posed for such systems with the Sommerfeld outgoing boundary condition. The proof relied on pseudo-differential operator techniques. Oscar Reula reported on a recent proof [35] of this fact which makes use only of partial integrations. The proof is much more transparent than the previous proof, and the argument based on partial integration is well adapted to the analysis of discretizations of the wave equation based on difference operators which have good “partial summation” properties. Let $(M,g_{ab})$ be a spacetime with timelike boundary $\mathcal{T}$. Let $M$ be foliated by spacelike surfaces $\Sigma_{t}$, with timelike unit normal $n^{a}$. Consider the initial boundary value problem for the wave equation $g^{ab}\nabla_{a}\nabla_{b}\phi=F$ with boundary condition of the form $(T^{b}+aN^{b})\nabla_{b}\phi\big{|}_{\mathcal{T}}=q$, $a>0$, where $T^{a}$ is a timelike unit vector field which is tangent to $\mathcal{T}$ and $N^{a}$ is the outward pointing normal to the boundary $\partial\Sigma_{t}=\Sigma_{t}\cap\mathcal{T}$ in $\Sigma_{t}$. The Sommerfeld condition is given by the choice $a=1$, which corresponds to a condition on the derivative of $\phi$ in an outgoing null direction. The proof proceeds by considering the fluxes of energy defined with respect to a vector field $u^{a}=T^{a}+\delta N^{a}$, where $\delta>0$ is a parameter to be chosen. Let $\Theta^{a}{}_{b}=\nabla^{a}\phi\nabla_{b}\phi-\frac{1}{2}\delta^{a}{}_{b}\nabla_{c}\phi\nabla^{c}\phi$ be the energy-momentum tensor for the scalar field. The energy $E$ and outward flux $\mathcal{F}$ are defined by integration over $\Sigma_{t}$ and $\partial\Sigma_{t}$, with respect to the densities $u^{b}\Theta^{a}{}_{b}n_{a}$ and $u^{b}\Theta^{a}{}_{b}N_{a}$, respectively. Using integration by parts one derives by a straightforward argument an energy estimate bounding $\partial_{t}E$ in terms of $E$, $F$ and $\mathcal{F}$. Expanding out the flux density $u^{b}\Theta^{a}{}_{b}N_{a}$ and using the boundary condition to eliminate terms, one finds that for an appropriate choice of $\delta=\delta(a)>0$, it contains a negative multiple of the square gradient of $\phi$ on $\partial\Sigma$ plus terms which can be estimated in terms of $E,F,\mathcal{F}$. Inserting this in the above mentioned energy estimate yields a form of the energy estimate which gives the boundary regularity required for well-posedness. Based on these ideas, the proof of well-posedness for the initial boundary can now be completed along standard lines. ## 3\. Moving punctures and black holes During the time since GRG17, Dublin 2004, there have been tremendous advances in numerical relativity, in particular in the binary black hole problem. The first breakthrough was the work by Frans Pretorius, using a constraint stabilized harmonic coordinate formulation, with excision. Using this code he was during the spring of 2005 able to evolve stably several orbits of a binary black hole system. This was followed shortly thereafter by announcements from several groups using different methods, that they were able to achieve similar results. One of the methods which has emerged as a successful alternative to the harmonic formulation is based on the Baumgarte-Shapiro-Shibata-Nakamura (BSSN) formulation of the Einstein evolution equations, cf. [49, 7], see also [48]. Due to various issues arising from a combination of several factors (gauge choice, excision etc.), codes based on BSSN initially suffered from instabilities which prevented more than partial orbits to be calculated. However when the proper techniques are applied one may successfully use the BSSN formulation to stably evolve black hole spacetimes, even without excision. The black holes (BHs) are thus included in the domain of computation, and one effectively treats the BHs as essentially weak singularities in the Cauchy slices. This approach is now one of the most popular in the numerical GR community. In spite of the success in using this formulation, the reason why it actually works has remained mysterious. Sascha Husa discussed recent work, cf. [28, 15], which sheds light on this issue. The moving puncture method starts with BHs represented by “filling” their interiors with wormholes, reduced to singular (“puncture”) points by spatial compactification. The moving-puncture method [17, 6] is only a minor modification of previous approaches, in that no attempt is made to factor out the singular puncture geometry during the simulation, but surprisingly in numerical evolutions of a Schwarzschild BH using the BAM code [15], Husa et al. found that this leads to a drastic change in the geometry of the time slices [28]. They have constructed an analytic solution for the stationary state of a nonspinning BH that matches the moving-puncture gauge, and found that the numerical data asymptote to this solution. The key result about the geometry of moving punctures is that for a Schwarzschild black hole the numerical evolutions approach a stationary slice that neither reaches an internal asymptotically flat end nor hits the physical singularity, as might be expected for a stationary slice with non-negative lapse. Rather, the slice ends at a throat at finite Schwarzschild radius ($R_{0}\approx 1.31M$), but infinite proper distance from the apparent horizon. This changes the singularity structure of the “puncture”. It is still a puncture in that there is a coordinate singularity at a single point in the numerical coordinates, but it does not correspond to an asymptotically flat end. In the course of Schwarzschild evolutions Husa et al. have found that the throat does collapse to the origin. Where one would have expected an inner and an outer horizon, they find only one zero in the norm of $(\frac{\partial}{\partial t})^{a}$, corresponding to the outer horizon. An under-resolved region does develop in the spacetime (it is the region between the throat and the interior spacelike infinity), but we are pushed out of causal contact with it. The throat itself has receded to infinite proper distance from the outer horizon. Matter fields or gravitational radiation will be trapped between the inner horizon and the throat, because unlike the gauge their propagation is limited by the speed of light; Husa et al. will consider this issue in future work. ## 4\. Isolated systems The analysis of the asymptotic structure of asymptotically flat spacetimes is of central importance in mathematical general relativity. The conformal compactification setting introduced by Penrose allows one to give a stringent analysis of the asymptotic properties of spacetimes, including conserved quantities. Further, the weak cosmic censorship conjecture has a clear formulation in this setting. The work of Klainerman et al. [18, 33] shows that asymptotic fall-off conditions on Cauchy data can be introduced so that the Cauchy development has a conformal compactification with any finite regularity. It is known that there are large classes of spacetimes which have a conformal compactification which is regular at spatial infinity. It is an interesting open question whether the assumption of smooth null infinity implies some type of rigidity for the spacetime [21, 19]. The regular conformal field equations of Friedrich give a clear cut formulation of this problem, and there is good hope of eventually finding sharp conditions for regularity. ### 4.1. Stability of Minkowski space The talk of Lydia Bieri addressed the global, nonlinear stability of solutions of the Einstein equations in General Relativity. In particular, she discussed results on the initial value problem for the Einstein vacuum equations, which generalizes the results of Christodoulou and Klainerman [18]. Every strongly asymptotically flat, maximal, initial data which is globally close to the trivial data gives rise to a solution which is a complete spacetime tending to the Minkowski spacetime at infinity along any geodesic. In [11], Bieri has considered the Cauchy problem with more general, asymptotically flat initial data. In particular, under the assumptions in [11], the spacetime curvature fails to be continuous. In order to show decay of the spacetime curvature and the corresponding geometrical quantities, the Einstein equations are decomposed with respect to adequate foliations of the spacetime. In the proof of Bieri, one encounters borderline estimates for certain quantities with respect to decay, indicating that the conditions concerning decay at infinity imposed on the initial data are sharp. ### 4.2. Rigidity for asymptotically simple spacetimes Juan Antonio Valiente Kroon reported on progress [53] towards a proof of the following conjecture concerning asymptotically simple spacetimes: If an asymptotically Euclidean, conformally flat, time symmetric initial data set for the Einstein vacuum equations gives rise to a development admitting a smooth conformal compactification at null infinity, then the initial data must be Schwarzschildean in a neighbourhood of infinity. It should be noted that the Schwarzschild spacetime is the only stationary spacetime admitting conformally flat slices. There are further indications of generalisations of this conjecture for more general classes of data, cf. [54, 55]. A possible proof of the above conjecture requires a detailed understanding of the structure of certain asymptotic expansions of the development of the initial data near null and spatial infinity. The tools to obtain these expansions are the “extended conformal Einstein equations” and the representation of spatial infinity known as “the cylinder at spatial infinity” which have been introduced by Friedrich in [25]. The asymptotic expansions are obtained by solving, for a given initial data set, a hierarchy of interior equations at the cylinder at spatial infinity. These interior equations allows to transport initial data from the initial hypersurface up to the “critical sets” where null infinity “touches” spatial infinity. The structure of the interior equations suggests that their solutions should contain very specific types of logarithmic divergences at the critical sets unless the initial data is Schwarzschildean. The explicit existence of these obstructions has been shown in [53]. ## 5\. Singularities ### 5.1. Stochastic aspects of generic singularities The proposal of Belinskiǐ, Khalatnikov and Lifshitz [9, 10, 32] (BKL) on the structure of generic cosmological singularities states roughly that the essential properties of the asymptotic dynamics of the gravitational field along typical timelines can be understood by considering certain spatially homogeneous models. Writing the Einstein equations in terms of Hubble normalized scale invariant frame variables due to Uggla et al. [52, 1, 47] or the approach of Damour et al. [22], based on Iwasawa decomposition, each of which relies on a long history of previous formulations, gives a description of the asymptotics of the gravitational field in terms of billiard dynamical systems. In the case of the Hubble normalized formulation, one gets a billiard in the Kasner plane, cf. [56], while for the Iwasawa formulation one gets for the case of 3+1 gravity, a billiard in a domain of hyperbolic space, which is analogous to the Misner-Chitre billiard. Claes Uggla presented work [30], building on [52, 1, 47], which generalizes and makes rigorous some aspects of the previous work of BKL and others on the stochastic aspects of the system which arise due to the chaotic nature of the asymptotic billiard systems. Using a combination of dynamical and statistical analyses and in part heuristic arguments, this work describes the generic properties of a “billiard attractor”. It turns out that several degrees of freedom, which a priori could have been of relevance for the asymptotic behavior are, generically, statistically suppressed. The dynamical and statistical arguments of Uggla et al. may be contrasted with results concerning asymptotic behavior in Bianchi type IX obtained using purely dynamical arguments [46]. The results presented by Uggla suggest that the role of Bianchi type IX as a model for the asymptotic dynamics of generic singularities should be re-evaluated, in view of the fact that generic singularities exhibit features that are not shared by Bianchi type IX. ### 5.2. Kinematic and Weyl singularities For expanding Bianchi cosmologies with a tilted perfect fluid with linear equation of state $p=(\gamma-1)\rho$, all timelike and null geodesics are complete into the future. However, the fluid congruence may be incomplete into the future (called a congruence singularity), accompanied by the blow-up of kinematic quantities associated with the fluid congruence. Woei-Chet Lim discussed recent work [20, 37] on the nature of such singularities. Much emphasis has been placed on the blow-up of the components of the Weyl tensor associated with the fluid congruence. However, as shown by examples due to Lim et al., this phenomenon is independent of the incompleteness of the fluid congruence. Hence it is necessary to differentiate congruence singularities into kinematic singularities and Weyl singularities. In particular, the fluid may become extremely titled or the kinematic or Weyl components may blow up, depending on whether $\gamma$ exceeds certain critical values. ### 5.3. Perturbations of naked singularity spacetimes The scalar field may be considered as a toy model for perturbations of a background spacetime. Brien Nolan discussed some rigorous mathematical results that probe the linear stability of naked singularities in self-similar collapse. He showed that the multipoles of a minimally coupled massless scalar field propagating on a spherically symmetric self-similar background space- time admitting a naked singularity maintain finite $H^{1,2}$ norm as they impinge on the Cauchy horizon. Further, each multipole obeys a point-wise bound at the horizon, as does its locally observed energy density [40]. The null energy condition is assumed to hold in the background spacetime. In order to view the scalar field as a toy model for perturbations of a background spacetime, the matter model must be specified. To study such perturbations, the matter content must be specified. Nolan considers what in this context is the simplest case: that of null dust - i.e. the stability of the Cauchy horizon in self-similar Vaidya space-time is studied. The results for the scalar field carry over to odd-parity linear perturbations at the level of both the metric and the curvature [41]. For even parity perturbations, the linearised Einstein equations can be reduced to a 5-dimensional first order symmetric hyperbolic system. The components of the state vector are gauge invariant metric and matter perturbation quantities. Nolan shows that the $L^{2}$ norm of the state vector blows up at the Cauchy horizon and that solutions for which the $L^{\infty}$ norm of the state vector does not blow up at the Cauchy horizon are unstable in the space of test function initial data. This provides strong evidence that the Cauchy horizon of the self-similar Vaidya space-time is unstable [39]. ### 5.4. Spacetime boundaries and properties of singularities In his talk, Benjamin Whale reported on joint work with Sue Scott, where the a-boundary and the a-boundary singularity theorem are applied to analyze the physical properties of singularities. ## 6\. Quasi-local aspects ### 6.1. Towards the quasi-localization of canonical GR Perhaps the most natural way of introducing the conserved quantities in asymptotically flat spacetimes is the canonical/Hamiltonian one. A key observation in the Hamiltonian formulation of GR, due to Arnowitt, Deser and Misner [5], is that the evolution parts of Einstein’s equations can be recovered formally as canonical equations of motion in the phase space, in which the constraint function (i.e. whose vanishing is just the constraint parts of Einstein’s equations) play the role of the Hamiltonian. However, as Regge and Teitelboim stressed, if we want to recover the correct evolution parts of Einstein’s equations as partial differential equations for smooth fields rather than some distributional generalization of them, then the Hamiltonian in the canonical equations of motion must be functionally differentiable with respect to the canonical variables [45]. Regge and Teitelboim showed that such a Hamiltonian can be obtained by adding an appropriate 2-surface integral to the constraint functions. The remarkable property of this Hamiltonian is that the ADM energy-momentum and angular momentum can be recovered as the value of the Hamiltonian on the constraint surface with appropriately chosen lapse and shift. On the other hand, the investigations of Beig and Ó Murchadha showed that for given boundary conditions on the canonical variables the asymptotic form of the lapse and the shift is already determined if we require that the evolution equations preserve the boundary conditions [8]. Moreover, they also showed that both the constraint functions and the Hamiltonians close to Poisson algebras, the former being an ideal in the latter, and their quotient, the so-called algebra of observables, is isomorphic to the Poincare algebra. The ADM conserved quantities then can be considered as appropriate coordinates on this algebra of observables. László Szabados considered in his talk the boundary conditions for canonical vacuum GR at the quasi-local level, i.e. when the spacelike hypersurface on which the canonical variables are defined is compact with smooth 2-boundary ${\mathcal{S}}$ [50] . The key ideas used by Szabados are found by analogy with the Hamiltonian analysis in the asymptotically flat case above, namely the requirement of the functional differentiability of all the functions whose Poisson bracket must be calculated, and the requirement of the compatibility of the evolution equations and the boundary conditions both on the canonical variables and the lapse and the shift. It has been shown by Szabados [51] that fixing the area element on the 2-surface ${\mathcal{S}}$ (rather than the whole induced 2-metric) is enough to have a well defined constraint algebra, a well defined Poisson algebra of basic Hamiltonians parameterized by lapses that are vanishing on ${\mathcal{S}}$ and shifts that are tangent to and divergence free on ${\mathcal{S}}$. Their quotient is a Lie algebra of a class of 2-surface observables. The evolution equations preserve these boundary conditions, and the observables (realized as the value of the basic Hamiltonians on the constraint surface) are 2+2–covariant, gauge-invariant and provide a representation of the Lie algebra of the divergence-free vector fields on ${\mathcal{S}}$. Szabados pointed out, that in special, standard situations the observables appear to behave as angular momentum. ### 6.2. The Einstein scalar field at finite infinity The ‘finite infinity’ paradigm of Ellis was proposed to study isolated gravitational systems, in our universe, where we experience the presence of other local matter fields and a cosmological background [24]. In this context it is not possible to make the assumption of asymptotic flatness at infinity, under which boundary conditions on matter have been studied in detail. Instead, a smooth timelike surface $\mathcal{F}$, is introduced. The timelike surface $\mathcal{F}$ should be located at a large finite distance from the centre of the local system of interest, with the aim to study the fields generated from this on $\mathcal{F}$. William Clavering in his talk examined the behaviour of the spherically symmetric Einstein scalar field at $\mathcal{F}$. He considered the evolution of a scalar field over a domain bounded by an initial hypersurface, characteristic curves, and $\mathcal{F}$. Using Hawking’s mass formula [29], he has studied the consequences of imposing conditions on the mass-energy flux at $\mathcal{F}$. Two examples were discussed; each a perturbation of a static exact solution. For a perturbation of Schwarzschild there is no mass-energy flux of the scalar field. Clavering conjectures that this is the case in all vacuum spacetimes. The analysis has been repeated for the pure scalar field case of the Wyman spacetime [57]. In this case, non-trivial expressions for the mass-energy flux in terms of the scalar field at $\mathcal{F}$ were obtained. ### 6.3. Quasi-local energy inside the event horizon Pointlike objects cause many of the divergences that afflict physical theories. For instance, the gravitational binding energy of a point particle in Newtonian mechanics is infinite. In general relativity, the analog of a point particle is a black hole and the notion of binding energy must be replaced by quasilocal energy. The quasilocal energy (QLE) derived by York, and elaborated by Brown and York [13], is finite outside the horizon but it was not considered how to evaluate it inside the horizon. Andrew Lundgren presented a prescription for finding the QLE inside a horizon, and showed that it is finite at the singularity for a variety of types of black hole. It turns out that the energy is typically concentrated just inside the horizon, not at the central singularity. It is impossible to localize gravitational energy due to the equivalence principle, so it is meaningless to define gravitational energy at a single point. This problem is avoided by considering the gravitational energy in a region of space. The boundary of the region is a two-dimensional surface, and in fact the quasilocal energy is defined only in terms of quantities defined on the enclosing surface (specifically the induced metric and extrinsic curvature). The QLE is defined only in terms of the gravitational degrees of freedom and makes no mention of any other fields that are present. However, since gravity couples to everything, the QLE counts all energy that is present in the region. Lundgren et al. [38] have considered only spherically-symmetric static black holes. The definition of the QLE in [13] was extended to be valid inside the event horizon, which is only a coordinate singularity. In the Schwarzschild case, the QLE at the central singularity is zero. The analogous quantity for a classical field of a point particle diverges at the center. The gravitational field itself carries energy which gravitates, causing the nonlinearity of general relativity. Lundgren remarked that the nonlinearity somehow provides a mechanism that cures the divergence expected of a point particle field. The definition of the QLE presented by Lundgren can be applied also to the deSitter case and the case with non-vanishing charge. ## 7\. Black holes ### 7.1. Black hole rigidity and spacetimes with compact Cauchy horizon Jim Isenberg reported on recent work with Vince Moncrief which studies analytic solutions of Einstein’s equations containing nondegenerate compact Cauchy horizons with non closed generators. If certain hypotheses are added, the spacetimes necessarily admit an isometry which acts along the generators of the horizon. Isenberg showed how to use these results to prove that stationary (non static) analytic black holes in arbitrary dimensions necessarily admit symmetries which are independent of the presumed stationarity symmetry. This work is closely related to work of Hollands et al. [31]. ### 7.2. Electromagnetic field on the background of high dimensional black holes In the last years various generalizations of black hole solutions to the high dimensional gravity have been found. One of them describes a generally rotating black hole with NUT charges. This solution possesses several nice properties as, e.g., the existence of the Killing-Yano tensor, the complete integrability of the geodesic motion or the separability of the Hamilton- Jacobi and Klein-Gordon equations. Its generalization including the acceleration of the black holes or the electromagnetic field is not, however, straightforward. In four dimensions it is possible to generalize the black hole solution to describe the accelerated black holes (the Plebański-Demiński solution). This transition is achieved by the conformal rescaling of the metric accompanied by a modification of some metric functions. Pavel Krtous showed in his talk that such a procedure is not sufficiently general in higher dimensions—only the maximally symmetric spacetimes in ‘accelerated’ coordinates are obtained. Further, he presented general algebraically special solutions of the Maxwell equations on the background of the mentioned high dimensional generally rotating black hole. They are adjusted to the algebraic structure of the background which is reflected by the existence of the principal Killing-Yano tensor. Such electromagnetic fields generalize the field of charged black holes in four dimensions. However, one may show that in higher dimensions such fields cannot be easily modified in such a way that they would satisfy full Maxwell-Einstein equations. ## 8\. Post-Newtonian expansions Post-Newtonian expansions (PNEs) are expansions of the Einstein equations in the parameter $1/c$, around $c=\infty$, where $c$ is lightspeed. The limit $c=\infty$ is the Newtonian limit of general relativity. Such expansions have been studied since the discovery of general relativity and there is a large literature available. Recently, post-Newtonian expansions have been used to calculate wave forms for the gravitational wave emissions from binary black hole inspirals. These have been compared with numerically calculated wave forms and turn out to be highly accurate until the last stage of the inspiral. The majority of results on PNEs are based on formal expansions which are used to calculate approximate values of physical quantities, see eg. [23, 12] and references therein. However, this formal approach does not deal with the question of convergence. In the absence of a precise notion of convergence, it becomes unclear to what extent the formal expansions actually approximate relativistic solutions. In view of the importance of the applications of PNEs, it is interesting to give a rigorous foundation to the procedure. Todd Oliynyk discussed his recent work on post-Newtonian expansions for the Einstein-Euler equations. Oliynyk considers expansions of solutions to these equations in the parameter $\epsilon=v_{T}/c$ about $\epsilon=0$, where $v_{T}$ is a characteristic velocity scale associated with the fluid matter. By rescaling the gravitational and matter variables, the Einstein-Euler equations can be written as (3) $G^{ij}=2\kappa\epsilon^{4}T^{ij}\quad\text{and}\quad\nabla_{i}T^{ij}=0$ where $\kappa$ $=$ $4\pi G\rho_{T}/v_{T}^{2}$, $v_{i}v^{i}$ $=$ $-\epsilon^{-2}$, $\rho_{T}$ is a characteristic value for the fluid density, and $t$ $=$ $x^{4}/v_{T}$ is a “Newtonian” time coordinate. In his talk, Oliynyk presented results on PNEs for a class of polytropes, which go beyond formal considerations [43, 44]. By introducing suitable renormalized variables for which the limit $\epsilon\to 0$ makes sense, and introducing a suitable gauge, Oliynyk derives a family of symmetric hyperbolic systems, depending on the parameter $\epsilon$. This system is studied in a class of weighted Sobolev spaces $H^{k}_{\delta,\epsilon}$, which interpolate between the weighted spaces $H^{k}_{\delta}$ and the standard spaces $H^{k}$. Using these spaces, it is possible to prove $\epsilon$-dependent energy estimates for solutions to the Einstein-Euler equations. These estimates are then used to prove the existence of convergent expansions in $\epsilon$ for suitably chosen initial data. Oliynyk discussed the relation between his convergent expansions and the formal PNEs. In order to recover the PNEs to a certain order requires the initial data to be chosen correctly. The method used to choose initial data is based on ideas in [14]. Oliynyk discussed how to recoved the post- Newtonian expansion to $2^{\text{nd}}$ order in his framework. ## 9\. Miscellaneous ### Wave tails in curved spacetimes Risto Tammelo reported on joint work with Laas and Mankinen dealing with the wave tails of scalar and electromagnetic fields. Their work shows that the intensity of the tail of the electromagnetic wave pulse emitted by a wave source within a compact binary in the vicinity of the geometric shadow of the source of gravitation can be of the same magnitude as the intensity of the direct pulse. The energy carried away by the tail can amount to approximately 10% of the energy of the low-frequency modes of the direct pulse. As an example of an exactly solvable model system, a scalar wave field on a particular Robertson-Walker spacetime has been considered. In the case of minimal coupling, if the metric describes the Friedman dust-dominated universe, the higher-order multipole solutions do not contain a wave tail term. A massless nonconformally-coupled scalar field is also considered in a class of Robertson-Walker spacetimes. Tammelo et al. show that an initially massless scalar field in the Robertson-Walker universe can obtain a mass in the corresponding asymptotic Minkowski space region. ### Ricci flow and the Einstein equations Eric Woolgar described work on the Ricci flow of asymptotically flat manifolds in the rotationally symmetric case [42]. This paper shows that if no minimal hypersphere is present initially, then one never forms. The flow then exists for all future time and converges to flat spacetime in the limit of infinite time, the limit being taken in the Cheeger-Gromov sense. The mass does not change during the flow, but the quasilocal mass within any fixed hypersphere about the origin (defined either by fixing the proper radius, the surface area, or the enclosed volume) evaporates away smoothly. Mohammad Akbar discussed relations between Ricci solitons and solutions to the Einstein- scalar field equations. ### Posters Several interesting abstracts had to be relegated to the poster session. Among these were abstracts by Håkan Andréasson on the formation of black holes in spherically symmetric gravitational collapse [4], Jim Isenberg on the AVTD behavior of $T^{2}$ symmetric solutions of the Einstein vacuum equations, Makoto Narita on cylindrically symmetric expanding spacetimes, Oscar Reula, [34], showing that the Robinson Trautman Maxwell equations do not constitute a well posed initial value problem, and Juan Antonio Valiente Kroon on a characterization of Schwarzschild initial data sets [26]. ## References * [1] L. Andersson, H. van Elst, W. C. Lim, and C. Uggla, _Asymptotic Silence of Generic Cosmological Singularities_ , Physical Review Letters 94 (2005), no. 5, 051101. * [2] H. Andréasson, _Sharp bounds on $2m/r$ of general spherically symmetric static objects_, 2007, arXiv.org:gr-qc/0702137. * [3] by same author, _Sharp bounds on the critical stability radius for relativistic charged spheres: I_ , 2007, arXiv.org:0708.0219. * [4] H. Andreasson, M. Kunze, and G. Rein, _The formation of black holes in spherically symmetric gravitational collapse_ , 2007, arXiv.org:0706.3787. * [5] R. Arnowitt, S. Deser, and C.W. Misner, _The dynamics of general relativity_ , Gravitation, an Introduction to Current Research (L. Witten, ed.), Wiley, New York, U.S.A., 1962, arXiv.org:gr-qc/0405109, pp. 227–265. * [6] J. G. Baker, J. Centrella, D.-I. Choi, M. Koppitz, and J. van Meter, _Gravitational wave extraction from an inspiraling configuration of merging black holes_ , Phys. Rev. Lett. 96 (2006), 111102. * [7] Thomas W. Baumgarte and Stuart L. Shapiro, _On the numerical integration of Einstein’s field equations_ , Phys. Rev. D59 (1999), 024007. * [8] R. Beig and N. Ó Murchadha, _The Poincaré group as the symmetry group of canonical general relativity_ , Ann. Phys. (N.Y.) 174 (1987), 463–498. * [9] V. A. Belinskiǐ, I. M. Khalatnikov, and E. M. Lifshitz, _Oscillatory approach to a singular point in the relativistic cosmology._ , Advances in Physics 19 (1970), 525–573. * [10] by same author, _A general solution of the Einstein equations with a time singularity._ , Advances in Physics 31 (1982), 639–667. * [11] L. Bieri, _An extension of the stability theorem of the Minkowski space in general relativity_ , Ph.D.thesis. ETH Zurich, Switzerland, 2007. * [12] L. Blanchet, G. Faye, and S. Nissanke, _Structure of the post-Newtonian expansion in general relativity_ , Phys. Rev. D (3) 72 (2005), no. 4, 044024, 10. * [13] J. D. Brown and J. W. York, Jr., _Quasilocal energy and conserved charges derived from the gravitational action_ , Phys. Rev. D47 (1993), 1407–1419. * [14] G. Browning and H.-O. Kreiss, _Problems with different time scales for nonlinear partial differential equations_ , SIAM J. Appl. Math. 42 (1982), no. 4, 704–718. * [15] B. Bruegmann, J. A. Gonzalez, M. Hannam, S. Husa, U. Sperhake, and W. Tichy, _Calibration of moving puncture simulations_ , 2006, arXiv.org:gr-qc/0610128. * [16] H. A. Buchdahl, _General Relativistic Fluid Spheres_ , Physical Review 116 (1959), 1027–1034. * [17] M. Campanelli, C. O. Lousto, P. Marronetti, and Y. Zlochower, _Accurate evolutions of orbiting black-hole binaries without excision_ , Phys. Rev. Lett. 96 (2006), 111101. * [18] D. Christodoulou and S. Klainerman, _The global nonlinear stability of the Minkowski space_ , Princeton Mathematical Series, vol. 41, Princeton University Press, Princeton, NJ, 1993. * [19] Piotr T. Chrusciel and Erwann Delay, _Existence of non-trivial, vacuum, asymptotically simple space-times_ , Class. Quant. Grav. 19 (2002), L71. * [20] A. A. Coley, S. Hervik, and W. C. Lim, _Fluid observers and tilting cosmology_ , Classical Quantum Gravity 23 (2006), no. 10, 3573–3591. * [21] Justin Corvino and Richard M. Schoen, _On the asymptotics for the vacuum Einstein constraint equations_ , J. Differential Geom. 73 (2006), no. 2, 185–217. * [22] T. Damour, M. Henneaux, and H. Nicolai, _Cosmological billiards_ , Classical Quantum Gravity 20 (2003), no. 9, R145–R200. * [23] J. Ehlers, _On limit relations between, and approximative explanations of, physical theories_ , Logic, methodology and philosophy of science, VII (Salzburg, 1983), Stud. Logic Found. Math., vol. 114, North-Holland, Amsterdam, 1986, pp. 387–403. * [24] G. F. R. Ellis, _Cosmology and local physics_ , New Astron. Rev. 46 (2002), 645–657. * [25] H. Friedrich, _Gravitational fields near space-like and null infinity_ , J. Geom. Phys. 24 (1998), no. 2, 83–163. * [26] A. Garcia-Parrado Gomez-Lobo and J. A. Valiente Kroon, _Initial data sets for the schwarzschild spacetime_ , Physical Review D 75 (2007), 024027\. * [27] J. Guven and N. Ó Murchadha, _Bounds on $2m/R$ for static spherical objects_, Phys. Rev. D (3) 60 (1999), no. 8, 084020, 8. * [28] M. Hannam, S. Husa, D. Pollney, B. Bruegmann, and N. O’Murchadha, _Geometry and regularity of moving punctures_ , 2006, arXiv.org:gr-qc/0606099. * [29] S. Hawking, _Gravitational radiation in an expanding universe_ , J. Math. Phys. 9 (1968), 598–604. * [30] J. M. Heinzle, C. Uggla, and N. Rohr, _The cosmological billiard attractor_ , 2007, arXiv.org:gr-qc/0702141. * [31] S. Hollands, A. Ishibashi, and R. M. Wald, _A higher dimensional stationary rotating black hole must be axisymmetric_ , Commun. Math. Phys. 271 (2007), 699–722. * [32] I. M. Khalatnikov, E. M. Lifshitz, K. M. Khanin, L. N. Shchur, and Ya. G. Sinaĭ, _On the stochastic properties of relativistic cosmological models near the singularity_ , General relativity and gravitation (Padova, 1983), Fund. Theories Phys., Reidel, Dordrecht, 1984, pp. 343–349. * [33] S. Klainerman and F. Nicolò, _Peeling properties of asymptotically flat solutions to the Einstein vacuum equations_ , Classical Quantum Gravity 20 (2003), no. 14, 3215–3257. * [34] C. Kozameh, O. Reula, and H.-O. Kreiss, _On the well posedness of robinson trautman maxwell solutions_ , 2007, arXiv.org:0708.1933. * [35] H. O. Kreiss, O. Reula, O. Sarbach, and J. Winicour, _Well-posed initial-boundary value problem for the harmonic einstein equations using energy estimates_ , 2007, arXiv.org:0707.4188. * [36] H.-O. Kreiss and J. Winicour, _Problems which are well posed in a generalized sense with applications to the Einstein equations_ , Classical Quantum Gravity 23 (2006), no. 16, S405–S420. * [37] W. C. Lim, A. A. Coley, and S. Hervik, _Kinematic and Weyl singularities_ , Classical Quantum Gravity 24 (2007), no. 3, 595–604. * [38] A. P. Lundgren, B. S. Schmekel, and Jr. J. W. York, _Self-renormalization of the classical quasilocal energy_ , Physical Review D (Particles, Fields, Gravitation, and Cosmology) 75 (2007), no. 8, 084026. * [39] B. C. Nolan, _Instability of the Cauchy horizon in self-similar Vaidya space-time_ , In preparation. * [40] by same author, _Bounds for scalar waves on self-similar naked-singularity backgrounds_ , Classical Quantum Gravity 23 (2006), no. 13, 4523–4537. * [41] by same author, _Odd-parity perturbations of self-similar Vaidya spacetime_ , Classical Quantum Gravity 24 (2007), no. 1, 177–200. * [42] T. Oliynyk and E. Woolgar, _Asymptotically flat ricci flows_ , to appear in Comm. Anal. Geom., 2006, arxiv:math/0607438. * [43] T.A. Oliynyk, _The newtonian limit for perfect fluids_ , to appear in Comm. Math. Phys. * [44] by same author, _Post-newtonian expansions for perfect fluids_ , in preparation. * [45] T. Regge and C. Teitelboim, _Role of surface integrals in the Hamiltonian formulation of general relativity_ , Ann. Phys. (N.Y.) 88 (1974), 286–318. * [46] H. Ringström, _The Bianchi IX attractor_ , Ann. Henri Poincaré 2 (2001), no. 3, 405–500. * [47] N. Röhr and C. Uggla, _Conformal regularization of Einstein’s field equations_ , Classical Quantum Gravity 22 (2005), no. 17, 3775–3787. * [48] Olivier Sarbach, Gioel Calabrese, Jorge Pullin, and Manuel Tiglio, _Hyperbolicity of the BSSN system of Einstein evolution equations_ , Phys. Rev. D66 (2002), 064002. * [49] M. Shibata and T. Nakamura, _Evolution of three-dimensional gravitational waves: Harmonic slicing case_ , Phys. Rev. D52 (1995), 5428–5444. * [50] L. B. Szabados, _Quasi-local energy-momentum and angular momentum in GR: A review article_ , Living Rev. Relativity 7 (2004), no. 4, 1–140, http://www.livingreviews.org/lrr-2004-4. * [51] by same author, _On a class of 2-surface observables in general relativity_ , Class. Quantum Grav. 23 (2006), 2291–2302, arXiv.org:gr-qc/0511059. * [52] C. Uggla, H. van Elst, J. Wainwright, and G. F. Ellis, _Past attractor in inhomogeneous cosmology_ , Phys. Rev. D 68 (2003), no. 10, 103502–+. * [53] J. A. Valiente Kroon, _A new class of obstructions to the smoothness of null infinity_ , Comm. Math. Phys. 244 (2004), no. 1, 133–156. * [54] by same author, _Time asymmetric spacetimes near null and spatial infinity. I. Expansions of developments of conformally flat data_ , Classical Quantum Gravity 21 (2004), no. 23, 5457–5492. * [55] by same author, _Time asymmetric spacetimes near null and spatial infinity. II. Expansions of developments of initial data sets with non-smooth conformal metrics_ , Classical Quantum Gravity 22 (2005), no. 9, 1683–1707. * [56] J. Wainwright and G. F. R. Ellis, _Dynamical Systems in Cosmology_ , Dynamical Systems in Cosmology, Edited by J. Wainwright and G. F. R. Ellis, pp. 357. ISBN 0521673526. Cambridge, UK: Cambridge University Press, June 2005., June 2005. * [57] M. Wyman, _Static spherically symmetric scalar fields in general relativity_ , Phys. Rev. D24 (1981), 839–841.

arxiv-papers

2024-09-04T02:48:54.643600

0803.4503

††thanks: Official contribution of the National Institute of Standards and Technology of the U.S. Department of Commerce; not subject to copyright. # Frequency evaluation of the doubly forbidden ${}^{1}S_{0}\rightarrow\,^{3}P_{0}$ transition in bosonic 174Yb N. Poli LENS and Dipartimento di Fisica, Università di Firenze, INFN - sezione di Firenze - 50019 Sesto Fiorentino, Italy Z. W. Barber N. D. Lemke also at University of Colorado, Boulder, CO, 80309 C. W. Oates L. S. Ma State Key Lab. of Precision Spectroscopy, ECNU, China J. E. Stalnaker present address: Department of Physics and Astronomy, Oberlin College, Oberlin OH 44074 T. M. Fortier S. A. Diddams L. Hollberg J. C. Bergquist A. Brusch S. Jefferts T. Heavner T. Parker National Institute of Standards and Technology 325 Broadway, Boulder, CO 80305 ###### Abstract We report an uncertainty evaluation of an optical lattice clock based on the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ transition in the bosonic isotope 174Yb by use of magnetically induced spectroscopy. The absolute frequency of the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ transition has been determined through comparisons with optical and microwave standards at NIST. The weighted mean of the evaluations is $\nu$(174Yb)=518 294 025 309 217.8(0.9) Hz. The uncertainty due to systematic effects has been reduced to less than 0.8 Hz, which represents $1.5\times 10^{-15}$ in fractional frequency. ###### pacs: 32.10.Dk, 06.30.Ft, 32.70.Jz, 39.30.+w Optical frequency references are now reaching fractional frequency uncertainties lower than those of the best Cs primary standards, which define the second. Intense efforts to reduce these levels to well below one part in $10^{16}$ are underway with many different atomic systems (neutral and ionic) for a variety of applications including tests of fundamental physics and the development of the next generation of primary frequency standards Rosenband et al. (2008); Ludlow et al. (2008). Here we report the first detailed evaluation and absolute frequency measurement (with an uncertainty of $1.5\times 10^{-15}$) of a clock system based on neutral 174Yb atoms confined to a one- dimensional optical lattice. Ytterbium is an attractive atom for optical clock studies due to its large mass and numerous abundant isotopes that offer a variety of nuclear spins (0,1/2,5/2). For this work, we focus on the potential of spin-zero isotopes for lattice clock work with the first evaluation of such a system at the $1\times 10^{-15}$ level. Spin-zero isotopes are appealing for frequency metrology due to their simple structure. Indeed atomic clocks based on $J=0\leftrightarrow J^{\prime}=0$ transition in these isotopes present a nearly pure two-level system in that it lacks Zeeman structure, first-order sensitivity to magnetic fields, optical pumping effects, and first-order vector/tensor sensitivity to the lattice laser intensity and polarization. Moreover, the spin-zero (even) isotopes in the alkaline earth-like atom systems are generally more abundant and easier to prepare (through laser cooling and trapping) for high resolution spectroscopy. To realize these advantages, however, it is necessary to use an external field such as a magnetic bias field (as in this work) to induce a nonzero transition strength for the otherwise truly forbidden transition Taichenachev et al. (2006); Barber et al. (2006). The presence of this external field leads to a small shift of the clock frequency that must be calibrated. Additionally, the weakness of the induced transition requires higher probe light intensities than are used with the odd (half-integer spin) isotopes, which lead to an AC Stark shift that also requires careful evaluation. This is the essential trade-off between the odd and even isotopes: one accepts two larger systematic effects in exchange for a considerably simpler atomic system that does not require optical pumping or hopping between multiple spectroscopic features in order to suppress first-order magnetic field and lattice polarization sensitivities. Optical lattice clock studies with Sr have thus far been focused on the odd isotope (87Sr, with nuclear spin 9/2), and the one evaluation with an even isotope (88Sr) had an uncertainty considerably higher than those of its odd counterparts Baillard et al. (2007). Here we show that the even isotopes, especially in the case of Yb, which has somewhat more favorable atomic properties for even isotope work Taichenachev et al. (2006), can achieve extremely low uncertainties as well, with no identified barriers to reaching the $10^{-17}$ level. Experimental details on the cooling and trapping setup have been reported in Barber et al. (2007). In brief, 174Yb atoms are trapped and cooled in a two- stage magneto-optical trap (MOT) working on the strong dipole allowed ${}^{1}S_{0}\leftrightarrow\\!^{1}P_{1}$ transition at 398.9 nm and then on the intercombination ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{1}$ transition at 555.8 nm. Typically 105 atoms are trapped at a temperature of less than $40~{}\mu$K in 350 ms. About $10^{4}$ atoms are then loaded to a 1D standing wave optical lattice operating at the magic wavelength of 759.35 nm for the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ clock transition. A lattice potential depth of about 500 $E_{r}$ (recoil energy $E_{r}/k_{B}$ = 100 nK for Yb at the lattice frequency) is realized by tightly focusing (waist $\approx 30$ $\mu$m) a 1 W injection-locked Ti:Sapphire laser. After a 25 ms delay that allows for the quadrupole magnetic MOT field to dissipate and the static magnetic field to be turned on for the spectroscopy, the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ clock transition at 578 nm is probed with a $\pi$-pulse lasting 50 to 150 ms. The probe laser is collinear with the lattice beam and is focused onto the atoms with a waist of 60 $\mu$m. Both the lattice polarization and the probe laser polarization are aligned with the vertical static magnetic field. The 578 nm clock laser is based on sum-frequency generation of a Nd:YAG laser at 1.32 $\mu$m, and an Yb fiber laser at 1.03 $\mu$m in a periodically-poled lithium niobate waveguide Oates et al. (2007). The 578 nm light produced is then frequency stabilized with a Pound-Drever-Hall lock to a vertically- mounted high finesse stable cavity. The short-term frequency stability is $2.2\times 10^{-15}$ at 1 to 2 s, with a residual frequency drift of less than 0.4 Hz/s. The residual drift is monitored and canceled to first order through feed-forwarding to an acousto-optical modulator interposed between the laser source and the reference cavity. The stabilized yellow light is then sent through phase-noise-compensated fibers to both a femtosecond frequency comb for clock comparison and to the atom trap. For a typical magnetic field strength of $B_{0}=1.3$ mT and a probe intensity of $I_{0}=300$ mW/cm2, the Rabi frequency is $\Omega\sim 5$ Hz, which generates a Fourier-limited linewidth of about $\Delta\nu=10$ Hz. To prevent spurious ac Stark shifts, the pre-cooling and trapping beams are switched off during the spectroscopy with acousto-optical modulators and mechanical shutters. The amount of population transfer via the clock transition is determined through fluorescence detection on the strong 399 nm transition of the atoms remaining in the ${}^{1}S_{0}$ ground state. To lock the frequency of yellow probe laser to the atomic transition, an error signal is derived by alternately probing the half-width points and then demodulating with a digital microprocessor, which provides frequency corrections to an acousto-optical modulator. For a typical observed S/N ratio of $\sim 10$ and a probe duty cycle of 15 %, the projected clock stability is $1\times 10^{-15}\,\tau^{-1/2}$. Evaluation of the systematic shifts of the Yb lattice clock was accomplished through comparisons with other optical frequency standards at NIST. While initial measurements were performed against a neutral Ca standard Oates et al. (2000), the measurements reported here were performed against the Hg+ optical standard Stalnaker et al. (2007). For the comparisons an octave-spanning Ti:Sapphire optical frequency comb Fortier et al. (2006) was locked to the Hg+ standard, and the beat frequency between the comb and the Yb standard was counted with 1 s gate times. We typically changed the value of the experimental parameter under study every 200 to 300 s in order to determine the eventual shift with a statistical uncertainty of about $3\times 10^{-16}$ (see Fig. 2). Figure 1: Systematic shifts of the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ transition in Yb optical lattice clock. a) Measurement of the second order Zeeman shift. The second order coefficient is evaluated $\beta=-6.12(10)$ Hz/mT2. b) Probe ac linear Stark shift. The coefficient $\kappa=15(3)$ mHz/(mW/cm2) is in agreement with the theoretical estimation (see text). c) Density shift. The data points are consistent with zero density shift with a total deviation of 0.6 Hz. d) Servo error. The frequency offset due to uncompensated linear drift of the reference cavity is shown (slope = 1.82(10) Hz/(Hz/s)). With periodic manual correction of the feed-forward system used to remove the linear cavity drift this uncertainty can be made negligible. In the case of spectroscopy of the clock transition with the even isotope, the linear Zeeman effect is zero, with only a second-order dependence on the magnetic field amplitude. In Fig. 1a we show an evaluation of the second-order Zeeman shift obtained by varying the static magnetic field between 1 mT and 3.5 mT. The resulting second-order Zeeman coefficient in 174Yb is $\beta=-6.12(10)$ Hz/mT2, found by a quadratic fit to zero field. For a field value of $1.6$ mT the shift is approximately 18 Hz and is determined with an uncertainty of less than $0.3$ Hz. The uncertainty in the systematic shift ($\Delta_{\textbf{B}}=2\beta|\textbf{B}|\delta\textbf{B}$) is limited mainly by the knowledge of the applied static magnetic field $B$, which is calibrated to better than 10 $\mu$T through spectroscopy of the magnetically sensitive ${}^{1}S_{0}\leftrightarrow\,^{3}P_{1}$ transition. Strategies for reducing the uncertainty of the second-order Zeeman shift include refining the absolute calibration of the magnetic field down to less than 1 $\mu$T (one possibility could be the use of the narrower ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ transition in the fermionic 171Yb isotope) and operation at lower bias field values. In a similar way, the ac Stark shift induced by the yellow probe light has been determined by measuring the shift as a function of intensity and extrapolating to zero intensity. Although the probe laser power is controlled with good precision ($<1$%) with a photodiode and fast servo control, the absolute intensity at the location of the atoms is not well determined. In the current setup, the clock excitation light is focused to a $\sim 60$ $\mu$m waist at the atoms, which is larger by only about a factor of two than the waist size of the lattice beam. This implies that the atoms experience an average intensity and not uniform illumination. In addition, drifts in alignment can cause the average intensity experienced by the atoms to vary by 10 % from day to day in the current setup. Fortunately, the intensity experienced by the atoms is directly proportional to the optical power, so the shift due to the yellow probe can be accurately extrapolated to zero intensity. As is reported in Fig. 1b, this effect shows a linear dependence on the probe intensity with a slope of $\kappa\approx 15(3)$ mHz/(mW/cm2), in agreement with the theoretical estimate Taichenachev et al. (2006). In this case, the typical $\sim$300 mW/cm2 probe intensity gives a total shift of about 7 Hz with an uncertainty of 0.2 Hz. This uncertainty could best be lowered by illuminating the atoms with a larger beam and/or operating at lower values of the intensity. It is important to emphasize that the present uncertainties of these two shifts do not pose fundamental limits for future evaluations. Both the value of the shifts and the consequent errors scale linearly with linewidths. In particular, an experimental linewidth of about 1 Hz could be produced with a reduced value of both the static magnetic field $B=0.3$ mT and the probe light intensity $I=70$ mW/cm2. In this case, with the present level of field calibrations the fractional frequency uncertainty would be less than $1\times 10^{-16}$, and a feasible improvement by a factor of 10 in the calibrations would reduce the overall uncertainty to less than $1\times 10^{-17}$. The frequency uncertainties due to lattice light have been studied in detail Barber et al. (2008). The current systematic uncertainties (polarizability and hyperpolarizability) due to the optical lattice (magic wavelength $\lambda_{magic}=759.3538$ nm) at a depth of 500 $E_{r}$ are given in Table 1, and considerable reduction of both is anticipated. Table 1: Frequency uncertainty budget for the 174Yb optical lattice clock. Some frequency shift values and their uncertainties depend on operating conditions and particular evaluation; typical values are given. Effect | Shift (10-15) | Uncertainty (10-15) ---|---|--- 2nd order Zeeman | -18 – -36 | 0.4 Probe light | 6 – 12.0 | 0.4 Lattice Polarizability | $<$ 1 | 0.6 Hyperpolarizability | 0.33 | 0.07 Density | -0.2 | 1.0 Blackbody shift | -2.5 | 0.25 Syst. Total | -15 – -27 | $<$ 1.5 An important systematic effect to be accounted for is any density shift due to cold collisions between atoms in the 1D optical lattice Ludlow et al. (2008). This has been evaluated by reducing the total atom number captured in the 399 nm MOT and subsequently the lattice, changing the density in the range 0.5-1$\times\rho_{0}$ where $\rho_{0}\sim 10^{11}$ cm-3 is the estimated mean density in the lattice (see Fig. 1c). The result, -0.1(0.6) Hz, is consistent with zero shift at this level of precision. The conservative uncertainty quoted in Table 1 is due to the small range of densities for which the system has been run reliably and the current operation of the clock at the maximum atom number. Further reduction in the uncertainty of the density shift will occur with better evaluations in a 1D lattice or through the use of an under- filled 2D or 3D lattice, for which any such shift should vanish. Finally, the Stark shift induced by blackbody radiation (BBR) has been estimated by measuring the mean value of the temperature of the MOT chamber (295(3) K) and calculating the shift from the theoretical estimates in Porsev and Derevianko (2006). The uncertainty is limited by knowledge of Yb polarizability at the 10 % level, or $2.5\times 10^{-16}$ fractionally. Precise measurements of the BBR shift or operation at cryogenic temperatures will be required to achieve sub-$10^{-17}$ uncertainty. Table 1 summarizes the systematic frequency shifts for magnetic field induced spectroscopy on ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ transition in 174Yb. The values of the individual shifts depend on the particular operating conditions chosen (i.e., lattice power and frequency, bias magnetic field strength, and probe laser intensity). The quoted uncertainties of each frequency bias represent typical values. The actual uncertainties for a given comparison depend on the operating conditions and the proximity to, and quality of, systematic shift calibrations. With improvements in the experimental setup and diagnostics, reduced long-term variations in operation conditions will allow for less frequent systematic shift calibrations. Figure 2: Frequency comparison of 174Yb against a maser calibrated to NIST-F1 (left) and Hg+ (right) through use of the optical fequency comb Stalnaker et al. (2007); Fortier et al. (2006). Histograms and Allan deviation generated from series of 1 s gate interval frequency counts. The linear fit on the two Allan deviations give a slope of $2\times$10${}^{-13}~{}\tau^{-1/2}$ and $5.5\times$10${}^{-15}~{}\tau^{-1/2}$ respectively. The statistical uncertainties for the two measurements are 0.9 Hz and 0.10 Hz. Four absolute determinations of the clock transition frequency in 174Yb were performed over several months against a maser calibrated with the NIST-F1 primary frequency standard Heavner et al. (2005) or against the optical Hg+ standard using the optical frequency comb Stalnaker et al. (2007); Fortier et al. (2006). As shown in Fig. 2, the noise contribution for both comparisons is mainly white in frequency, with a fractional instability of $2\times 10^{-13}~{}\tau^{-1/2}$ vs. the maser and $5.5\times 10^{-15}~{}\tau^{-1/2}$ vs. Hg+. Recent improvements to the 578 nm stabilization cavity have resulted in an improvement of the Yb vs. Hg+ fractional instability to less than $3\times 10^{-15}~{}\tau^{-1/2}$, with the stability reaching to below $1\times 10^{-16}$ in a couple of thousand seconds. For the comparisons against the maser, statistical and calibration uncertainties are significant, and contribute a fractional uncertainty of about $2.5\times 10^{-15}$, depending on the length of the comparison. For comparisons against the optical Hg+ standard, statistical uncertainty can be made negligible ($\sim 2\times 10^{-16}$) with less than 20 minutes of averaging, and the current calibration of Hg+ to NIST-F1 is $0.65\times 10^{-15}$. The absolute frequency for all the comparisons is then derived by correcting the statistical value with the calibration of the reference against NIST-F1 and for the gravitational shift caused by relative height differences. The final results of the frequency measurements of the clock transition in 174Yb are reported in Fig. 3. Two values were obtained from optical comparisons against the Hg+ standard and the other two were obtained though comparison against a NIST-F1 calibrated maser. The weighted mean of all the evaluations is $\nu$(174Yb) = 518 294 025 309 217.8(0.9) Hz. Figure 3: Frequency evaluation of Yb clock transition. Two measurements were against Hg+ ($\bullet$) and the other two measurements were against a calibrated maser ($\blacktriangle$). The weighted mean of all the evaluations is $\nu(^{174}\textrm{Yb})=518\,294\,025\,309\,217.8(0.9)$ Hz (shaded area is the one $\sigma$ confidence interval). We have presented a frequency evaluation of the ${}^{1}S_{0}\leftrightarrow\\!^{3}P_{0}$ clock transition in the 174Yb isotope at a level approaching that of the best atomic primary standards. Fractional frequency uncertainty below that of the NIST-F1 standard ($\sim 4\times 10^{-16}$) will be acheivable in the near-term with further measurements of the systematic frequency shifts. At this level, the current uncertainty in the room temperature BBR shift of Yb would become the limiting factor, and would need to be addressed for high accuracy ratio measurements against other optical standards Rosenband et al. (2008); Ludlow et al. (2008). Then straightforward modifications to the experiment (i.e. better vacuum to improve the lifetime of the atoms in lattice, or multi-dimensional lattices) should lead to significant reductions in the other systematic frequency shifts and uncertainties. Importantly, we foresee no significant barrier to achieving sub-$10^{-17}$ accuracy with an optical lattice clock based on magnetically induced spectroscopy of an even isotope of Yb. The authors would like to thank Jun Ye and the Sr clock team at JILA for our continuing collaboration. A. Brusch acknowledges support from the Danish Natural Science Research Council. L.S. Ma is supported by NSFC(60490280) & STCSM(07JC14019). ## References * Rosenband et al. (2008) T. Rosenband, D. B. Hume, P. O. Schmidt, C. W. Chous, A. Brusch, L. Lorini, W. H. Oskay, R. E. Drullinger, T. M. Fortier, J. E. Stalnaker, et al., Science 319, 1808 (2008). * Ludlow et al. (2008) A. D. Ludlow, T. Zelevinsky, G. K. Campbell, S. Blatt, M. M. Boyd, M. H. G. de Miranda, M. J. Martin, J. W. Thomsen, S. M. Foreman, J. Ye, et al., Science 319, 1805 (2008). * Taichenachev et al. (2006) A. V. Taichenachev, V. I. Yudin, C. W. Oates, C. W. Hoyt, Z. W. Barber, and L. Hollberg, Physical Review Letters 96, 083001 (2006). * Barber et al. (2006) Z. W. Barber, C. W. Hoyt, C. W. Oates, L. Hollberg, A. V. Taichenachev, and V. I. Yudin, Physical Review Letters 96, 083002 (2006). * Baillard et al. (2007) X. Baillard, M. Fouché, R. L. Targat, P. G. Westergaard, A. Lecallier, Y. L. Coq, G. D. Rovera, S. Bize, and P. Lemonde, Optics Letters 32, 1812 (2007). * Barber et al. (2007) Z. W. Barber, C. W. Hoyt, J. E. Stalnaker, N. Lemke, C. W. Oates, T. M. Fortier, S. Diddams, and L. Hollberg, Proceedings of SPIE 6673, E 1 (2007). * Oates et al. (2007) C. Oates, Z. Barber, J. Stalnaker, C. H. T. Fortier, S. Diddams, and L. Hollberg, Proc. 2007 Joint Mtg. IEEE Intl. Freq. Cont. Symp. and EFTF Conf. (2007). * Oates et al. (2000) C. W. Oates, E. A. Curtis, and L. Hollberg, Optics Letters 25, 1603 (2000). * Stalnaker et al. (2007) J. E. Stalnaker, S. A. Diddams, T. M. Fortier, K. Kim, L. Hollberg, J. C. Bergquist, W. M. Itano, M. J. Delany, L. Lorini, W. H. Oskay, et al., Applied Physics B pp. 10.1007/s00340–007–2762–z (2007). * Fortier et al. (2006) T. Fortier, A. Bartels, and S. Diddams, Optics Letters 31, 1011 (2006). * Barber et al. (2008) Z. W. Barber, J. E. Stalnaker, N. D. Lemke, N. Poli, C. W. Oates, T. M. Fortier, S. A. Diddams, L. Hollberg, C. W. Hoyt, A. V. Taichanchev, et al., PRL 100, 103002 (2008). * Porsev and Derevianko (2006) S. G. Porsev and A. Derevianko, Physical Review A 74, 020502(R) (2006). * Heavner et al. (2005) T. P. Heavner, S. R. Jefferts, E. A. Donley, J. H. Shirley, and T. E. Parker, Metrologia 42, 411 (2005).

arxiv-papers

2024-09-04T02:48:54.650040

0803.4514

# Impurity resonance states in noncentrosymmetric superconductor $CePt_{3}Si$: a probe for Cooper-pairing symmetry Bin Liu1, and Ilya Eremin1,2 1 Max-Planck-Institut für Physik komplexer Systeme, D-01187 Dresden, Germany 2 Institute für Mathematische und Theoretische Physik, TU-Braunschweig, D-38106 Braunschweig, Germany ###### Abstract Motivated by the recent discovery of noncentrosymmetric superconductors, such as $CePt_{3}Si$, $CeRhSi_{3}$ and $CeIrSi_{3}$, we investigate theoretically the impurity resonance states with coexisting $s$\- and $p$-wave pairing symmetries. Due to the nodal structure of the gap function, we find single nonmagnetic impurity-induced resonances appearing in the local density of state (LDOS). In particular, we analyze the evolution of the local density of states for coexisting isotropic $s$-wave and $p$-wave superconducting states and compare with that of anisotropic $s$-wave and $p$-wave symmetries of the superconducting gap. Our results show that the scanning tunneling microscopy can shed light on the particular structure of the superconducting gap in non- centrosymmetric superconductors. ###### pacs: 74.20.Rp, 74.90.+n, 74.25.Jb ## I Introduction Recent discoveries of superconductivity in the systems that posses a lack of inversion symmetry such as $CePt_{3}Si$bauer with $T_{c}\simeq 0.75K$ and more recently $CeRhSi_{3}$kimura , $CeIrSi_{3}$sugitani , $Li(Pd_{1-x},Pt_{x})_{3}B$togano , $UIr$akazawa , $Y_{2}C_{3}$amano have raised an interest in the theoretical investigation of superconductivity in these systems. Among interesting questions the most important one concerns the underlying symmetry of the superconducting order parameter. In particular, in all these materials, there is a nonzero potential gradient $\nabla V$ averaged in the unit cell due to lack of inversion symmetry, which results in the anisotropic spin-orbit interaction. Its general form can be determined by a group theoretical argumentsamokhin and, as it has been found, leads to many interesting properties edel ; rashba ; yip ; sigrist ; fujimoto1 ; yanase ; eremin . For example, on general grounds there is a mixing of the spin-singlet and spin triplet superconducting states due to the lack of inversion. In CePt3Si the pairing symmetry has been studied theoreticallyrashba ; yip ; sigrist ; fujimoto1 ; yanase and it is believed that the $s+p$-wave superconducting state is realized. Frigeri et al.sigrist pointed out that the spin-orbit interaction could determine the direction of the ${\bf d}$-vector as ${\bf d}||\vec{l}$ ($\vec{l}$ is the vector of the Rashba spin-orbit coupling) for which the highest transition temperature was obtained. A microscopic calculation with the detailed structure of the Fermi surfaceyanase seems to confirm that the $s+p$ wave state is the most probable one. The experimental studies of the temperature dependencies of the spin-lattice relaxationyogi , the magnetic penetration depthbonalde , and the thermal conductivity measurementsizawa are also consistent with this conjecture. It is known that the non-magnetic as well as the magnetic impurities in the conventional and unconventional superconductors already have been proven to be a useful tool to distinguish between various symmetries of the superconducting statebalatskyREV . For example, in the conventional isotropic $s$-wave superconductor the single magnetic impurity induced resonance state is located at the gap edge, which is known as Yu-Shiba-Rusinov stateyu . In the case of unconventional superconductor with $d_{x^{2}-y^{2}}$-wave symmetry of the superconducting state the non-magnetic impurity-induced bound state appears near the Fermi energy as a hallmark of $d_{x^{2}-y^{2}}$-wave pairing symmetrybalatsky . The origin of this difference is understood as being due to the nodal structure of two kinds of SC order: in the $d_{x^{2}-y^{2}}$-wave case the phase of Cooper-pairing wave function changes sign across the nodal line which yields finite density of states below the superconducting gap, while in the isotropic $s$-wave case the density of states is gapped up to energies of about $\Delta_{0}$ and thus the bound state can appear only at the gap edge. In principle the formation of the impurity resonance states can also occur in unconventional superconductors if the nodal line or point does not exist at the Fermi surface of a superconductor like it occurs for isotropic nodeless $p$-wave and/or $d_{x}+id_{y}$-wave superconductors for the large value of the potential strengthwang . Therefore, STM measurements of the impurity states can provide important messages about the pairing symmetry in the revelent systems. In the noncentrosymmetric superconductor with the possible coexistence of s-wave and p-wave pairing symmetry, it is very interesting to see what is the nature of the impurity state, and whether a low energy resonance state can still occur around the impurity through changing the dominant role played by each of the pairing components. Previously the effect of the non-magnetic impurity scattering has been studied in the non- centrosymmetric superconductors with respect to the suppression of TcfrigeriEPB and the behavior of the upper critical fieldminsam . In this paper we investigate theoretically the impurity resonance states where both $s$-wave and $p$-wave Cooper-pairing coexist. Due to the nodal structure of gap function as a result of the interference between the spin triplet and the spin singlet components of the superconducting order parameters, we find that a single nonmagnetic impurity-induced resonance state appears in the local density of state. In particular, we analyze the evolution of the local density of states for coexisting isotropic $s$-wave and $p$-wave superconducting states and compare with that of anisotropic $s$-wave and $p$-wave symmetries of the superconducting gap. Our results show that the scanning tunneling microscopy can shed light on the particular structure of the superconducting gap in non-centrosymmetric superconductors. ## II the Model and T-matrix formulation Theoretical models of the superconducting state in CePt3Si are based upon the existence of a Rashba type spin-orbit coupling (RSOC)rashba . Therefore, following previous considerationsigrist we start from an single orbital model with RSOC $\displaystyle H$ $\displaystyle=$ $\displaystyle\sum_{{\bf k}s}\varepsilon_{\bf k}c_{{\bf k}s}^{\dagger}c_{{\bf k}s}+\alpha\sum_{{\bf k}ss^{\prime}}{\bf g_{k}}\cdot{\bf\sigma_{ss^{\prime}}}c_{{\bf k}s}^{\dagger}c_{{\bf k}s^{\prime}},$ (1) where $c_{{\bf k}s}^{\dagger}$ ($c_{{\bf k}s}$) is the fermion creation (annihilation) operator with spin $s$ and momentum ${\bf k}$. Here, $\varepsilon_{\bf k}$ is the tight-binding energy dispersion $\displaystyle\varepsilon_{\bf k}$ $\displaystyle=$ $\displaystyle 2t(\cos(k_{x})+\cos(k_{y}))+4t_{1}\cos(k_{x})\cos(k_{y})$ (2) $\displaystyle+$ $\displaystyle 2t_{2}(\cos(2k_{x})+\cos(2k_{y}))$ $\displaystyle+$ $\displaystyle[2t_{3}+4t_{4}(\cos(k_{x})+\cos(k_{y}))$ $\displaystyle+$ $\displaystyle 4t_{5}(\cos(2k_{x})+\cos(2k_{y}))]\cos(k_{z})$ $\displaystyle+$ $\displaystyle 2t_{6}\cos(2k_{z})-\mu$ which reproduces the so-called $\beta$-band of $CePt_{3}Si$ as obtained from the band structure calculationssamokhin ; yanase . The electron hopping parameters are $(t,t_{1},t_{2},t_{3},t_{4},t_{5},t_{6},n)=(1,-0.15,-0.5,-0.3,-0.1,-0.09,-0.2,1.75)$ and the electron density per site $n$ is used to determine the chemical potentialyanase . The second term of Eq.(1) is the RSOC interaction where $\alpha$ denotes the coupling constant and the vector function ${\bf g_{k}}$ is assumed in the following form ${\bf g_{k}}=(-\sin k_{y},\sin k_{x},0)$. This term removes the usual Kramers degeneracy between the two spin states at a given ${\bf k}$, and leads to a quasiparticle dispersion $\epsilon_{\bf k}=\varepsilon_{\bf k}\pm\alpha|{\bf g_{k}}|$ with ${|{\bf g_{k}}|=\sqrt{{\bf g_{k}}^{2}_{x}+{\bf g_{k}}^{2}_{y}+{\bf g_{k}}^{2}_{z}}}$, splitting the Fermi surface (FS) into two sheets. Based on the above hopping parameters and RSOC constant $\alpha=0.3t$, the resulting FS is shown in Fig.1, where the main characteristic features of the FS has been successfully reproducedsamokhin . Figure 1: (color online) The calculated Fermi surface using the Eq. (1) and the spin-orbit coupling constant $\alpha=0.3t$. In the superconducting state, the presence of RSOC breaks the parity and, therefore, mixes the singlet (even parity) and triplet (odd parity) Cooper- pairing states. A full symmetry analysisyanase ; samokhin shows that $s$-wave pairing $\Delta_{s}=\Delta_{0}(\cos(k_{x})+\cos(k_{y}))$ and a $p$-wave triplet pairing state with order parameter ${\bf d_{k}}$ parallel to the ${\bf g_{k}}$ vector, ${\bf d_{k}}=d_{0}{\bf g_{k}}$ are able to coexist. Following previous estimationsyanase we have taken the odd parity component ${\bf d_{k}}=d_{0}(-\sin k_{y},\sin k_{x},0)$. Then the mean field BCS Hamiltonian for this system has the matrix form $\displaystyle H_{\bf k}=\left(\matrix{\varepsilon_{\bf k}&\alpha g&-d^{\ast}&\Delta_{\bf k}\cr\alpha g^{\ast}&\varepsilon_{\bf k}&-\Delta_{\bf k}&d\cr-d&-\Delta^{\ast}_{\bf k}&-\varepsilon_{\bf k}&\alpha g^{\ast}\cr\Delta^{\ast}_{\bf k}&d^{\ast}&\alpha g&-\varepsilon_{\bf k}\cr}\right).$ (3) Where for briefly, $g=({\bf g_{k}}_{x}-i{\bf g_{k}}_{y})$ and $d=({\bf d_{k}}_{x}+i{\bf d_{k}}_{y})$. The inverse of the single-particle Green’s function is defined as $\displaystyle g^{-1}({\bf k},i\omega_{n})=i\omega_{n}I-H_{\bf k},$ (4) where $I$ is the $4\times 4$ identity matrix. Taking the inverse of Eq. (3) we find $\displaystyle g({\bf k},i\omega_{n})=\left(\matrix{G({\bf k},i\omega_{n})&F({\bf k},i\omega_{n})\cr F^{\dagger}({\bf k},i\omega_{n})&-G^{t}(-{\bf k},-i\omega_{n})\cr}\right)$ (5) where $\displaystyle G({\bf k},i\omega_{n})$ $\displaystyle=$ $\displaystyle\sum_{\tau=\pm 1}\frac{1+\tau({\bf\vec{g}_{{\bf k}}}\cdot{\bf\sigma})}{2}G_{\tau}({\bf k},i\omega_{n}),$ (6) $\displaystyle F({\bf k},i\omega_{n})$ $\displaystyle=$ $\displaystyle\sum_{\tau=\pm 1}\frac{1+\tau({\bf\vec{g}_{\bf k}}\cdot{\bf\sigma})}{2}i\sigma_{y}F_{\tau}({\bf k},i\omega_{n}),$ (7) and $\displaystyle G_{\tau}({\bf k},i\omega_{n})$ $\displaystyle=$ $\displaystyle\frac{i\omega_{n}+\epsilon_{\tau}}{(i\omega_{n})^{2}-E^{2}_{{\bf k}\tau}},$ (8) $\displaystyle F_{\tau}({\bf k},i\omega_{n})$ $\displaystyle=$ $\displaystyle\frac{\Delta_{\tau}}{(i\omega_{n})^{2}-E^{2}_{{\bf k}\tau}}.$ (9) Here, the single-particle excitation energy is $\displaystyle E_{{\bf k}\tau}=\sqrt{\epsilon_{\tau}^{2}+|\Delta_{\tau}|^{2}}$ (10) with $\displaystyle\epsilon_{\tau}=\varepsilon_{\bf{\bf k}}+\tau\alpha|{\bf g_{k}}|;\Delta_{\tau}=\Delta_{\bf k}+\tau|{\bf d_{k}}|,$ (11) and the unit vector is ${\bf\vec{g}_{\bf k}}={\bf g_{\bf k}}/{|{\bf g_{\bf k}}|}$. For completion the equations above have to be supplemented by the self- consistency equation that determines the symmetry of the superconducting gap and the superconducting transition temperature. For the sake of simplicity and also because this is not critical for our further analysis we consider the superconducting order parameter as a given parameter. At the same time, recent studies based on the helical spin fluctuation mediated Cooper-pairing find two stable superconducting phases with either dominantly $s+p$-wave or $p+d+f$-wave symmetry of superconducting order parameteryanase ; sigrist08 ; fujimoto . In the following we adopt the former one for our calculation. The next step is to obtain the Green’s function in the presence of a single impurity site. The impurity scattering is given by $\displaystyle H_{imp}=U_{0}\sum_{\sigma}c_{0\sigma}^{\dagger}c_{0\sigma},$ (12) where without loss of generality we have taken a single-site nonmagnetic impurity of strength $U_{0}$ located at the origin, $r_{i}=0$. Then the site dependent Green’s function can be written in terms of the T-matrix formulationwang ; fisher ; wang1 as $\displaystyle\zeta(i,j;i\omega_{n})$ $\displaystyle=$ $\displaystyle\zeta_{0}(i-j;i\omega_{n})$ (13) $\displaystyle+$ $\displaystyle\zeta_{0}(i,i\omega_{n})T(i\omega_{n})\zeta_{0}(j,i\omega_{n}),$ where $\displaystyle T(i\omega_{n})$ $\displaystyle=$ $\displaystyle\frac{U_{0}\rho_{3}}{1-U_{0}\rho_{3}\zeta_{0}(0,0;i\omega_{n})}$ (14) $\displaystyle\zeta_{0}(i,j;i\omega_{n})$ $\displaystyle=$ $\displaystyle\frac{1}{N}\sum_{\bf k}e^{i\bf k\cdot\bf R_{ij}}g(k,i\omega_{n}),$ (15) with $\rho_{i}$ being the Pauli spin operator, and $\bf R_{i}$ is the lattice vector, $\bf R_{ij}=\bf R_{i}-\bf R_{j}$. Finally, the local density of state which can be measured in the STM experiment has been obtained as $\displaystyle N(r,\omega)=-\frac{1}{\pi}\sum_{i}{\rm Im}\zeta_{ii}(r,r;\omega+i\eta),$ (16) where $\eta$ denotes an infinitely small positive number. ## III Numerical Results and Discussions ### III.1 The density of state Before considering the effect of the impurity it is useful to analyze first the density of state (DOS) in the superconducting state, which is expressed as, $\displaystyle\rho(\omega)=-\frac{1}{\pi}{\rm Im}\sum_{i,{\bf k}}g_{ii}({\bf k},\omega)$ (17) Figure 2: (color online) The evolution of the local density of states for various ratio between coexisting isotropic $s$-wave and $p$-wave Cooper- pairing state. The left and right panels refer to the different values of the damping constant $\Gamma$. The dashed and the dotted curve denote the contribution of the different bands and the straight curve refers to the total density of states. The parameters of the gaps and the damping $\Gamma$ are given in terms of hopping integral $t$. As we already have mentioned above it is not necessary to calculate the magnitude of the gap functions self-consistently since we are mainly interested in the qualitative properties arising from the gap structure. We first consider the situation when the $s$-wave part of the total superconducting gap is momentum independent, $\Delta_{s}=\Delta_{0}$. In Fig.2 we show the evolution of the density of state for positive frequencies for various values of the $s$-wave component of the superconducting gap. In particular, for zero value of the $s$-wave component the superconducting gap is purely determined by the $p$-wave superconducting gap with the point node at the Fermi surfaces of the corresponding bands at ($k_{x}=0,k_{y}=0$). This gap structure is the same for both bands splitted by the spin-orbit coupling. With increasing value of the isotropic $s$-wave gap one finds that the total superconducting gap in one of the bands increasing with the total superconducting gap $\Delta_{s}+|{\bf d}_{\bf k}|$ while it decreases effectively for the other band for which the total gap is $\Delta_{s}-|{\bf d}_{\bf k}|$. Once both $s-$wave and $p$-wave superconducting gaps are the same, the accidental node forms at one of the band and the behavior of the density of states changes to a linear at low energy reflecting the formation of the line of node. We further note that density of states shows only slight electron-hole asymmetry. In Fig. 3 we show a similar evolution of the density of states, however, now the $s$-wave component of the superconducting gap is momentum dependent, $\Delta_{s}=\Delta_{0}(\cos k_{x}+\cos k_{y})=\Delta_{0}\gamma_{\bf k}$ comment . Figure 3: (color online) The evolution of the local density of states for various ratio between coexisting anisotropic $s$-wave and $p$-wave Cooper- pairing state. The left and right panels refer to the different values of the damping constant $\Gamma$. The dashed and the dotted curve denote the contribution of the different bands and the straight curve refers to the total density of states. The parameters of the gaps and the damping $\Gamma$ are given in terms of hopping integral $t$. Interestingly enough, here the node in the density of states forms already when the $p$-wave superconducting gap component is zero (see also Fig.5) and is the result of the initial momentum structure of the $s$-wave superconducting gap that yields point nodes on the Fermi surface. This is unique to the anisotropic $s$-wave superconducting gap. By introducing the interference between $s$-wave and $p$-wave gap the position of the node is shifted to the different points of the Brillouin Zone, however, here the nodal structure of the superconducting gap is not a result of the interference effect between $p$-wave and $s$-wave of the superconducting gap but arises already in the pure anisotropic $s$-wave symmetry and shifted by introducing the moderate component of the $p$-wave gap. ### III.2 Impurity resonance states Figure 4: (color online) The LDOS for coexisting isotropic $s$-wave and $p$-wave Cooper pairing states for various ratio of the parameters. The straight (red) curves refer to the calculated density of states without impurity and the dashed (green) curves refer to the LDOS at the $(0,1,0)$ position. Here, we use $U_{0}=5t$. Figure 5: (color online) The LDOS for coexisting anisotropic $s$-wave ($\Delta$)and $p$-wave Cooper pairing states for various ratio of the parameters. The straight (red) curves refer to the calculated density of states without impurity and the dashed (green) curves refer to the LDOS at the $(0,1,0)$ position. Here, we use $U_{0}=5t$. In view of complicated band structure arising in CePt3Si from the Rashba spin- orbit coupling and the corresponding interference effect for the superconducting gap the density of states in a clean case that can be accessed by the tunneling experiments cannot give a precise information on the exact structure of the superconducting gap in the non-centrosymmetric superconductors. At the same time, an introduction of the non-magnetic impurity can give additional important information on the symmetry of the superconducting gap in such a material. In terms of Eq.(16), the T-matrix can be written as $\displaystyle T^{-1}(i\omega_{n})=U^{-1}_{0}-\rho_{3}\zeta_{0}(0,0;i\omega_{n}),$ (18) and the position of the impurity resonant state is given by $\det T^{-1}=0$. We first study the situation of the isotropic $s$-wave superconducting gap coexisting with $p$-wave. In Fig.4 we show the calculated density of states without impurity and also the local density of states with an impurity on the nearest neighbor site $(0,1,0)$. Without the $s$-wave component the density of states shows the formation of the impurity induced resonant bound states that appear symmetrically in energy at the positive and negative sides of the LDOS. Clearly these resonant bound states arise due to unconventional nature of the $p$-wave superconducting gap and the nodal points at the Fermi surface. One clearly sees that upon increasing of the isotopic $s$-wave contribution the bound state shifts towards the edge of the superconducting gap implying the zero density of states for energies lower than $\Delta_{0}$. In Fig.5 we show the corresponding local density of states for the coexisting anisotropic $s$-wave and $p$-wave superconducting gaps.In the present case, for any value of the $s$-wave and $p$-wave gap there are nodal points at the Fermi surface resulting either from the internal structure of the anisotropic $s$-wave gap, point nodes from the $p$-wave state, or a nodal line at one of the bands that arises due to interference of the $p$-wave and $s$-wave gap. Therefore, the impurity induced bound state occurs for all ratios between the $p$-wave and $s$-wave gap. Note, that in case of pure anisotropic $s$-wave gap due to the nodal structure on both of the bands the impurity induced bound state becomes visible only for a very large values of the potential scattering strength $U_{0}$. ## IV Summary In summary, we have investigated theoretically the non-magnetic impurity induced resonance bound states in the superconductors without inversion symmetry using as an example $CePt_{3}Si$, which is believed to have a line node in the energy gap arising from the coexistence of $s$-wave and $p$-wave pairing symmetry. Analyzing the local density of states we find that the nodal structure of gap function, we find that a single nonmagnetic impurity-induced resonance states is highly probable in non-centrosymmetric superconductors. We show that further STM experiments may reveal the exact symmetry of the superconducting gap in these systems. We thank Jun Chang for useful discussions. ## References * (1) See e.g., E. Bauer, G. Hilscher, H. Michor, CH. Paul, E.W. Scheidt, A. Gribanov, Yu. Seropegin, H. No$\ddot{e}$l, M. Sigrist, and P. Rogl, Phys. Rev. Lett. 92, 027003 (2004). * (2) N. Kimura, K. Ito, K. Saitoh, Y. Umeda, H. Aoki, and T. Terashima, Phys. Rev. Lett. 95, 247004 (2005). * (3) I. Sugitani, Y. Okuda, H. Shishido, T. Yamada, A. Thamizhavel, E. Yamamoto, T.D. Matsuda, Y. Haga, T. Takeuchi, R. Settai, and Y. Onuki, J. Phys. Soc. Jpn. 75, 043703 (2006). * (4) K. Togano, P. Badica, Y. Nakamori, S. Orimo, H. Takeya, and K. Hirata, Phys. Rev. Lett. 93, 247004 (2004); P. Badica, T. Kondo, and K. Togano, J. Phys. Soc. Jpn. 74, 1014 (2005). * (5) T. Akazawa, H. Hidaka, T. Fujiwara, T.C. Kobayashi, E. Yamamoto, Y. Haga, R. Settai, and Y. Onuki, J. Phys.: Condens. Matter 16, L29 (2004). * (6) G. Amano, S. Akutagawa, T. Muranaka, Y. Zenitani, and J. Akimitsu, J. Phys. Soc. Jpn. 73, 530 (2004). * (7) K. Samokhin, Phys. Rev. Lett. 94 024515 (2005). * (8) V. M. Edelstein, JETP 68, 1244 (1989); V. M. Edelstein, Phys. Rev. Lett. 75 2004 (1995) * (9) L.P. Gor kov and E. Rashba, Phys. Rev. Lett. 87 037004 (2001). * (10) S.K. Yip, Phys. Rev. B 65, 144508 (2002). * (11) P.A. Frigeri, D.F. Agterberg, A. Koga, and M. Sigrist, Phys. Rev. Lett. 92, 097001 (2004); R.P. Kaur, D.F. Agterberg, and M. Sigrist, Phys. Rev. Lett. 94 137002 (2005). * (12) S. Fujimoto, Phys. Rev. B. 72, 024515 (2005); S. Fujimoto, J. Phys. Soc. Jpn. 75, 083704 (2006); ibid. 76 051008 (2007). * (13) Y. Yanase, and M. Sigrist, J. Phys. Soc. Jpn. 76, 043712 (2007). * (14) I. Eremin, and J.F. Annett, Phys. Rev. B 74, 184524 (2006). * (15) M. Yogi, Y. Kitaoka, S. Hashimoto, T. Yasuhida, R. Settai, T.D. Matsuda, Y. Haga, Y. Onuki, P. Rogl, and E. Bauer, Phys. Rev. Lett. 93, 027003 (2004). * (16) I. Bonalde, W. Brämer-Escamilla, and E. Bauer, Phys. Rev. Lett. 94, 207002 (2005). * (17) K. Izawa, Y. Kasahara, Y. Matsuda, K. Behnia, T. Yasuda, R. Settai, and Y. Onuki, Phys. Rev. Lett. 94, 197002 (2005). * (18) see for example A.V. Balatsky, I. Vekhter, and J.-X. Zhu, Rev. Mod. Phys. 78, 373 (2006). * (19) L. Yu, Acta. Phys. Sin. 21, 75 (1965); H. Shiba, Prog. Theor. Phys. 40, 435 (1968); A.I. Russinov, Sov. Phys. JETP 29, 1101 (1969). * (20) A.V. Balatsky, and M.I. Sakola, Phys. Rev. B 51, 15547 (1995); M.I. Sakola, A.V. Balatsky, and D.J. Scalapino, Phys. Rev. Lett. 77, 1841 (1996). * (21) Q.H. Wang and Z.D. Wang, Phys. Rev. B 69, 092502 (2004). * (22) P.A. Frigeri, D.F. Agterberg, I. Milat, and M. Sigrist, Eur. Phys. J. B 54, 435 (2006). * (23) V.P. Mineev, and K.V. Samokhin, Phys. Rev. B 75, 184529 (2007). * (24) Y. Yanase, and M. Sigrist, arXiv:0805.2791 (unpublished). * (25) Y. Tada, N. Kawakami, and S. Fujimoto, J. Phys. Soc. Jpn. 77, 054707 (2008). * (26) Ø. Fischer, M. Kugler, I. Maggio-Aprile, C. Berthod, and C. Renner, Rev. Mod. Phys 79, 353 (2007). * (27) Q.H. Wang, Phys. Rev. Lett. 88, 057002 (2002). * (28) Note that both the nodal and the isotropic $s$-wave components of the total superconducting gap may coexist in the microscopic theory of the Cooper-pairing. Moreover, the on-site Coulomb repulsion will tend to favor the nodal $s$-wave gap function.

arxiv-papers

2024-09-04T02:48:54.655305

0804.0010

# Triangle Area Numbers and Solid Rectangular Numbers Konstantine D. Zelator Department of Mathematical Sciences University of Northern Colorado Greeley, CO 80639 The subject matter of this paper is the area of triangles with integer side lengths and area that is a rational number. We have the following definition: Definition 1: A triangle area number is the area number of a triangle whose sides have integer lengths and whose area is a rational number. As it will be established in the last section of this paper, Section 6, every triangle area number is an even integer. Furthermore, every triangle area number is divisible by 3. A well known subset of the set of triangle area numbers is the set of Pythagorean numbers: those are the areas of Pythagorean triangles (right triangles with integer sides). Another interesting subset of the set of triangle area numbers is that of solid rectangular numbers (the precise defintion is given in Section 6); those numbers arise in connection with rectangular solids (or parallelepipeds) with integer edges and integer inner diagonals; more specifically, the solid rectangular numbers arise from the integer solutions to the diophantine equation (1) (see below). The formulas obtained in Result 2 of Section 5 do generate solid rectangular numbers exclusively (also see table below Result 2). In Section 6, where a complete description of triangle area numbers is given, it is shown that the formulas of Result 2 actually generate all the solid rectangular numbers and none other. At the end of this paper (end of Section 6) we list all triangle area numbers not exceeding 999, all Pythagorean numbers and all solid rectangular numbers. There are exactly 96 triangle area numbers not exceeding 999; thirty-one of them are Pythagorean numbers, two are solid-rectangular numbers, but not Pythagorean, and sixty-three of them are triangle area numbers that are neither Pythagorean nor solid rectangular numbers. ## 1 Introduction As it will become evident, there is a close relationship between triangles with integer side lengths and whose area $A$ is a rational number on the one hand, and the diophantine equation $x^{2}+y^{2}+z^{2}=t^{2}$ (or its more general version, see (17) in Section 6) on theother hand. Consider $x^{2}+y^{2}+z^{2}=t^{2}.$ (1) There is a specific geometric meaning behind the equation. If a quadruple of positive integers $(x_{0},y_{0},z_{0},t_{0})$ satisfies equation (1), then the rectangular solid whose length, width and height (twelve edges, three distinct edge length values) are the natural numbers $x_{0},y_{0},z_{0}$, will have two inner diagonals both having length equal to the natural number $t_{0}$. In Section 6, we establish a process wherein solutions of (1) are used to generate integer-sided triangle with rational area $A$. In order to clearly see how the relationship between such triangles and equation (1) arises, we must make use of the Heron-formula for the area $A$ of the triangle. Let $a,b,c$ be positive integers, the side lengths of a triangle. Heron’s formula: $A=\sqrt{S\cdot(s-a)(s-b)(s-c)}$ where $s={\displaystyle\frac{N}{2}}={\displaystyle\frac{a+b+c}{2}}$, the semi- perimeter, $N$ the perimeter of the triangle. Squaring, we obtain $A^{2}=s\cdot(s-a)\cdot(s-b)\cdot(s-c)$. If $\theta_{1},\ \theta_{2},\ \theta_{3}$ are the degree (or radian) measures of the triangle’s three angles, we know that $A={\displaystyle\frac{1}{2}}a\cdot b\cdot\sin(\theta_{3})={\displaystyle\frac{1}{2}}a\cdot c\cdot\sin(\theta_{2})={\displaystyle\frac{1}{2}}b\cdot c\cdot\sin(\theta_{1}).$ One can immediately observe that a triangle with integer sides $a,b,c$ will have rational area $A$, if and only if all three sine values $\sin(\theta_{1}),\ \sin(\theta_{2}),$ $\sin(\theta_{3})$ are rational numbers. Also, any triangle with integer sides $a,b,c$ will have rational cosine values $\cos(\theta_{1}),\cos(\theta_{2}),\cos(\theta_{3})$ (regardless of whether or not $A$ is a rational number); this is an immediate consequence of the Law of Cosines If we set $\left.\begin{array}[]{rcl}-a+b+c&=&N_{1}\\\ \\\ a-b+c&=&N_{2}\\\ \\\ a+b-c&=&N_{3}\end{array}\right\\}$ (2) We also obtain from solving for $a,b,c$ the following expressions: $\begin{array}[]{lll}a={\displaystyle\frac{N_{2}+N_{3}}{2}},&\ \ b={\displaystyle\frac{N_{1}+N_{3}}{2}},&\ \ c={\displaystyle\frac{N_{1}+N_{2}}{2}}\end{array}$ (3) From $A^{2}=s\cdot(s-a)\cdot(s-b)\cdot(s-c),\ s={\displaystyle\frac{N}{2}},\ \ N=a+b+c$, and (3) we obtain, $16A^{2}=N\cdot N_{1}\cdot N_{2}\cdot N_{3}$ (4) Let $\delta$ be the greatest common divisor of the integers $a,b,c$. And $d$ the greatest common divisor of the positive integers $N_{1},N_{2}$, and $N_{3}$. Note that $N_{1},N_{2}$, and $N_{3}$ are indeed positive since $b+c>a,a+c>b,a+b>c$, these are the triangle inequalities. In the language of number theory, $\left.\begin{array}[]{l}(a,b,c)=\delta\\\ \\\ (N_{1},N_{2},N_{3})=d\end{array}\right\\}$ (5) ## 2 The Relationship Between $\delta$ and $d$ (Irrespective of Whether the Area is Rational) How do integers $\delta$ and $d$ relate? As it will be evident below, there are only two possibilities: either $\delta=d$ or $2\delta=d$. First note, by inspection, that (2) clearly shows that $\delta$ must be a divisor of $d$. Also, observe that depending on the combination of parities of $a,b$, and $c$, either all three $N_{1},N_{2}$, and $N_{3}$ are odd; or they are all even. In case all three $N_{1},N_{2}$, and $N_{3}$ are odd, it is easy to see from (3) that $d$ must also be a common divisor of $a,b$, and $c$; and hence a divisor of $\delta$. So when $N_{1},N_{2}$, and $N_{3}$ are odd, and $\delta$ and $d$ must divide each other, and since they are positive, they must be equal. However, when $N_{1},N_{2}$, and $N_{3}$ are all even, the situation becomes significantly more complicated. Without elaborating on the details, an analysis modulo 4 demonstrates that the following possibilities or combinations occur: (Parenthetically we note here that in the above observations, an underlying assumption was made- a fact very well known in number theory: every common divisor of any number of integers, must divide their greatest common divisor.) Below, we include all possibilities, including the case where $N_{1},N_{2}$, and $N_{3}$ are odd. This analysis is the result of equations (2) and (3), and considerations modulo 4. In total, we have seven broad occurrences or combinations: Case 1: $a,b,c$ are odd: $N_{1},N_{2}$, and $N_{3}$ are odd, $\delta=d$. Case 2: One of $a,b,c$ is even and the other two odd: one of $N_{1},N_{2}$, and $N_{3}$ is congruent to 0 modulo 4; the other two are congruent to 2(mod 4). In this case, $d=2\delta$. Case 3: Two of $a,b,c$ are even, the other odd: $N_{1},N_{2},N_{3}$ are odd, and $\delta=d$. Case 4: All three of $a,b,c$ are even, with either all three being congruent to 2(mod 4); or with two of them congruent to 0(mod 4), the third one 2(mod 4). In either (sub)-case, all three $N_{1},N_{2},N_{3}$ must be congruent to 2(mod 4); $\delta=d$. Case 5: Two of $a,b,c$ are congruent to 2(mod 4), the third one congruent to 0(mod 4): Then, all three $N_{1},N_{2},N_{3}$ are multiples of 4; and $d=2\delta$. Case 6: All three $a,b,c$, are multiples of 4; and so must $N_{1},N_{2}$, and $N_{3}$ be. And with the exponents $\alpha_{i_{1}},\alpha_{i_{2}},\alpha_{i_{3}}$ satisfying $\alpha_{i_{1}}\geq\alpha_{i_{2}}\geq\alpha_{i_{3}}$, with at least one inequality sign being strict (so that not all $\alpha_{i_{1}},\alpha_{i_{2}},\alpha_{i_{3}}$ are equal), where $(i_{1},i_{2},i_{3})$ is a permutation of $(1,2,3)$ and $2^{\alpha_{i_{1}}},2^{\alpha_{i_{2}}},2^{\alpha_{i_{3}}}$, are the highest powers of 2 dividing $N_{i_{1}},N_{i_{2}},N_{i_{3}}$, respectively. In this case $d=2\delta$. Case 7: All three $a,b,c$, are divisible by 4; and thus, so are $N_{1},N_{2}$, and $N_{3}$. And with $N_{1},N_{2},N_{3}$ being divisible by the same highest power of 2 (that is, $\alpha_{1}=\alpha_{2}=\alpha_{3}$ in the notation of Case 6). In this case $\delta=d$. Numerical Examples 1. 1) For Case 1: Any three odd natural numbers $a,b,c$, will do, as long as $a+b>c$, $a+c>b$, and $b+c>a$. 2. 2) For Case 2: $a=6,b=5,c=9,N_{1}=8,N_{2}=10,N_{3}=2;\ \delta=1,\d{=}2$ 3. 3) For Case 3: $a=6,b=12,c=9,N_{1}=15,N_{2}=3,N_{3}=9;\ \delta=d=3$ 4. 4) For Case 4: $a=6,b=14,c=18,N_{1}=26,N_{2}=10,N_{3}=2;\ \delta=d=2$ 5. 5) For Case 5: $a=34,b=38,c=44,N_{1}=48,N_{2}=40,N_{3}=28;\ \delta=4,d=8$ 6. 6) For Case 6: $a=34,b=38,c=44,N_{1}=48,N_{2}=40,N_{3}=28;\ \delta=4,d=8,\ \alpha_{1}=3,\alpha_{2}=5,\alpha_{3}=4$ 7. 7) For Case 7: $a=24,b=28,c=36,N_{1}=40,N_{2}=32,N_{3}=16;\ \delta=d=8,\ \alpha_{1}=\alpha_{2}=\alpha_{3}$ ## 3 The Primitive Case and Result 1 According to (5), we have $\left({\displaystyle\frac{a}{\delta},\frac{b}{\delta},\frac{c}{\delta}}\right)=1=\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)$. Furthermore, note that since the perimeter $N$ is equal to $N=a+b+c$, it is also true that $N=N_{1}+N_{2}+N_{3}$ (by adding the three equations in (2); or in (3)). In what is to follow, we will make the assumption below: $\left({\displaystyle\frac{N_{1}}{d}\cdot\frac{N_{2}}{d}\cdot\frac{N_{3}}{d},\frac{N}{d}}\right)=1$ (6) In other words, we will assume that $\left({\displaystyle\frac{N}{d}}\right)$ is relatively prime to each of the integers ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$. Remark 1: It can be proven that (6) implies $\left({\displaystyle\frac{a}{\delta}\cdot\frac{b}{\delta}\cdot\frac{c}{\delta},\frac{N}{d}}\right)=1$ in Cases 1, 2, 3, 4, 5, and 7, listed in the previous section; also in Case 6, except when $\alpha_{i_{1}}=\alpha_{i_{2}}>\alpha_{i_{3}}$ or when $\alpha_{i_{1}}>\alpha_{i_{2}}=\alpha_{i_{3}}$; in which sub-cases (of Case 6), we have $\left({\displaystyle\frac{a}{\delta}\cdot\frac{b}{\delta}\cdot\frac{c}{\delta},\frac{N}{\delta}}\right)=2$. We leave this as an exercise for the reader to prove. Now, let us revisit equation (4). We have, $16A^{2}=(4A)^{2}=d^{4}\cdot\left({\displaystyle\frac{N}{d}}\right)\cdot\left({\displaystyle\frac{N_{1}}{d}}\right)\cdot\left({\displaystyle\frac{N_{2}}{d}}\right)\cdot\left({\displaystyle\frac{N_{3}}{d}}\right)$ (7) The right-hand side of (7) is an integer, while the left-side is the square of a rational number. It is a standard exercise in elementary number theory to prove that if the square of a rational number is an integer, then the rational number itself must be an integer. In our situation this says that if $4A$ is a rational number (which is equivalent to saying that $A$ is rational), then in fact, $4A$ must be an integer. Hence, the left side of (7), will be an integer square when $A$ is rational. Now apply condition (6): Since the two factors ${\displaystyle\frac{N}{d}}$ and ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$ are relatively prime and the product is an integral square, each of them must be an integer square. (Another standard exercise in elementary number theory.) Thus, we must have ${\displaystyle\frac{N_{1}}{d}\cdot\frac{N_{2}}{d}\cdot\frac{N_{3}}{d}}=k^{2}$ (8) and ${\displaystyle\frac{N}{d}}=t^{2}$ for some integers $k$ and $t$. But we know that $N_{1}+N_{2}+N_{3}=N$ (refer to (2) or (3)), so that combined with ${\displaystyle\frac{N}{d}}=t^{2}$ we obtain, ${\displaystyle\frac{N_{1}}{d}+\frac{N_{2}}{d}+\frac{N_{3}}{d}}=t^{2}$ (9) Let $d_{12}$ be the greatest common divisor of ${\displaystyle\frac{N_{1}}{d}}$ and ${\displaystyle\frac{N_{2}}{d}}$; likewise let $d_{13}$ and $d_{23}$ be the greatest common divisors of ${\displaystyle\frac{N_{1}}{d}}$ and ${\displaystyle\frac{N_{3}}{d}}$, and ${\displaystyle\frac{N_{2}}{d}}$, and ${\displaystyle\frac{N_{3}}{d}}$; respectively. A cursory observation reveals that these three natural numbers must be mutually relatively prime (that is any two of them must be relatively prime). Why? Because if a prime divisor divided any two of them, it would have to divide all three ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d}}$, and ${\displaystyle\frac{N_{3}}{d}}$ thus violating the condition $\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)=1$. We can now write, $\left.\begin{array}[]{l}\left(d_{12},d_{13}\right)=\left(d_{12},d_{23}\right)=\left(d_{13},d_{23}\right)=1,\ {\displaystyle\frac{N_{1}}{d}}=d_{12}\cdot d_{13}\cdot M_{1}\\\ \\\ {\displaystyle\frac{N_{2}}{d}}=d_{12}\cdot d_{23}\cdot M_{2},{\displaystyle\frac{N_{3}}{d}}=d_{13}\cdot d_{23}\cdot M_{3},\\\ \\\ {\rm for\ some\ positive\ integers}\\\ \\\ M_{1},M_{2},M_{3}\ {\rm with}\ \left(M_{1},M_{2}\right)=\left(M_{1},M_{3}\right)=\left(M_{2},M_{3}\right)=1\end{array}\right\\}$ (10) If we combine the formulas of (10) with (8) we obtain $d^{2}_{12}\cdot d^{2}_{13}\cdot d^{2}_{23}\cdot M_{1}\cdot M_{2}\cdot M_{3}=k^{2}$ The latter equation clearly shows that the product $M_{1}\cdot M_{2}\cdot M_{3}$ must be an integer square $M_{1}=k^{2}_{1},\ \ \ \ M_{2}=k^{2}_{2},\ \ \ \ M_{3}=k^{2}_{3}$ (11) for natural numbers $k_{1},k_{2}$, and $k_{3}$. Combining (11) with the formulas in the second line of (10), and with (9) we arrive at the following necessary condition: $d_{12}\cdot d_{13}\cdot k^{2}_{1}+d_{12}\cdot d_{23}\cdot k^{2}_{2}+d_{13}\cdot d_{23}\cdot k^{2}_{3}=t^{2}$ (12) An immediate consequence of this necessary condition is the following result: As part of the hypothesis of Result 1, assume condition (6) Result 1: Let $a,b,c$ be the integer side-lengths of a triangle; and, $N_{1}=-a+b+c,N_{2}=a-b+c,\ N_{3}=a+b-c,\ d=(N_{1},N_{2},N_{3}),$ $d_{12}=\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d}}\right),d_{13}=\left({\displaystyle\frac{N_{1}}{d},\frac{N_{3}}{d}}\right),d_{23}=\left({\displaystyle\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)$. Also let $i_{1},i_{2},i_{3}$ be a permutation of $(1,2,3)$. (a) There is no triangle with all three $a,b,c,$ being odd or with , two of them even the other odd; and with the area $A$ being a rational number. Moreover, there exists no triangle with the area being a rational number and the three integers $N_{1},N_{2},N_{3}$ being divisible by the same highest power of $2$. (b) There exists no triangle with rational area, ${\displaystyle\frac{N_{i_{1}}}{d}}$ even, ${\displaystyle\frac{N_{i_{2}}}{d}}$ and ${\displaystyle\frac{N_{i_{3}}}{d}}$ both odd, and $d_{i_{1}i_{2}}\equiv 1\equiv d_{i_{1}i_{3}}$ (mod 4); or with $d_{i_{1}i_{2}}\equiv 3\equiv d_{i_{1}i_{3}}$ (mod 4). (c) If $d_{12}=d_{23}=d_{13}=1$, then there is no triangle with rational area $A$. Note: In part (b), since ${\displaystyle\frac{N_{i_{2}}}{d},\frac{N_{i_{3}}}{d}}$, are assumed to be odd, it follows from (10) that all three $d_{i_{1}i_{2}},d_{i_{1}i_{3}},d_{i_{2}i_{3}}$ must be odd. Proof: 1. (a) Since $a,b,c$ are odd or two of them even the other odd; or since $N_{1},N_{2},N_{3}$ are divisible by the same highest power of $2$; the integers ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$ are odd, we argue by contradiction: If such a triangle existed, then the necessary condition (12) would be satisfied. Because ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$ are odd, (10) and (11) imply that $d_{12},d_{13},d_{23},k_{1},k_{2}$ and $k_{3}$ must all be odd; thus (12) implies that $t$ must also be odd, since $t^{2}$ is the sum of three odd integers. But the square of any odd integer must be congruent to 1 modulo 4. Hence (12) would imply, $d_{12}\cdot d_{13}+d_{12}\cdot d_{23}+d_{13}\cdot d_{23}\equiv 1$ (mod 4). This congruence though is impossible, no matter what the three odd numbers $d_{12},d_{13},d_{23}$ are. Why? Because $d_{12}\cdot d_{13}+d_{12}\cdot d_{23}+d_{13}\cdot d_{23}$ is always congruent to 3 modulo 4. We leave it to the reader to verify that this is the case in each of the following combinations: $d_{12}\equiv d_{13}\equiv d_{23}$ (mod 4) (so either all three are $\equiv 1$ (mod 4); or $\equiv 3$ (mod 4); two of $d_{12},d_{13},d_{23}$ are congruent to 1 (mod 4); the other one congruent 3 (mod 4). Or vise versa. 2. (b) ${\displaystyle\frac{N_{i_{2}}}{d},\frac{N_{i_{3}}}{d}}$ are odd, it follows from (10) that $d_{i_{1}i_{2}},d_{i_{1}i_{3}},d_{i_{2}i_{3}}$, are all odd; and from (10) and (11) that $k_{i_{2}},k_{i_{3}}$ are both odd while $k_{i_{1}}$ is even by (10), (11), and the assumption that ${\displaystyle\frac{N_{i_{1}}}{d}}$ is even. Again, the argument is by contradiction: since $k^{2}_{i_{1}}\equiv 0$ (mod 4); and since the integer $t$ in (12) must be even, (12) would imply, $d_{i_{2}i_{3}}\cdot\left(d_{i_{1}i_{2}}+d_{i_{1}i_{3}}\right)\equiv 0$ (mod 4); and since $d_{i_{2}i_{3}}$ is odd, the last congruence would imply $d_{i_{1}i_{2}}+d_{i_{1}i_{3}}\equiv 0$ (mod 4), which is impossible, since the hypothesis in part (b) implies $d_{i_{1}i_{2}}+d_{i_{1}i_{3}}\equiv 2$ (mod 4). 3. (c) Since $d_{12}=d_{23}=d_{13}=1$, (12) implies $k^{2}_{1}+k^{2}_{2}+k^{2}_{3}=t^{2}$. But the integers $k_{1},k_{2},k_{3}$ are mutually relatively prime; hence either all three are odd or one of them is even, the other two odd. Thus $k^{2}_{1}+k^{2}_{2}+k^{2}_{3}\equiv 3$ or 2(mod 4); but $t^{2}\equiv 1$ or 0(mod 4), and so we have a contradiction. An Immediate Consequence of Result 1(a): If an integer-sided triangle satisfying (6) has rational area $A$, then $A$ must be, in fact, an even integer. To see why, consider the numbers $N_{1},N_{2},N_{3}$, as defined in (2). Clearly either all three are odd or all three are even. If they are all odd, then that means, either the three sides $a,b,c$ are odd; or two are even, the third odd; but then, according to Result 1(a), such a triangle cannot have rational area. Therefore, if $A$ is rational, all three $N_{1},N_{2},N_{3}$ must be even, and consequently so must be the perimeter $N=N_{1}+N_{2}+N_{3}$. According to (4), $A=\sqrt{\frac{N\cdot N_{1}\cdot N_{2}\cdot N_{3}}{16}}$; however, the product $N\cdot N_{1}\cdot N_{2}\cdot N_{3}$ will be divisible by 16; thus, $A$ will be the square root of the integer; so if $A$ is rational, it must be an integer. To see that $A$ must be even, observed that $A$ would be odd only when each of $N_{1},N_{2},N_{3},N$ is exactly divisible by $2$; so that the product $N_{1}N_{2}N_{3}N$ is exactly divisible by 16; but that would mean, in particular, that the highest power of $2$ dividing all three $N_{1},N_{2},N_{3}$ would be $2^{1}$; this is precluded by Result 1a though. ## 4 The General Solution to Diophantine Equation (1): This can be found in some number theory books. The derivation of the general solution (1) can be found in W. Sierpinski’s book Elementary Theory of Numbers see [1]. Every solution $(x,y,z,t)$, in positive integers $x,y,z$, and $t$, of equation (1) must be of the form, $x={\displaystyle\frac{l^{2}+m^{2}-n^{2}}{n}},y=2l,\ z=2m,\ t={\displaystyle\frac{l^{2}+m^{2}+n^{2}}{n}}$ where $m,n,l$ are positive integers, $n$ is a divisor of $l^{2}+m^{2}$ and $n<\sqrt{l^{2}+m^{2}}$ (13) And conversely, any quadruple $(x,y,z,t)$ that satisfies (13) must be a solution in positive integers $x,y,z,t$, of equation (1). Below, we make the following remarks or observations: Observation 1: If $(x,y,z,t)$ is a (positive) integer solution to (1), note that an argument modulo 4 shows that at least two of $x,y,z,t$ must be even integers. Observation 2: If $(x,y,z,t)$ is a solution to (1) with $y$ and $z$ even, then the numbers $l={\displaystyle\frac{y}{2}},\ m={\displaystyle\frac{z}{2}},n={\displaystyle\frac{t-x}{2}}$ are clearly uniquely determined; which shows that every solution of the equation (1), in positive integer $x,y,z,t$ with $y,z$ even, is obtained exactly once by the use of formulas (13). Observation 3: In order to eliminate solutions of (1) with interchanged unknowns, we may reject pairs, $l,m$ for which $m>l$ and take only these $n$ for which the numbers $x$ are odd. Thus, eliminating also all the solutions of (1) for which $x,y,z$, and $t$ are even. To include them again, it is sufficient to multiply each of the solutions with odd $x$ by the powers of 2 successively. Below, we present a sample of fifteen solutions to (1), with $m\leq l$ and $x$ odd. A Sample of Solutions (with $m\leq l$ and $x$ odd) $\begin{array}[]{|c|c|c|c|c|c|c|c|}\hline\cr\ \ \ L&\ \ \ m&\ \ \ l^{2}+m^{2}&\ \ \ n&\ \ \ x&\ \ \ y&\ \ \ Z&\ \ \ t\\\ \hline\cr 1&1&2&1&1&2&2&3\\\ \hline\cr 3&3&18&3&3&6&6&9\\\ \hline\cr 5&1&26&1&25&10&2&27\\\ \hline\cr 5&3&34&1&33&10&6&35\\\ \hline\cr 5&5&50&1&49&10&10&51\\\ \hline\cr 6&4&52&1&51&12&8&53\\\ \hline\cr 7&3&58&1&57&14&6&59\\\ \hline\cr 7&7&98&1&97&14&14&99\\\ \hline\cr 8&2&68&1&67&16&4&69\\\ \hline\cr 9&3&90&5&13&18&6&23\\\ \hline\cr 9&3&90&6&9&18&6&21\\\ \hline\cr 9&5&106&2&51&18&10&55\\\ \hline\cr 9&7&130&10&3&18&14&23\\\ \hline\cr 9&9&162&1&161&18&18&163\\\ \hline\cr 9&9&162&9&9&18&18&27\\\ \hline\cr\end{array}$ ## 5 Result 2 and Generating Integer-sided Triangles with Rational Area: We state the result: Result 2: If $(x,y,z,t)$ is any solution, in positive integers $x,y,z$, and $t$, to equation (1), then any triangle with integer sides $a,b,c$, defined below, has rational area; $a={\displaystyle\frac{D(y^{2}+z^{2})}{2}},b={\displaystyle\frac{D(x^{2}+z^{2})}{2}},c={\displaystyle\frac{D(x^{2}+y^{2})}{2}},$ where $D$ is a positive integer, if all three $x,y,z$ are even; and $D$ is an even positive integer, if two of $x,y,z$ are even, the third one odd. Note: Recall that any solution to (1), requires that at least two of $x,y,z$ must be even. Also note that when one of $x,y,z$ is odd, $D$ must be even, otherwise two of $a,b,c$, would not be integers. Proof: We compute four quantities in terms of $x,y,z$, and $t$: $s={\displaystyle\frac{a+b+c}{2}}={\displaystyle\frac{D\cdot(x^{2}+y^{2}+z^{2})}{2}}={\displaystyle\frac{D\cdot t^{2}}{2}}$, since $x^{2}+y^{2}+z^{2}=t^{2}$ according to hypothesis. Next, we compute $s-a={\displaystyle\frac{D\cdot(x^{2}+y^{2}+z^{2})}{2}}={\displaystyle\frac{D\cdot(y^{2}+z^{2})}{2}}={\displaystyle\frac{D\cdot x^{2}}{2}}$. Likewise $s-b={\displaystyle\frac{D\cdot(x^{2}+y^{2}+z^{2})}{2}}-{\displaystyle\frac{D\cdot(x^{2}+z^{2})}{2}}={\displaystyle\frac{D\cdot y^{2}}{2}}$. And $s-c={\displaystyle\frac{D\cdot(x^{2}+y^{2}+z^{2}}{2}}-{\displaystyle\frac{D\cdot(x^{2}+y^{2})}{2}}={\displaystyle\frac{D\cdot z^{2}}{2}}$. Now apply Heron’s formula: $A=\sqrt{s\cdot(s-a)\cdot(s-b)\cdot(s-c)};\ A={\sqrt{\frac{D^{4}\cdot x^{2}\cdot y^{2}\cdot z^{2}\cdot t^{2}}{2^{4}}}}={\displaystyle\frac{D^{2}\cdot x\cdot y\cdot z\cdot t}{4}}$, which proves that $A$ is a rational number, in fact an integer since at least two of the $x,y,z$ are even. Remark 2: Note that if in (12), each of the numbers $d_{12}\cdot d_{13},d_{12}$ $\cdot d_{23},d_{13}\cdot d_{23}$ is an integer square, then by starting with an integer -sided triangle with rational area, one can produce a solution to (1). Also, since $d_{12},d_{13},d_{23}$, are pairwise relatively prime, then all three products $d_{12}\cdot d_{13},d_{12}\cdot d_{23},d_{13}\cdot d_{23}$ will be perfect squares, if and only if each of the three numbers $d_{12},d_{13},d_{23}$ is a square itself; the smallest such values are (without loss of generality) $d_{12}=4$ and $d_{13}=d_{23}=1$. This can happen with $k_{3}$ odd in (12); more specifically, if we $k_{1}=k_{2}=k_{3}=1$ and $t=3$ then (12) is satisfied. A calculation shows that (see (11) and (10)), that if we also take $d=2$, then $N_{1}=8,N_{2}=8,N_{3}=2$; so that (from (3)) $a=8,b=5,c=5$, and the area is $A=12$. Below, we use the table of sample solutions of equation (1) (found in the previous section) to generate integer-sided triangles with rational area. For this we use the formulas, found in Result 2, that generate the three sides $a,b,$ and $c$. We take $D=2$, since in the table of sample solution to (1), $x$ is odd. As it will become clear, in the next section (Section 6), all the numbers listed below are solid rectangular numbers. $\begin{array}[]{|c|c|c|c|c|c|c|l|}\hline\cr x&y&z&t&a&b&c&\begin{array}[]{rcl}A{\rm(area)}&=&{\displaystyle\frac{D^{2}\cdot x\cdot y\cdot z\cdot t}{4}}\\\ &=&x\cdot y\cdot z\cdot t(D=2)\end{array}\\\ \hline\cr 1&2&2&3&8&5&5&12=2^{2}\cdot 3\\\ \hline\cr 3&6&6&9&72&45&45&972=2^{2}\cdot 3^{5}\\\ \hline\cr 25&10&2&27&104&627&725&13,500=2^{2}\cdot 3^{3}\cdot 5^{3}\\\ \hline\cr 33&10&6&35&136&1,125&1,189&69,300=2^{2}\cdot 3^{2}\cdot 5\cdot 7\cdot 11\\\ \hline\cr 49&10&10&51&200&2,501&2,501&249,900=2^{2}\cdot d\cdot 5^{2}\cdot 7^{2}\cdot 17\\\ \hline\cr 51&12&8&53&208&2,665&2,745&259,488=2^{5}\cdot 3^{2}\cdot 17\cdot 53\\\ \hline\cr 57&14&6&59&232&3,285&3,445&282,492=2^{2}\cdot 3^{2}\cdot 7\cdot 19\cdot 59\\\ \hline\cr 97&14&14&99&392&9,605&9,605&1,882,188=2^{3}\cdot 3^{4}\cdot 7^{2}\cdot 11^{2}\\\ \hline\cr 67&16&4&69&272&4,505&4,745&295,872=2^{6}\cdot 3\cdot 23\cdot 67\\\ \hline\cr 13&18&6&23&360&204&493&32,292=2^{2}\cdot 3^{3}\cdot 13\cdot 23\\\ \hline\cr 9&18&6&21&360&81&405&20,412=2^{2}\cdot 3^{6}\cdot 7\\\ \hline\cr 51&18&10&55&424&2,701&2,925&504,900=2^{2}\cdot 3^{3}\cdot 5^{2}\cdot 11\cdot 17\\\ \hline\cr 3&18&14&23&520&205&333&17,388=2^{2}\cdot 3^{3}\cdot 7\cdot 23\\\ \hline\cr 161&18&18&163&648&26,245&26,245&8,502,732=2^{2}\cdot 3^{4}\cdot 7\cdot 23\cdot 163\\\ \hline\cr 9&18&18&27&648&415&415&78,732=2^{2}\cdot 3^{9}\\\ \hline\cr\end{array}$ ## 6 A complete description of triangle area numbers and solid rectangular numbers. In this section we drop condition (6) of Section 3. The purpose of (6) was to simplify the situation in order to establish a connection with equation (1). So, we will only assume the validity of equations (1) through (5) which were established in Section 1. Let us reconsider the integers ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$, and ${\displaystyle\frac{N}{d}}$. Because of (5), if $D_{1}$ is the greatest common divisor of ${\displaystyle\frac{N_{1}}{d}}$ and ${\displaystyle\frac{N}{d}}$, then $D_{1}$ must be relatively prime to both ${\displaystyle\frac{N_{2}}{d}}$, and ${\displaystyle\frac{N_{3}}{d}}$; for if, say $D_{1}$ had a prime divisor in common with ${\displaystyle\frac{N_{2}}{d}}$, then (recall that) ${\displaystyle\frac{N_{1}}{d}+\frac{N_{2}}{d}+\frac{N_{3}}{d}=\frac{N}{d}}$ would imply that that prime divisor would also be a divisor of ${\displaystyle\frac{N_{3}}{d}}$, contrary to the fact that $\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)=1$. Likewise, by applying the same reasoning, we see that $D_{2}$, the greatest common divisor of ${\displaystyle\frac{N_{2}}{d}}$ and ${\displaystyle\frac{N}{d}}$, must be relatively prime to both ${\displaystyle\frac{N_{1}}{d}}$ and ${\displaystyle\frac{N_{3}}{d}}$. And if $D_{3}=\left({\displaystyle\frac{N_{3}}{d},\frac{N}{d}}\right)$, we must have $\left(D_{3},{\displaystyle\frac{N_{1}}{d}}\right)=1=\left(D_{3},{\displaystyle\frac{N_{2}}{d}}\right)$. Next, let $d_{12}=\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d}}\right),\ d_{13}=\left({\displaystyle\frac{N_{1}}{d},\frac{N_{3}}{d}}\right),\ d_{23}=\left({\displaystyle\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)$. Again as before, by virtue of ${\displaystyle\frac{N}{d}=\frac{N_{1}}{d}+\frac{N_{2}}{d}+\frac{N_{3}}{d}}$ and $\left({\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)=1$, we conclude that the three integers $d_{12},d_{13},d_{23}$, are mutually relatively prime. We can write, ${\displaystyle\frac{N_{1}}{d}}=D_{1}d_{12}d_{13}M_{1},{\displaystyle\frac{N_{2}}{d}}=D_{2}d_{12}d_{23}M_{2},\ {\displaystyle\frac{N_{3}}{d}}=D_{3}d_{13}d_{23}M_{3}$, and ${\displaystyle\frac{N}{d}}=D_{1}D_{2}D_{3}M$; where the positive integers $D_{1},\ D_{2},\ D_{3},\ d_{12},\ d_{13},\ d_{23},\ M_{1},\ M_{2},\ M_{3}$, and $M$ satisfy the following seven coprimeness conditions: (i) $(D_{1},\ D_{2}\ d_{23}\ M_{2})=(D_{1},\ D_{3}\ d_{23}\ M_{3})=1$ (ii) $(D_{2},\ D_{1}\ d_{13}\ M_{1})=(D_{2},\ D_{3}\ d_{13}\ M_{3})=1$ (iii) $(D_{3},\ D_{1}\ d_{12}\ M_{1})=(D_{3},\ D_{2}\ d_{12}M_{2})=1$ (iv) $(d_{12},\ d_{13})=(d_{12},\ d_{23})=(d_{13},\ d_{23})=1$ (14) (v) $(M,\ d_{12}\ d_{13}\ d_{23}\ M_{1}\ M_{2}\ M_{3})=1$ (vi) $(D_{1}\ D_{2}\ D_{3},\ d_{12}\ d_{13},\ d_{23})=1$ (vii) $(M_{1},\ M_{2})=(M_{1},\ M_{3})=(M_{2},\ M_{3})=1$ According to (4), $16A^{2}=N\cdot N_{1}\cdot N_{2}\cdot N_{3}$; using the four formulas in (14) we obtain, $16A^{2}=d^{4}\cdot D^{2}_{1}D^{2}_{2}D^{2}_{3}d^{2}_{12}d^{2}_{13}d^{2}_{23}\cdot M_{1}M_{2}M_{3}M$ (15) Since $4A$ is a rational number and the right-hand side of (15) is an integer, $(4A)^{2}$ must be an integer; which means that $4A$ must actually be an integer (we have already come across this argument in Section 3 - see below equation (7)). Therefore $(4A)^{2}$ is an integer square so that (15) implies that $M_{1}M_{2}M_{3}M$ must be an integer square as well: but according to the coprimeness conditions of (14), the integers $M_{1},M_{2},M_{3}$, and $M$ are actually mutually relatively prime, thus each of them must be an integer square. We have, $\left.\begin{array}[]{l}M_{1}=k^{2}_{1},\ M_{2}=k^{2}_{2},\ M_{2}=k^{2}_{3},\ M=k^{2}\\\ \\\ {\rm for\ positive\ integers}\ k_{1},k_{2},k_{3},k\ {\rm satisfying}\\\ \\\ (k_{1},k_{2})=(k_{1},k_{3})=(k_{2},k_{3})=1=(k,k_{1}k_{2}k_{3}).\end{array}\right\\}$ (16) Applying the four formulas in (14) and those in (16), and combining these with the equation ${\displaystyle\frac{N_{1}}{d}+\frac{N_{2}}{d}+\frac{N_{3}}{d}=\frac{N}{d}}$ we arrive at, $D_{1}d_{12}d_{13}k^{2}_{1}+D_{2}d_{12}d_{23}k^{2}_{2}+D_{3}d_{13}d_{23}k^{2}_{3}=D_{1}D_{2}D_{3}\cdot k^{2}$ (17) We need two Lemmas: Lemma 1: If $A,B,C$ are odd integers, then $AB+BC+AC\equiv 3$(mod 4). Proof: Apply the identity $(A+B+C)^{2}=A^{2}+B^{2}+C^{2}+2(AB+AC+BC)$. Since $A+B+C\equiv 1$(mod 2), we have $(A+B+C)^{2}\equiv 1$(mod 8). Hence, from $(A+B+C)^{2}\equiv A^{2}\equiv B^{2}\equiv C^{2}\equiv 1$(mod 8) and the above identity we obtain $(A+B+C)^{2}-(A^{2}+B^{2}+C^{2})\equiv 2(AB+AC+BC)$(mod 8) $\Rightarrow 1-3\equiv 2(AB+AC+BC)$(mod 8); $6\equiv 2(AB+AC+BC)$(mod 8) $\Rightarrow AB+BC+AC\equiv 3$(mod 4) Lemma 2: In equation (17), at least one of the following nine integers $D_{1},D_{2},D_{3},d_{12},d_{13},d_{23},k_{1},k_{2},k_{3}$, must be even. Proof: If all these nine integers were odd, then so would the integer $k$, as a congruence modulo 2 in (17) shows. We would have $k^{2}_{1}\equiv k^{2}_{2}\equiv k^{2}_{3}\equiv k^{2}\equiv 1$(mod 4) and if we multiply both sides of (17) by the product $D_{1}D_{2}D_{3}$ and also make use of $D^{2}_{1}D^{2}_{2}D^{2}_{3}\equiv 1\equiv D^{2}_{1}\equiv D^{2}_{2}\equiv D^{2}_{3}$(mod 4), we end up with the congruence, $D_{2}D_{3}d_{12}d_{13}+D_{1}D_{3}d_{12}d_{23}+D_{1}D_{2}d_{13}d_{23}\equiv{\rm(mod4)}$ (18) However, by applying Lemma 1, with $A=D_{1}D_{2}d_{12},B=D_{2}D_{3}d_{23}$, and $C=D_{1}D_{3}d_{13}$ we obtain $D^{2}_{2}D_{1}D_{3}d_{12}d_{23}+D^{2}_{1}D_{2}D_{3}d_{12}d_{13}+D_{1}D_{2}d_{13}d_{23}\equiv 3{\rm(mod\ 4)},$ And since $D^{2}_{1}\equiv D^{2}_{2}\equiv D^{2}_{3}\equiv 1$(mod 4), the last congruence implies $D_{1}D_{3}d_{12}d_{23}+D_{2}D_{3}d_{12}d_{13}+D_{1}D_{2}d_{13}d_{23}\equiv 3\rm{(mod4)},$ contradicting congruence (18). We now go back to the four formulas found in (14), combining them with (16) and (3) to obtain the following set of formulas: $\begin{array}[]{cr}{\displaystyle\frac{N_{1}}{d}}=D_{1}d_{12}d_{13}k^{2}_{1},\ {\displaystyle\frac{N_{2}}{d}}=D_{2}d_{12}d_{23}k^{2}_{2},\ {\displaystyle\frac{N_{3}}{d}}=D_{3}d_{13}d_{23}k^{2}_{3},&\\\ \\\ {\displaystyle\frac{N}{d}}=D_{1}D_{2}D_{3}k^{2};\ {\rm and\ area}\ A=\sqrt{{\displaystyle\frac{N\cdot N_{1}N_{2}N_{3}}{16}}};&\\\ \\\ 4A=D_{1}D_{2}D_{3}d_{12}d_{23}d_{13}k_{1}k_{2}k_{3}k\cdot d^{2};&\\\ \\\ {\rm and\ sides}\ a,b,c\ {\rm given\ by},&\hskip 43.36243pt(19)\\\ \\\ a={\displaystyle\frac{d\cdot d_{23}\cdot(D_{2}d_{12}k^{2}_{2}+D_{3}d_{13}k^{2}_{3})}{2}}&\\\ \\\ b={\displaystyle\frac{d\cdot d_{13}\cdot(D_{1}d_{12}k^{2}_{1}+D_{3}d_{23}k^{2}_{3})}{2}}&\\\ \\\ c={\displaystyle\frac{d\cdot d_{12}\cdot(D_{1}d_{13}k^{2}_{1}+D_{2}d_{23}k^{2}_{2})}{2}}&\end{array}$ Now, apply Lemma 2. We know that at least one of the numbers $k_{1},k_{2},k_{3},d_{12}$, $d_{13},d_{23},D_{1},D_{2},D_{3}$, must be even. This clearly shows in (19) that one or two of ${\displaystyle\frac{N_{1}}{d},\frac{N_{2}}{d},\frac{N_{3}}{d}}$ will be even, $\left(\\!\\!{\rm it\,can\,not\,be\,all\,of\,them\,because}\,\left({\displaystyle\frac{N_{1}}{d},}\right.\right.$ $\left.{\displaystyle\frac{N_{2}}{d},\frac{N_{3}}{d}}\right)$ $\left.\begin{array}[]{c}\\\ =1\\\ \\\ \end{array}\right)$. This implies that $d$ must be an even integer. Why? Because if $d$ were odd, then one or two of the integers $N_{1},N_{2},N_{3}$ would be even; but the formulas in (3) would then imply that not all $a,b,c$ are integers, contrary to the central assumption of this paper, namely that $a,b,c,$ are the integer side lengths of a triangle whose area is rational. Hence, $d$ must be even. Since $d$ is even, and at least one of $k_{1},k_{2},k_{3},d_{12},$ $d_{13},d_{23},D_{1},D_{2}D_{3}$ is even, the formula that gives the area $A$ in (19) implies that the area $A$ is in fact an even integer. Note 1: If we take a close look at the coprimeness conditions in (14), the formulas in (16), and equation (17), we see that at most two of the nine integers $k_{1},k_{2},k_{3},d_{12},d_{13},d_{23},D_{1},D_{2},D_{3}$, can be even. In fact if two of these nine integers are even, then one is $d_{i_{1}i_{2}}$ and the other $k_{i_{1}}$; or alternatively one is $D_{i_{1}}$ and the other $k_{i_{1}}$ where $(i_{1},i_{2},i_{3})$ is a permutation of $(1,2,3)$. As the listing of all (not exceeding 999) triangle area numbers (at the end of this section) shows, some area numbers are multiples of 4 and some are congruent to 2 modulo 4. But as it turns out, all of them are divisible by 3. We have the following. Result 3 (a) Every triangle area number $A$ is an even integer. (b) If $A\equiv 2$(mod 4), then either exactly one of $d_{12},d_{13},d_{23}$ is congruent to 2 modulo 4; the other two are both odd; all the integers $k_{1},k_{2},k_{3},k,D_{1},D_{2},D_{3}$ are odd; and $d\equiv 2$(mod 4); or alternatively, exactly one of $D_{1},D_{2},D_{3}$ is congruent to 2(mod 4); the other two are both odd; and $k_{1},k_{2},k_{3},k,d_{12},d_{13},d_{23}$ are odd; and $d\equiv 2$(mod 4). (c) Each triangle area number $A$ is divisible by $3$. Proof: 1. (a) This has already been shown and explained above (see explanation below (19)). 2. (b) Suppose $A\equiv 2$(mod 4). Then by the formula for $4A$ in (19), the fact that $d$ is even, and Lemma 2, it follows that $d\equiv 2$(mod 4) and that exactly one of the nine integers $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3},k_{1},k_{2},k_{3}$, must be congruent to 2 modulo 4; the other eight of them must all be odd, as well as the integer $k$. To conclude the proof, it suffice to demonstrate that none of the integers $D_{1},D_{2},D_{3},k_{1},k_{2},k_{3}$ can be congruent to 2 mod 4. Suppose that one of $k_{1},k_{2},$ or $k_{3}$ is congruent to 2 mod 4; without loss of generality, say $k_{1}\equiv 2$(mod 4), and $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3}$, and $k$ are all odd. We have a contradiction modulo 2, since the left-hand side of (17) will be an even integer, but the right-hand side will be odd. 3. (c) To establish that $A$ must be divisible by $3$, we show that (at least) one of the ten integers $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3},k_{1},k_{2},k_{3},k$ must be divisible by 3; it would then follow by the formula for $4A$ in (19), that $A$ must be a multiple of $3$. We argue by contradiction; let us suppose that none of the ten numbers $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3},k_{1},k_{2},k_{3},k$, is divisible by 3. We have $k^{2}_{1}\equiv k^{2}_{2}\equiv k^{2}_{3}\equiv k^{2}\equiv 1$(mod 3); applying this to (17) considered modulo 3 we obtain, $D_{1}D_{12}d_{13}+D_{2}d_{12}d_{23}+D_{3}d_{13}d_{23}\equiv D_{1}D_{2}D_{3}{\rm(mod3)}$ (20) Since $(d_{12}d_{13})\cdot(d_{12}d_{13})\cdot(d_{12}d_{13})=d^{2}_{12}d^{2}_{13}d^{2}_{23}\equiv 1$(mod 3); 1. (i) either each of the three products $d_{12}d_{13},d_{12}d_{23},d_{13}d_{23}$, is congruent ot 1 modulo 3 or 2. (ii) two of these three products are congruent to $-1$ modulo 3, the third congruent to 1 modulo 3. If (i) is the case then (20) implies, $D_{1}+D_{2}+E_{3}\equiv D_{1}D_{2}D_{3}$(mod 3). Each of $D_{1},D_{2},D_{3}$ can be either $1$ or $-1$ (mod 3). No combination of nonzero values modulo 3 of $D_{1},D_{2},D_{3}$ can satisfy the latter congruence; it is impossible. If (ii) is the case, we can take, without loss of generality, $d_{12}d_{13}\equiv 1$ and $d_{12}d_{23}\equiv d_{13}d_{23}\equiv-1$(mod 3). Using (21) we obtain $D_{1}-D_{2}-D_{3}\equiv D_{1}D_{2}D_{3}$(mod 3). Again this has no solution with $D_{1}D_{2}D_{3}\not\equiv 0$(mod 3). To see this more clearly consider the fact that at least two of $D_{1}D_{2}D_{3}$ must have the same value mod 3: $D_{1}\equiv D_{2}$ or $D_{1}D_{2}D_{3}$ or $D_{2}\equiv D_{3}$ (the “or” is not exclusive here); each of these three possibilities combined with $D_{1}-D_{2}-D_{3}\equiv D_{1}D_{2}D_{3}$(mod 3), implies $D_{1}D_{2}D_{3}\equiv 0$(mod 3), in violation of $D_{1}D_{2}D_{3}\not\equiv 0$(mod 3); we omit the details. End of proof. $\square$ We now turn our attention to (17): If each of the integers $D_{1}d_{12}d_{13},D_{2}d_{12}d_{23},$ $D_{3}d_{13}d_{23}$, and $D_{1}D_{2}D_{3}$ is an integer square, equation (17) procudes a solution to (1); also, according to Note 2 (above Result 3), the six integers $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3}$ are mutually relatively prime; thus $D_{1}d_{12}d_{13},D_{2}d_{12}d_{23},$ $D_{3}d_{13}d_{23}$, and $D_{1}D_{2}D_{3}$ will all be the integer squares if and only if each integer $d_{12},d_{13},d_{23},D_{1},D_{2},D_{3}$ is an integer square. We set $D_{1}=L^{2}_{1},\ D_{2}=L^{2}_{2},\ D_{3}=L^{2}_{3},\ d_{12}=\delta^{2}_{12},\ d_{13}=\delta^{2}_{13},\ d_{23}=\delta^{2}_{23}$ (21) with $L_{1},L_{2},L_{3},\delta_{12},\delta_{13},\delta_{23}$ mutually relatively prime. By (21) and (17) we obtain, $(L_{1}\cdot\delta_{12}\cdot\delta_{13}\cdot k_{1})^{2}+(L_{2}\cdot\delta_{12}\cdot\delta_{23}\cdot k_{2})^{2}+(L_{3}\cdot\delta_{13}\cdot\delta_{23}\cdot k_{3})^{2}=(L_{1}L_{2}L_{3}k)^{2}$ (22) Now it is clear if $(x_{0},y_{0},z_{0},t_{0})$ is a solution to (1), we must have (up to symmetry) according to (22), $x_{0}=L_{1}\delta_{12}\delta_{13}k_{1},\ \ y_{0}=L_{2}\delta_{12}\delta_{23}k_{2},\ \ z_{0}=L_{3}\delta_{13}\delta_{23}k_{3},\ \ t_{0}=L_{1}L_{2}L_{3}k.$ (23) By considering the coprimeness conditions of (14) and Note 2 (above Result 3), we can make the following observation: 1. (i) The solution $(x_{0},y_{0},z_{0},t_{0})$ to (1) must satisfy $(x_{0},y_{0},z_{0})=1$ 2. (ii) $\delta_{12}=(x_{0},y_{0}),\ \delta_{13}=(x_{0},z_{0}),\ \delta_{23}=(y_{0},z_{0})$ 3. (iii) $L_{1}=\left({\displaystyle\frac{x_{0}}{(x_{0},y_{0})\cdot(x_{0},z_{0})}},t_{0}\right),\ L_{2}=\left({\displaystyle\frac{y_{0}}{(y_{0},x_{0})\cdot(y_{0},z_{0})}},t_{0}\right)$, and $L_{3}=\left({\displaystyle\frac{z_{0}}{(z_{0},x_{0})\cdot(z_{0},y_{0})}},t_{0}\right)$ The three conditions (i), (ii), and (iii) clearly shows that given any solution $\\{x_{0},y_{0},z_{0},t_{0}\\}$ to (1), with $(x_{0},y_{0},z_{0})=1$, the integers $\delta_{12},\delta_{13},\delta_{23},L_{1},L_{2}$, and $L_{3}$ are uniquely determined up to symmetry or permutations; hence by (23), it follows that the $k_{1},k_{2},k_{3},k$ are likewise uniquely determined. If we make use of the area formula (19), combined with (21) and (23), a computation yields $A={\displaystyle\frac{x_{0}\cdot y_{0}\cdot z_{0}\cdot d^{2}}{4}}$ (24) Note that according to Observation 1 of Section 3, in any solution $(x,y,z,t)$ to (1), at least two of $x,y,z$ must be even. thus, since $d$ is even, (24) clearly shows that the area number $A$ must be divisible by 4. Definition 2: Let $(x_{0},y_{0},z_{0},t_{0})$ be a positive integer solution to (1) that satisfies $(x_{0},y_{0},z_{0})=1$, and $d$ any even positive integer. The positive integer $A={\displaystyle\frac{x_{0}\cdot y_{0}\cdot z_{0}\cdot d^{2}}{4}}$ is called a solid rectangular number. It is clear by now that every solid rectangular number arises directly from equation (1); moreover every solid rectangular number is the area of a unique triangle (up to the geometric transformations of reflection and symmetry): Indeed, by using the formulas for the sides $a,b$, and $c$ in (19) in conjunction with (21) and (23) a computation implies, $\left\\{\begin{array}[]{ll}a={\displaystyle\frac{d\cdot(y^{2}_{0}+z^{3}_{0})}{2}},\ b={\displaystyle\frac{d\cdot(x^{2}_{0}+z^{2}_{0})}{2}},\ c={\displaystyle\frac{d\cdot(x^{2}_{0}+y^{2}_{0})}{2}}\\\ \\\ {\rm with}\ (x_{0},y_{0},z_{0})=1\end{array}\right\\}$ (25) Note that the formulas in (25) are equivalent with the formulas of Result 2 in Section 5. Indeed, since the formulas of (25) hold under condition $(x_{0},y_{0},z_{0})=1$, they obviously satisfy the formulas of Result 2. Conversely, if we factor out the greatest common divisor $(x_{0},y_{0},z_{0})$ of $x_{0},y_{0},z_{0}$ in the formulas of Result 2, we immediately see that (25) is satisfied. Hence these formulas in (25) (or in Result 2) generate and describe all the triangles whose areas satisfy (24), i.e., they are solid rectangular numbers. We end this paper by listing all triangle area numbers not exceeding 999; also listing two important subsets; the Pythagorean numbers and the solid rectangular numbers. We must note here that there are many, in fact infinitely many, triangle area numbers that correspond to different triangles. It has been know for quite some time that there exist infinitely many pairs of Pythagorean triangles or triples with the same area number (Pythagorean numbers). In fact, there exist infinitely many triples of Pythagorean triangles with the same area number, explicitly parametrically described, refer to [2] for more details. There are exactly 96 triangle area, one digit, two digit, and three digit numbers. Triangle Area Numbers $\leq$ 999: 6, 12, 24, 30, 36, 42, 48, 54, 60, 66, 72, 84, 90, 96, 108, 114, 120, 126, 132, 144, 150, 156, 168, 180, 192, 198, 204, 210, 216, 234, 240, 252, 264, 270, 288, 294, 300, 306, 324, 330, 336, 360, 378, 384, 390, 296, 408, 420, 432, 456, 462, 468, 480, 486, 504, 510, 522, 528, 540, 546, 570, 576, 588, 594, 600, 624, 630, 648, 660, 672, 684, 690, 714, 720, 726, 744, 750, 756, 768, 780, 792, 798, 810, 816, 840, 864, 876, 900, 924, 930, 936, 960, 966, 972, 984, 990. Triangle Area Numbers that are also Pythagorean: 6, 24, 30, 54, 60, 84, 96, 120, 150, 180, 210, 216, 270, 294, 330, 384, 420, 480 486, 504, 540, 546, 600, 630, 726, 750, 840, 864, 924, 960, 990. Triangle Area Numbers that are also solid Rectangular: 12, 972. Note that for example, the Pythagorean number 210 corresponds to different right triangles or triples: the triangle with sides $a=37,b=35$, and $c=12$; and the triangle with sides $a=20,b=21$, and $c=20$; they both have area number 210. If we look at the list of 96 area triangle numbers not exceeding 999, we will see (by a quick count) that exactly twenty-one of them are congruent to 2 mod 4 (refer to Result 2 for the special conditions that must be satisfied when $A\equiv 2$(mod 4), in addition to the coprimeness conditions (14)). So when $A\equiv 2$(mod 4), we have the following list: 6, 42, 66, 90, 114, 126, 198, 234, 306, 390, 462, 510, 522, 570, 594, 690, 714, 798, 810, 966, 990. Below we list, for each value of $A\equiv 2$(mod 4) (only for the first seven values of $A$), the corresponding values of the integers $D_{1},D_{2},D_{3},d_{12},d_{13},d_{23},$ $k_{1},k_{2},k_{3},k$ and $d$, as well as the side lengths $a,b,c$ of the triangle(s) whose area is the given number $A$. Recall that $D_{1},D_{2},D_{3},d_{12},d_{13},d_{23},k_{1},k_{2},k_{3},k$, must satisfy the coprimeness conditions (14) and equation (17). Table $\begin{array}[]{||c|c|c|c|c|c|c|c|c|c|c|c|c|c|c||}\hline\cr\hline\cr D_{1}&D_{2}&D_{3}&d_{12}&d_{13}&d_{23}&k_{1}&k_{2}&k_{3}&k&d&a&b&c&A\\\ \hline\cr 1&2&3&1&1&1&1&1&1&1&2&5&4&3&6=2\cdot 3\\\ \hline\cr 1&3&7&1&1&2&1&1&1&1&2&20&15&7&42=2\cdot 3\cdot 7\\\ \hline\cr 2&1&11&1&1&1&1&3&1&1&2&20&13&11&66=2\cdot 3\cdot 11\\\ \hline\cr 1&1&3&2&1&5&1&1&1&3&2&25&17&12&90=2\cdot 3^{2}\cdot 5\\\ \hline\cr 1&1&19&1&1&1&1&3&1&1&2&37&20&19&114=2\cdot 3\cdot 19\\\ \hline\cr 2&3&1&1&1&1&1&1&7&3&2&52&51&3&126=2\cdot 3^{2}\cdot 7\\\ \hline\cr 1&11&6&1&1&1&1&1&3&1&2&65&55&12&198=2\cdot 3^{2}\cdot 11\\\ \hline\cr\hline\cr\end{array}$ Prove or Disprove Conjecture: No Pythagorean number is a solid rectangular number. ## References * [Sierpinski] ierpinski, W., Elementary Theory of Numbers, p. 67, Warsaw, 1964. * [Beiler] ieler, A. H., Recreations in the Theory of Numbers, Dover Publications, Inc., New York, 1966. AMS Classification Numbers 11A99, 11A25, 11D9.

arxiv-papers

2024-09-04T02:48:54.662842

0804.0053

To find all my favorite papers: ]Please search for author ”Wanng” in the arXiv database. # The Unification of Four Fundamental Interactions Hai-Jun Wang [ Center for Theoretical Physics and School of Physics, Jilin University, Changchun 130023, China ###### Abstract In this paper, we first incorporate the weak interaction into the theory of General Nonlocality [J. Math. Phys. 49, 033513 (2008)] by finding a appropriate metric for it. Accordingly, we suggest the theoretical frame of General Nonlocality as the candidate theory of grand unification. In this unifying scenario, the essential role of photon field is stressed. ## I Introduction It has been decades since scientists commenced their efforts to unify the four fundamental interactions. The efforts focused mainly on how to quantize the gravity. In that context some new theories such as String theory, Quantum Loop Gravity and Non-Commutative Geometry have been developed. In contrast, as we know, no attempt has been practiced in the opposite direction: to unify the quantum theory under the frame of General Relativity. Now such an opposite scheme is feasible. Since we find our previous theory of General Nonlocality [1] can also include the weak interaction as a part. The remainder of the paper is arranged as follows. in section 2, we first present the detailed analysis on where the fermion mass should come from –the problem unresolved in our previous manuscript[1], and subsequently the weak interaction is incorporated into the theory of General Nonlocality in section 3. Finally, we suggest the theoretical frame of General Nonlocality as universal law of fundamental forces. ## II The origin of mass of material particles In the restored Dirac equation (8.12) in Ref. [1], the required mass term in Eq. (8.5) is missing. And we have tried to remedy this by two methods there, but they are not satisfactory. One method is to add a mass term to the right hand side of Eq. (8.9) directly, then Eq. (8.2) would have a nonzero term on its right hand side too. That obviously contradicts the hypothesis that motion equation is just the geodetic line. The other method is to accept the term $i\,\int A_{\nu}\partial_{\lambda}\partial^{\nu}{\psi\,}dx^{\lambda}$ as the mass term. However, on one hand this term is path-dependent (and thus nonlocal) if the integration is not over a closed loop; on the other hand, even if the closed loop is performed, one notes that the coupling of $A_{\nu}$ and ${\psi}$ would give the mass value no more than $m/\sqrt{137}$ ($m$ is the electron mass). So this method is also infessible. The origin of fermion mass is still a problem. The mass problem arises after the projection from geodetic equation (8.2) to its space-time representation (8.12) via the replacement $d\rightarrow\gamma_{\mu}\partial^{\mu}$. To respect the hypothesis that motion equation is the geodetic line we may not make any alteration to the starting point Eq. (8.2). Also, the phrase ”plane wave” appears in the introduction of section VIII–A should not be confused with the conventional term in quantum mechanics. In former case the ”plane wave” means the local plane wave $d{\psi\,}$, after the electron wave is observed by another fermion locally in a _particular complex-frame_ (such frame is assumed to always exist). So putting a plane wave $e^{i\vec{k}\cdot\vec{x}-i\omega t}$ (in the common sense of quantum mechanics) into Eq. (8.2) to check the mass term is inappropriate. Obviously, the replacement $d\rightarrow\gamma_{\mu}\partial^{\mu}$ is invariant under Lorentz transformation. However, if the same Lorentz group is viewed as the structure group obeyed by local plane wave $d{\psi}$, then the parallel displacement as well as the motion equation (8.2) may not exist any longer, since the existence of the _particular frame_ can not be guaranteed under the mere transformations of complex representation of Lorentz group $SL(2,\not C)$–$D^{(1/2,1/2)}$. Therefore, the transformation of Eq. (8.2) has nothing to do with the mass term either. Summarizing the above analyses, the mass problem must lie in the replacement $d\rightarrow\gamma_{\mu}\partial^{\mu}$ and the form of plane wave we substitute in Eq. (8.12). Therefore the mass origin has to do with real space instead of complex space. In complex space, the mass differences become trivial for material particles. So, the origin of the fermion mass may lie in the more delicate replacement than $d\rightarrow\gamma_{\mu}\partial^{\mu}$. If we refine the replacement by $\displaystyle d^{+}$ $\displaystyle\rightarrow$ $\displaystyle\gamma_{\mu}\partial^{\mu}-im$ $\displaystyle d$ $\displaystyle\rightarrow$ $\displaystyle\gamma_{\mu}\partial^{\mu}+im$ $\displaystyle dd$ $\displaystyle\rightarrow$ $\displaystyle d^{+}d$ then we obtain the satisfactory dominating terms $(\Box+m^{2})\psi,$ but in the second term of Eq. (8.9) several other redundant terms would appear. Even if these redundant terms are not the troublesome, we still face the fact of applying consistently the same replacement to the derivation of the boson field equations, which undoubtedly will ruin the previous formalism of the obtained field equation as well as the perfect understanding of the boson mass origin. So, the refinement of the replacement is also limited. Now let’s return to physics. It can be noticed that fermion masses always accompany the appearance of charges, with the exceptions such as the neutrino and neutron, but the neutrino has almost no mass and neutron has a relatively larger magnetic moment. So mostly, if a fermion of spin-$1/2$ has nonzero mass, it must be nontrivially relevant to the electromagnetism. In formalism concerning special relativity, this relevance is best expressed by the relationship between $\not\partial=\gamma_{\mu}\partial^{\mu}$ and $m$. So, we pose the hypothesis that the appearence of the term $\gamma_{\mu}\partial^{\mu}$ is always accompanied by the mass term, both before the wave function $\psi$. The further understanding of this judgement roots in the fact that in micro-world, the fermion mass is detected almost completely via electromagnetic interaction, in contrast to our knowledge that in our everyday life mostly we evaluate masses with the aids of gravitation. Based on this understanding of the operator $\gamma_{\mu}\partial^{\mu}$, we add a mass term $-m^{2}\psi$ at will to the right hand of the equation (8.9) once we begin to project it to space-time, as a part of projection. By ruling out several possibilities of adding mass term, finally we are subject to a physical manner. So far, all the terms in quadratic form of Dirac equation [1] can be restored from our geodetic equation. The understanding of origination of boson mass in the paper [1] is consistent and perfect. ## III The metric for the weak interaction Naturally, the method of nonlocality [1] is expected to describe the weak interaction too. Here we use the known rules–the definition of two sides of the boundary of physical region–to carry out what the metric matrix should be for the weak interaction. In the attempt we use the criteria (11.6) $\mid A_{\bar{\alpha}\beta}\mid=1-A_{0}^{2}+\vec{A}^{2}=\\{\begin{array}[]{cc}0&\text{bound states}\\\ 1&\text{ asymptotically--free states}\end{array}$ (1) and approximation form (11.14a) $(A_{\bar{\alpha}\beta}^{ab})=\Gamma^{0}\otimes I_{2\times 2}+A_{\mu}\Gamma^{1}\gamma^{\mu}\otimes\vec{A}_{a}\tau^{a}\text{ , }\tau^{a}\text{ are the Pauli matrices}$ (2) in Ref. [1] as the starting point (In this paper these two equations denoted as Eq. (1) and Eq. (2) respectively.). The $\Gamma^{0}$ in the above equation is the initial metric matrix we are searching for. To make the metric matrix satisfy the boundary condition Eq. (1), we have to search for the appropriate form of $\Gamma^{0}$ in order to include matrix factor $\gamma^{\mu}\gamma^{5}$ in the $\Gamma^{1}$ of Eq. (2). In what follows we list the possibilities and give the discussions. 1\. First we examine the possibility that in the absence of interaction, if the (initial) metric matrix form is $\gamma_{0}\,\gamma_{5}$, then after a period of interaction the matrix evolves into $\gamma_{0}\gamma_{5}+\gamma_{0}\gamma_{\mu}\gamma_{5}A^{\mu}$, i.e. the interaction vertex $\bar{\psi}\gamma_{5}\psi\rightarrow\bar{\psi}\gamma_{\mu}\gamma_{5}\psi A^{\mu}$ (apart from a coupling constant). The explicit form of the metric matrix is $(\gamma_{0}+\gamma_{0}\gamma_{\mu}A^{\mu})\gamma_{5}=\left(\begin{array}[]{cc}-\vec{\sigma}\cdot\vec{A}&1+A_{0}\\\ -1+A_{0}&-\vec{\sigma}\cdot\vec{A}\end{array}\right)\text{ ,}$ (3) which leads to the determinant $\det(\gamma_{0}\gamma_{5}+\gamma_{0}\gamma_{\mu}\gamma_{5}A^{\mu})=\vec{A}^{2}+1-A_{0}^{2}\text{ ,}$ (4) which is just the form of electrodynamics, meeting the requirements of both strong interaction side (bound state) and weak interaction side (asymptotically-free state). The choice of $\gamma_{0}\,\gamma_{5}$ is a promising candidate for $\Gamma^{0}$. Whereas the actual vertex of weak interaction has the form like $\bar{\psi}\gamma_{\mu}(1\pm\gamma_{5})\psi A^{\mu}$, we should combine this vertex with initial metric $\gamma_{0}\,\gamma_{5}$. 2\. Examine the possibility $\bar{\psi}\gamma_{5}\psi\rightarrow\bar{\psi}\gamma_{\mu}(1\pm\gamma_{5})\psi A^{\mu}$: the metric matrix is $\gamma_{0}\gamma_{5}+\gamma_{0}\gamma_{\mu}A^{\mu}(1\pm\gamma_{5})=\left(\begin{array}[]{cc}X&1\pm X\\\ -1\pm X&X\end{array}\right)\text{ ,}$ (5) where $X=A_{0}\pm\vec{\sigma}\cdot\vec{A}$. The corresponding determinant yields $\det(\begin{array}[]{cc}X&1\pm X\\\ -1\pm X&X\end{array})=1\text{ ,}$ (6) which meets the requirement of asymptotically-free side only. As well known, the weak interaction gives nonsupport of bound states. So the Eq. (5) and the initial metric form $\gamma_{0}\,\gamma_{5}$ are just the perfect ones. The Eq. (6) automatically holds regardless of the details of operator $X$, so we cannot infer any further conclusions from it, even if substituting the concrete form of $\vec{A}$ $\vec{A}=\vec{A}_{a}\tau^{a}\text{ .}$ (7) The other determinants such as $\det(\gamma_{0}+\gamma_{0}\gamma_{\mu}(1\pm\gamma_{5})A^{\mu})$ and $\det(\gamma_{0}(1\pm\gamma_{5})+\gamma_{0}\gamma_{\mu}(1\pm\gamma_{5})A^{\mu})$ are also examined, but they present no required boundary properties. Additionally, extension of the group from $SU(2)$ to $U(2)$ is still necessary while considering the curving effect of weak interaction. ## IV Universal law for fundamental forces So far, we have included all of the three microscopic interactions in the theoretical frame of General Nonlocality. Additionally, in view of the well- known puzzles of conservation law of energy-momentum tensor in General Relativity–which is understandable if the energy-momentum tensor is also nonlocal, we can put forward the hypothesis that _the dynamics for all of the fundamental forces should be described with the common motion equation_ (geodetic line) _and field equation_ ($R=0$) _in geometrical manner_ , which automatically induces the Nonlocality. The history of the development of description of dynamics, after Newton equation, have underwent several stages, as follows $\text{Hamitonian}\longrightarrow\text{Lagarangian}\longrightarrow\text{Action}\longrightarrow\text{Geometry},$ each of the succeeding one is more general than its preceding one. The Lagrangian is able to include the covariant form of special relativity, and the action form can incorporate the gauge fixing condition naturally and make the quantization process more fluent. As for geometry, we hope it can circumvent the renormalization processes, as well as provide other nonperturbative methods. The curvings (Geometries) corresponding to the four fundamental interactions are all relevant to effects of photon. It is well known that the curvature of space-time (in General Relativity) has been confirmed by the curving effect of light. But for the microscopic interactions, one may not be aware of how does the photon field curve in complex space–unitary space. We assert here that the unifying picture of electrodynamics, weak interaction (the weak doublets), and quark dynamics (color triplet) may be achieved by regarding the curving of photon’s field, just as implied in Ref. [1]: The photon exists as a pure energy form, which can decay into lepton- antilepton pair$\rightarrow$proton-antiproton pair$\rightarrow$quark-antiquark pair. The processes successively take place with the increase of the photon energy. The remarkable feature of the successive processes is that the lepton- antilepton is in the $U(1,\not C)$ region of the gauge fields, and proton- antiproton is in $U(2,\not C)$ hadron region [forms the weak doublet together with neutron], and quark antiquark in $U(3,\not C)$ color region. This feature leads to the ansätz that the photon field can be curved into the $U(n,\not C)$-space with its energy increasing. So, we can add another curving to the curvings of solely in $GL(n,\not C)$-space and/or solely in $U(n,\not C)$-space. Now different sections of complex spaces can be linked by the photon-field curving: $U(1,\not C)\rightarrow U(2,\not C)\rightarrow U(3,\not C)$. In formalism, this kind of curving can be interpreted by generalizing the meaning of Eq. (7.3) in Ref. [1], with the indices $a,c$ being not confined to one unitary group, but able to change continuously among $U(1,\not C)$, $U(2,\not C)$ and $U(3,\not C)$. Possibly, the dynamics of General Nonlocality also governs other realms of nature, such as Thermodynamics or Hydrodynamics, if only we find the appropriate space, as well as the metric forms and the structure groups for the space. Then using geodetic line and the field equation $R=0$ can mean all. ## References * (1) H. J. Wang, General Nonlocality in Quantum Fields, J. Math. Phys. 49, 033513 (2008), also at arXiv: quant-ph/0512191.

arxiv-papers

2024-09-04T02:48:54.668466

0804.0070

# An ideal mass assignment scheme for measuring the Power Spectrum with FFTs Weiguang Cui11affiliation: Shanghai Astronomical Observatory, the Partner Group of MPA, Nandan Road 80, Shanghai 200030, China; E-mail: wgcui@shao.ac.cn 44affiliation: Joint Institute for Galaxy and Cosmology (JOINGC) of Shanghai Astronomical Observatory and University of Science and Technology of China , Lei Liu11affiliation: Shanghai Astronomical Observatory, the Partner Group of MPA, Nandan Road 80, Shanghai 200030, China; E-mail: wgcui@shao.ac.cn 44affiliation: Joint Institute for Galaxy and Cosmology (JOINGC) of Shanghai Astronomical Observatory and University of Science and Technology of China , Xiaohu Yang11affiliation: Shanghai Astronomical Observatory, the Partner Group of MPA, Nandan Road 80, Shanghai 200030, China; E-mail: wgcui@shao.ac.cn 44affiliation: Joint Institute for Galaxy and Cosmology (JOINGC) of Shanghai Astronomical Observatory and University of Science and Technology of China , Yu Wang11affiliation: Shanghai Astronomical Observatory, the Partner Group of MPA, Nandan Road 80, Shanghai 200030, China; E-mail: wgcui@shao.ac.cn 44affiliation: Joint Institute for Galaxy and Cosmology (JOINGC) of Shanghai Astronomical Observatory and University of Science and Technology of China , Longlong Feng22affiliation: Purple Mountain Observatory, Nanjing 210008, China , Volker Springel33affiliation: Max-Planck-Institut für Astrophysik, Karl- Schwarzschild-Strasse 1, 85748 Garching, Germany ###### Abstract In measuring the power spectrum of the distribution of large numbers of dark matter particles in simulations, or galaxies in observations, one has to use Fast Fourier Transforms (FFT) for calculational efficiency. However, because of the required mass assignment onto grid points in this method, the measured power spectrum $\langle|\delta^{f}(k)|^{2}\rangle$ obtained with an FFT is not the true power spectrum $P(k)$ but instead one that is convolved with a window function $|W(\mathbf{k})|^{2}$ in Fourier space. In a recent paper, Jing (2005) proposed an elegant algorithm to deconvolve the sampling effects of the window function and to extract the true power spectrum, and tests using N-body simulations show that this algorithm works very well for the three most commonly used mass assignment functions, i.e., the Nearest Grid Point (NGP), the Cloud In Cell (CIC) and the Triangular Shaped Cloud (TSC) methods. In this paper, rather than trying to deconvolve the sampling effects of the window function, we propose to select a particular function in performing the mass assignment that can minimize these effects. An ideal window function should fulfill the following criteria: (i) compact top-hat like support in Fourier space to minimize the sampling effects; (ii) compact support in real space to allow a fast and computationally feasible mass assignment onto grids. We find that the scale functions of Daubechies wavelet transformations are good candidates for such a purpose. Our tests using data from the Millennium Simulation show that the true power spectrum of dark matter can be accurately measured at a level better than 2% up to $k=0.7k_{N}$, without applying any deconvolution processes. The new scheme is especially valuable for measurements of higher order statistics, e.g. the bi-spectrum, where it can render the mass assignment effects negligible up to comparatively high $k$. (cosmology:) large-scale structure of universe - cosmology: theory \- methods: numerical - methods: data analysis ## 1 INTRODUCTION In studies of the cosmic large-scale structure, a number of different statistical methods are routinely used to extract various information of interest (e.g., regarding the cosmology, the initial perturbation, etc) that is embedded in the distribution of the dark matter particles (in the case of simulations) or the galaxies (in observations). The power spectrum $P(k)$ is one of the most powerful and basic statistical measures that describes the distribution of mass and light in the Universe, and one of the most throughly investigated quantities in modelling the structure formation process. The initial primordial power spectrum of the mass fluctuations is usually assumed to follow a power law, $P_{0}(k)=Ak^{n}$. The linearly processed power spectrum $P_{\rm lin}(k)$ can be well predicted by codes such as CMBFAST (Seljak & Zaldarriaga 1996), or approximated by various fitting formula (e.g. Bardeen et al. 1986; Efstathiou, Bond & White 1992; Eisenstein & Hu 1998) for different matter and energy content. Using N-body simulations, the non-linear power spectrum $P_{\rm NL}(k)$ has been modelled by various authors (e.g. Peacock & Dodds 1996; Ma & Fry 2000; Smith et al. 2003). Apart from these theoretical models, direct measurement of the power spectrum from observations plays an extremely important role both in cosmology and galaxy formation theories. Although there are different biases relative to the mass power spectrum, one can roughly say that $P(k)$ on very large scales measures the primordial density fluctuations, which is closely connected with the cosmology models (e.g., Spergel et al. 2007), while $P(k)$ on small scales characterizes the later non-linear evolution (e.g., Peacock & Dodds 1996). As an essential statistical measure for the distribution of galaxies, the power spectrum $P(k)$ has been estimated and modeled from most of the redshift surveys. Recent investigations along this direction include the CfA and Perseus-Pisces redshift surveys (Baumgart & Fry 1991), the radio galaxy survey (Peacock & Nicholson 1991), the IRAS QDOT survey (Kaiser 1991), the 2Jy IRAS survey (Jing & Valdarnini 1993), the 1.2Jy IRAS survey (Fisher et al. 1993), the Las Campanas Redshift Survey (Lin et al. 1996; Yang et al. 2001), the 2dF Galaxy Redshift Survey (Percival et al., 2001; Tegmark, Hamilton & Xu 2002; Sánchez et al. 2006) and the Sloan Digital Sky Survey (Tegmark et al. 2004; Percival et al., 2007). Among these works, the galaxy power spectra are measured either using the Fast Fourier Transform (FFT) technique or direct summation, or other advanced techniques (e.g., Yang et al. 2001; Tegmark et al. 2004). Apart from these observational probes, the power spectrum is also widely measured from N-body simulations (e.g. Davis et al. 1985). For these measurements, one has to use FFTs since there are too many particles in the simulations to apply direct summation. Before performing the FFT, one therefore needs to assign the particle distribution $\rho(\mathbf{r})$ onto grids $\rho(\mathbf{r}_{g})$ (usually onto $2^{3i}$ grid cells, where $i$ is an integer). As pointed out in a recent paper by Jing (2005), such an assignment process is equivalent to a convolution of the real density field with a given assignment window function $W(\mathbf{r})$, and sampling the convolved density field at the $2^{3i}$ grid points. Thus the power spectrum based on the FFT of $\rho(\mathbf{r}_{g})$ is not equal to that based on the Fourier transform (FT) of $\rho(\mathbf{r})$. In order to obtain the true power spectrum to an accuracy of a few percent, the sampling effects should be carefully corrected (Jing 2005; and references therein). To this end, Jing (2005) proposed an elegant algorithm to iteratively deconvolve the power spectrum measurement for the impact of the mass assignment and to extract the true power spectrum. Tests using N-body simulations show that their algorithm works extremely well for the three commonly used mass assignment functions, i.e., the Nearest Grid Point (NGP), the Cloud In Cell (CIC) and the Triangular Shaped Cloud (TSC) methods. In this paper, rather than trying to correct for the influence of the window function, we seek to minimize the effects of the mass assignment by selecting special window functions. An ideal window function should fulfill the following criteria: (i) compact top-hat like support in Fourier space to avoid the sampling effects; (ii) compact support in real space to allow computationally efficient mass assignment onto grids. We find that the scale functions of the Daubechies wavelet transformations are good candidates for simultaneously matching both requirements. In fact, as we will demonstrate they allow an accurate measurement of the power spectrum with FFTs without the need for a deconvolution procedure. This is of great help especially for accurate measurements of higher order spectra, like the bi-spectrum, where FFTs are needed but the de-aliasing methods are not available yet. We will discuss this application to accurate measurements and modelling of the bi- spectrum in a subsequent paper. This paper is organized as follows. In Section 2 we give a brief description of the methodology for measuring the power spectrum from the discrete distribution of dark matter particles. In Section 3 we first present the commonly used window functions in both real and Fourier spaces, and then and introduce our new mass assignment scheme. In Section 4 we compare the power spectra extracted from the Millennium Run using different methods. Finally, we summarize our results in Section 5. ## 2 Measuring the power spectrum In this section we outline the methods used to measure the power spectrum from the distribution of dark matter particles (Peebles 1980). Unless stated otherwise we shall follow Jing (2005), and we refer readers who are interested in a more detailed and complete set of formulae to this paper and references therein. We start from the definition of the power spectrum. Let $\rho({\mathbf{r}})$ be the cosmic density field and $\overline{\rho}$ the mean density. Then the density contrast $\delta({\mathbf{r}})$ can be expressed as, $\delta({\mathbf{r}})={\rho({\mathbf{r}})-\overline{\rho}\over\overline{\rho}}.$ (1) Based on the cosmological principle, we assume that $\rho(\mathbf{r})$ in a very large volume $V_{\mu}$ fairly represents the overall cosmic density field, and that it can be taken to be periodic. The FT of $\delta(\mathbf{r})$ can be defined as: $\delta({\mathbf{k}})={1\over V_{\mu}}\int_{V_{\mu}}\delta({\mathbf{r}})e^{i{\mathbf{r}}\cdot{\mathbf{k}}}{\rm d}{\mathbf{r}}\,.$ (2) And by definition, its power spectrum $P(k)$ is simply related to ${\delta({\mathbf{k}})}$ as $P(k)\equiv\langle\mid{\delta({\mathbf{k}})}\mid^{2}\rangle\,,$ (3) where $\langle\cdot\cdot\cdot\rangle$ means the ensemble average. However, in practice, the cosmic density field is usually traced by the distribution of galaxies or dark matter particles. In these cases, the density field $\rho({\mathbf{r}})$ is replaced by the number density distribution of objects $n(\mathbf{r})=\sum_{j}\delta^{D}(\mathbf{r}-\mathbf{r}_{j})$, where $\mathbf{r}_{j}$ is the coordinate of object $j$ and $\delta^{D}(\mathbf{r})$ is the Dirac-$\delta$ function. And the FT of the related number density contrast $\delta(\mathbf{r})$ can be expressed as, ${\delta^{d}({\mathbf{k}})}={1\over V_{\mu}\overline{n}}\int_{V_{\mu}}n(\mathbf{r})e^{i{\mathbf{r}}\cdot{\mathbf{k}}}{\rm d}{\mathbf{r}}-\delta^{K}_{{\mathbf{k}},{\mathbf{0}}}\,,$ (4) where $\overline{n}$ is the mean number density, the superscript $d$ represents the discrete case of $\rho({\mathbf{r}})$, and $\delta^{K}$ is the Kronecker delta. If we divide the volume $V_{\mu}$ into infinitesimal elements $\\{{\rm d}V_{i}\\}$ within which there are either 0 or 1 objects, then the above equation can be written as: ${\delta^{d}({\mathbf{k}})}={1\over N}\sum_{i}n_{i}e^{i{\mathbf{r}}_{i}\cdot{\mathbf{k}}}-\delta^{K}_{{\mathbf{k}},{\mathbf{0}}}\,,$ (5) where $N$ is the total number of objects in $V_{\mu}$ and $n_{i}$ is either 0 or 1. After a bit of algebra, it is seen that the true power spectrum can be measured via $P(k)\equiv\langle\mid{\delta({\mathbf{k}})}\mid^{2}\rangle\ =\langle\mid{\delta^{d}({\mathbf{k}})}\mid^{2}\rangle-{1\over N}\,.$ (6) Obviously, when the FT is directly applied to the distribution of the galaxies or dark matter particles, one needs to correct for the discreteness (or shot noise) effect, which introduces an additional term $1/N$ to the power spectrum $\langle\mid{\delta^{d}({\mathbf{k}})}\mid^{2}\rangle$. The above method of using a direct summation in the FT can be used to measure the power spectrum from the distribution of galaxies, when the number of objects is not very large. However, because of the huge number of particles involved in N-body simulations, it is almost impossible to be applied to the dark matter particles of cosmological density fields. Instead, a computationally attractive approach is to use an FFT. The density contrast in Fourier space using a FFT is, $\delta^{f}({\mathbf{k}})={1\over N}\sum_{\mathbf{g}}n^{f}({\mathbf{r}}_{g})e^{i{\mathbf{r}}_{g}\cdot{\mathbf{k}}}-\delta^{K}_{{\mathbf{k}},{\mathbf{0}}}\,,$ (7) where the superscript $f$ represents the FFT. $n^{f}({\mathbf{r}}_{g})$ is the convolved density value on the $\mathbf{g}$-th grid point $\mathbf{r}_{g}=\mathbf{g}H$ (where $\mathbf{g}$ is an integer vector; $H$ is the grid spacing), $n^{f}({\mathbf{r}}_{g})=\int n({\mathbf{r}})W(\mathbf{r}-{\mathbf{r}}_{g})\,{\rm d}{\mathbf{r}}\,,$ (8) where $W(\mathbf{r})$ is the mass assignment function. Note that Eqs. (5) and (7) are different in that the summations carried out in the former equation is over the objects and the latter over space (the grid points). After several steps (see also Hockney and Eastwood 1981), Jing (2005) derived the following power spectrum estimator, $\displaystyle\langle|\delta^{f}({\mathbf{k}})|^{2}\rangle$ $\displaystyle=$ $\displaystyle\sum_{\mathbf{n}}|W(\mathbf{k}+2k_{N}\mathbf{n})|^{2}P(\mathbf{k}+2k_{N}\mathbf{n})$ (9) $\displaystyle+$ $\displaystyle{1\over N}\sum_{\mathbf{n}}|W(\mathbf{k}+2k_{N}\mathbf{n})|^{2}\,,$ where $W(\mathbf{k})$ is the FT of the window function $W(\mathbf{r})$, $k_{N}=\pi/H$ is the Nyquist wavenumber, and the summation is over all 3D integer vectors $\mathbf{n}$. According to equation (9), one can easily identify the impact of the mass assignment onto the measured power spectrum. First, the mass assignment introduces the factor $W^{2}(\mathbf{k})$ to both the true power spectrum and the shot noise ($1/N$) terms. Second, the quantity $\langle|\delta^{f}({\mathbf{k}})|^{2}\rangle$ is a measure for a convolved power spectrum (i.e., the sums over $\mathbf{n}$) which suffers from sampling effects. As pointed out in Jing (2005) and will be shown in Section 3, the sampling effects are significant near the Nyquist wavenumber $k_{N}$ and should be carefully corrected in an accurate measure of the power spectrum. ## 3 The role of the mass assignment function As shown by equation (9), the mass assignment window function plays an important role in measuring the power spectrum using an FFT. We separate its impact into two parts: one on the shot noise term (second term of the r.h.s. of Eq. 9) and the other on the true power spectrum term (first term of the r.h.s. of Eq. 9). Hereafter we refer to the impact on the true power spectrum term as the sampling effects. Usually, the impact on the shot noise term can be handled analytical according to the FT of the window function. However, because of the convolution with the true power spectrum, the sampling effects can not be corrected easily. There are basically two strategies for handling the sampling effects. One can either try to correct for them by deconvolving the impact of the window function (which is carried out in Jing 2005) or try to use an optimal window function that minimizes the sampling effects from the outset (the purpose of this work). Below we discuss a few commonly used window functions as well as the particular mass assignment proposed here both in real and Fourier spaces, and then discuss their impact on measuring the true power spectrum with an FFT in detail. Figure 1: Left panel: the three commonly used mass assignment window functions, NGP, CIC, and TSC, as indicated. Right panel: the square of the window functions in Fourier space. ### 3.1 Traditional mass assignment functions The most popular mass assignment functions used in measuring the power spectrum are the NGP, CIC and TSC methods. Their forms in real space can be described by $W(\mathbf{x})=\Pi_{i}W(x_{i})$, with $W(x_{i})=\cases{1&{$|x_{i}|<0.5$}\cr 0&{else}\cr}~{}~{}~{}~{}~{}~{}~{}{\rm NGP,}$ (10) $W(x_{i})=\cases{1-|x_{i}|&{$|x_{i}|<1$}\cr 0&{else}\cr}~{}~{}~{}~{}~{}{\rm CIC,}$ (11) and $W(x_{i})=\cases{0.75-x_{i}^{2}&{$|x_{i}|<0.5$}\cr(1.5-|x_{i}|)^{2}\over 2&{$0.5<|x_{i}|<1.5$}\cr 0&{else}\cr}~{}~{}~{}~{}~{}{\rm TSC,}$ (12) where $x_{i}$ ($i=1,2,3$) is the $i$-th component of $\mathbf{x}$. In the left panel of Fig. 1, we show these window functions in real space, with solid, dotted and dashed lines corresponding to the NGP, CIC and TSC methods, respectively. Their impact on the measurement of the power spectrum using an FFT (Eq. 9) can be understood most easily based on their Fourier space behavior. According to Hockney & Eastwood (1981), these three mass assignment window functions can be described in Fourier space by $W(\mathbf{k})=\Pi_{i}W(k_{i})$, with $W(k_{i})=\Bigl{[}{\sin({\pi k_{i}\over 2k_{N}})\over({\pi k_{i}\over 2k_{N}})}\Bigr{]}^{p}\,,$ (13) where $k_{i}$ ($i=1,2,3$) is the $i$-th component of $\mathbf{k}$, and $p=1$ for NGP, $p=2$ for CIC, and $p=3$ for TSC. We show in the right panel of Fig. 1 (the square of) the related window functions in Fourier space. These window functions peak at $k=0$ with $W^{2}(k)=1$ and decrease sharply with $k\gtrsim 0$, especially for the CIC and TSC kernels. According to Eq. (9), the impact of the window functions can be separated into two parts, one on the shot noise and one on the true power spectrum. It is quite easy to model and correct the impact on the shot noise term, $D^{2}(\mathbf{k})\equiv{1\over N}\sum_{\mathbf{n}}W^{2}(\mathbf{k}+2k_{N}\mathbf{n})\,.$ (14) For the NGP, CIC, and TSC assignments, the shot noise term can be expressed as, $D^{2}(\mathbf{k})={1\over N}\cases{1,&NGP,\cr\Pi_{i}[1-{2\over 3}\sin^{2}({\pi k_{i}\over 2k_{N}})],&CIC,\cr\Pi_{i}[1-\sin^{2}({\pi k_{i}\over 2k_{N}})+{2\over 15}\sin^{4}({\pi k_{i}\over 2k_{N}})].&TSC.\cr}$ (15) In practice, for the latter two cases, one often uses the following isotropic approximation to model the shot noise term, $D^{2}(\mathbf{k})\approx{1\over N}\cases{[1-{2\over 3}\sin^{2}({\pi k\over 2k_{N}})],&CIC,\cr[1-\sin^{2}({\pi k\over 2k_{N}})+{2\over 15}\sin^{4}({\pi k\over 2k_{N}})],&TSC,\cr}$ (16) where $k=|\mathbf{k}|$. As has been shown in Jing (2005), this approximation works very well for $k\leq 0.7k_{N}$, however, it can underestimate the true value by about 40% at $k\sim k_{N}$. Nevertheless, compared to the power spectrum term we are trying to measure in a CDM cosmology, this error in the shot noise term is usually negligible. Now we turn to the impact of the window functions on the first term of the r.h.s of Eq. (9), the sampling effects. There are three aspects that need to be taken into account in measuring the true power spectrum if an accuracy of a few percent is required. * • Smearing effect In the summation of the true power spectrum over $\mathbf{n}$, only the term $\mathbf{n}=0$ is what we intend to measure. However, according to the results shown in the right panel of Fig. 1, the $W^{2}(k)$ term decreases sharply from $W^{2}(0)=1$ at $k\gtrsim 0$, especially for the CIC and TSC methods. Thus the contribution from the related true power spectrum $P(\mathbf{k})$ is greatly suppressed. This effect is the so-called smearing or smoothing effect, which has been discussed in the literature (e.g., Baumgart & Fry 1991; Jing & Valdarnini 1993; Scoccimarro et al. 1998) * • Anisotropy effect In pratice, one may use the average of the $\langle|\delta^{f}({\mathbf{k}})|^{2}\rangle$ over different directions for a given $k$ to estimate the power spectrum $P(k)$. However, the window function $W^{2}(\mathbf{k})$ is not isotropic for different directions for a given $k$, that is, $W^{2}(\mathbf{k})$ is different, e.g., for $\mathbf{k}=k(1/\sqrt{3},1/\sqrt{3},1/\sqrt{3})$, $\mathbf{k}=k(1/\sqrt{2},1/\sqrt{2},0)$, $\mathbf{k}=k(1,0,0)$, etc. This effect is small for the NGP method, but quite significant for the CIC and TSC methods, especially at $k\sim k_{N}$ (eg. Baumgart & Fry 1991, Jing 2005). * • Aliasing effect The power spectrum estimator $\langle|\delta^{f}({\mathbf{k}})|^{2}\rangle$ contains not only the contribution from $P(\mathbf{k})$ where $\mathbf{n}=0$ but also from $P(2k_{N}\mathbf{n}+\mathbf{k})$ where $\mathbf{n}\neq 0$. The latter contribution, which is called the alias effect, prevents us from obtaining the true power spectrum $P(\mathbf{k})$ straightforwardly. This effect, which is prominent near the Nyquist wavenumber $k_{N}=0.5\times(2\pi/H)$ (significant for NGP method and less significant for TSC method), has been discussed and handled using an iterative correction method in Jing (2005). The smearing and anisotropy effects are easy to be corrected. For instance, one can directly normalize the density contrast in Fourier space, $\delta^{f}({\mathbf{k}})$, with the window function $W({\mathbf{k}})$ (e.g., Baumgart & Fry 1991). Thus, the corrected density contrast $\delta^{f}({\mathbf{k}})/W({\mathbf{k}})$ obviously no longer suffer from these two effects at $k\leq k_{N}$, however at the price of a much enhanced aliasing effect (i.e., the ${\mathbf{n}}\neq 0$ terms in Eq. 9). Because of the aliasing effect the power spectrum measured at $k=k_{N}$ can become a factor of 2 larger than the true value. Such kind of aliasing effects also exist in radio imaging analyses based on FFTs, and various perticular mass assignment schemes have been discussed in order to minimize their impact (e.g. Briggs et al. 1999). Using an elegant iterative correction methods, Jing (2005) has properly corrected the impact of the aliasing effect, and illustrated its success in obtaining the true power spectrum. On the other hand, his method can only be applied to the estimation of the power spectrum. For measurements of higher order spectra, e.g. the bi-spectrum, there is so far no straightforward method that can correct the aliasing effect. In what follows, rather than trying to correct the above three kinds of effects, we attemp to find a mass assignment window function that does not or only to very small degree suffer from these effects. Figure 2: Left panel: the scaling functions 12 (D12) and 20 (D20) of Daubechies, used here as mass assignment window functions. Right panel: the square of these two scaling functions in Fourier space. For comparison, we also show a top hat window function in Fourier space, which corresponds to $W(x)=\sin(\pi x)/(\pi x)$ in real space. ### 3.2 Daubechies window functions An ideal window function that does not suffer from the sampling effects mentioned above is obviously a top-hat in Fourier space. Using such a window function, the power spectrum measurement Eq. (9) can be reduced to Eq. (6). However such a mass assignment function, $W(x)=\sin(\pi x)/(\pi x)$, is not a compact localized function in real space. In the mass assignment onto the grid, one may then have to distribute each particle’s mass to too many grid cells. In fact, if we want to maintain an accuracy of 1% in the mass assignment, the mass of each particle should be distributed to $60^{3}$ grid cells! Such an assignment scheme may eliminate most if not all of the computational advantage that an FFT can bring us. Thus, a suitable mass assignment window function should be localized both in real and Fourier space. A good candidate that features these properties is the scale function of the wavelet transformation. The wavelet transformation has been previously introduced to astrophysical studies and has been applied successfully in the analysis of various astrophysical observations (c.f., Fang & Thews 1998), e.g., on the distributions of galaxies (e.g. Martinez et al. 1993; Fang & Feng 2000; Yang et al., 2001; 2002a,b; Feng & Fang 2004), on the properties of Ly$\alpha$ absorption lines (e.g. Pando & Fang 1997; Meiksin 2000), on the galaxy clusters (e.g., Slezak et al. 1994; Grebenev et al. 1995; Gambera et al. 1997; Schäfer et al. 2005), etc. Here, we introduce the scale function $\phi(x)$ of the Daubechies wavelet transformation for use in power spectrum measurements, which has the following properties (e.g., Daubechies 1992), $\int\phi(x)~{}{\rm d}x\equiv 1\,,$ (17) $\sum_{n}\phi(x+n)\equiv 1\,,$ (18) and its Fourier transform, $\phi(k)$, satisfies $\int\phi^{2}(k)~{}{\rm d}k\equiv 1\,,$ (19) $\sum_{n}\phi^{2}(k+2\pi n)\equiv 1\,.$ (20) In this paper, we use the Daubechies D12 and D20 scale functions (Daubechies 1988, 1992) as our new mass assignment window functions, $W(x)=\phi(x)$, which are shown in the left panel of Fig. 2. In the right panel of Fig. 2, the squares of these two window functions in Fourier space are shown as dotted and dashed lines, as indicated. For comparison, we also show in the right panel the ideal case of a top-hat Fourier window function as the solid line. The D12 and D20 window functions in Fourier space $W^{2}(k)$ resemble the ideal case very well, especially in the D20 case whose deviation from the ideal case at $k=0.35$ (i.e., $0.7k_{N}$) is smaller than 2%. Note that these particular mass assignment window functions are different from the traditional schemes, e.g. NGP, CIC and TSC in that: (1) they are not symmetric; (2) they are not positive definite. However these two features will not induce any undesirable consequences in our application. First, since the overall shifting of the window function will not impact the amplitude of $\delta(k)$, the window function $\phi(x)$ shown in the left panel of Fig. 2 can be treated as symmetric components centered at $x\sim 1.75$ and $x\sim 2.5$, respectively, with additional fluctuations at nearby grid cells. Second, the window function needs not necessarily be positive definite, as we are measuring the density contrast $\delta(x)$, and even the ideal window function $W(x)=\sin(\pi x)/(\pi x)$ is not positive definite. Before we turn to a discussion of their impact on measuring the true power spectrum, let us consider the computational cost for the mass assignment using the D12 and D20 scale functions. According to their real space behavior, at much better than 0.5% accuracy, each mass particle should be distributed onto $6^{3}$ (D12) or $8^{3}$ (D20) grid cells, respectively, which is a factor of 10 or 20 times more than the TSC method with $3^{3}$ grid cells. However, we argue that this cost is worthwhile given the following positive features of the Daubechies assignment. First, according to Eq. (20), the shot noise term in Eq. (9) for these mass assignment algorithms is, $D^{2}(k)\equiv 1/N\,.$ (21) Second, by comparing the Fourier-space behaviors of the D12 and D20 functions with those of the traditional mass assignment methods, NGP, CIC and TSC, it becomes clear that the three sampling effects of smearing, alias and anisotropy are greatly suppressed. Moreover, for the D20 window function, if we only measure the power spectrum up to $k=0.7k_{N}$, we do not need to take into account any of those three kinds of effects explicitly, because the true power spectrum is recovered with better than 2% accuracy! Another very important aspect is that such a mass assignment scheme can be fruitfully applied to the measurement of the higher-order spectra using an FFT. For instance, in measuring the bi-spectrum using an FFT with the D20 mass assignment, we do not need to consider the sampling effects up to $k=0.7k_{N}$ at all, since here the bi-spectrum can be measured directly with an accuracy level better than 3%. Note that so far there is no other approach known to accurately correct for the sampling effects in measuring the bi-spectrum with an FFT. We defer an application of our new technique and a theoretical modeling of the higher order spectra to a forthcoming paper. Figure 3: Upper-Left panel: the FFT power spectra measured using the commonly employed mass assignment functions: NGP, CIC and TSC, as indicated. In this panel, only the shot noise term has been corrected. Lower-left panel: the same measurements as shown in the upper-left panel, but now corrected for the sampling effects using the iterative correcting method proposed by Jing (2005). Upper-right panel: the FFT power spectra measured using the Daubechies scale functions D12 and D20 as mass assignment functions. Only the short noise term $1/N$ has been corrected. In these three panels, for reference we also plot the power spectrum prediction by Smith et al. (2003) using the Millennium Run’s cosmological parameters. Lower-right panel: a comparison of the ratios between the measured power spectra and the ‘halofit’ prediction by Smith et al. (2003), for the NGP, CIC, TSC, D12 and D20 mass assignment window functions. The vertical lines mark the locations of $k=k_{N}$ and $k=0.7k_{N}$, respectively. ## 4 Tests using N-body simulations Having discussed the impact of the mass assignment window functions on the measurement of the true power spectrum using an FFT, and having introduced the D12 and D20 scale functions, we proceed to demonstrate their performance when applied to the measurement of the mass power spectrum of a large N-body simulation. Here we briefly describe the simulation, the Millennium Run, used for this project. The Millennium Run is a very large dark matter simulation of the concordance $\Lambda$CDM cosmology with $2160^{3}\simeq 1.0078\times 10^{10}$ particles in a periodic box of $500\,h^{-1}$Mpc on a side (Springel et al. 2005). The simulation was carried out by the Virgo Consortium using a customized version of the GADGET2 code. The cosmological parameters used in this simulation are $\Omega_{\rm m}=\Omega_{\rm dm}+\Omega_{\rm b}=0.25$, $\Omega_{\rm b}=0.045$, $h=0.73$, $\Omega_{\Lambda}=0.75$, $n=1$, and $\sigma_{8}=0.9$. For our test investigation, we randomly select 10% of the dark matter particles (because of practical limits in computer memory) and measure their power spectra using the different window functions we described in the previous section. To measure the power spectrum, we employ an FFT of the density distribution of dark matter particles assigned to a grid with $1024^{3}$ cells using the mass assignment algorithms discussed in Section 3. Thus the corresponding Nyquist wavenumber is $k_{N}=1024\pi/500\,h\,{\rm Mpc}^{-1}$. In the upper-left panel of Fig 3, we show the FFT power spectra measured using the traditional mass assignment functions, NGP, CIC and TSC, where only the shot noise term has been subtracted. In this figure, the power spectrum is presented in terms of $\Delta^{2}(k)\equiv 2\pi k^{3}P(k)$. For comparison, we show the theoretical prediction of the non-linear power spectrum by Smith et al. (2003) as the solid line, based on the Millennium Run’s cosmological parameters. Obviously, because of the sampling effects we discussed in Section 3.1, the power spectra are quite different at $k\gtrsim 1\,h{\rm Mpc}^{-1}$ ($\sim 0.2k_{N}$). The power spectra measured without correcting the sampling effects, especially for the TSC method, significantly under-predict the true power spectrum. Using the methods proposed by Jing (2005), we can iteratively correct for the sampling effects and extract estimates of the true power spectrum. The corrected power spectra for the NGP, CIC and TSC mass assignment methods are shown in the lower-left panel of Fig. 3. Comparing these power spectra among themselves and with the ‘halofit’ prediction of Smith et al. (2003), we are convinced that the true power spectrum is well recovered at all scales $k\leq k_{N}$, and is roughly consistent with the prediction by Smith et al. Now we turn to use the Daubechies scale functions D12 and D20 as our mass assignment window functions. The resulting power spectra after correcting for the shot noise term $1/N$ are shown in the upper-right panel of Fig. 3, as indicated. Without any correction for the sampling effects, the measured power spectra look very nice and match the theoretical predictions by Smith et al. (2003) on all scales up to $k\leq k_{N}$. This is very different from the results shown in the upper-left panel of Fig. 3 for the classical assignment functions. In fact, at a low resolution view, there is no visible difference between these results and the corrected measurements shown in the lower-left panel of Fig. 3. Finally, we take more accurate comparisons between the power spectra measured with these different methods by showing their ratios with respect to the ‘halofit’ prediction of Smith et al. (2003). The de-convolved power spectra based on the NGP, CIC and TSC mass assignments, the directly measured power spectra using D12 and D20 mass assignments are ploted together for comparison in the bottom right panel of Fig. 3. Here are a few observations that can be made: (1) the three de-convolved power spectra are very well consistent with each other at a level better than 2% at $k\leq 0.7k_{N}$, and at a level of about 5% at $k\sim 1.0k_{N}$; (2) the directly measured power spectra based on the D12 and D20 (the latter slightly better) functions have an accuracy of better than 2% at $k\leq 0.7k_{N}$ and at a level of about 10% at $k\sim 1.0k_{N}$; (3) there is about 20% under-prediction on large scales (with $k<1~{}h{\rm Mpc^{-1}}$) and 5% over-prediction on small scales by Smith et al. (2003) for the power spectrum of the Millennium Simulation. According to these findings, we may conclude that both the deconvolution method and the direct measurement based on the Daubechies scale functions, especially for D20, can recover the true power spectrum with better than 2% accuracy at $k\leq 0.7k_{N}$. Moreover, as a conservative prediction, the bi-spectrum can be measured at a level better than 3% at $k\leq 0.7k_{N}$ if the D20 window function is used in the mass assignment for the FFT. This should be very useful for accurate studies of the bi-spectrum. ## 5 SUMMARY To quantify the large-scale structure in the distributions of a large population of dark matter particles or galaxies, one may measure their power (or higher order) spectra using a FFT. However, the required mass assignment onto the points of the FFT-grid can introduce sampling effects in the measured power spectra. Most of these effects have been noticed and discussed in the literature before (e.g., Baumgart & Fry 1991; Jing & Valdarnini 1993; Jing 2005). Among these, Jing (2005) was the first to use an iterative correction method to compensate for all of these sampling effects, especially the alias effect. In this paper, we follow Jing (2005) and discuss the impact of the mass assignment on measuring the power spectrum with an FFT. There are two components that the employed window function can impact: one is the shot noise term and the other is the term involving the true power spectrum. With respect to the influence on the true power spectrum term, there are three different sampling effects that need to be considered: the smearing effect, the aliasing effect and the anisotropy effect. Rather than trying to deconvolve for the sampling effects, we propose to use a special window function: the Daubechies wavelet scale function that can minimize these sampling effects. In particular, the D12 and D20 scale functions considered here are compact in real space, which allows a fast mass assignment onto the grid cells, while at the same time their top-hat like shape in Fourier space leads only to very small sampling effects. According to the Fourier transform $W^{2}(k)$ of the D20 function, at $k\leq 0.7k_{N}$, all the sampling effects induced by the mass assignment can only affect the measured power spectrum at less than a level of 2%. This is confirmed by the tests we carried out with the Millennium Run simulation. More importantly, as a conservative prediction, the new method proposed here can measure the bi-spectrum of dark matter particles at better than 3% accuracy for $k\leq 0.7k_{N}$, without the need to apply any correction for the sampling effects, apart from a simple substraction of the shot noise term. We thank Yipeng Jing for helpful discussions, Olaf Wucknitz and the anonymous referee for helpful comments that greatly improved the presentation of this paper. This work is supported by the One Hundred Talents project, Shanghai Pujiang Program (No. 07pj14102), 973 Program (No. 2007CB815402), 863 program (No. 2006AA01A125), the CAS Knowledge Innovation Program (Grant No. KJCX2-YW-T05) and grants from NSFC (Nos. 10533030, 10673023, 10373012, 10633049). ## References * (1) Bardeen J.M., Bond J.R., Kaiser N., Szalay A.S., 1986, ApJ, 304, 15 (BBKS). * (2) Baumgart, D.J. & Fry, J.N. 1991, ApJ, 375, 25 * (3) Briggs D.S., Schwab F.R., Sramek R.A., 1999, ASPC, 180, 127 * (4) Daubechies I., 1988, Comm. Pure Appl. Math., 41 (7), 909 * (5) Daubechies I., Ten Lectures on Wavelets, SIAM, 1992. * (6) Davis, M., Efstathiou, G., Frenk, C.S., & White, S.D.M. 1985, ApJ, 292, 371 * (7) Efstathiou G., Bond J.R., White S.D.M., 1992, MNRAS, 258, 1 (EBW) * (8) Eisenstein D.J., Hu W., 1998, ApJ, 496, 605 * (9) Fang, L. Z., & Thews, R. 1998, Wavelet in Physics (Singapore : World Scientifc) * (10) Fang, L.Z., & Feng, L.L., 2000, ApJ, 539, 5 * (11) Feng, L.L. & Fang, L.Z., 2004, ApJ, 601, 54 * (12) Fisher, K., Davis, M., Strauss, M.A., Yahil, A., & Huchra, J.P. 1993, ApJ, 402, 42 * (13) Gambera, M., Pagliaro, A., Antonuccio-Delogu, V., Becciani, U., 1997, ApJ, 488, 136 * (14) Grebenev, S. A., Forman, W., Jones, C., Murray, S., 1995, ApJ, 445, 607 * (15) Hockney, R.W. & Eastwood, J.W. 1981, Computer simulations using particles. Mc Graw-Hill * (16) Jing, Y.P. 2005, ApJ, 620, 559 * (17) Jing, Y.P. & Valdarnini, R. 1993, ApJ, 406, 6 * (18) Kaiser, N. 1991, in Texas/ESO-CERN Symposium on Relativistic Astrophysics, eds. J. Barrow et al. (New York: New York Academic Science), 295 * (19) Lin H., Kirshner R.P., Shectman S.A., Landy S.D., Oemler A., Tucker D.L., Schechter P.L., 1996, ApJ, 471, 671 * (20) Ma C.P, Fry J.N., ApJ, 543, 503 * (21) Martinez, V.J., Paredes, S., Saar, E., 1993, MNRAS, 260, 365 * (22) Meiksin, A., 2000, MNRAS, 314, 566 * (23) Pando, J., Fang, L.Z., 1996, ApJ, 459, 1 * (24) Peacock J.A., Dodds S.J., 1996, MNRAS, 280, 19 * (25) Peacock, J.A. & Nicholson, D. 1991, MNRAS, 253, 307 * (26) Peebles, P.J.E. 1980, The large scale structure of the universe. (Princeton: Princeton University Press) * (27) Percival W.J., et al., 2001, MNRAS, 327, 1297 * (28) Percival W.J., et al., 2007, ApJ, 657, 645 * (29) Sánchez A.G., Baugh C.M., Percival W.J., Peacock J.A., Padilla N.D., Cole S., Frenk C.S., Norberg P., 2006, MNRAS, 366, 189 * (30) Schäfer, B. M., Pfrommer, C., Zaroubi, S., 2005, MNRAS, 362, 1418 * Scoccimarro et al. (1998) Scoccimarro, R., Colombi, S., Fry, J. N., Frieman, J. A., Hivon, E., & Melott, A. 1998, ApJ, 496, 586 * (32) Seljak U., Zaldarriaga M., 1996, ApJ, 469, 437 * (33) Slezak, E, Durret, F., Gerbal, D., 1994, AJ, 108, 1996 * (34) Smith R.E., et al., 2003, MNRAS, 341, 1311 * (35) Spergel D.N., et al., 2007, ApJS, 170, 377 * (36) Springel V. et al., 2005, Nature, 435, 629 * (37) Tegmark M., Hamilton A.J.S., Xu Y., 2002, MNRAS, 335, 887 * (38) Tegmark M., et al., 2004, ApJ, 606, 702 * (39) Yang X., Feng L.L., Chu Y.Q., Fang L.Z., 2001, ApJ, 553, 1 * (40) Yang X., Feng L.L., Chu Y.Q., Fang L.Z., 2002a, ApJ, 560, 549 * (41) Yang X., Feng L.L., Chu Y.Q., Fang L.Z., 2002b, ApJ, 566, 630

arxiv-papers

2024-09-04T02:48:54.672832

0804.0091

# A family of conformally flat Hamiltonian-minimal Lagrangian tori in $\mathbb{CP}^{3}$ Andrey E. Mironov and Dafeng Zuo Mironov, Sobolev Mathematical institute, Siberian Branch of the Russian Academy of Sciences and Novosibirsk state university, Novosibirsk Zuo, School of Mathematics, Korea Institute for Advanced Study 207-43 Cheongnyangni 2-dong, Dongdaemun-gu Seoul, 130-722 Korea and Department of Mathematics, University of Science and Technology, Hefei 230026, P.R.China mironov@math.nsc.ru, dfzuo@kias.re.kr ###### Abstract. In this paper by reduction we construct a family of conformally flat Hamiltonian-minimal Lagrangian tori in $\mathbb{CP}^{3}$ as the image of the composition of the Hopf map $\mathcal{H}:\mathbb{S}^{7}\to\mathbb{CP}^{3}$ and a map $\psi:\mathbb{R}^{3}\to\mathbb{S}^{7}$ with certain conditions. MSC(2000): 05C42, 53D12, 35Q51; Key words: conformally flat, Hamiltonian-minimal Lagrangian tori ## 1\. Introduction A Lagrangian submanifold of a Kähler manifold is said to be Hamiltonian- minimal (briefly, H-minimal) if it is a critical point of the volume under Hamiltonian deformations. This notion was introduced by Oh in [17], who also gave an example, that is, Clifford tori in $\mathbb{C}^{n}$ with standard Hermitian metric $\mathbb{S}^{1}(r_{1})\times\dots\times\mathbb{S}^{1}(r_{n})\subset{\mathbb{C}}^{n},$ where $\mathbb{S}^{1}(r_{j})$ is a circle of radius $r_{j}$ in ${\mathbb{C}}$. In [7], Hélein and Romon studied a general construction of H-minimal tori in $\mathbb{C}^{2}$ from the point of view of completely integrable systems. They provided new explicit nontrivial examples of H-minimal Lagrangian tori which include the examples previously constructed by Castro and Urbano in $\mathbb{C}^{2}$ [3]. A similar construction has been generalized to the cases of Hermitian symmetric spaces, e.g., in $\mathbb{CP}^{2}$, see [8, 11, 10] for details. However, in the non-flat cases, the underlying equations are no longer linear, which makes the problem much harder. Although in [10], Ma introduced a spectral parameter $\lambda\in\mathbb{S}^{1}$, as she pointed out, a description of H-minimal Lagrangian tori in $\mathbb{CP}^{2}$ in terms of theta functions seems to be possible. But owing to this spectral parameter $\lambda\not\in\mathbb{C}$, thus it is still open about the integrability of this problem in classical sense. In [13, 14] it is shown that if a Lagrangian conformal map from ${\mathbb{R}}^{2}$ to ${\mathbb{C}}P^{2}$ is given as composition of maps $\varphi:={\mathcal{H}}\circ\psi:{\mathbb{R}}^{2}\rightarrow S^{5}\rightarrow{\mathbb{C}}P^{2},$ where ${\mathcal{H}}$ is the Hopf map, then components of $\varphi$ satisfy Shrödinger equation $\Delta\varphi_{j}+i(\theta_{x}\partial_{x}\varphi_{j}+\theta_{y}\partial_{y}\varphi_{j})+4e^{u}\varphi_{j}=0,$ where $ds^{2}=2e^{u}(dx^{2}+dy^{2})$ is an induced metric and $\theta(x,y)$ is Lagrangian angle. In the case of H-minimal Lagrangian tori $\theta$ is a linear function. So in order to construct finite gap tori it is necessary to use spectral data of finite gap Shrödiger operators, see [6], [18] for details. In [2], Castro and Urbano constructed a family of minimal Lagrangian tori in $\mathbb{CP}^{2}$, which are characterized by their invariance under a one- parameter group of holomorphic isometries of $\mathbb{CP}^{2}$. By using this idea, in [13, 9], they independently reduced this problem to a one dimensional system and obtained an equivariant solution in terms of elliptic functions, and then constructed H-minimal Lagrangian tori in $\mathbb{CP}^{2}$. In high dimensional case one of the present authors constructed some examples of H-minimal and minimal Lagrangian immersions and embeddings in ${\mathbb{C}}^{n}$ and $\mathbb{CP}^{n}$. Recently we can find some of works about this topic, see [1, 4, 5] and references therein. But it is far from the complete characterization. In this paper we address to construct a family of conformally flat H-minimal Lagrangian tori in $\mathbb{CP}^{3}$ by reduction methods, which generalizes the results in [13, 15, 9], with the metric $ds^{2}=e^{u}(dx^{2}+dy^{2}+dz^{2}).$ In $\mathbb{CP}^{3}$, it seems to be a little harder. We thus restrict to discuss a very special case, that is, $u=u(z)$ and the Lagrangian angle $\theta=ax+by$, where $a$ and $b$ are arbitrary real constants. ## 2\. Preliminaries In this section, we review some well-known facts without proofs and sketch our strategy about the construction of H-minimal Lagrangian cone or tori in $\mathbb{C}^{4}$ and $\mathbb{CP}^{3}$. ### 2.1. Notations Let $\mathbb{C}^{4}$ be the canonical complex space of dimension $4$ endowed with an Hermitian product $\langle u,v\rangle=\displaystyle\sum_{k=1}^{4}u_{k}\bar{v}_{k}.$ Let us denote $\omega=\mbox{Im}\langle~{},~{}\rangle$ and $(~{},~{})=\mbox{Re}\langle~{},~{}\rangle$. Let $\psi:\mathbb{R}^{3}\to\mathbb{S}^{7}$ be an oriented Lagrangian immersion $L$, i.e. $\psi^{*}\omega=0$, where $\mathbb{S}^{7}$ is the unit sphere in ${\mathbb{C}}^{4}$. Wolfson in [19] introduced a Lagrangian angle $\theta$ and obtained a criterion of H-minimality of $L$ in terms of $\theta$, that is, the Lagrangian immersion $L$ is H-minimal if and only if the Lagrangian angle $\theta$ is a harmonic function on $L$. The Lagrangian angle $\theta$ of $L$ in $\mathbb{C}^{4}$ is defined by the formula $e^{i\,\,\theta(p)}=dz_{1}\wedge\dots\wedge dz_{4}(\Psi),\quad p\in L,$ where $z_{j},j=1,\cdots,4$ are coordinates on ${\mathbb{C}}^{4}$ and $\Psi$ is an orthonormal tangent frame at $p\in L$ with the same orientation of $L$. For the general case, see [19] for details. Let $\mathcal{H}:S^{7}\rightarrow{\mathbb{C}}P^{3}$ be the Hopf map. An induced Hermitian product on ${\mathbb{C}P}^{3}$ is called to be the Fubini- Study metric defined by $<\zeta_{1},\zeta_{2}>:=\langle\tilde{\zeta}_{1},\tilde{\zeta}_{2}\rangle,$ where $\zeta_{i},~{}i=1,2$ are tangent to ${\mathbb{C}P}^{3}$ and $\tilde{\zeta}_{i}$ are the corresponding horizontal lifting by $\mathcal{H}$. Let $\mathcal{C}$ be a Lagrangian cone in ${\mathbb{C}}^{4}$ with the vertex at the origin. It follows from the definition of $<~{},~{}>$ that $\mathcal{H}(\widetilde{\mathcal{C}})$ is a Lagrangian submanifold in ${\mathbb{C}P}^{3}$, where $\widetilde{\mathcal{C}}=\mathcal{C}\cap S^{7}$. Moreover, if the cone $\mathcal{C}$ is H-minimal in $\mathbb{C}^{4}$, then $\mathcal{H}(\widetilde{\mathcal{C}})$ is also H-minimal in ${\mathbb{C}P}^{3}$, see [12, 1] for details. ### 2.2. On conformally flat Lagrangian immersions In the following we only consider conformally flat immersions in $\mathbb{C}^{4}$ and $\mathbb{CP}^{3}$. Let $\psi=(\psi^{1},\psi^{2},\psi^{3},\psi^{4}):\mathbb{R}^{3}\to\mathbb{S}^{7}\subset\mathbb{C}^{4}$ be an oriented immersion with a conformally flat metric $ds^{2}=e^{u(x,y,z)}(dx^{2}+dy^{2}+dz^{2})$ satisfying the following properties $\langle\psi,\psi_{x}\rangle=\langle\psi,\psi_{y}\rangle=\langle\psi,\psi_{z}\rangle=0,$ $\langle\psi_{x},\psi_{y}\rangle=\langle\psi_{y},\psi_{z}\rangle=\langle\psi_{z},\psi_{x}\rangle=0,$ by the above arguments, thus $\mathcal{H}\circ\psi$ is a Lagrangian immersion in $\mathbb{CP}^{3}$ and (2.1) $\Phi=(\psi,e^{-u}\psi_{x},e^{-u}\psi_{y},e^{-u}\psi_{z})^{t}\in\mbox{U}(4)$ and the Lagrangian angle $\theta(x,y,z)$ is given by (2.2) $e^{i\,\,\theta}={\rm det}(\Phi).$ By using (2.1) and (2.2), we have (2.3) $\Psi=(e^{i\,\,\theta}\psi,e^{-u}\psi_{x},e^{-u}\psi_{y},e^{-u}\psi_{z})^{t}\in\mbox{SU}(4)$ and denote (2.4) $\mathcal{U}:=\Psi_{x}\Psi^{-1},\quad\mathcal{V}:=\Psi_{y}\Psi^{-1},\quad\mathcal{W}:=\Psi_{z}\Psi^{-1}\in\mbox{SU}(4).$ The compatibility condition of (2.4) is (2.5) $\mathcal{U}_{y}-\mathcal{V}_{x}+[\mathcal{U},\mathcal{V}]=0,\,\mathcal{V}_{z}-\mathcal{W}_{y}+[\mathcal{V},\mathcal{W}]=0,\,\mathcal{W}_{x}-\mathcal{U}_{z}+[\mathcal{W},\mathcal{U}]=0.$ Moreover, if $\triangle\theta(x,y,z)=0$, then the immersion $\mathcal{H}\circ\psi$ is a conformally flat H-minimal Lagrangian immersion in $\mathbb{CP}^{3}$, where $\triangle$ is the corresponding Lapalacian operator given by the formula $\triangle:=-e^{-u(z)}\,[\,\frac{\partial^{2}}{\partial x^{2}}+\frac{\partial^{2}}{\partial y^{2}}+\frac{\partial^{2}}{\partial z^{2}}+\frac{u_{x}}{2}\frac{\partial}{\partial x}+\frac{u_{y}}{2}\frac{\partial}{\partial y}+\frac{u_{z}}{2}\frac{\partial}{\partial z}\,].$ Conversely, given $\mbox{SU}(4)$-valued $\mathcal{U},\ \mathcal{V},\ \mathcal{W}$ satisfying (2.5), we solve the system (2.4) with (2.3) for $\Psi\in\mbox{SU}(4)$ and then obtain the immersion $\psi:\mathbb{R}^{3}\to\mathbb{S}^{7}$ and the Lagrangian angle $\theta$. If $\triangle\theta=0$ and $\psi$ satisfies certain periodic conditions, then the immersion $\mathcal{H}\circ\psi$ gives a conformally flat H-minimal Lagrangian torus in $\mathbb{CP}^{3}$. Notice that for general case, this method does not work. ## 3\. Conformally flat H-minimal Lagrangian tori in $\mathbb{CP}^{3}$ In this section, by using the above method, we will construct a special class of conformally flat H-minimal Lagrangian tori in $\mathbb{CP}^{3}$ with $u=u(z)$ and the Lagrangian angle $\theta=ax+by$, where $a$ and $b$ are arbitrary real constants. ### 3.1. The first step is to choose $\mathcal{U}$, $\mathcal{V}$ and $\mathcal{W}$. By a direct calculation, we have the following lemma. ###### Lemma 3.1. Suppose $\mathcal{U}=(u_{kl})$, $\mathcal{V}=(v_{kl})$, $\mathcal{W}=(w_{kl})$ are $\hbox{SU}(4)$-valued matrices satisfying (2.4) and $u_{kl},v_{kl},w_{kl}$ depend only one variable $z$ for $2\leq k,l\leq 4$, then they must be the following form $\small{\mathcal{U}=\left(\begin{array}[]{cccc}ia&e^{i\,\theta+u}&0&0\\\ -e^{-i\,\theta+u}&-i(a+c_{1})&ic_{2}&ic_{3}e^{-3u}-u^{\prime}\\\ 0&ic_{2}&ic_{1}&0\\\ 0&ic_{3}e^{-3u}+u^{\prime}&0&0\\\ \end{array}\right),}$ and $\small{\mathcal{V}=\left(\begin{array}[]{cccc}ib&0&e^{i\,\theta+u}&0\\\ 0&ic_{2}&ic_{1}&0\\\ -e^{-i\,\theta+u}&ic_{1}&-i(b+c_{2})&ic_{3}e^{-3u}-u^{\prime}\\\ 0&0&ic_{3}e^{-3u}+u^{\prime}&0\\\ \end{array}\right),}$ and $\small{\mathcal{W}=\left(\begin{array}[]{cccc}0&0&0&e^{i\,\theta+u}\\\ 0&ic_{3}e^{-3u}&0&0\\\ 0&0&ic_{3}e^{-3u}&0\\\ -e^{-i\,\theta+u}&0&0&-2ic_{3}e^{-3u}\\\ \end{array}\right)},$ where $u=u(z)$ satisfies the equation (3.1) $u^{\prime}\,{}^{2}+e^{2u}+c_{3}^{2}e^{-6u}-\mathfrak{C}=0,~{}~{}u^{\prime}=\frac{du}{dz}.$ and $c_{1}$, $c_{2}$, $c_{3}$ are arbitrary real constants and $\mathfrak{C}=ac_{1}+bc_{2}+2c_{1}^{2}+2c_{2}^{2}$. ### 3.2. The second step is to solve the following system (3.2) $\Psi_{x}=\mathcal{U}\Psi,\quad\Psi_{y}=\mathcal{V}\Psi,\quad\Psi_{z}=\mathcal{W}\Psi.$ Write (3.3) $\Psi=(e^{i\,\,\theta}\psi,e^{-u}\psi_{x},e^{-u}\psi_{y},e^{-u}\psi_{z})^{t},$ where $\psi=\psi(x,y,z)$ is a smooth function. By using (3.3), the system (3.2) can be rewritten as (3.4) $\displaystyle\psi_{xz}-(u^{\prime}+ic_{3}e^{-3u})\psi_{x}=0,$ (3.5) $\displaystyle\psi_{yz}-(u^{\prime}+ic_{3}e^{-3u})\psi_{y}=0,$ (3.6) $\displaystyle\psi_{xy}-i(c_{2}\psi_{x}+ic_{1}\psi_{y})=0,$ and (3.7) $\displaystyle\psi_{xx}+e^{2u}\psi+i(a+c_{1})\psi_{x}-ic_{2}\psi_{y}+((u^{\prime}-ic_{3}e^{-3u})\psi_{z}=0,$ (3.8) $\displaystyle\psi_{yy}+e^{2u}\psi- ic_{1}\psi_{x}+i(c_{2}+b)\psi_{y}+((u^{\prime}-ic_{3}e^{-3u})\psi_{z}=0,$ (3.9) $\displaystyle\psi_{zz}+e^{2u}\psi+(2ic_{3}e^{-3u}-u^{\prime})\psi_{z}=0.$ From (3.4) and (3.5), we know that $\psi$ must be of the form (3.10) $\psi=P(z)\varphi(x,y)+Q(z),\quad\varphi(x,y)\neq\mbox{constant},$ and then (3.4) and (3.5) reduce to (3.11) $P(z)-(u^{\prime}+ic_{3}e^{-3u})P(z)=0.$ The solution of (3.11) is (3.12) $P(z)=a_{1}e^{u+ic_{3}\int e^{-3u}dz},\quad a_{1}\in\mathbb{R}.$ Substituting (3.10) and (3.12) into (3.6), we get (3.13) $\varphi_{xy}-i(c_{2}\varphi_{x}+c_{1}\varphi_{y})=0.$ Substituting (3.10) into (3.7) and (3.8), and then using (3.1) and (3.13), we get the following (3.14) $\displaystyle Q^{\prime}(z)=\frac{e^{5u}Q(z)}{ic_{3}-u^{\prime}e^{3u}},$ (3.15) $\displaystyle\varphi_{xx}+i(a+c_{1})\varphi_{x}-ic_{2}\varphi_{y}+\mathfrak{C}\,\varphi=0,$ (3.16) $\displaystyle\varphi_{yy}-ic_{1}\varphi_{x}+i(b+c_{2})\varphi_{y}+\mathfrak{C}\,\varphi=0.$ Write (3.17) $Q(z)=H(z)e^{iG(z)},$ where $G(z)$ and $H(z)$ are real smooth functions. By differentiating (3.1), we obtain (3.18) $u^{\prime\prime}+e^{2u}-3c_{3}^{2}e^{-6u}=0.$ By using (3.18), and separating the real part and the imaginary part of (3.14), we obtain $H^{\prime}(z)(\mathfrak{C}-e^{2u})+u^{\prime}e^{2u}H(z)=0,\quad G^{\prime}(z)(e^{2u}-\mathfrak{C})-c_{3}e^{-u}=0,$ thus $H(z)=a_{2}\sqrt{\mathfrak{C}-e^{2u}},~{}~{}a_{2}\in\mathbb{R},\quad G(z)=\int\frac{c_{3}e^{-u}}{e^{2u}-\mathfrak{C}}dz+a_{3},~{}~{}a_{3}\in\mathbb{R}.$ From (3.13), without loss of generality, we could assume that $\varphi(x,y)$ has the form $\varphi(x,y)=\sum a_{\alpha_{j}\beta_{j}}e^{i(x\alpha_{j}+y\beta_{j})},~{}~{}\beta_{j}=\frac{c_{2}\alpha_{j}}{\alpha_{j}-c_{1}},~{}~{}a_{\alpha_{j}\beta_{j}}\in\mathbb{C},~{}~{}a_{00}=0.$ It follows from (3.17),(3.15) and (3.16) that $\alpha=\alpha_{j}$ is a root of the equation ${\alpha}^{3}+a{\alpha}^{2}-\mathfrak{B}\alpha+{c_{{1}}}\mathfrak{C}=0.$ where $\mathfrak{B}=2\,c_{{1}}a+3\,{c_{{1}}}^{2}+c_{{2}}b+3\,{c_{{2}}}^{2}$. Notice that up to now we only use (3.4)—(3.8) to obtain an explicit form of $\psi$ as follows (3.19) $\displaystyle\psi(x,y,z)$ $\displaystyle=\sum a_{1}a_{\alpha_{j}\beta_{j}}e^{u(z)+ic_{3}\int{e^{-3u(z)}dz}}e^{i(x\alpha_{j}+y\beta_{j})}$ $\displaystyle\quad+~{}a_{2}e^{ia_{3}}\sqrt{\mathfrak{C}-e^{2u(z)}}e^{ic_{3}\int{\dfrac{e^{-u(z)}}{e^{2u(z)}-\mathfrak{C}}}dz}.$ Furthermore, it is easy to check that this function $\psi$ also satisfies (3.9). Thus we solve the system (3.2). Summarizing the above discussions, we have the following proposition. ###### Proposition 3.2. If we suppose that $u=u(z)$ is a smooth solution of $u(z)^{\prime}\,{}^{2}+e^{2u(z)}+c_{3}^{2}e^{-6u(z)}-\mathfrak{C}=0$; and $\alpha$ is a root of the equation (3.20) ${\alpha}^{3}+a\,{\alpha}^{2}-\mathfrak{B}\,\alpha+{c_{{1}}}\mathfrak{C}=0$ Then $\Psi=(e^{i\,\,\theta}\psi,e^{-u}\psi_{x},e^{-u}\psi_{y},e^{-u}\psi_{z})^{t}$ is a solution of the system (3.2) with $\theta=a\,x+b\,y$ and $\psi(x,y,z)=\kappa_{1}e^{i(\alpha\,x+\beta\,y)}P(z)+\kappa_{2}Q(z),\quad\beta=\dfrac{c_{2}\alpha}{\alpha- c_{1}},$ where $\kappa_{1}$ and $\kappa_{2}$ are arbitrary complex constants and $P(z)=e^{u(z)+ic_{3}\int{e^{-3u(z)}dz}},\quad Q(z)=\sqrt{\mathfrak{C}-e^{2u(z)}}e^{ic_{3}\int{\dfrac{e^{-u(z)}}{e^{2u(z)}-\mathfrak{C}}}dz}.$ ### 3.3. Main results We now state our main theorem. ###### Theorem 3.3. Suppose that the equation (3.20) has three distinct roots, denoted by $\alpha_{1}$, $\alpha_{2}$ and $\alpha_{3}$. Write $\beta_{j}=\frac{c_{2}\alpha_{j}}{\alpha_{j}-c_{1}},~{}j=1,2,3$. Then the map $\mathcal{H}\circ\psi:\mathcal{R}^{3}\to\mathbb{CP}^{3}$ defines a conformally flat H-minimal Lagrangian immersion in $\mathbb{CP}^{3}$, where $\mathcal{H}:\mathbb{S}^{7}\to\mathbb{CP}^{3}$ is the Hopf map and the map $\psi:\mathcal{R}^{3}\to\mathbb{S}^{7}\subset\mathbb{C}^{4}$ is given by the formula $\psi=(\gamma_{1}P(z)e^{i(x\alpha_{1}+y\beta_{1})},\ \gamma_{2}P(z)e^{i(x\alpha_{2}+y\beta_{2})},\ \gamma_{3}P(z)e^{i(x\alpha_{3}+y\beta_{3})},\ \gamma_{4}Q(z)).$ Here $\gamma_{4}=\sqrt{\frac{1}{\mathfrak{C}}}$ and $\tiny{\gamma_{1}=\sqrt{\frac{\mathfrak{C}+\alpha_{2}\alpha_{3}}{\mathfrak{C}(\alpha_{1}-\alpha_{2})(\alpha_{1}-\alpha_{3})}}\,,\,\gamma_{2}=\sqrt{\frac{\mathfrak{C}+\alpha_{1}\alpha_{3}}{\mathfrak{C}(\alpha_{2}-\alpha_{1})(\alpha_{2}-\alpha_{3})}}\,,\,\gamma_{3}=\sqrt{\frac{\mathfrak{C}+\alpha_{1}\alpha_{2}}{\mathfrak{C}(\alpha_{3}-\alpha_{1})(\alpha_{3}-\alpha_{2})}}}\,.$ ###### Proof. It suffices to check that $\Psi=(e^{i\,\,\theta}\psi,e^{-u}\psi_{x},e^{-u}\psi_{y},e^{-u}\psi_{z})^{t}\in\mbox{SU}(4).$ By using (3.20), we have (3.21) $\alpha_{1}+\alpha_{2}+\alpha_{3}=-a,~{}~{}\alpha_{1}\alpha_{2}+\alpha_{1}\alpha_{3}+\alpha_{2}\alpha_{3}=-\mathfrak{B},~{}~{}\alpha_{1}\alpha_{2}\alpha_{3}=-\mathfrak{C}.$ It follows from (3.21) and the explicit forms of $\gamma_{j}$ that $\displaystyle\displaystyle\sum_{j=1}^{3}\gamma_{j}^{2}=\gamma_{4}^{2},$ $\displaystyle\quad\sum_{j=1}^{3}\gamma_{j}^{2}\alpha_{j}=0,\quad\displaystyle\sum_{j=1}^{3}\gamma_{j}^{2}\alpha_{j}^{2}=1,$ $\displaystyle\displaystyle\sum_{j=1}^{3}\gamma_{j}^{2}\alpha_{j}\beta_{j}=0,$ $\displaystyle\quad\displaystyle\sum_{j=1}^{3}\gamma_{j}^{2}\beta_{j}=0,\quad\sum_{j=1}^{3}\gamma_{j}^{2}\beta_{j}^{2}=1.$ These identities yield that $\displaystyle\langle\psi,\psi\rangle=1,\langle\psi_{x},\psi_{x}\rangle=\langle\psi_{y},\psi_{y}\rangle=e^{2u},$ $\displaystyle\langle\psi,\psi_{x}\rangle=\langle\psi,\psi_{y}\rangle=\langle\psi,\psi_{z}\rangle=0,$ $\displaystyle\langle\psi_{x},\psi_{y}\rangle=\langle\psi_{y},\psi_{z}\rangle=\langle\psi_{z},\psi_{x}\rangle=0,$ $\displaystyle\langle\psi_{z},\psi_{z}\rangle=P^{\prime}(z)\overline{P^{\prime}(z)}\sum_{j=1}^{3}\gamma_{j}^{2}+\gamma_{4}^{2}Q^{\prime}(z)\overline{Q^{\prime}(z)}$ $\displaystyle\qquad=\gamma_{4}^{2}[e^{2u}(u^{\prime}2+c_{3}^{2}e^{-6u})+\frac{e^{4u}(u^{\prime}2+c_{3}^{2}e^{-6u})}{\mathfrak{C}-e^{2u}}]$ $\displaystyle\qquad=e^{2u}.\qquad\mbox{by using \eqref{MZ3.3}}$ That is to say, $\Psi\in\mbox{SU}(4)$. Thus we complete the proof of the theorem.∎ We finish this section to discuss how to obtain conformally flat H-minimal Lagrangian tori in $\mathbb{CP}^{3}$. Notice that in (3.1) if we make the following change (3.22) $u=u(z):=-\log(2\sqrt{-q(z)}\,),$ then we have (3.23) $q^{\prime}(z)^{2}=256c_{3}^{2}q(z)^{5}+4\mathfrak{C}q(z)^{2}+q(z).$ Thus if we choose three real constants $c_{1}$, $c_{2}$ and $c_{3}$ such that the equation $256c_{3}^{2}t^{5}+4\mathfrak{C}t^{2}+t=0$ has two negative roots and does not have multiple roots, then this assures that (3.23) has a smooth periodic solution of the period $\tau$, see [16] for details. It follows from (3.22) that so is (3.1). We here remark that in this case $\mathfrak{C}=ac_{1}+bc_{2}+2c_{1}^{2}+2c_{2}^{2}>0$. We next discuss the condition such that the function $\psi$ in (3.19) is a periodic function of $x$, $y$,$z$ repesctively. According to the form of $\psi$ in (3.19), if we assume that $c_{1}\in\mathbb{Q}$ and $\alpha_{2}$, $\alpha_{2}$ and $\alpha_{3}$ are three distinct rational roots of (3.20), then $\psi$ is periodic w.r.t. $x$ and $y$. Notice that $u(z+\tau)=u(z)$ and there exists a periodic function $h(z)$ of the periodic $\tau$ such that $\int\dfrac{e^{-3u(z)}\mathfrak{C}}{e^{2u(z)}-\mathfrak{C}}dz=h(z)+z\int_{0}^{\tau}\dfrac{e^{-3u(z)}\mathfrak{C}}{e^{2u(z)}-\mathfrak{C}}dz.$ This implies that if we assume $\frac{c_{3}\mathfrak{C}\tau}{2\pi}\int_{0}^{\tau}\dfrac{e^{-3u(z)}}{e^{2u(z)}-\mathfrak{C}}dz\in\mathbb{Q}\,,$ then $\psi$ is periodic in $z$ with the period $n\tau$ for some $n\in\mathbb{N}$. Thus, combining with Theorem 3.3, we have ###### Theorem 3.4. If we suppose that $1$. $u=u(z)$ is a periodic solution of (3.1) with the period $\tau$; $2$. $c_{1}\in\mathbb{Q},\quad\dfrac{c_{3}\mathfrak{C}\tau}{2\pi}\int_{0}^{\tau}\dfrac{e^{-3u(z)}}{e^{2u(z)}-\mathfrak{C}}dz\in\mathbb{Q}\,$; $3$. $\alpha_{1}$, $\alpha_{2}$ and $\alpha_{3}$ are distinct rational roots of (3.20). Then the map $\mathcal{H}\circ\psi:\mathbb{R}^{3}\to\mathbb{CP}^{3}$ defines a conformally flat H-minimal Lagrangian torus in $\mathbb{CP}^{3}$. ## Acknowledgment A.E.Mironov is grateful to generous supports by KIAS, where part of the work was done in KIAS. The work of A.E.Mironov was partially supported by Russian Federation foundation for basic research (grant no.06-01-00094a) and grant MK-9651.2006.1 of the president of Russian Federation. Zuo would like to thank Qing Chen and Bumsig Kim and Youjin Zhang for their constant guidance and supports. Zuo also thanks Chengmin Bai for hospitality during stay at Chern Institute of mathematics in Nankai university, where part of the work was done. The work of Zuo was partially supported by a post-doc fellowship from KIAS and the Natural Science Foundation of China (grant no. 10501043). Both of the authors thank organizers for the hospitality during the ISLAND-3 on Islay, Scotland. ## References * [1] Castro I., Li H. and Urbano F., Hamiltonian-minimal Lagrangian submanifolds in Complex space forms, Pacific J. Math., 227 (2006) 43–63. * [2] Castro I. and Urbano F., New examples of minimal Lagrangian tori in the complex projective plane, Manuscripta Math., 85(1994) 265–281. * [3] Castro I. and Urbano F., Examples of unstable Hamiltonian-minimal Lagrangian tori in $\mathbb{C}^{2}$., Compositio Math., 111 (1998)1–14. * [4] Chen B.-Y. and Garay O.J., Classification of Hamiltonian-stationary Lagrangian sumbanifolds of constant in ${\mathbb{C}}P^{3}$ with positive nullity, Nonlinear Analysis (2007) doi:10.1016/j.na.2007.06.005. * [5] Dong Y.-X., Hamiltonian-minimal Lagrangian submanifolds in Kaehler manifolds with symmetries, Nonlinear Anal., 67 (2007) 865–882. * [6] Dubrovin B.A., Krichever I.M. and Novikov S.P., The Schrödinger equation in a periodic field and Riemann surfaces, Sov. Math. Dokl., 17 (1976) 947–951. * [7] Hélen F. and Romon P.; Hamiltonian stationary Lagrangian surfaces in $\mathbb{C}^{2}$., Comm. Anal. Geom. 10 (2002)79–126. * [8] Hélen F. and Romon P. Hamiltonian stationary tori in ${\mathbb{C}P}^{2}$, Comm. Anal. Geom., 10 (2002) 79–126. * [9] Ma Hui and Schmies M., Examples of hamiltonian Stationary Lagrangian tori in ${\mathbb{C}}P^{2}$, Geometriae Dedicata, 118(2006) 173–183. * [10] Ma Hui, Hamiltonian stationary Lagrangian surfaces $\mathbb{CP}^{2}$ , Ann. Global. Anal Geom., 27(2005) 1–16. * [11] Ma Hui and Ma Y.-J., Totally real minimal minimal tori in $\mathbb{CP}^{2}$, Math. Z. 249(2005) 241–267. * [12] Mironov A.E., New examples of Hamiltionian-minimal and minimal Lagrangian manifolds in $\mathbb{C}^{n}$ and $\mathbb{CP}^{n}$, (Russian), Mat. Sb. 195 (2004) 89–102. * [13] Mironov A.E., On the Hamiltonian-minimal Lagrangian tori in the ${\mathbb{C}}P^{2}$, Siberian Math. J., 44(2003) 1039–1042. * [14] Mironov A.E., The Veselov–Novikov hierarchy of equations, and integrable deformations of minimal Lagrangian tori in ${\mathbb{C}}P^{2}$ (in Russian), Sib. Elektron. Mat. Izv., 1(2004) 38–46. * [15] Mironov A.E., On family of conformal flat minimal Lagrangian tori in ${\mathbb{C}P}^{3}$, Math. Notes., 81 (2007) 329–337. * [16] Novikov S.P.(Ed.), Theory of Solitons. The Inverse Sacttering Method, Plenum, New York, 1984. * [17] Oh,Y-G., Volume minimization of Lagrangian submanifolds under Hamiltonian deformations, Math. Z.,212(1993) 175–192. * [18] Veselov A.P. and Novikov, S.P., Finite-zone two-dimensional potential Schrödinger operators. Explicit formulas and evolution equations, Sov. Math. Dokl., 30(1984) 588–591. * [19] Wolfson J., Minimal Lagrangian diffeomophisms and the Monge-Ampére equation, J. Differential Geom. 45 (1997) 335–373.

arxiv-papers

2024-09-04T02:48:54.678304

0804.0114

# Spectral data for Hamiltonian-minimal Lagrangian tori in ${\mathbb{C}}P^{2}$ A. E. Mironov This work was supported by the RFBR (grant 06-01-00094a), by the SB RAS (complex integration project No. 2.15), and by IHES, France ## 1 Introduction In this work, we find spectral data that allow to find Hamiltonian-minimal Lagrangian tori in ${\mathbb{C}}P^{2}$ in terms of theta functions of spectral curves. A Lagrangian submanifold in a Kähler manifold is called Hamiltonian-minimal, if the variations of volume along the Hamiltonian fields are equal to zero. The simplest example of a Hamiltonian-minimal Lagrangian submanifold (HML- submanifold) is a Clifford torus [1] $S^{1}(r_{1})\times\dots\times S^{1}(r_{n})\subset{\mathbb{C}}^{n},$ where $S^{1}(r_{i})\subset{\mathbb{C}}$ is a circle of radius $r_{i}$. See [2] for more examples of closed HML-submanifolds in ${\mathbb{C}}^{n}$ and ${\mathbb{C}}P^{n}$. Helein and Romon [3] give a description of HML-tori in ${\mathbb{C}}^{2}$ using a Weierstrass representation of Lagrangian surfaces. McIntosh and Romon [4] find spectral data for such surfaces. Until recently, only a few examples of Hamiltonian-minimal (but not minimal) Lagrangian tori in ${\mathbb{C}}P^{2}$ were known (see [2], [5], [6]). The authors of [7], [8] show that HML-tori in ${\mathbb{C}}P^{2}$ are finite type surfaces. Below we show that HML-tori in ${\mathbb{C}}P^{2}$ are described by the following system of equations (Lemma 3) $V_{y}+U_{x}=0,$ $2U_{y}-2V_{x}=(bv_{x}+av_{y})e^{v},$ $\partial_{x}^{2}v+\partial_{y}^{2}v=4(U^{2}+V^{2})e^{-2v}-4e^{v}-2(Ua- Vb)e^{-v},$ where $ds^{2}=2e^{v(x,y)}(dx^{2}+dy^{2})$ is an induced metric on a torus, $U(x,y)$, $V(x,y)$ are periodic real functions, $a,b$ are some real constants. We do not know whether this system has a Lax representation with a spectral parameter. This is the main difficulty for constructing quasi-periodic solutions in terms of theta functions of spectral curves. Minimal Lagrangian tori correspond to the condition $a=b=0$. In this case, the indicated system of equations is reduced to the Tzizeica equation $\partial_{x}^{2}v+\partial_{y}^{2}v=4e^{-2v}-4e^{v}.$ The Tzizeica equation allows the Lax representation with a spectral parameter found in [9]. Quasi-periodic solutions of this equation were found in [10]. In [11], the authors show that there are smooth periodic solutions corresponding to smooth minimal Lagrangian tori. We briefly present the basic elements of our construction. Define a Lagrangian surface $\Sigma\subset{\mathbb{C}}P^{2}$ using a composition of the mappings $\varphi:{\mathbb{R}}^{2}\rightarrow S^{5}\subset{\mathbb{C}}^{3}$ and the Hopf projection ${\cal H}:S^{5}\rightarrow{\mathbb{C}}P^{2},$ where $S^{5}$ is a unite sphere in ${\mathbb{C}}^{3}$. Suppose that the induced metric on $\Sigma$ has a conformal form $ds^{2}=2e^{v(x,y)}(dx^{2}+dy^{2}),$ where $x,y$ are coordinates on ${\mathbb{R}}^{2}$. As follows from the Lagrangianity and the conformality of mapping ${\cal H}\circ\varphi$, $<\varphi,\varphi_{x}>=<\varphi,\varphi_{y}>=<\varphi_{x},\varphi_{y}>=0,\ |\varphi_{x}|^{2}=|\varphi_{y}|^{2}=2e^{v},$ $None$ where $<.,.>$ is a Hermitian product in ${\mathbb{C}}^{3}$ (see [5]). We are looking for the components $\varphi^{i},i=1,2,3$ of the vector function $\varphi$ in the form $\varphi^{i}=\alpha_{i}\psi(x,y,Q_{i}),$ where $\psi(x,y,Q)$ is a two-points Baker-Akhiezer function on an auxiliar Riemannian surface $\Gamma$ named spectral curve, $Q_{i}\in\Gamma$ are some points, $\alpha_{i}$ — some constants. In theorem 1, we indicate spectral data, i.e. some sufficient constraints on the Baker-Akhiezer function $\psi$, such that the equations $<\varphi,\varphi>=1,\ <\varphi,\varphi_{x}>=<\varphi,\varphi_{y}>=<\varphi_{x},\varphi_{y}>=0$ hold. Note that equation $<\varphi_{x},\varphi_{y}>=\varphi^{1}_{x}\bar{\varphi}^{1}_{y}+\varphi^{2}_{x}\bar{\varphi}^{2}_{y}+\varphi^{3}_{x}\bar{\varphi}^{3}_{y}=0,$ is ”similar” to equation $\frac{\partial x^{1}}{\partial u^{1}}\frac{\partial x^{1}}{\partial u^{2}}+\frac{\partial x^{2}}{\partial u^{1}}\frac{\partial x^{2}}{\partial u^{2}}+\frac{\partial x^{3}}{\partial u^{1}}\frac{\partial x^{3}}{\partial u^{2}}=0,$ that needs to be solved in order to construct orthogonal curvilinear systems of coordinates in ${\mathbb{R}^{3}}$. Krichever [12] indicates spectral data allowing to restore $n$-orthogonal curvilinear coordinates in ${\mathbb{R}^{n}}$. The idea of theorem 1 is close to [12]. In lemmas 5 and 6 we indicate constraints on spectral data that let mapping ${\cal H}\circ\varphi$ be conformal, i.e. condition $|\varphi_{x}|^{2}=|\varphi_{y}|^{2}=2e^{v}$ holds. The key point of our construction is assertion (lemma 2) stating that the components of the vector function $\varphi$ satisfy the Schrödinger equation $\partial_{x}^{2}\varphi^{j}+\partial_{y}^{2}\varphi^{j}-i(\beta_{x}\partial_{x}\varphi^{j}+\beta_{y}\partial_{y}\varphi^{j})+4e^{v}\varphi^{j}=0,$ where $\beta$ is a Lagrangian angle (for definition see below). The Lagrangian angle defines the mean curvature vector $H=J\nabla\beta,$ $\nabla\beta$ being the gradient of function $\beta$ in the induced metric on a torus, $J$ being the complex structure operator on ${\mathbb{C}}P^{2}$. In case that surface $\Sigma$ is a HML-torus, the Lagrangian angle is a linear function, i.e. $\beta_{x}$ and $\beta_{y}$ are constants (see [2]). If surface $\Sigma$ is a minimal Lagrangian torus, the Lagrangian angle is constant, i.e. the components $\varphi^{j}$ satisfy the potential Schrödinger equation $\partial_{x}^{2}\varphi^{j}+\partial_{y}^{2}\varphi^{j}+4e^{v}\varphi^{j}=0.$ Thus, the Schrödinger equation allows to distinguish HML-tori and minimal Lagrangian tori amongst Lagrangian tori. In the main theorem 2 we indicate spectral data corresponding to such surfaces. The method used to distinguish Lagrangian surfaces with a conformal metric (lemma 5) was first applied by Novikov and Veselov [13] for distinguishing two-dimensional potential Schrödinger operators that are finite-gap on one level of energy from finite-gap Schrödinger operators with a magnetic field. All in all, the spectral data found in theorem 2 are a generalization of spectral data found in [13]. The rest of this paper is organized as follows. In paragraph 2 we find equations for HML-tori in ${\mathbb{C}}P^{2}$. In paragraph 3 we prove the main theorem. In paragraph 4 we give examples of spectral curves that satisfy the conditions stated in the main theorem. Note that our solutions are, in the general case, quasi-periodic, and we do not ask whether there exist periodic solutions. This problem requires extra research which we are planning to conduct in the future. This work is dedicated to Sergey Petrovich Novikov on the occasion of his 70th birthday. ## 2 Equations of Hamiltonian-minimal Lagrangian tori in ${\mathbb{C}}P^{2}$ As follows from (1), matrix $\tilde{\Phi}=(\varphi,\frac{1}{\sqrt{2}}e^{-\frac{v}{2}}\varphi_{x},\frac{1}{\sqrt{2}}e^{-\frac{v}{2}}\varphi_{y})^{\top}$ belongs to the group $U(3)$. A Lagrangian angle is a function defined from the equality $e^{i\beta(x,y)}={\rm det}\tilde{\Phi}.$ From the definition of a Lagrangian angle we get $\Phi=\left(\begin{array}[]{ccc}\varphi^{1}&\varphi^{2}&\varphi^{3}\\\ \frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{1}_{x}&\frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{2}_{x}&\frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{3}_{x}\\\ \frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{1}_{y}&\frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{2}_{y}&\frac{1}{\sqrt{2}}e^{-\frac{v}{2}-i\frac{\beta}{2}}\varphi^{3}_{y}\\\ \end{array}\right)\in{\rm SU(3)},$ Matrix $\Phi$ satisfies equations $\Phi_{x}=A\Phi,\ \Phi_{y}=B\Phi,$ $None$ where $A$ and $B$ have the form $A=\left(\begin{array}[]{ccc}0&\sqrt{2}e^{\frac{v}{2}+i\frac{\beta}{2}}&0\\\ -\sqrt{2}e^{\frac{v}{2}-i\frac{\beta}{2}}&if&-\frac{v_{y}}{2}+i(h+\frac{\beta_{y}}{2})\\\ 0&\frac{v_{y}}{2}+i(h+\frac{\beta_{y}}{2})&-if\\\ \end{array}\right)\in{\rm su(3)},$ $B=\left(\begin{array}[]{ccc}0&0&\sqrt{2}e^{\frac{v}{2}+i\frac{\beta}{2}}\\\ 0&ih&\frac{v_{x}}{2}+i(-f+\frac{\beta_{x}}{2})\\\ -\sqrt{2}e^{\frac{v}{2}-i\frac{\beta}{2}}&-\frac{v_{x}}{2}+i(-f+\frac{\beta_{x}}{2})&-ih\\\ \end{array}\right)\in{\rm su(3)},$ $f(x,y)$ and $h(x,y)$ being some real functions. The zero-curvature equation $A_{y}-B_{x}+[A,B]=0$ implies the following lemma (see [14]) ###### Lemma 1 Lagrangian surfaces are defined by a system of equations $2V_{y}+2U_{x}=(\beta_{xx}-\beta_{yy})e^{v},$ $2U_{y}-2V_{x}=(\beta_{y}v_{x}+\beta_{x}v_{y})e^{v},$ $\Delta v=4(U^{2}+V^{2})e^{-2v}-4e^{v}-2(U\beta_{x}-V\beta_{y})e^{-v},$ where $U=fe^{v},V=he^{v}.$ From (2), get equalities $\varphi_{xx}^{j}=\frac{1}{2}(-4e^{v}\varphi^{j}+\varphi_{x}^{j}(2if+v_{x}+i\beta_{x})+\varphi_{y}^{j}(2ih- v_{y}+i\beta_{y})),$ $\varphi_{yy}^{j}=\frac{1}{2}(-4e^{v}\varphi^{j}+\varphi_{x}^{j}(-2if- v_{x}+i\beta_{x})+\varphi_{y}^{j}(-2ih+v_{y}+i\beta_{y})),$ implying lemma (see [14]). ###### Lemma 2 The components $\varphi^{j}$ of the vector function $\varphi$ satisfy the Schrödinger equation $\partial_{x}^{2}\varphi^{j}+\partial_{y}^{2}\varphi^{j}-i(\beta_{x}\partial_{x}\varphi^{j}+\beta_{y}\partial_{y}\varphi^{j})+4e^{v}\varphi^{j}=0.$ Thus, there is a two-dimensional Schrödinger operator connected in a natural way to every Lagrangian surface given by the mapping $\varphi$. In of [14] (see also [15]) it is shown that in case of minimal tori, the isospectral deformations of the Schrödinger operator given by the Veselov-Novikov equation correspond to integrable deformations of minimal tori. If $\beta$ is a harmonic function, then mapping ${\cal H}\circ\varphi$ defines the HML-surface [2]. In particular, if mapping ${\cal H}\circ\varphi$ is doubly periodic, then the Lagrangian angle is the linear function $\beta(x,y)=ax+by+c,\ a,b,c\in{\mathbb{R}}.$ From Lemma 1 follows ###### Lemma 3 HML-tori in ${\mathbb{C}}P^{2}$ are described by the following equations $V_{y}+U_{x}=0,$ $2U_{y}-2V_{x}=(bv_{x}+av_{y})e^{v},$ $\Delta v=4(U^{2}+V^{2})e^{-2v}-4e^{v}-2(Ua-Vb)e^{-v}.$ Note that after replacing $V=-\frac{u_{x}}{2},U=\frac{u_{y}}{2}$ the equations in lemma 3 reduce to a system on functions $u$ and $v$ $\Delta u=e^{v}\nabla vA,$ $\Delta v=e^{-2v}(\nabla u)^{2}-4e^{v}-e^{-v}\nabla uA,$ $A=(b,a)$, that passes the Painleve test for integrability such that it could well be that there exists a Lax representation with a spectral parameter for the equations of lemma 3. The fact that there exist solutions in theta functions points into the same direction. ## 3 Main Theorem ### 3.1 Baker-Akhiezer function Let $\Gamma$ be a Riemannian surface of genus $g$. Suppose that the following set of divisors be given on $\Gamma$ $\gamma=\gamma_{1}+\dots+\gamma_{g+l},\ r+r_{1}+\dots+r_{l},$ a pair of marked points $P_{1}$, $P_{2}\in\Gamma$, and also the local parameters $k_{1}^{-1},k_{2}^{-1}$ in the neighborhood of points $P_{1}$ and $P_{2}$. The two-points Baker-Akhiezer function, corresponding to the spectral data $\\{\Gamma,P_{1},P_{2},k_{1},k_{2},\gamma,r+r_{1}+\dots+r_{l}\\},$ is called a function $\psi(x,y,P),P\in\Gamma$, with the following characteristics: 1) in the neighborhood of $P_{1}$ and $P_{2}$ function $\psi$ has essential singularities of the following form $\psi=e^{k_{1}x}\left(f_{1}(x,y)+\frac{g_{1}(x,y)}{k_{1}}+\frac{h_{1}(x,y)}{k_{1}^{2}}+\dots\right),$ $\psi=e^{k_{2}y}\left(f_{2}(x,y)+\frac{g_{2}(x,y)}{k_{2}}+\frac{h_{2}(x,y)}{k_{2}^{2}}+\dots\right).$ 2) on $\Gamma\backslash\\{P_{1},P_{2}\\}$, function $\psi$ is meromorphic with simple poles on $\gamma$. The space of such functions has dimension $l+1$, consequently, the Baker- Akhiezer function is uniquely defined if we demand the following normalization conditions to be fulfilled 3) $\psi(x,y,r)=d,\ \psi(x,y,r_{i})=0,\ i=1,\dots,l,$ where $d$ is a non-zero constant. Express the Baker-Akhiezer function explicitly by the theta function of surface $\Gamma$. On surface $\Gamma$, choose a base of cycles $a_{1},\dots,a_{g},\ b_{1},\dots,b_{g}$ with the following intersection indices $a_{i}\circ a_{j}=b_{i}\circ b_{j}=0,\ a_{i}\circ b_{j}=\delta_{ij}.$ By $\omega_{1},\dots,\omega_{g}$ we denote a base if holomorphic differentials on $\Gamma$, normalized by conditions $\int_{a_{j}}\omega_{i}=\delta_{ij}.$ Let the matrix of $b$-periods of differentials $\omega_{j}$ with components $B_{ij}=\int_{b_{i}}\omega_{j}$ be denoted by $B$. This matrix is symmetric and has a positively defined imaginary part. The Riemannian theta function is defined by an absolutely converging series $\theta(z)=\sum_{m\in{\mathbb{Z}}}e^{\pi i(Bm,m)+2\pi i(m,z)},\ z=(z_{1},\dots,z_{g})\in{\mathbb{C}}^{g}.$ The theta function has the following characteristics $\theta(z+m)=\theta(z),\ m\in{\mathbb{Z}},$ $\theta(z+Bm)={\rm exp}(-\pi i(Bm,m)-2\pi i(m,z))\theta(z),\ m\in{\mathbb{Z}}.$ Let $X$ denote a Jacoby variety of surface $\Gamma$ $X={\mathbb{C}}^{g}/\\{{\mathbb{Z}}^{g}+B{\mathbb{Z}}^{g}\\}.$ Let $A:\Gamma\rightarrow X$ be an Abelian mapping defined by formula $A(P)=\left(\int_{q_{0}}^{P}\omega_{1},\dots,\int_{q_{0}}^{P}\omega_{g}\right),\ P\in\Gamma,$ $q_{0}\in\Gamma$ being a fixed point. For points $\gamma_{1},\dots,\gamma_{g}$ in the general case and according to the Riemann theorem, the function $\theta(z+A(P)),$ where $z=K-A(\gamma_{1})-\dots-A(\gamma_{g})$, has exactly $g$ zeros on $\Gamma$ $\gamma_{1},\dots,\gamma_{g}$. Let $\Omega^{1}$ and $\Omega^{2}$ denote meromorphic differentials on $\Gamma$ with the only simple poles in points $P_{1}$ and $P_{2}$, respectively and normalized by the conditions $\int_{a_{j}}\Omega^{1}=\int_{a_{j}}\Omega^{2}=0,\ j=1,\dots,g.$ Let $U$ and $V$ denote their vectors of $b$-periods $U=\left(\int_{b_{1}}\Omega^{1},\dots,\int_{b_{g}}\Omega^{1}\right),\ V=\left(\int_{b_{1}}\Omega^{2},\dots,\int_{b_{g}}\Omega^{2}\right).$ To begin with, consider the case $l=0$. Let $\tilde{\psi}$ denote function $\tilde{\psi}(x,y,P)=\frac{\theta(A(P)+xU+yV+z)}{\theta(A(P)+z)}{\rm exp}(2\pi ix\int_{q_{0}}^{P}\Omega^{1}+2\pi iy\int_{P_{1}}^{P}\Omega^{2}).$ $None$ Then, the Baker-Akhiezer function sought for has the form $\psi(x,y,P)=f(x,y)\tilde{\psi}(x,y,P),$ where function $f$ is defined by the condition $\psi(x,y,r)=d$. For $l>0$, the Baker-Akhiezer function can be found in the following form $\psi=f\tilde{\psi}+f_{1}\tilde{\psi}_{1}+\dots+f_{l}\tilde{\psi}_{l},$ where $\tilde{\psi}_{j}$ is a function of form (3), constructed by the divisor $\gamma_{1}+\dots+\gamma_{g-1}+r_{j}.$ Find functions $f,f_{j}$ from the normalization conditions 3). ### 3.2 Lagrangian immersions Let $\varphi^{1},\varphi^{2},\varphi^{3}$ denote the following functions $\varphi^{i}=\alpha_{i}\psi(x,y,Q_{i}),$ where $Q_{1},Q_{2},Q_{3}\in\Gamma$ is an additional set of points, $\alpha_{i}$ are some constants. In the following theorem, we state sufficient conditions for the vector- function $\varphi=(\varphi^{1},\varphi^{2},\varphi^{3})$ to define a Lagrangian imbedding of the plane ${\mathbb{R}}^{2}$ in ${\mathbb{C}}P^{2}$. Suppose that surface $\Gamma$ has a holomorphic involution $\sigma$ and an antiholomorphic involution $\mu$ $\sigma:\Gamma\rightarrow\Gamma,\ \mu:\Gamma\rightarrow\Gamma,$ for which points $P_{1}$, $P_{2}$ and $r$ are fixed, and $k_{i}(\sigma(P))=-k_{i}(P),\ k_{i}(\mu(P))=\bar{k}_{i}(P).$ Let $\tau$ denote involution $\sigma\mu$. Involution $\tau$ acts on local parameters as follows $k_{i}(\tau(P))=-\bar{k}_{i}(P).$ The following theorem holds: ###### Theorem 1 Let $Q_{i}$ be fixed points of the antiholomorphic involution $\tau$. Suppose that on $\Gamma$ there exists a meromorphic 1-form $\Omega$ with the following set of divisors of zeros and poles $(\Omega)_{0}=\gamma+\tau\gamma+P_{1}+P_{2},$ $None$ $(\Omega)_{\infty}=Q_{1}+Q_{2}+Q_{3}+r+R+\tau R,$ $None$ where $R=r_{1}+\dots+r_{l}$. Then, function $\varphi^{i}$ satisfies the equations $\varphi^{1}\bar{\varphi}^{1}A_{1}+\varphi^{2}\bar{\varphi}^{2}A_{2}+\varphi^{3}\bar{\varphi}^{3}A_{3}+|d|^{2}\rm{Res}_{r}\Omega=0,$ $\varphi^{1}\bar{\varphi}^{1}_{x}A_{1}+\varphi^{2}\bar{\varphi}^{2}_{x}A_{2}+\varphi^{3}\bar{\varphi}^{3}_{x}A_{3}=0,$ $\varphi^{1}\bar{\varphi}^{1}_{y}A_{1}+\varphi^{2}\bar{\varphi}^{2}_{y}A_{2}+\varphi^{3}\bar{\varphi}^{3}_{y}A_{3}=0,$ $\varphi^{1}_{x}\bar{\varphi}^{1}_{y}A_{1}+\varphi^{2}_{x}\bar{\varphi}^{2}_{y}A_{2}+\varphi^{3}_{x}\bar{\varphi}^{3}_{y}A_{3}=0,$ $\varphi^{1}_{x}\bar{\varphi}^{1}_{x}A_{1}+\varphi^{2}_{x}\bar{\varphi}^{2}_{x}A_{2}+\varphi^{3}_{x}\bar{\varphi}^{3}_{x}A_{3}-|f_{1}|^{2}c_{1}=0,$ $None$ $\varphi^{1}_{y}\bar{\varphi}^{1}_{y}A_{1}+\varphi^{2}_{y}\bar{\varphi}^{2}_{y}A_{2}+\varphi^{3}_{y}\bar{\varphi}^{3}_{y}A_{3}-|f_{2}|^{2}c_{2}=0,$ $None$ with $A_{k}=\frac{\rm{Res}_{Q_{k}}\Omega}{|\alpha_{k}|^{2}},\ k=1,2,3$, $c_{1},c_{2}$ being the coefficients of form $\Omega$ in the neigborhood of points $P_{1}$ and $P_{2}$: $\Omega=(c_{1}w_{1}+iaw_{1}^{2}+\dots)dw_{1},\ w_{1}=1/k_{1},$ $\Omega=(c_{2}w_{2}+ibw_{2}^{2}+\dots)dw_{2},\ w_{2}=1/k_{2}.$ ###### Corollary 1 If $\rm{Res}_{Q_{i}}\Omega>0,\ \rm{Res}_{r}\Omega<0$, then for $\alpha_{i}=\sqrt{\rm{Res}_{Q_{i}}\Omega},\ d=\sqrt{\frac{-1}{\rm{Res}_{r}\Omega}},$ the following equality holds: $<\varphi,\varphi>=1,\ <\varphi,\varphi_{x}>=<\varphi,\varphi_{y}>=<\varphi_{x},\varphi_{y}>=0,$ i.e. the mapping ${\cal H}\circ r:{\mathbb{R}}^{2}\rightarrow{\mathbb{C}}P^{2}$ is Lagrangian, with the induced metric on $\Sigma$ having a diagonal form $ds^{2}=|f_{1}|^{2}|c_{1}|dx^{2}+|f_{2}|^{2}|c_{2}|dy^{2}.$ Proof of theorem 1. Consider the 1-form $\Omega_{1}=\psi(P)\overline{\psi(\tau P)}\Omega$. By virtue of the definition of involution $\tau$, function $\overline{\psi(\tau P)}$ has the following form in the neighborhoods of points $P_{1}$ and $P_{2}$ $\overline{\psi(\tau P)}=e^{-k_{1}x}\left(\bar{f}_{1}(x,y)-\frac{\bar{g}_{1}(x,y)}{k_{1}}+\frac{\bar{h}_{1}(x,y)}{k_{1}^{2}}+\dots\right),$ $\overline{\psi(\tau P)}=e^{-k_{2}y}\left(\bar{f}_{2}(x,y)-\frac{\bar{g}_{2}(x,y)}{k_{2}}+\frac{\bar{h}_{2}(x,y)}{k_{2}^{2}}+\dots\right).$ Consequently, form $\Omega_{1}$ has no essential singularities in points $P_{1}$ and $P_{2}$. The simple poles $\gamma+\tau\gamma$ of function $\psi(P)\overline{\psi(\tau P)}$ reduce with the zeros in these points of form $\Omega$. The zeros $R+\tau R$ of function $\psi(P)\overline{\psi(\tau P)}$ reduce with the simple poles of form $\Omega$. Thus, form $\Omega_{1}$ has only simple poles in points $Q_{1},Q_{2},Q_{3}$ and $r$ with equal residues corresponding to $\psi(Q_{1})\overline{\psi(Q_{1})}{\rm Res}_{Q_{1}}\Omega=\varphi^{1}\bar{\varphi}^{1}A_{1},\ \varphi^{2}\bar{\varphi}^{2}A_{2},\ \varphi^{3}\bar{\varphi}^{3}A_{3},\ |d|^{2}\rm{Res}_{r}\Omega.$ Consequently, the sum of these residues is equal to zero, and this proves the first equality. Form $\psi(P)\overline{\psi(\tau P)_{x}}\Omega$ also has no real singularities in points $P_{1}$ and $P_{2}$. This form has only simple poles in points $Q_{1},Q_{2}$ and $Q_{3}$ with residues equal to $\varphi^{1}\bar{\varphi}^{1}_{x}A_{1},\ \varphi^{2}\bar{\varphi}^{2}_{x}A_{2},\ \varphi^{3}\bar{\varphi}^{3}_{x}A_{3}.$ The second equality is proven. The proof of the following two equalities is analogical. For this, consider the forms $\psi(P)\overline{\psi(\tau P)_{y}}\Omega,\ \psi(P)_{x}\overline{\psi(\tau P)_{y}}\Omega,$ that also have only simple poles in points $Q_{1},Q_{2}$ and $Q_{3}$. In order to prove the last two equalities, consider forms $\psi(P)_{x}\overline{\psi(\tau P)_{x}}\Omega,\ \psi(P)_{y}\overline{\psi(\tau P)_{y}}\Omega.$ These forms have simple poles in points $Q_{1},Q_{2}$, $Q_{3},P_{1}$ and $Q_{1},Q_{2}$, $Q_{3},P_{2}$ with the residues $\varphi^{1}_{x}\bar{\varphi}^{1}_{x}A_{1},\ \varphi^{2}_{x}\bar{\varphi}^{2}_{x}A_{2},\ \varphi^{3}_{x}\bar{\varphi}^{3}_{x}A_{3},\ -|f_{1}|^{2}c_{1}$ and $\varphi^{1}_{y}\bar{\varphi}^{1}_{y}A_{1},\ \varphi^{2}_{y}\bar{\varphi}^{2}_{y}A_{2},\ \varphi^{3}_{y}\bar{\varphi}^{3}_{y}A_{3},\ -|f_{2}|^{2}c_{2}.$ Theorem 1 is proven. Remark. Actually, in corollary 1 it is sufficient to demand that only the following inequalities be fulfilled $\frac{{\rm Res}_{Q_{2}}\Omega}{{\rm Res}_{Q_{1}}\Omega}>0,\ \frac{{\rm Res}_{Q_{3}}\Omega}{{\rm Res}_{Q_{1}}\Omega}>0,\ \frac{{\rm Res}_{r}\Omega}{{\rm Res}_{Q_{1}}\Omega}<0.$ The entire construction easily applies to this case. ### 3.3 Hamiltonian-minimal Lagrangian immersions In this section, we find spectral data such that the mapping constructed in the preceding section is conformal and Hamiltonian-minimal. Consider function $F(x,y,P)=\partial_{x}^{2}\psi+\partial_{y}^{2}\psi+A(x,y)\partial_{x}\psi+B(x,y)\partial_{y}\psi+C(x,y)\psi.$ Chose functions $A(x,y),B(x,y)$ and $C(x,y)$ such that $F(x,y,Q_{i})=0,\ i=1,2,3.$ If mapping ${\cal H}\circ\varphi$ is conformal, then by lemma 2, the Lagrangian angle is expressed by the functions $A(x,y)$ and $B(x,y)$. Thus, we need to find spectral data such that the metric on surface $\Sigma$ has a conformal form, and the coefficients $A,B$ are constants. ###### Lemma 4 The following equality holds: $A(x,y)=\frac{1}{c_{1}|f_{1}|^{2}}(-c_{1}\bar{f}_{1}(g_{1}+2{f_{1}}_{x})+f_{1}(-ia\bar{f}_{1}+c_{1}(\bar{g}_{1}+\bar{f_{1}}_{x}))+c_{2}f_{2}\bar{f_{2}}_{x}),$ $B(x,y)=\frac{1}{c_{2}|f_{2}|^{2}}(-c_{2}\bar{f}_{2}(g_{2}+2{f_{2}}_{y})+f_{2}(-ib\bar{f}_{2}+c_{2}(\bar{g}_{2}+\bar{f_{2}}_{y}))+c_{1}f_{1}\bar{f_{1}}_{y}),$ $C(x,y)=-\frac{1}{|d|^{2}{\rm Res}_{r}\Omega}(c_{1}|f_{1}|^{2}+c_{2}|f_{2}|^{2}).$ Proof of lemma 4. Consider three forms $\omega_{1}=F(P)\overline{\psi(\tau P)}\Omega,$ $\omega_{2}=F(P)\overline{\psi(\tau P)_{x}}\Omega,$ $\omega_{3}=F(P)\overline{\psi(\tau P)_{y}}\Omega.$ By the construction of involution $\tau$, forms $\omega_{1}$, $\omega_{2}$ and $\omega_{3}$ have no essential singularities in points $P_{1}$ and $P_{2}$. Form $\omega_{1}$ has only simple poles in points $P_{1},P_{2}$ and $r$ with the sum of residues $c_{1}|f_{1}|^{2}+c_{2}|f_{2}|^{2}+C(x,y)|d|^{2}{\rm Res}_{r}\Omega=0.$ Form $\omega_{2}$ has only a pole of second order in point $P_{1}$ and a simple pole in point $P_{2}$, whose sum of residues equals $-c_{1}\bar{f}_{1}(g_{1}+2{f_{1}}_{x})+f_{1}((-ia- c_{1}A(x,y))\bar{f_{1}}+c_{1}(\bar{g}_{1}+\bar{f_{1}}_{x}))+c_{2}f_{2}\bar{f_{2}}_{x}=0.$ Analogically, form $\omega_{3}$ has only a pole of second order in point $P_{2}$ and a simple pole in point $P_{1}$ with a sum of residues $-c_{2}\bar{f}_{2}(g_{2}+2{f_{2}}_{y})+f_{2}((-ib- c_{2}B(x,y)\bar{f_{2}}+c_{2}(\bar{g}_{2}+\bar{f_{2}}_{y}))+c_{1}f_{1}\bar{f_{1}}_{y}=0.$ This yields formulas for $A(x,y),B(x,y),C(x,y)$, indicated in lemma 4. Thus, lemma 4 is proven. The following lemma holds. ###### Lemma 5 Suppose that on surface $\Gamma$ there exists a meromorphic form $\omega$ with the following divisors of zeros and poles $(\omega)_{0}=\gamma+\sigma\gamma,\ (\omega)_{\infty}=P_{1}+P_{2}+R+\sigma R,$ $None$ where ${\rm Res}_{P_{1}}\omega+{\rm Res}_{P_{2}}\omega=0$. Then, $f_{1}^{2}=f_{2}^{2},$ i.e. the induced metric on surface $\Sigma$ has the form $ds^{2}=|f_{1}|^{2}(|c_{1}|dx^{2}+|c_{2}|dy^{2}).$ Proof of lemma 5. Consider the form $\omega_{1}=\psi(P)\psi(\sigma P)\omega.$ This form has only simple poles in points $P_{1}$ and $P_{2}$, consequently, the sum of residues in these points equals zero $f_{1}^{2}{\rm Res}_{P_{1}}\omega+f_{2}^{2}{\rm Res}_{P_{2}}\omega=0.$ Lemma 5 is proven. Remark. Condition ${\rm Res}_{P_{1}}\omega+{\rm Res}_{P_{2}}\omega=0$ is fulfilled, for instance, if $l=0$ or when $\sigma\omega=-\omega$. ###### Lemma 6 Suppose that $\mu(\gamma)=\gamma,\ \mu(R)=R,\ d\in{\mathbb{R}}.$ are real. Then, $\psi(x,y,\mu(P))=\overline{\psi(x,y,P)}.$ For the proof of this standard lemma it is sufficient to note that function $\overline{\psi(x,y,\mu(P))}$ fulfills conditions 1)–3) in the definition of the Beker-Akhiezer function. Consequently the functions $\overline{\psi(x,y,\mu(P))}$ and $\psi(x,y,P)$ are coincide. Below we suppose the conditions of lemma 6 to be fulfilled. In particular, this means that $\tau(\gamma)=\sigma(\gamma),\ \tau(R)=\sigma(R),$ $None$ and also, that functions $f_{i},g_{i}$, participating in decomposition $\psi$ in the neighborhood of points $P_{1}$ and $P_{2}$ are real. Remark. As follows from the existence of the meromorphic forms $\omega$ and $\Omega$ on surface $\Gamma$, there is a meromorphic function $\lambda=\frac{\omega}{\Omega}$ with the following divisors of zeros and poles $(\lambda)_{0}=Q_{1}+Q_{2}+Q_{3}+r,\ (\lambda)_{\infty}=2P_{1}+2P_{2}$ on $\Gamma$. To begin with, suppose that in the decomposition of form $\Omega$, in neighborhoods of points $P_{1}$ the coefficient $c_{1}$ is equal to 1\. By virtue of forms $\Omega$ and $\overline{\Omega(\tau P)}$ possessing the same set of zeros and poles, forms $\Omega$ and $\overline{\Omega(\tau P)}$ are proportional. From $\tau w_{1}=-\overline{w}_{1}$ and from the fact that in the neighborhood of point $P_{1}$ decompositions $\Omega(P)=(w_{1}+iaw_{1}^{2}+\dots)dw_{1},$ $\overline{\Omega(\tau P)}=(w_{1}+i\bar{a}w_{1}^{2}+\dots)dw_{1},$ hold, we get $\Omega(P)=\overline{\Omega(\tau P)}.$ In the neighborhood of point $P_{2}$ we have $\Omega(P)=(c_{2}w_{2}+ibw_{2}^{2}\dots)dw_{2}=\overline{\Omega(\tau P)}=(\bar{c}_{2}w_{2}+i\bar{b}w_{2}^{2}\dots)dw_{2},$ consequently, $c_{2},a,b$ are real constants. Suppose that the inequalities of residues in corollary 1 be fulfilled. From (6) and (7) follows that $c_{1}$ and $c_{2}$ have the same sign, and by virtue of assumption $c_{1}=1$, $c_{2}$ is a positive constant. Thus, replacing $w_{2}\rightarrow\frac{w_{2}}{\sqrt{c_{2}}}$, consider that $c_{1}=c_{2}=1.$ Thus, the metric on surface $\Sigma$ has a conformal form $ds^{2}=f_{1}^{2}(dx^{2}+dy^{2}).$ The main theorem holds. ###### Theorem 2 Suppose that the conditions of theorem 1 and lemmas 5 and 6 are fulfilled. Then, the Lagrangian angle looks as follows $\beta=ax+by+c,$ where $c$ is some real constant, i.e. surface $\Sigma$ is Hamiltonian-minimal. If also $\sigma Q_{i}=Q_{i}$, then surface $\Sigma$ is minimal. Proof of theorem 2. Since $f_{1}=\bar{f_{1}},\ f_{2}=\bar{f_{2}},\ f_{1}^{2}=f_{2}^{2},\ g_{1}=\bar{g}_{1},\ g_{2}=\bar{g}_{2},\ c_{1}=c_{2}=1$ by lemma 4, coefficients $A$ and $B$ have the form $A=-ia,\ B=-ib,$ consequently, the Lagrangian angle has the form indicated in theorem 2, consequently, surface $\Sigma$ is Hamiltonian-minimal. What’s left is to prove the last point of the theorem. Consider two forms $\Omega(P)$ and $\Omega(\sigma P)$. According to our assumptions, these forms have the same set of zeros and poles. From their decompositions in the neighborhood of points $P_{1}$ and $P_{2}$ $\Omega(P)=(w_{1}+iaw_{1}^{2}+\dots)dw_{1},$ $\Omega(\sigma P)=(w_{1}-iaw_{1}^{2}+\dots)dw_{1},$ $\Omega(P)=(w_{2}+ibw_{2}^{2}+\dots)dw_{2},$ $\Omega(\sigma P)=(w_{2}-ibw_{2}^{2}+\dots)dw_{2}$ follows that forms $\Omega(P)$ and $\Omega(\sigma P)$ coincide and $a=b=0$, i.e. surface $\Sigma$ is minimal. Thus, theorem 2 is proven. ## 4 Examples ### 4.1 Spectral curve of genus 0: Clifford tori Suppose that $\Gamma={\mathbb{C}}P^{1},P_{1}=\infty,P_{2}=0$. The Riemannian surface $\Gamma$ allows the holomorphic involution $\sigma:\Gamma\rightarrow\Gamma,\ \sigma(z)=-z$ and the antiholomorphic involutions $\mu:\Gamma\rightarrow\Gamma,\ \mu(z)=\bar{z},$ $\tau:\Gamma\rightarrow\Gamma,\ \tau(z)=-\bar{z}.$ Let $l=0$, then the Baker-Akhiezer function has the form $\psi(x,y,z)=e^{xz+\frac{y}{z}}f(x,y).$ Let points $Q_{1},Q_{2},Q_{3}$ have the coordinates $iq_{1},iq_{2},iq_{3}\in i{\mathbb{R}}$, and point $r$ — a coordinate $i\tilde{r}\in i{\mathbb{R}}$. In this case $\tau(Q_{j})=Q_{j},\ \tau(r)=r.$ From the normalization condition follows $\psi(x,y,r)=d.$ This leads us to $\psi(x,y,z)=de^{xz+\frac{y}{z}}e^{-i\tilde{r}x+\frac{iy}{\tilde{r}}}.$ The form $\Omega$ looks as follows: $\Omega=\frac{szdz}{(z-iq_{1})(z-iq_{2})(z-iq_{3})(z-i\tilde{r})}.$ The decomposition coefficients $c_{1}$, $c_{2}$ of form $\Omega$ in the neighborhood of points $P_{1}$, $P_{2}$ equal $c_{1}=-s,\ c_{2}=\frac{s}{q_{1}q_{2}q_{3}\tilde{r}}.$ Consequently, $s=-1,\ \tilde{r}=-\frac{1}{q_{1}q_{2}q_{3}}.$ The vector-function $\varphi$ has the following components $\varphi^{1}=\alpha_{1}de^{\frac{-i(q_{1}-\tilde{r})(q_{1}\tilde{r}x+y)}{q_{1}\tilde{r}}},\ \varphi^{2}=\alpha_{2}de^{\frac{-i(q_{2}-\tilde{r})(q_{2}\tilde{r}x+y)}{q_{2}\tilde{r}}},\ \varphi^{3}=\alpha_{3}de^{\frac{-i(q_{3}-\tilde{r})(q_{3}\tilde{r}x+y)}{q_{3}\tilde{r}}},$ where $\alpha_{1}=\sqrt{\frac{-q_{1}}{(q_{2}-q_{1})(q_{1}-q_{3})(q_{1}-\tilde{r})}},\ \alpha_{2}=\sqrt{\frac{-q_{2}}{(q_{1}-q_{2})(q_{2}-q_{3})(q_{2}-\tilde{r})}},\ $ $\alpha_{3}=\sqrt{\frac{-q_{3}}{(q_{1}-q_{3})(q_{3}-q_{2})(q_{3}-\tilde{r})}},\ d=\sqrt{\frac{(\tilde{r}-q_{1})(\tilde{r}-q_{2})(\tilde{r}-q_{3})}{-\tilde{r}}}.$ For example, for $q_{1}=2,q_{2}=-2,q_{3}=1/2$ we have $\varphi^{1}=\frac{3}{\sqrt{14}}e^{\frac{1}{2}i(2x+y)},\ \varphi^{2}=\frac{1}{\sqrt{6}}e^{\frac{3}{2}i(-2x+y)},\ \varphi^{3}=\frac{2}{\sqrt{21}}e^{-3i(\frac{x}{4}+y)}.$ The induced metric looks as follows: $ds^{2}=\frac{7}{4}(dx^{2}+dy^{2}),$ with functions $\varphi^{j}$ satisfying equation $\Delta\varphi^{j}+i\frac{11}{4}\partial_{x}\varphi^{j}+i\partial_{y}\varphi^{j}+\frac{9}{2}\varphi^{j}=0,$ the Lagrangian angle has the form $\beta=-\left(\frac{11}{4}x+y-\pi\right).$ ### 4.2 Spectral curves of genus $g>0$: HML-tori Let $\Gamma_{0}$ be a hyperelliptic surface of genus $g$ given by equation $y^{2}=P(x),$ where $P(x)$ is a polynom of degree $2g+2$ with real coefficients without multiple roots. Let $f$ denote a meromorphic function on surface $\Gamma_{0}$ $f=\frac{x-\beta}{x-\alpha},$ where $\alpha,\beta$ are some real numbers such that $P(\alpha)\neq 0,P(\beta)\neq 0$. Function $f$ has two simple zeros in points $\tilde{P_{1}}=(\beta,\sqrt{P(\beta)}),\ \tilde{P_{2}}=(\beta,-\sqrt{P(\beta)})$ and two simple poles in points $\tilde{r}=(\alpha,\sqrt{P(\alpha)}),\ \tilde{Q}_{1}=(\alpha,-\sqrt{P(\alpha)}).$ Let $\Gamma$ denote the Riemannian surface of function $\sqrt{f}$. The affine part of surface $\Gamma$ is given in ${\mathbb{C}}^{3}$ with coordinates $x,y,z$ by the system of equations $y^{2}=P(x),$ $z^{2}=\frac{x-\beta}{x-\alpha}.$ Surface $\Gamma$ allows a holomorphic involution $\sigma:\Gamma\rightarrow\Gamma,\ \sigma(x,y,z)=(x,y,-z)$ with four fixed points $P_{1},P_{2},Q_{1},r$ — the inverse images of points $\tilde{P_{1}},\tilde{P_{2}},\tilde{Q}_{1},\tilde{r}$ for the projection $\Gamma\rightarrow\Gamma_{0}$. Surface $\Gamma$ also allows antiholomorphic involutions $\mu:\Gamma\rightarrow\Gamma,\ \mu(x,y,z)=(\bar{x},\bar{y},\bar{z}),$ $\tau:\Gamma\rightarrow\Gamma,\ \tau(x,y,z)=(\bar{x},\bar{y},-\bar{z}).$ Choose $\alpha$ and $\beta$ such that $P(\alpha)>0,P(\beta)>0$. Then, points $P_{1},P_{2},Q_{1},r$ are fixed with respect to involution $\tau$. Consider the meromorphic function $\lambda$ on surface $\Gamma$ $\lambda=z^{2}-cz,$ where $c$ is some imaginary constant. This function has two multiple poles in points $P_{1},P_{2}$, two simple zeros in points $Q_{1},r$ and two zeros in points $Q_{2}=\left(\frac{\beta-c^{2}\alpha}{1-c^{2}},\sqrt{P\left(\frac{\beta-c^{2}\alpha}{1-c^{2}}\right)},c\right),$ $Q_{3}=\left(\frac{\beta-c^{2}\alpha}{1-c^{2}},-\sqrt{P\left(\frac{\beta-c^{2}\alpha}{1-c^{2}}\right)},c\right).$ Choose $c$ to be purely imaginary such that $P\left(\frac{\beta-c^{2}\alpha}{1-c^{2}}\right)>0.$ Then, points $Q_{2}$ and $Q_{3}$ are invariant with respect to $\tau$, but not invariant with respect to $\sigma$. Thus, if form $\omega$ has a divisor of zeros and poles of form (8) with property $\sigma\omega=-\omega$ with real divisors $\gamma$ and $R$ (see (9)), then form $\Omega=\frac{\omega}{\lambda}$ has the required set of zeros and poles (4) and (5). If needed, divide $\Omega$ by a suitable constant such that the decomposition $\Omega$ in the neighborhood of $P_{1}$ has the form $\Omega=(w_{1}+\dots)dw_{1}.$ Forms $\overline{\Omega(\tau P)}$ and $\Omega(P)$ have the same set of zeros and poles. Consequently, by virtue of decomposition $\Omega$ in the vicinity of $P_{1}$ and by virtue of $\tau(w_{1})=-\bar{w}_{1}$ get $\overline{\Omega(\tau P)}=\Omega(P)$. Chose a local parameter $v_{i}$ in the neighborhood of point $Q_{i}$ such that $\tau(v_{i})=-\bar{v}_{i}$, consequently, the decomposition coefficient $q_{i}$ $\Omega=\left(\frac{q_{i}}{v_{i}}+\dots\right)dv_{i},$ by virtue of equality $\overline{\Omega(\tau P)}=\Omega(P)$, is real, i.e. ${\rm Res}_{Q_{i}}\Omega\in{\mathbb{R}},\ i=1,2,3.$ Analogically, ${\rm Res}_{r}\Omega\in{\mathbb{R}}.$ Now chose the construction’s parameters such that ${\rm Res}_{Q_{i}}\Omega>0,\ {\rm Res}_{r}\Omega<0$ and obtain spectral data satisfying all requirements of theorem 2. ### 4.3 Spectral curves of genus $g>0$: minimal Lagrangian tori As in the preceding example, let $\Gamma_{0}$ denote a hyperelliptic curve of genus $g$ given by equation $y^{2}=P(x).$ Let $\Gamma$ denote a Riemannian surface of function $\sqrt{g}$, where $g=\frac{(x-\beta_{1})(x-\beta_{2})}{(x-\alpha_{1})(x-\alpha_{2})},\ \beta_{1},\beta_{2}\in{\mathbb{R}},$ where $\alpha_{1},\alpha_{2}$ are real roots of the polynomial $P(x)$. The affine part of surface $\Gamma$ is given in ${\mathbb{C}}^{3}$ with coordinates $x,y,z$ by the system of equations $y^{2}=P(x),$ $z^{2}=\frac{(x-\beta_{1})(x-\beta_{2})}{(x-\alpha_{1})(x-\alpha_{2})}.$ Let $P_{1},P_{2}\in\Gamma$ denote inverse images of points $(\alpha_{1},0),(\alpha_{2},0)\in\Gamma_{0}$ for a natural projection $\Gamma\rightarrow\Gamma_{0}$, let $Q_{1},Q_{2},Q_{3},r\in\Gamma$ denote points $Q_{1}=(\beta_{1},\sqrt{P(\beta_{1})},0),\ Q_{2}=(\beta_{1},-\sqrt{P(\beta_{1})},0),$ $Q_{3}=(\beta_{2},\sqrt{P(\beta_{2})},0),\ r=(\beta_{2},-\sqrt{P(\beta_{2})},0).$ All these points are invariant under a holomorphic involution $\sigma:\Gamma\rightarrow\Gamma,\ \sigma(x,y,z)=(x,y,-z).$ Surface $\Gamma$ allows antiholomorphic involutions $\mu:\Gamma\rightarrow\Gamma,\ \mu(x,y,z)=(\bar{x},\bar{y},\bar{z}),$ $\tau:\Gamma\rightarrow\Gamma,\ \tau(x,y,z)=(\bar{x},\bar{y},-\bar{z}).$ Chose $\beta_{1}$ and $\beta_{2}$ such that $P(\beta_{1})>0,P(\beta_{2})>0$. Then, points $P_{1},P_{2},Q_{1},$ $Q_{2}$, $Q_{3},r$ are fixed for involution $\tau$. Function $\lambda=z$ has the following set of zeros and poles $(\lambda)_{0}=Q_{1}+Q_{2}+Q_{3}+r,\ (\lambda)_{\infty}=2P_{1}+2P_{2}.$ Thus, if form $\omega$ has simple poles in points $P_{1}$ and $P_{2}$ and $\sigma\omega=-\omega$, then form $\Omega=\frac{\omega}{\lambda}$ has a set of zeros and poles satisfying the condition of theorem 2. Since $\sigma(Q_{i})=Q_{i}$, according to theorem 2 surface $\Sigma$ is minimal. ## References * [1] Y. Oh. Volume minimization of Lagrangian submanifolds under Hamiltonian deformations // Math. Z. 1993. V. 212. P. 175-192. * [2] A.E. Mironov. On new examples of Hamiltonian-minimal and minimal Lagrangian submanifolds in ${\mathbb{C}}^{n}$ and ${\mathbb{C}}P^{n}$ // Sb. Math. 2004. V. 195. N.1. P.85-96. * [3] F. Helein, P. Romon. Hamiltonian stationary Lagrangian surfaces in ${\mathbb{C}}^{2}$ // Comm. Anal. Geom. 2002. V. 10. P. 79–126. * [4] I. McIntosh, P. Romon. The spectral data for Hamiltonian stationary Lagrangian tori in ${\mathbb{R}}^{4}$// arxiv:0707.1767 * [5] A.E. Mironov. On Hamiltonian-minimal Lagrangian tori in ${\mathbb{C}}P^{2}$ // Siberian Math. Journal. 2003. V. 44. N. 6. P. 1039-1042. * [6] H. Ma, M. Schmies. Examples of Hamiltonian Stationary Lagrangian tori in ${\mathbb{C}}P^{2}$ // Geometriae Dedicata. 2006. V. 118. P. 173–183. * [7] F. Helein, P. Romon. Hamiltonian stationary tori in the complex projective plane // Proceedings of the London Mathematical Society. 2005. V. 90. P. 472–496 * [8] H. Ma. Hamiltonian stationary Lagrangian surfaces $\mathbb{CP}^{2}$ // Ann. Global. Anal Geom. 2005. V. 27. P. 1–16. * [9] A.V. Mikhailov. The reduction problem and the scatering method //Physica 3D. 1981. N. 1. P. 73–117. * [10] R.A. Sharipov. Minimal tori in five-dimensional sphere in ${\mathbb{C}}^{3}$ // Theor. and Math. Phys. 1991. V. 87. N. 1. P. 48–56. * [11] E. Carberry, I. McIntosh. Minimal Lagrangian 2-tori in ${\mathbb{C}}P^{2}$ come in real families of every dimention // J. London Math. Soc. 2004. V. 69. P. 531–544. * [12] I.M. Krichever. Algebraic-geometric $n$-orthogonal curvilinear coordinate systems and the solution of associativity equations // Funct. Anal. Appl. 1997. V. 31. N. 1, P. 25–39. * [13] A.P. Veselov, S.P. Novikov. Finite-zone two-dimensional Schrödinger operators. Potential operators. // Sov. Math. Dokl. 1984. V. 279. N.1. P. 784–788. * [14] A.E. Mironov. The Novikov-Veselov hierarchy of equations and integrable deformations of minimal Lagrangian tori in $\mathbb{C}P^{2}$ // Siberian Electronic Mathematical Reports. 2004. V. 1. P. 38-46 * [15] A.E. Mironov. Relationship between symmetries of the Tzitzeica equation and the Novikov–Veselov hierarchy. Math. Notes. 2007. V. 82\. N. 4. P. 569-572. Sobolev Institute of Mathematics, Novosibirsk State University, E-mail address: mironov@math.nsc.ru

arxiv-papers

2024-09-04T02:48:54.683829

0804.0117

# On a Commutative Ring of Two Variable Differential Operators with Matrix Coefficients A.E. Mironov ## 1 Introduction In this work, we construct commutative rings of two variable matrix differential operators that are isomorphic to a ring of meromorphic functions on a rational manifold obtained from the ${\mathbb{C}}P^{1}\times{\mathbb{C}}P^{1}$ by identification of two lines with the pole on a certain rational curve. The commutation condition for differential operators is equivalent to a system of non-linear equations in the operators’ coefficients. For selected operator coefficients, the commutation equations reduce to known soliton equations such as the Korteweg-de Vries equation, the Kadomtsev-Petviashvili equation, the $\sin$-Gordon equation and others. The problem of classifying commuting ordinary differential operators was solved in [1]. If two differential operators $L_{1}=\partial_{x}^{n}+u_{n-1}\partial_{x}^{n-1}+\dots+u_{0}(x),\ L_{2}=\partial_{x}^{m}+v_{m-1}\partial_{x}^{m-1}+\dots+v_{0}(x)$ commute, then by the Burchnall-Chaundy lemma [2] there exists a non-zero polynomial $Q(\lambda,\mu)$ of two commuting variables $\lambda$ and $\mu$ such that $Q(L_{1},L_{2})=0.$ The smooth compactification of the curve given in ${\mathbb{C}}^{2}$ by the equation $Q(\lambda,\mu)=0$ is called spectral curve. If $\psi$ — is a common eigen-function, and $\lambda$ and $\mu$ — the corresponding eigen-values $L_{1}\psi=\lambda\psi,\ L_{2}\psi=\mu\psi,$ then point $P=(\lambda,\mu)$ lies on the spectral curve . In this way, the spectral curve parametrizes the commom eigen-functions of the operators $L_{1}$ and $L_{2}$. The function $\psi(x,P)$ (the Baker-Akhiezer function) has a unique essential singularity on the spectral curve, and outside of this point it is meromorphic. The Baker-Akhiezer function one can find uniquely from its spectral data, i.e. from the set of poles, the essential singular point and certain relations of the residues in the poles. In this way, the commutative rings of differential operators correspond to the sets of spectral data. There is no such classification for operators depending on more than one variable. In the multidimensional case we have the Burchnall-Chaundy lemma analogue proven by Krichever [3]. If the operators $L_{1},\dots,L_{n+1}$ in $n$ variables whose lead symbols have constant coefficients, do commutate pairwise, then there exists a polinomial $Q(\lambda_{1},\dots,\lambda_{n+1})$ in the commutative variables $\lambda_{1},\dots,\lambda_{n+1}$, such that $Q(L_{1},\dots,L_{n+1})=0.$ As in the one-dimensional case, the manifold given in ${\mathbb{C}}^{n+1}$ by equation $Q(\lambda_{1},\dots,\lambda_{n+1})=0$ parametrizes the common eigen- functions of operators $L_{1}\psi=\lambda_{1}\psi,\dots,L_{n+1}\psi=\lambda_{n+1}\psi.$ The question of how to correctly compactify this spectral manifold and to correctly assign the spectral data for unique recovery of the Baker-Akhiezer function, and, consequently, the commutative ring of differential operators, remains entirely open. In [4], [5], the authors find a formal generalization of the Krichever construction in the two-dimensional case. There are several approaches to the construction of multidimensional commutative operators ([6] offers insight into some of them). Sometimes, for operators of a certain kind the commutation equation can be integrated (see e.g. [7]). In [8], [9] (see also the references in these papers ) the authors propose a method to recover the Baker-Akhiezer function by the affine part of certain rational manifolds. The enumerated examples are very interesting since the commutative rings contain Schrödinger operators. Nakayashiki [10] (see also [11]) found spectral data corresponding to commuting differential operators in $g$ variables with matrix coefficients (the matrix size being $g!\times g!$) , with principle polarized Abelian manifolds with a non-singular theta-divisor serving as spectral manifolds. For $g=2$ these operators have been studied in [12], [13]. Note the main distinction between the one-dimensional case from the multidimentional one: Even if we could construct operators $L_{1}$ and $L_{2}$, with a sufficiently large family of common eigen-functions, e.g. a family of parametrized points of a certain algebraic manifold, then, in contrast to the one-dimensional case this does not imply that operators $L_{1}$ and $L_{2}$ commute, because the kernel of the commutator $[L_{1},L_{2}]$ is in the general case infinite dimensional. This difficulty is overcome in [10] as follws. Nakayashiki constructs a module $M$ (the Baker-Akhiezer module) above a ring of differential operators ${\cal D}_{g}={\cal O}[\partial_{x_{1}},\dots,\partial_{x_{g}}],$ where ${\cal O}$ is a ring of analytic functions in variables $x_{1},\dots,x_{g}$ in the neighborhood of $0\in{\mathbb{C}}^{g}$, which consists of functions that depend on $x=(x_{1},\dots,x_{g})$ and $P\in X$, and which possesses the following two remarkable characteristics: 1\. $M$ is a free module above ${\cal D}_{g}$ of rank $g!$ 2\. For any meromorphic function $\lambda$ on $X$ with pole on a theta-divisor and for any function $\varphi\in M$, function $\lambda\varphi$ also lies in $M$. From this construction follows that there exists an imbedding of the ring $A_{\theta}$ of meromorphic functions on $X$ with pole on the theta-divisor into a ring of differential operators on $g$ variables with matrix coefficients. Take base $\psi_{1}(x,P),\dots,\psi_{g!}(x,P)$ in the ${\cal D}_{g}$-module $M$. Denote the vector function $\psi(x,P)$ $(\psi_{1}(x,P),\dots,\psi_{g!}(x,P))^{\top}$ by $\psi(x,P)$. Then for any meromorphic function $\lambda(P)\in A_{\theta}$ there exists a uniqe differential operator $D(\lambda)$ such that $D(\lambda)\psi(x,P)=\lambda(P)\psi(x,P).$ Since $\lambda$ and $\mu$ do not depend on $x$, equality $D(\lambda)D(\mu)\psi=D(\mu)D(\lambda)\psi=D(\mu\lambda)\psi=\mu\lambda\psi,$ holds for any two functions $\lambda,\mu\in A_{\theta}$. $D(\lambda)D(\mu)\psi=D(\mu)D(\lambda)\psi=D(\mu\lambda)\psi=\mu\lambda\psi,$ consequently, by virtue of the differential operators’ uniqueness we have $D(\lambda)D(\mu)=D(\mu)D(\lambda)=D(\mu\lambda),$ i.e., operators $D(\lambda)$ and $D(\mu)$ commute. Rothstein [14] (see also [15]) shows that a ring of meromorphic functions on a Fano surface with a certain fixed pole can be imbedded into a ring of matrix differential operators. In this work, we construct an analogue of Nakayashiki’s construction for the following spectral manifold. Let $\Gamma$ denote a manifold derived from the ${\mathbb{C}}P^{1}\times{\mathbb{C}}P^{1}$ identification of two straight lines $p_{1}\times{\mathbb{C}}P^{1}\sim{\mathbb{C}}P^{1}\times p_{2}.$ Namely, upon assigning coordinates $(z_{1}:w_{1},z_{2}:w_{2})$ on ${\mathbb{C}}P^{1}\times{\mathbb{C}}P^{1}$, we identify the following points: $(a_{1}:b_{1},t_{1}:t_{2})\sim(t_{1}:t_{2},a_{2}:b_{2}),$ with $(a_{i}:b_{i})$ being coordinates of $p_{i}$, $p_{1}\neq p_{2}$, $(t_{1}:t_{2})\in{\mathbb{C}}P^{1}$. Let $f(P)$ denote the following form on ${\mathbb{C}}P^{1}\times{\mathbb{C}}P^{1}$ $f(P)=\alpha z_{1}z_{2}+\beta z_{1}w_{2}+\gamma z_{2}w_{1}+\delta w_{1}w_{2},\ P=(z_{1}:w_{1},z_{2}:w_{2})\in\Gamma,$ $\alpha,\beta,\gamma,\delta\in{\mathbb{C}},$ and suppose that identity $f(a_{1}:b_{1},t_{1}:t_{2})-Af(t_{1}:t_{2},a_{2}:b_{2})=0,\ A\in{\mathbb{C}}^{*}.$ $None$ holds for all $(t_{1}:t_{2})$ Equality (1) restricts the choice of constants $a_{i},b_{i},\alpha,\beta,\gamma,\delta$ and means that form $f(P)$ is a section of a line bundle on $\Gamma$. Let $A_{f}$ denote a ring of meromorphic functions on $\Gamma$ with a pole on a curve given by equation $f(P)=0$. The main result of this work is ###### Theorem 1 There exists an embedding $D:A_{f}\rightarrow Mat(2,{\cal D})$ of a ring of meromorphic functions $A_{f}$ on $\Gamma$ into the ring $2\times 2$-matrix differential operators in the variables $x$ and $y$. In section 2, we introduce a Baker-Akhiezer module corresponding to the manifold $\Gamma$. In section 3, we present theorem 2 and show that this module is free of rank 2. Theorem 1 follows directly from theorem 2. In section 3 we also give explicit examples of commuting operators. It is remarkable that by rationality of $\Gamma$, the operator coefficients are elementary functions. Acknowledgement The author would like to thank ESF Research Network MISGAM for partial support of the ISLAND-3 conference, where the main results of this work were obtained, and also appreciate to Oleg Chalyh and Alexander Veselov for useful discussions of this work. The author is thankful to Atsushi Nakayashiki for the invitations in Kyushu University, useful discussions of our results, promotional interest, and important remark proposed to this work (see paragraph 2.1). The author is also grateful to Grant-in-Aid for Scientific Research (B) 17340048 for financial support of the visits in Kyushu University. ## 2 The Baker-Akhiezer module In this section we point to some obvious and necessary conditions that should be fulfilled by the spectral data of commutative rings of multidimensional differential operators. Also, we construct a Baker–Akhiezer module on manifold $\Gamma$. ### 2.1 General construction As noted before, in many significant examples of commutative rings of differential operators the common eigen-functions are parametrized by points of a spectral manifold $X$, the ring of operators itself being isomorphic to a ring $A_{Y}$ of meromorphic functions on $X$ with pole on a hypersurface $Y\subset X$. Let the operators have matrix size $k\times k$ coefficients, with $\psi(x,P)=(\psi_{1}(x,P),\dots,\psi_{k}(x,P))^{\top}$ being a common vector-eigen-function (for sake of simplicity we consider the case of rank 1 operators, i.e. where to every point of the spectral manifold there corresponds a unique common eigen-function) and let $x=(x_{1},\dots,x_{n})$, $P\in X$ be a spectral parameter. For all examples, the unique differential operator $D(\lambda)$ can be recovered by means of the meromorphic function $\lambda\in A_{Y}$ such that $D(\lambda)\psi=\lambda\psi.$ $None$ Let’s consider module $M$ over ${\cal D}_{n}$, generated by functions $\psi_{1},\dots,\psi_{k}$ $M=\\{d_{1}\psi_{1}+\dots+d_{k}\psi_{k},\ d_{i}\in{\cal D}_{n}\\}.$ As follows from the uniqueness of operator $D(\lambda)$, which satisfies equality (2), the ${\cal D}_{n}$-module $M$ is free. Actually, suppose that there exist operators $d_{1},\dots,d_{k}$ such that $d_{1}\psi_{1}+\dots+d_{k}\psi_{k}=0.$ Let $d$ denote an matrix operator with operators $d_{1},\dots,d_{k}$ in all rows. Then we have $(D(\lambda)+d)\psi=\lambda\psi,$ which contradicts the uniqueness of $D(\lambda)$. We show that if $\psi_{0}\in M,\lambda\in A_{Y},$ then $\lambda\psi_{0}\in M$. Let $\psi_{0}=d_{1}\psi_{1}+\dots+d_{k}\psi_{k}.$ By multiplying equality (2) from the left by a vector-row made up by operators $d_{1},\dots,d_{k}$, we get $(d_{1},\dots,d_{k})D(\lambda)\psi=(d_{1},\dots,d_{k})\lambda\psi=\lambda(d_{1},\dots,d_{k})\psi=\lambda\psi_{0},$ consequently, $\lambda\psi_{0}\in M$. Thus, apparently to every commutative ring there connects a Baker-Akhiezer module which satisfies the conditions 1,2 (see introduction). In all examples the Baker-Akhiezer function has an essential singularity on $Y$ and has the form $\psi(x,P)=g(x,P)\exp(x_{1}F_{1}(P)+\dots+x_{n}F_{n}(P)),$ where $F_{i}(P),i=1,\dots,n$ are meromorphic (in the general case many-valued functions) on $X$ with pole on $Y$, $g(x,p)=(g_{1}(x,P),\dots,g_{k}(x,P))^{\top},$ $g_{i}(x,P)$ are meromorphic sections of the line bundle $L$ on $U\times X$, $U$ is an open subset in ${\mathbb{C}}^{n}$ with pole on $\tilde{Y}=U\times Y.$ The line bundle $L$ possesses connections $\nabla_{i}=\partial_{x_{i}}+F_{i}(P).$ It is clear that $\nabla_{j}g_{i}(x,P)$ is a section of bundle $L\otimes[\tilde{Y}]^{s},s>0$. In this way, we represent module $M$ in the form $M=\cup_{m=1}^{\infty}M(m),$ where $M(m)$ consists of functions $M(m)=\\{\tilde{g}(x,P)\exp(x_{1}F_{1}(P)+\dots+x_{n}F_{n}(P)),\ \tilde{g}(x,P)\in{\rm H^{0}}(L\otimes[\tilde{Y}]^{m})\\}.$ What is more, mappings $\partial_{x_{j}}:M(m)\rightarrow M(m+1)$ hold. Further, for simplicity we assume that $n=2$. First, we consider the case $k=1$. Let the ${\cal D}_{2}$-module $M$ be generated by the function $\psi$. We can assume that $\psi\in M(1)$. Since the rank of the ${\cal O}$-module of differential operators is of an order no higher than $m-1$ equal to $\frac{m(m+1)}{2}$, by virtue of freeness of $M$ we have the obvious equality ${\rm rank}_{\cal O}M(m)=\frac{m(m+1)}{2}.$ Since for any meromorphic function $\lambda$ with pole of order $m$ on $Y$ there should be a unique differential operator $D(\lambda)$ of order $m$ such that equality (2) is satisfied, so in the general case equality ${\rm rank}_{\cal O}{\rm H}^{0}(U\times X,L\otimes[\tilde{Y}]^{m})=\frac{m(m+1)}{2}$ should be fulfilled. In this way, for the theory of commuting two variable differential operators with scalar coefficients, the following problem is important: Classification of algebraic manifolds $X$ of dimension 2, such that ${\rm dim}{\rm H}^{0}(X,E\otimes[Y]^{m})=\frac{m(m+1)}{2},\ m>0,$ where $E$ is a certain line bundle over $X$. Let $k=2$, $\psi_{1},\psi_{2}$ be the basis in the ${\cal D}_{2}$-module $M$. We can assume that $\psi_{1}\in M(1)$, $\psi_{2}\in M(m_{0}),m_{0}\geq 1$. Then, by virtue of freeness of $M$ we have ${\rm rank}_{\cal O}M(m)=\frac{m(m+1)}{2}+\frac{(m-m_{0}+1)(m-m_{0}+2)}{2},\ m>m_{0}.$ Analogically, in order to find commuting two variable differential operators with matrix coefficients of size $2\times 2$, the following problem is important: $2\times 2$: Classification of algebraic manifolds $X$ of dimension 2 such that ${\rm dim}{\rm H}^{0}(X,E\otimes[Y]^{m})=\frac{m(m+1)}{2}+\frac{(m-m_{0}+1)(m-m_{0}+2)}{2},$ $None$ $m>m_{0}$. Exactly in the same way we can examine the general case for arbitrary $n$ and $k$. Example (A. Nakayashiki [10]). Let $X^{g}={\mathbb{C}}^{g}/\\{{\mathbb{Z}}^{g}+\Omega{\mathbb{Z}}^{g}\\}$ be a principle polarized Abelian variety, with $\Omega$ being a symmetric complex matrix with a positively defined imaginary part. As subset $Y\subset X^{g}$ we consider a theta-divisor given by the zeroes of a theta function that is defined by the absolutely convergent series $\theta(z)=\sum_{n\in{\mathbb{Z}}^{g}}\exp(\pi i\langle m,\Omega m\rangle+2\pi i\langle m,z\rangle),\ z\in{\mathbb{C}}^{g}.$ Assume that the theta divisor is a non-singular subvariety. The Baker-Akhiezer module has the form $M=\cup_{m=1}^{\infty}M(m),$ $M(m)=\left\\{g(z,x)\exp\left(-x_{1}\partial_{z_{1}}\log\theta(z)\dots- x_{g}\partial_{z_{g}}\log\theta(z)\right)\\!\right\\},$ where $f(x,z)$ is a meromorphic function on ${\mathbb{C}}^{g}$ with a pole of order not larger than $m$ on the theta-divisor, and also periodic: $g(x,z+\Omega m+n)=\exp(-2\pi i\langle m,c+x\rangle)g(x,z),\ m,n\in{\mathbb{Z}}^{g},$ $c\in{\mathbb{C}}^{g}\backslash\\{0\\}$ being a fixed point * • $M$ is a free ${\cal D}_{g}$-module of rank $g!$. Besides, ${\rm rank}_{\cal O}M(m)=m^{g}.$ In this way, for $g=2$ in formula (3) we have $m_{0}=2$. Remark. Speaking more precisely, the conditions formulated on increasing of ${\rm dim}{\rm H}^{0}(X,E\otimes[Y]^{m})$ are necessary for the ${\rm gr}{\cal D}$-module ${\rm gr}M$ to be free, where the graduation is respectively induced by the degree of the operator and the order of pole on $Y$ (see details in [10]). Maybe it is possible that free ${\cal D}$ modules $M$ exists, but ${\rm gr}{\cal D}$-modules ${\rm gr}M$ at the same time are not free. ### 2.2 The Baker-Akhiezer module corresponding to the manifold $\Gamma$ Now turn to our construction. The form $f(P)$ (see introduction) is the section of the line bundle $\tilde{E}$ on ${\mathbb{C}}P^{1}\times{\mathbb{C}}P^{1}$, if identity (1) holds, it can be interpreted as a section of the line bundle on $\Gamma$ derived from $\tilde{E}$ by identification of the fibres over double points. Let’s introduce two forms $f_{i}(P)=\alpha_{i}z_{1}z_{2}+\beta_{i}z_{1}w_{2}+\gamma_{i}z_{2}w_{1}+\delta_{i}w_{1}w_{2},\ \alpha_{i},\beta_{i},\gamma_{i},\delta_{i}\in{\mathbb{C}},\ i=1,2,$ such that for $(t_{1}:t_{2})\in{\mathbb{C}}P^{1}$ identity $\frac{f_{i}(a_{1}:b_{1},t_{1}:t_{2})}{f(a_{1}:b_{1},t_{1}:t_{2})}-\frac{f_{i}(t_{1}:t_{2},a_{2}:b_{2})}{f(t_{1}:t_{2},a_{2}:b_{2})}-c_{i}=0,\ c_{i}\in{\mathbb{C}}.$ $None$ holds. The dimension of spaces of such forms equals 3. By virtue of (1) identity (4) is equivalent to $f_{i}(a_{1}:b_{1},t_{1}:t_{2})-Af_{i}(t_{1}:t_{2},a_{2}:b_{2})-Ac_{i}f(t_{1}:t_{2},a_{2}:b_{2})=0.$ Chose $f_{1}$ and $f_{2}$ such $f_{1},f_{2}$ and $f$ are linearly independent (any other form that satisfies condition (4) with a certain constant $c_{i}$ is a linear combination of $f_{1}$, $f_{2}$ and $f$). Let $F_{1}(P)=\frac{f_{1}(P)}{f(P)},\ F_{2}(P)=\frac{f_{2}(P)}{f(P)}.$ Let $M(n)$ denote a set of functions of the form $M(n)=\left\\{\psi=\frac{\tilde{f}(P,x,y)}{f^{n}(P)}\exp\left(xF_{1}(P)+yF_{2}(P)\right)\right\\},$ for which the identity $\psi(a_{1}:b_{1},t_{1}:t_{2},x,y)-\Lambda\psi(t_{1}:t_{2},a_{2}:b_{2},x,y)=0,\ \Lambda\in{\mathbb{C}},$ $None$ is fulfilled for all $(t_{1}:t_{2})\in{\mathbb{C}}P^{1}$, where $\tilde{f}$ is a form of the following kind: $\tilde{f}=\sum_{k,l=0}^{n}z_{1}^{k}w_{1}^{n-k}z_{2}^{l}w_{2}^{n-l}f_{kl}(x,y).$ $None$ The set of functions from $M(n)$ forms a module above a ring of analytic functions ${\cal O}$ on variables $x,y$ in the nighbourhood of $0\in{\mathbb{C}}^{2}$. The rank of module $M(n)$ above ${\cal O}$ (dimension of the space of functions from $M(n)$ for fixed $x,y$) is equal to ${\rm rank}_{\cal O}M(n)=n(n+1).$ Actually, the dimension of the space of forms of kind (6) is equal to $(n+1)^{2}$. Note that by virtue of (1) and (4) identity (5) is equivalent to $\tilde{f}(a_{1}:b_{1},t_{1}:t_{2},x,y)-\tilde{f}(t_{1}:t_{2},a_{2}:b_{2},x,y)\Lambda A^{n}\exp(-xc_{1}-yc_{2})=0.$ The last equality signifies that the coefficients of the homogeneous polinomial on $t_{1},t_{2}$ of $n$-th power standing in the left-hand part equal zero. This imposes a $(n+1)$ restriction on the choice of the coefficients of form $\tilde{f}$. In this way, ${\rm rank}_{\cal O}M(n)=(n+1)^{2}-(n+1)=n(n+1)$ and, consequently, in formula (3) for the manifold $\Gamma$ $m_{0}$=1. Note that since $\Lambda$ does not depend on $x,y$ identity (5) preserves its form at differentiation of $\psi$ by $x$ $y$. Consequently, we have two mappings $\partial_{x}:M(n)\rightarrow M(n+1),\ \partial_{y}:M(n)\rightarrow M(n+1).$ In this way, the structure of the Baker-Akhiezer module above the ring of differential operators ${\cal D}={\cal O}[\partial_{x},\partial_{y}].$ is given on the set $M=\cup_{n=1}^{\infty}M(n)$ ## 3 Proof of theorem 1 ### 3.1 Proof of freeness of the Baker-Akhiezer module Choose two functions $\psi_{1}$ and $\psi_{2}$ in $M(1)$ $\psi_{i}=\frac{h_{i}(P,x,y)}{f(P)}\exp\left(xF_{1}(P)+yF_{2}(P)\right),$ which are independent over ${\cal O}$ where $h_{i}(P,x,y)=a_{i}(x,y)z_{1}z_{2}+b_{i}(x,y)z_{1}w_{2}+c_{i}(x,y)w_{1}z_{2}+d_{i}(x,y)w_{1}w_{2},\ i=1,2.$ Functions $h_{i}$ satisfy the identity $h_{i}(a_{1}:b_{1},t_{1}:t_{2},x,y)-h_{i}(t_{1}:t_{2},a_{2}:b_{2},x,y)\Lambda A\exp(-xc_{1}-yc_{2})=0.$ Let $P_{1}$ and $P_{2}$ denote the intersection points of the curves given by equations $f_{1}(P)=0$ and $f(P)=0$, and by $Q_{1}$ and $Q_{2}$ the intersection points of the curves given by equations $f_{2}(P)=0$ and $f(P)=0$. For their bulkiness, we do not cite the explicit formulae for the point coordinates. $P_{1},P_{2}$ and $Q_{1},Q_{2}$. By small movements of $c_{1}$ and $c_{2}$ we can achieve these points to be mutually distinct. The following theorem holds: ###### Theorem 2 For any function $\varphi\in M$ there exist two unique differential operators $d_{1},d_{2}\in{\cal D}$ such that $d_{1}\psi_{1}+d_{2}\psi_{2}=\varphi,$ i.e. $M$ is a free module over the ring of differential operators ${\cal D}$ of rank two, generated by functions $\psi_{1},\psi_{2}$. Theorem 1 follows directly from theorem 2. Let $N$ denote the module over ${\cal D}$, generated by the functions $\psi_{1},\psi_{2}$ $N=\\{d_{1}\psi_{1}+d_{2}\psi_{2},d_{1},d_{2}\in{\cal D}\\}.$ Theorem 2 follows from lemma 1 and lemma 2. In lemma 1 we show that the ${\cal D}$-module $N$ is free of rank 2, and in lemma 2 we show that ${\cal D}$-moduli $M$ and $N$ coincide. The following lemma holds: ###### Lemma 1 The module $N$ is free over the ring of differential operators ${\cal D},$ generated by functions $\psi_{1},\psi_{2}$. Proof of lemma 1. Suppose that this module is not free, i.e. there exist two differential operators $d_{1},d_{2}\in D$ of order $n$ and $k$ $d_{1}=\alpha_{n}(x,y)\partial_{x}^{n}+\alpha_{n-1}(x,y)\partial_{x}^{n-1}\partial_{y}+\dots+\alpha_{0}(x,y)\partial_{y}^{n}+\dots,$ $d_{2}=\beta_{k}(x,y)\partial_{x}^{k}+\beta_{k-1}(x,y)\partial_{x}^{k-1}\partial_{y}+\dots+\beta_{0}(x,y)\partial_{y}^{k}+\dots,$ such that $d_{1}\psi_{1}+d_{2}\psi_{2}=0.$ $None$ First, consider the case of $n\neq k$. For sake of definiteness, let $n>k$. Divide equality (7) by $\exp\left(xF_{1}(P)+yF_{2}(P)\right)$, multiply by $f^{n+1}(P)$ and after this contrain the resulting equality to a curve $f(P)=0$. Get $h_{1}(P,x,y)(\alpha_{n}(x,y)f_{1}^{n}(P)+\alpha_{n-1}(x,y)f_{1}^{n-1}(P)f_{2}+\dots+$ $\alpha_{0}(x,y)f_{2}^{n}(P))=0.$ $None$ Substitute the resulting equality with point $P_{1}$ $h_{1}(P_{1},x,y)\alpha_{0}(x,y)f_{2}^{n}(P_{1})=0.$ Direct verification shows that function $h_{1}(P_{1},x,y)$ is not identically equal to 0, consequently, $\alpha_{0}(x,y)=0.$ Divide (8) by $f_{1}$ and substitute again point $P_{1}$ $h_{1}(P_{1},x,y)\alpha_{1}(x,y)f_{2}^{n-1}(P_{1})=0,$ consequently, $\alpha_{1}=0$. Analogically, we can show that $\alpha_{2}=\dots=\alpha_{n}=0.$ Now consider the case of $k=n$. For $P\in\\{f(P)=0\\}$ in place of (8) get $h_{1}(P,x,y)(\alpha_{n}(x,y)f_{1}^{n}(P)+\alpha_{n-1}(x,y)f_{1}^{n-1}(P)f_{2}+\dots+\alpha_{0}(x,y)f_{2}^{n}(P))+$ $h_{2}(P,x,y)(\beta_{n}(x,y)f_{1}^{n}(P)+\beta_{n-1}(x,y)f_{1}^{n-1}(P)f_{2}+\dots+\beta_{0}(x,y)f_{2}^{n}(P)).$ $None$ In (9) replace points $P=P_{1}$ and $P=P_{2}$. From this $h_{1}(P_{1},x,y)\alpha_{0}(x,y)f_{2}^{n}(P_{1})+h_{2}(P_{1},x,y)\beta_{0}(x,y)f_{2}^{n}(P_{1})=0,$ $h_{1}(P_{2},x,y)\alpha_{0}(x,y)F_{2}^{n}(P_{2})+h_{2}(P_{2},x,y)\beta_{0}(x,y)f_{2}^{n}(P_{2})=0.$ Consequently, $\frac{h_{1}(P_{1},x,y)}{h_{2}(P_{1},x,y)}=\frac{h_{1}(P_{2},x,y)}{h_{2}(P_{2},x,y)}=\frac{\beta_{0}(x,y)}{\alpha_{0}(x,y)}.$ By direct verification find that $\frac{h_{1}(P_{1},x,y)}{h_{2}(P_{1},x,y)}\neq\frac{h_{1}(P_{2},x,y)}{h_{2}(P_{2},x,y)}$ $None$ (here, for their bulkiness we do not cite explicit formulae for $\frac{h_{1}(P_{i},x,y)}{h_{2}(P_{i},x,y)}$. For simplicity we do this for a concrete example (see below)). Consequently, $\alpha_{0}=\beta_{0}=0$. Dividing (9) by $f_{1}$ yields $\alpha_{1}=\beta_{1}=0$. Analogically $\alpha_{2}=\dots=\alpha_{n}=\beta_{2}=\dots=\beta_{n}=0.$ Consequently, there exist no such operators $d_{1}$ and $d_{2}$. Lemma 1 is proven The following holds: ###### Lemma 2 Modules $M$ and $N$ coincide. Proof of Lemma 2 Let $N(n)$ denote the following subset in $N$ $N(n)=\\{d_{1}\psi_{1}+d_{2}\psi_{2},\ d_{1},d_{2}\in{\cal D},\ {\rm ord}\ d_{1},{\rm ord}\ d_{2}\leq n-1\\}.$ Since the ${\cal D}$-module $N$ is free, ${\rm rank}_{\cal O}N(n)=2\ {\rm rank}_{\cal O}\\{d\psi_{i},\ d\in{\cal D},\ {\rm ord}\ d\leq n-1\\}=n(n+1).$ Consequently, ${\rm rank}_{\cal O}M(n)={\rm rank}_{\cal O}N(n).$ By virtue of the obvious inclusion $N(n)\subset M(n),$ get $M=N.$ Thus, Lemma 2 is proven, jointly with theorem 2. ### 3.2 Explicit formulas for commuting differential operators In this section, we give an example of commuting differential operators and common vector eigen-functions which are parametrized by the spectral manifold $\Gamma$. Let points $p_{1},p_{2}\in{\mathbb{C}}P^{1}$ have the following coordinates $p_{1}=(1:0),\ p_{2}=(0:1).$ We introduce three forms $f(P)=z_{1}z_{2}+z_{1}w_{2}+w_{1}w_{2},$ $f_{1}(P)=z_{1}z_{2}+2w_{1}z_{2}-w_{1}w_{2},$ $f_{2}(P)=-z_{1}z_{2}+2w_{1}z_{2}+w_{1}w_{2}.$ By direct verification find that form $\theta(P)$ satisfies identification (1) for $A=1$, and forms $f_{1}(P)$ and $f_{2}(P)$ satisfy identity (4), with $c_{1}=1$ and $c_{2}=-1$, respectively. Curves $f(P)=0$ and $f_{1}(P)=0$ intersect in points $P_{1}=(-2-\sqrt{2}:1,-\frac{1}{\sqrt{2}}:1),\ P_{2}=(-2+\sqrt{2}:1,\frac{1}{\sqrt{2}}:1),$ and curves $f(P)=0$ and $f_{2}(P)=0$ — in points $Q_{1}=(-\sqrt{2}:1,-1+\frac{1}{\sqrt{2}}:1),\ Q_{2}=(\sqrt{2}:1,-1-\frac{1}{\sqrt{2}}:1),$ Take base $\psi_{1},\psi_{2}$ in the ${\cal D}$-module $M$ $\psi_{1}=\frac{w_{1}z_{2}}{f(P)}\exp\left(xF_{1}(P)+yF_{2}(P)\right),$ $\psi_{2}=\frac{z_{1}z_{2}e^{y-x}+z_{1}w_{2}+w_{1}w_{2}e^{x-y}}{f(P)}\exp\left(xF_{1}(P)+yF_{2}(P)\right).$ Then, $\frac{h_{1}(P_{1},x,y)}{h_{2}(P_{1},x,y)}=-\frac{e^{x+y}}{\sqrt{2}(e^{y}-e^{x})(-e^{x}+(1+\sqrt{2})e^{y})},$ $\frac{h_{1}(P_{2},x,y)}{h_{2}(P_{2},x,y)}=-\frac{e^{x+y}}{\sqrt{2}(e^{y}-e^{x})(e^{x}+(-1+\sqrt{2})e^{y})},$ thus, inequality (10) holds. The four most simple meromorphic functions on $\Gamma$ with poles on the curve $f(P)=0$ have the form $\lambda_{1}=\frac{w_{1}z_{2}}{f(P)},\ \lambda_{2}=\frac{z_{1}w_{1}z_{2}^{2}}{f(P)^{2}},\ \lambda_{3}=\frac{z_{1}z_{2}w_{1}w_{2}}{f(P)^{2}},\ \lambda_{4}=\frac{z_{1}w_{1}w_{2}^{2}+z_{1}^{2}z_{2}w_{2}}{f(P)^{2}}.$ The pairwise commutating operators corresponding to this function have the form $D(\lambda_{1})=\left(\begin{array}[]{cc}\frac{1}{4}(\partial_{x}+\partial_{y})&0\\\ 0&\frac{1}{4}(\partial_{x}+\partial_{y})\\\ \end{array}\right).$ $[D(\lambda_{2})]_{11}=\frac{e^{x}}{8(e^{x}-e^{y})}(\partial_{x}^{2}-\partial_{y}^{2})-\frac{e^{x+y}}{4(e^{x}-e^{y})^{2}}(\partial_{x}+\partial_{y}),$ $[D(\lambda_{2})]_{12}=\frac{e^{x+y}}{16(e^{x}-e^{y})^{2}}(\partial_{x}+\partial_{y})^{2},$ $[D(\lambda_{2})]_{21}=\frac{1}{8}(e^{y-x}-e^{x-y}-2)\partial_{x}^{2}+\frac{1}{8}(e^{x-y}-e^{y-x}-2)\partial_{y}^{2}+\frac{1}{2}\partial_{x}\partial_{y}+$ $\frac{e^{x}+e^{2x-y}+5e^{y}-e^{2y-x}}{4(e^{x}-e^{y})}\partial_{x}+\frac{3e^{x}-e^{2x-y}+3e^{y}+e^{2y-x}}{4(e^{y}-e^{x})}\partial_{y}-\frac{e^{y}(2e^{x}+e^{y}}{(e^{x}-e^{y})^{2})},$ $[D(\lambda_{2})]_{22}=\frac{e^{x}}{8(e^{y}-e^{x})}\partial_{x}^{2}-\frac{1}{4}\partial_{x}\partial_{y}+\frac{e^{x}-2e^{y}}{8(e^{y}-e^{x})}\partial_{y}^{2}+\frac{e^{y}(2e^{x}+e^{y})}{8(e^{x}-e^{y})^{2}}(\partial_{x}+\partial_{y}).$ The operator corresponding to the function $\lambda_{3}$ has the form $[D(\lambda_{3})]_{11}=\frac{(e^{x}+e^{y}}{8(e^{y}-e^{x})}(\partial_{x}^{2}-\partial_{y}^{2})+\frac{(e^{2x}+e^{2y}}{4(e^{y}-e^{x})^{2}}(\partial_{x}+\partial_{y}),$ $[D(\lambda_{3})]_{12}=\frac{e^{x+y}}{8(e^{y}-e^{x})^{2}}(\partial_{x}-\partial_{y})^{2},$ $[D(\lambda_{3})]_{21}=\frac{1}{4}(2+e^{x-y}-e^{y-x})\partial_{x}^{2}-\partial_{x}\partial_{y}+\frac{1}{4}(2-e^{x-y}+e^{y-x})\partial_{y}^{2}+$ $\frac{2e^{x}+e^{2x-y}+4e^{y}-e^{2y-x}}{2(e^{y}-e^{x})}\partial_{x}+\frac{4e^{x}-e^{2x-y}+2e^{y}+e^{2y-x}}{2(e^{x}-e^{y})}\partial_{y}+\frac{e^{2x}+e^{2y}+4e^{x+y}}{(e^{x}-e^{y})^{2}},$ $[D(\lambda_{3})]_{22}=\frac{3e^{x}-e^{y}}{8(e^{x}-e^{y})}\partial_{x}^{2}+\frac{1}{2}\partial_{x}\partial_{y}+\frac{e^{x}-3e^{y}}{8(e^{x}-e^{y})}\partial_{y}^{2}-\frac{3e^{x+y}}{2(e^{y}-e^{x})^{2}}(\partial_{x}+\partial_{y}).$ The operator corresponding to the function $\lambda_{4}$ has the form $[D(\lambda_{4})]_{11}=\frac{e^{x}+3e^{y}}{4(e^{x}-e^{y})}\partial_{x}^{2}+\frac{1}{2}\partial_{x}\partial_{y}-\frac{3e^{x}+e^{y}}{4(e^{x}-e^{y})}\partial_{y}^{2}-$ $\frac{e^{2x}+3e^{2y}}{2(e^{x}-e^{y})^{2}}\partial_{x}-\frac{3e^{2x}+e^{2y}}{2(e^{x}-e^{y})^{2}}\partial_{y}-\frac{2e^{x+y}}{(e^{y}-e^{x})^{2}},$ $[D(\lambda_{4})]_{12}=\frac{e^{x+y}}{2(e^{y}-e^{x})^{2}}((\partial_{x}+\partial_{y})^{2}+\partial_{x}+\partial_{y}),$ $[D(\lambda_{4})]_{21}=(e^{y-x}-e^{x-y}-2)\partial_{x}^{2}+4\partial_{x}\partial_{y}+(e^{x-y}-e^{y-x}-2)\partial_{y}^{2}+$ $\frac{5e^{x}+e^{2x-y}+9e^{y}-3e^{2y-x}}{e^{x}-e^{y}}\partial_{x}+\frac{9e^{x}-3e^{2x-y}+5e^{y}+e^{2y-x}}{e^{x}-e^{y}}\partial_{y}+$ $\frac{2e^{-x-y}(e^{4x}+e^{4y}-6e^{2(x+y)}-4e^{3x+y}-4e^{x+3y})}{(e^{x}-e^{y})^{2}},$ $[D(\lambda_{4})]_{22}=\frac{7e^{x}-3e^{y}}{4(e^{y}-e^{x})}\partial_{x}^{2}-\frac{3}{2}\partial_{x}\partial_{y}+\frac{3e^{x}-7e^{y}}{4(e^{y}-e^{x})}\partial_{y}^{2}-$ $\frac{e^{2x}+3e^{2y}-16e^{x+y}}{2(e^{x}-e^{y})^{2}}\partial_{x}-\frac{3e^{2x}+e^{2y}-16e^{x+y}}{2(e^{x}-e^{y})^{2}}\partial_{y}.$ ## References * [1] I.M. Krichever. Commutative rings of ordinary linear differential operators // Funct. Anal. Appl. 1978. V. 12. N. 4. P. 41–52. * [2] J.L. Burchnall, I.W. Chaundy. Commutative ordinary differential operators // Proc. London Math. Society. 1923. Ser. 2. V. 21. P. 420–440. * [3] I.M. Krichever. Methods of algebraic geometry in the theory of nonlinear equations (Russian) // Uspehi Mat. Nauk. 1977. V. 32. N. 6\. P. 183–208. * [4] A.N. Parshin. The Krichever correspondence for algebraic surfaces // Funct. Anal. Appl. 2001. V. 35. N. 1. P. 74–76 * [5] A.B. Zheglov, D. V. Osipov. On some questions related to the Krichever correspondence // Math. Notes. 2007. V. 81. N. 4. 2007. P. 467–476. * [6] A. Kasman, E. Previato. Commutative partial differential operators. Advances in nonlinear mathematics and science // Phys. D 2001. 152/153. P. 66–77. * [7] V.M. Bukhshtaber, S.Yu. Shorina. The $w$-function of a solution of the $g$th stationary KdV equation // Russian Math. Surveys. 2003. V. 58. N. 4. P. 780–781 * [8] O.A. Chalykh, A.P. Veselov. Commutative rings of partial differential operators and Lie algebras. Comm. Math. Phys. 1990. V. 126, N. 3. P. 597–611. * [9] O.A. Chalykh, M.V. Feigin, A.P. Veselov. Multidimensional Baker-Akhiezer functions and Huygens’ principle. Comm. Math. Phys. 1999\. V. 206. N. 3. P. 533–566. * [10] A. Nakayashiki. Structure of Baker–Akhiezer Modules of Principally Polarized Abelian varieties, Commuting Partial Differential Operators and Associated Integrable Systems // Duke Math. J. 1991. V. 62. N. 2. P. 315–358. * [11] A. Nakayashiki. Commuting partial differential operators and vector bundles over Abelian varieties. Amer. J. Math // 1994. V. 116\. P. 65–100. * [12] A.E. Mironov. Commutative rings of differential operators connected with two-dimensional abelian varieties // Siberian Math. Journal. 2000. V. 41. N. 6. P. 1148-1161. * [13] A.E. Mironov. Real commuting differential operators connected with two-dimensional abelian varieties // Siberian Math. Journal. 2002. V. 43. N. 1. P. 97-113. * [14] M. Rothstein. Sheaves with connection on abelian varieties // Duke Math. J. 1996. V. 84. N. 3 P. 565–598. * [15] M. Rothstein. Dynamics of the Krichever construction in several variables // J. Reine Angew. Math. 2004. V. 572. P. 111–138. Sobolev Institute of Mathematics, Novosibirsk State University, E-mail address: mironov@math.nsc.ru

arxiv-papers

2024-09-04T02:48:54.689001

0804.0124

# Dynamical electro-weak symmetry breaking from deformed AdS: vector mesons and effective couplings. Marco Fabbrichesi INFN, Sezione di Trieste, Trieste, Italy. Maurizio Piai Department of Physics, University of Washington, Seattle, WA 98195 Swansea University, School of Physical Sciences, Singleton Park, Swansea, Wales, UK Luca Vecchi INFN, Sezione di Trieste, Trieste, Italy. Scuola Internazionale Superiore di Studi Avanzati, Via Beirut 4, Trieste, Italy. ###### Abstract We study a modification of the five-dimensional description of dynamical electro-weak symmetry breaking inspired by the AdS/CFT correspondence. Conformal symmetry is broken in the low-energy region near the IR brane by a power-law departure from the pure AdS background. Such a modification—while not spoiling the identification of the IR brane with the scale of confinement— has a dramatic effect on both the coupling of the first composite states to the standard model currents and their self-couplings. Chiral symmetry breaking can take place at a scale larger than the IR cut-off. This study shows that observables, such as the precision parameter $\hat{S}$, which depend on the couplings of the lightest composite states to the currents are very sensitive to the details of the dynamics in the low energy region where conformal symmetry is lost and electro-weak symmetry is broken just above the scale of confinement. Therefore results of calculations of these observables in AdS/CFT inspired scenarios should be interpreted conservatively. The most important phenomenological consequence for physics at the LHC is that the bound on the mass scale of the heavy excitations (technirho mesons) in a realistic model is in general lower than in the pure AdS background with a simple hard-wall cut- off in the IR. ###### pacs: 11.10.Kk, 12.15.Lk, 12.60.Nz ## I Introduction The mechanism responsible for electro-weak symmetry breaking in the standard model will be tested at the Large Hadron Collider (LHC). One important goal for this experimental program is to understand whether the interactions responsible for electro-weak symmetry breaking are strong or weak. It is essential to identify theoretically clean, measurable quantities that can help distinguish these two possibilities unambiguously. One might think that this is an easy task: after all, a strongly coupled model in the spirit of technicolor TC predicts the existence of towers of broad, strongly coupled composite resonances, with a rich spectroscopy at the TeV scale, in analogy with what is known about QCD at the GeV scale. Yet, indirect experimental data about electro-weak symmetry breaking PT ; Barbieri cannot be easily reconciled with this framework and already suggest that, if electro- weak symmetry breaking is a dynamical effect, the low-energy effective field theory description of the new strongly-coupled sector has to exhibit features that are not generic. Walking technicolor walking is a plausible candidate for such a strongly coupled model, based on conformal behavior and large anomalous dimensions in the IR, but calculability within this framework has been a challenge AS . In recent years, based on the ideas of the AdS/CFT correspondence AdS/CFT , and on the pioneering work of Randall and Sundrum RS1 ; pheno , many models have been investigated which exhibit in the low energy region the basic properties expected in a walking theory, while being calculable. Examples now exist of models that are compatible (within the errors) with the precision data and can be discovered at the LHC. The literature on the subject is already extensive AdS/TC ; HS ; MP ; higgsless ; composite . Most of these models assume a conformal behavior of the strongly coupled sector in the energy region spanning few orders of magnitude above the electro-weak scale and the existence of a weakly-coupled effective field theory description of the low-energy dynamics of the resonances. The construction of the effective field theory is derived by writing a weakly coupled extra-dimension model with a non-trivial gravity background, and by using the dictionary of the AdS/CFT correspondence to relate back to four dimensions. A generic phenomenological feature of all these models is that, unless a clever mechanism arranging for non-trivial (often fine-tuned) cancellations is implemented, a quite severe lower bound on the mass $M_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}$ 2.5–3 TeV of the lightest spin-1 resonance (techni-rho) results, in particular from the bounds on the electro-weak parameter $\hat{S}$ PT ; Barbieri . This result, together with the assumption that the effective field theory be weakly coupled (and hence calculable), gives rise to a spectacular signature (a sharp resonance peak) at the LHC MP2 . Unfortunately, it is very difficult to distinguish it from the signature of a generic, weakly-coupled extension of the standard model with an extended gauge group, predicting a new massive $Z^{\prime}$ gauge boson. Indisputable evidence proving that a strongly-coupled sector is responsible for electro-weak symmetry breaking would be the discovery of at least the first two spin-1 resonances, hence proving that these new particles are not elementary, but higher energy excitations of a composite object. The major obstacle against this scenario is the unfortunate numerology emerging from the combination of precision data and LHC high-energy discovery reach. If $M_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}$ 2.5–3 TeV, than it follows that the mass of the second resonance must be $M_{2}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}$ 5–6 TeV and just beyond the region where LHC data are expected to give convincing evidence AtlasTDR . Yet, a pretty mild relaxation of the experimental bounds would be enough to change this situation radically, since $M_{1}\sim 1.5$ TeV would imply $M_{2}\approx$ 2.5–4 TeV, well within reach even at moderate luminosity MP2 . It is hence timely, just before LHC starts collecting data, to question how accurate the AdS/CFT description of realistic dynamical electro-weak symmetry breaking is, and whether some of the approximations implied by this description could account for the desired softening of the bounds, without at the same time spoiling the calculability of the effective field theory. In analogy with RS1 , the five-dimensional picture usually contains two hard boundaries representing the UV and IR cut-off between which the theory is conformal. This is the weakest link with the idea that electro-weak symmetry breaking be triggered by a non-abelian gauge theory with an approximate IR fixed point. Taken literally, this picture means that, both in the UV and in the IR, conformal symmetry is lost instantaneously, via a sharp transition. As for the UV cut-off, this is not a real problem from the low-energy effective field theory point of view. The details of how an asymptotically-free fundamental theory in the far UV enters a quasi-conformal phase below the UV cut-off, can always be reabsorbed (via holographic renormalization HR ; MP ) in the definition of otherwise divergent low-energy parameters of the effective field theory, defined at a given order in the perturbative expansion of the effective field theory itself. Rather different is the case against using a hard-wall regulator in the IR. There is no sense in which IR effects decouple and can be renormalized away, and hence the low-energy effects we are interested in, when comparing the effective field theory to the experimental data, are inherently sensitive to the choice of the IR regulator. On the one hand, the very validity of the effective field theory description based on the AdS/CFT dictionary requires that the hard-wall cut-off be at least a reasonable leading order approximation (otherwise the effective field theory itself would be strongly coupled, and not admit a controllable expansion). On the other hand, corrections are expected to be present, and estimating their size and understanding their phenomenological consequences is crucial, at the very least in order to know what to expect in experiments such as those at the LHC, which is going to test precisely the energy range close to the IR cut-off. To be more specific. In the IR, three different phase transitions are taking place: electro-weak symmetry breaking, conformal symmetry breaking and confinement. These cannot define three parametrically separate scales, since they are all triggered by the same physical effect, namely the fact that the underlying (unknown) theory possesses an approximate fixed point in the IR. Hence the RG flow of the underlying dynamics is not going to reach the IR fixed point (which is only approximate), but will drift away from it at low energies, after spending some time (walking) in its proximity. Yet, there is no reason to expect these three effects to arise precisely at the same energy (temperature), and they might define three distinct critical scales (temperatures) that differ by $O(1)$ coefficients. An illustration of this point can be obtained by considering an ${\cal N}=1$ supersymmetric QCD model with $N_{c}$ colors and $N_{f}$ fermions. At least at large-$N_{c}$, for $3N_{c}/2<N_{f}<3N_{c}$, the theory is asymptotically free, but has a fixed point in the IR seiberg ; cascade (for recent progress towards the rigorous construction of the gravity dual see, for instance, Nunez ). If $N_{f}$ is not far from the lower bound, so that the theory is strongly coupled at distances larger than a UV cut-off $1/L_{0}$, then the theory might be approximately described by a large-$N_{c}$ conformal field theory at strong ’t Hooft coupling. Suppose now that at some smaller energy, characterized by a length scale $\bar{L}\gg L_{0}$, for some reason (for example the existence of a suppressed symmetry-breaking higher-order operator, which acquires a large anomalous dimension in the IR turning it into a relevant deformation) a symmetry-breaking condensate forms, reducing further $N_{f}$ to a value $N_{f}^{\prime}$ closer to or below $3N_{c}/2$. Symmetry-breaking drives the theory away from the original fixed point, and induces the loss of conformal symmetry. The coupling now runs fast (because the coupling itself was already big and large anomalous dimensions are present), and (depending on $N_{f}^{\prime}$) the theory either enters a new conformal phase at stronger coupling or confines. The breaking of the global $SU(N_{f})_{L}\times SU(N_{f})_{R}$, conformal symmetry-breaking and confinement take all place approximately at the same scale. Yet, the energy at which the coupling reaches its upper bound defines a new scale $L_{1}$ which might well be some numerical factor away from $\bar{L}$, the scale at which the RG-flow trajectory departed away from the fixed point. If this is the qualitative behavior of the UV-complete dynamical model that is ultimately responsible for electro-weak symmetry breaking , describing it as a slice of AdS space between two hard walls is a good leading-order approximation. Nevertheless, we may wonder whether a factor of 3 or 4 separating the scales of conformal symmetry-breaking and confinement can be completely ignored, in the light of the phenomenological consequences at the LHC that a mere factor of two might have. In this paper, we study the effect of such a factor. We consider the simplest possible effective field theory description of dynamical electro-weak symmetry breaking as a 5D weakly-coupled system (see also MP ), introduce (besides the UV brane at $L_{0}$ and the IR brane at $L_{1}$) a new discontinuity at the scale $\bar{L}$, very close to the IR scale $L_{1}$, and assume that the background deviates from the AdS case for $\bar{L}<z<L_{1}$. As for the origin and description of electro-weak symmetry-breaking, we will treat it as a completely non-dynamical effect localized in the IR, somehow in the spirit of Higgless models. The breaking could take place at $L_{1}$ as well as at $\bar{L}$ (or anywhere in between), as suggested by the SQCD example above. We compare the effects on the electro-weak precision parameter $\hat{S}$ in these two cases, as illustrative of two extreme possibilities, without committing ourselves to either of them. The idea that chiral symmetry breaking might, for a generic model, take place at a scale higher than confinement has been in the literature for a while chiral/confinement , has been supported by lattice evidence in some special case lattice , and has recently been discussed also in string-inspired models SS . A realistic model should also implement a dynamical mechanism generating the mass of the standard model fermions. This can be done either via extended technicolor higher-order interactions between the standard model fermions and the new strong sector ETC ; APS (represented in the 5D picture by Yukawa interactions localized at the UV, with the symmetry-breaking vacuum expectation value not localized, but exhibiting a non-trivial power-law profile in the bulk), or via the assumption that standard model fields are themselves (partially) composite, in the spirit of topcolor and related models topcolor (which would imply the fermions be allowed to propagate in the bulk of the 5D model). A detailed discussion of how the global family symmetry of the standard model is broken would be required in order to study how the phenomenology of flavor-changing transitions and the physics of the third generation would be affected by the proposed modification of the background. In this paper, we treat the standard model fermions as non-dynamical fields, described by a set of external currents, and do not address the problem of their mass generation. ## II Preliminaries A non-trivial departure of the dynamics of the spin-1 resonances, with respect to that on pure AdS geometry, may be either due to a modification of the gravity background or to the presence of a non-dynamical background (dilaton). Since we consider an effective field theory where only spin-1 states are dynamical, it is not possible to distinguish between these two effects at this level. We choose to describe the model in terms of a deformation of the gravity background, for simplicity. Consider the five-dimensional space described by the metric $\displaystyle\mbox{d}s^{2}$ $\displaystyle=$ $\displaystyle a(z)^{2}\left(\eta_{\mu\nu}\mbox{d}x^{\mu}\mbox{d}x^{\nu}\,-\,\mbox{d}z^{2}\right)\,,$ (1) where $L_{0}<z<L_{1}$. We will assume that the geometry approaches pure AdS in the UV region, $a(z)\rightarrow L/z$ as $z\rightarrow L_{0}$, and departs from it at a scale $z\sim\bar{L}$. In most of the calculations we take $L_{0}=L$ for simplicity. We are interested in describing a model that at low-energy (below $1/L_{1}$) can be matched to the electro-weak chiral Lagrangian EWCL . This requires to introduce a 5-dimensional gauge group which is at least $SU(2)_{L}\times U(1)_{Y}$, but may be enlarged to accommodate custodial symmetry. Irrespectively of the details, the model contains a vectorial sector (the neutral part of which consists of the photon and its excitations) and an axial sector (containing the $Z$ boson and its excitations). In this paper we describe only the phenomenology connected with the neutral gauge bosons, hence we dispense with the details of the complete symmetry group. For concreteness, we take the vectorial sector to be described by the pure Yang-Mills $SU(2)$ theory with the following action: $\displaystyle{\cal S}$ $\displaystyle=$ $\displaystyle\int\mbox{d}^{4}x\int_{L_{0}}^{L_{1}}\mbox{d}z\sqrt{G}\left[G^{MR}G^{NS}\left(-\frac{1}{2}{\rm Tr}F_{MN}F_{RS}\right)+2g\delta(z-L_{0})G^{MN}{\rm Tr}J_{M}A_{N}\right]\,,$ (2) where $F_{MN}=\partial_{M}A_{N}-\partial_{N}A_{M}+ig[A_{M},A_{N}]$ is the field strength tensor of $A_{M}=A_{M}^{a}T^{a}$ with $T^{a}=\tau^{a}/2$ the generators of $SU(2)$, where $g$ is the (dimensionful) gauge coupling, and where $J_{M}=(J_{\mu}(x),0)$ is the four-dimensional external current localized on the UV-brane. Quantization requires to add appropriate gauge-fixing terms, canceling the mixing terms between spin-0 and spin-1 fields, which in unitary gauge implies $A_{5}(x,z)=0$. After Fourier transforming in 4D, $A_{\mu}(x,z)\equiv\int\frac{\mbox{d}^{4}q}{(2\pi)^{2}}e^{iqx}A_{\mu}(q)v(q,z)$, the free bulk equations read $\displaystyle\partial_{5}\left[a(z)\partial_{5}v(q,z)\right]$ $\displaystyle=$ $\displaystyle-q^{2}a(z)v(q,z)\,.$ (3) Substituting the solutions in the action, and canceling the boundary terms at $z=L_{1}$, without breaking the gauge symmetry, requires to impose Neumann boundary conditions: $\displaystyle\partial_{5}v(q,L_{1})$ $\displaystyle=$ $\displaystyle 0\,.$ (4) This set of equations admits always a constant, massless zero mode. Finally, the action can be rewritten as a pure boundary term at the UV, from which one can read the vector two-point correlator, that for $L_{0}\rightarrow L$ is $\displaystyle\Sigma_{V}(q^{2})$ $\displaystyle=$ $\displaystyle g^{2}\frac{v(q,L_{0})}{\partial_{5}v(q,L_{0})}\,,$ (5) which can be expanded as $\displaystyle\Sigma_{V}(q^{2})$ $\displaystyle=$ $\displaystyle e^{2}\left(\frac{1}{q^{2}}\,+\,\sum_{i}\frac{R_{i}}{q^{2}-M_{i}^{2}}\right)\,,$ (6) where $M_{i}$ ($i=1,2,\dots$) are the masses of the excited states, and $R_{i}$ define their effective couplings to the four-dimensional currents, normalized to the coupling $e^{2}$ of the massless mode (to be identified with the electro-magnetic coupling of the photon). The (dimensionful) bulk coupling $g$ controls the perturbative expansion used to extract this correlator. It is not directly related to the effective coupling $e$ of the standard model gauge boson (photon), but rather is related to the strength of the effective interactions among its heavy (composite) excitations. The relation between these two effective couplings depends on how the theory is regularized in the UV, and is not a calculable quantity, because of the divergences in the $L_{0}\rightarrow 0$ limit. A rigorous treatment requires to introduce appropriate counterterms and treat the ratio $e^{2}L/g^{2}$ as a free parameter. For the purposes of this paper, which primarily require comparing identical UV settings with different IR deformations, we can simplify this procedure by assuming that $L_{0}\ll L_{1}$ be finite and fixed, and express this ratio as a function of the scales and couplings in the model. We discuss later how good the perturbative expansion is by estimating the size of the effective self-coupling of the composite states. In order to compute $\hat{S}$ one has to introduce also the axial-vector excitations, and a symmetry-breaking mechanism. For the purposes of this paper, we only consider the Higgsless limit, defined by the introduction of a localized, infinitely massive Higgs scalar which assumes a non-trivial symmetry-breaking vacuum expectation value. The axial-vector modes $v_{A}(q,z)$ still satisfy Eq (3), but their boundary conditions (and the gauge fixing action) are modified. We consider two cases in the following. In the first, symmetry-breaking takes place on the boundary $L_{1}$ so that the axial-vector profiles $v_{A}(q,z)$ obey generalized Neumann boundary condition: $\displaystyle\partial_{5}v_{A}(q,L_{1})\,+\,mv_{A}(q,L_{1})$ $\displaystyle=$ $\displaystyle 0\,.$ (7) The effective symmetry-breaking parameter $m$ has dimension of a mass. In the limit $m\rightarrow 0$ one recovers the symmetric case, while for $m\rightarrow+\infty$ one recovers the Dirichelet boundary conditions. The mass of the $Z$ boson depends on $m$ is such a way that it vanishes for vanishing $m$, but is determined by $L_{1}$ for arbitrarily large $m$. In the second case we consider a symmetry-breaking vacuum expectation value localized at a different point $\bar{L}<L_{1}$ in the fifth dimension. The modifications to be implemented in this case will be discussed in the next sections. All of this allows to define the axial-vector correlator $\Sigma_{A}(q^{2})$ by replacing in Eq. (5) $v_{A}(q,z)$ and its derivative to $v(q,z)$. After these manipulations, the precision parameter $\hat{S}$ is given by $\displaystyle\hat{S}$ $\displaystyle=$ $\displaystyle\left.e^{2}\cos^{2}\theta_{W}\,\frac{\mbox{d}}{\mbox{d}q^{2}}\left(\frac{1}{\Sigma_{V}(q^{2})}-\frac{1}{\Sigma_{A}(q^{2})}\right)\right|_{q^{2}=0}\,,$ (8) where $e$ has been defined before, and corresponds to the electro-magnetic coupling, while $\theta_{W}$ is the effective weak-mixing angle. We recall here that an approximate extrapolation to large Higgs masses yields the experimental limit $\hat{S}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}0.003$ at the $3\sigma$ level Barbieri . ## III Pure AdS background We summarize here the results of the case in which the background is purely AdS with $a(z)=L/z$, and assume for simplicity that $L_{0}=L$. The vector correlator is $\displaystyle\Sigma^{(0)}_{V}(q^{2})$ $\displaystyle=$ $\displaystyle\frac{g^{2}\left(J_{0}(L_{1}q)Y_{1}(Lq)-J_{1}(Lq)Y_{0}(L_{1}q)\right)}{q\left(J_{0}(L_{1}q)Y_{0}(Lq)-J_{0}(Lq)Y_{0}(L_{1}q)\right)}\,.$ (9) In order to discuss the spectrum and couplings, the following approximations can be used: $\displaystyle\Sigma^{(0)}_{V}(q^{2})$ $\displaystyle\simeq$ $\displaystyle\frac{g^{2}J_{0}(L_{1}q)}{Lq^{2}\left(\frac{\pi}{2}Y_{0}(L_{1}q)-J_{0}(qL_{1})\left(\gamma_{E}+\log\frac{Lq}{2}\right)\right)}$ (10) $\displaystyle\simeq$ $\displaystyle\frac{g^{2}}{Lq^{2}\left(\frac{\pi}{2}\tan(L_{1}q-\frac{\pi}{4})-\left(\gamma_{E}+\log\frac{Lq}{2}\right)\right)}\,,$ (11) the first of which is valid for $L\ll L_{1}$, and the second for $qL_{1}>1$. From (10) one can read the coupling of the zero mode: $\displaystyle e^{2}$ $\displaystyle=$ $\displaystyle\frac{g^{2}}{L\log L_{1}/L}\,.$ (12) From (11) one can look for the poles and the residues $R_{i}$. The poles (besides the pole at zero) are in the vicinity of those of $\tan(L_{1}q-\pi/4)$: $\displaystyle M_{i}$ $\displaystyle\simeq$ $\displaystyle\frac{\pi}{4L_{1}}\left((4i-1)-\frac{2}{\gamma_{E}+\log(4i-1)\pi L/(8L_{1})}\right)\,,$ (13) while the residues are approximately given by: $\displaystyle R_{i}$ $\displaystyle\simeq$ $\displaystyle\frac{4\log(L_{1}/L)}{-2+\frac{2\pi L_{1}M_{i}}{1+\sin(2L_{1}M_{i})}},$ (14) with $i=1,2,\dots$. These approximations are acceptably accurate as long as $L\ll L_{1}$. A numerical calculation will be performed later on, when discussing the phenomenology for some relevant choice of parameters. The axial correlator can be computed exactly: $\displaystyle\Sigma^{(0)}_{A}(q^{2})$ $\displaystyle=$ $\displaystyle\frac{g^{2}\left(\left(qJ_{0}(L_{1}q)+mJ_{1}(L_{1}q)\right)Y_{1}(Lq)-J_{1}(Lq)\left(qY_{0}(L_{1}q)+mY_{1}(L_{1}q)\right)\right)}{q\left(qJ_{0}(L_{1}q)Y_{0}(Lq)+mJ_{1}(L_{1}q)Y_{0}(Lq)-J_{0}(Lq)\left(qY_{0}(L_{1}q)+mY_{1}(L_{1}q)\right)\right)}\,,$ (15) and, for $L_{0}\ll L_{1}$, yields $\displaystyle\hat{S}$ $\displaystyle=$ $\displaystyle\frac{\cos^{2}\theta_{W}L_{1}m(3L_{1}m+8)}{4(L_{1}m+2)^{2}\log\left(\frac{L_{1}}{L}\right)}\,.$ (16) In the limit $m\rightarrow+\infty$ we have $\displaystyle\hat{S}$ $\displaystyle=$ $\displaystyle\frac{3\cos^{2}\theta_{W}}{4\log{L_{1}/L}}\,.$ (17) Imposing the ($3\sigma$-level) experimental limit we find that $\displaystyle\frac{g^{2}}{L}$ $\displaystyle=$ $\displaystyle e^{2}\log\frac{L_{1}}{L}\,=\,e^{2}\frac{3\cos^{2}\theta_{W}}{4\hat{S}}\,\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}\,20\,,$ (18) where $e^{2}\simeq 0.1$ is the effective coupling of the electro-magnetic $U(1)_{Q}$ in the standard model. Since, as discussed later, $g^{2}/L$ gives a measure of the effective strength of the self interactions between resonances (and the dimensionful coupling $g$ is the expansion parameter in the 5D action) the experimental bounds are satisfied only at the price of loosing calculability, as is the unfortunate case also when trying to build QCD-like technicolor models in 4D, either using the large-$N$ expansion, hidden local symmetry, or deconstruction (see for instance deconstructTC ). We do not discuss further this limit. In the more interesting and realistic case in which $mL_{1}\ll 1$, the axial- vector spectrum and couplings are approximately the same as the vectorial sector. In this framework $m$ is just a free parameter, and we treat it as such. With finite $mL_{1}\ll 1$, the mass of the lightest axial-vector state is approximately $M_{Z}^{2}\simeq m/(L_{1}\log(L_{1}/L))$, and hence $\displaystyle\hat{S}\simeq\frac{\cos^{2}\theta_{W}}{2\log L_{1}/L}\,mL_{1}\simeq\frac{\cos^{2}\theta_{W}}{2}M_{Z}^{2}L_{1}^{2}\,$ (19) satisfies the bounds on $\hat{S}$ for $1/L_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}1$ TeV, which depending on the value of $L_{0}/L_{1}$ translates into a bound $M_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}2.5-4$ TeV. For instance, for $g^{2}/L<1/2$ it requires $M_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}2.8$ TeV, and consequently $M_{2}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}6$ TeV, which is beyond the projected reach of the LHC searches. ## IV Departure from AdS We now consider the possibility that conformal invariance be violated at some energy regime above the confinement scale and suppose there exists a hierarchy of scales $L_{0}=L<\bar{L}<L_{1}$ such that the space is the usual AdS for $L_{0}<z<\bar{L}$, but departs from it in the IR region $\bar{L}<z<L_{1}$. Our aim is to model this behavior without affecting the approximate description of confinement provided by the IR brane (different motivations lead the authors of HS to other parameterizations). The simplest form one can choose in order to achieve this goal is a power-law warp factor $\displaystyle a(z)$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{cc}\frac{L}{z}&z<\bar{L}\cr\frac{L}{z}\left(\frac{\bar{L}}{z}\right)^{n-1}&z>\bar{L}\end{array}\right.\,.$ (22) This parameterization may be viewed as a leading order approximation of a smooth background describing the appearance of some relevant deformation in the conformal field theory before the underlying fundamental theory confines. We will see later that a power-law avoids generating an explicit mass gap from the bulk equations, so that the quantity $1/L_{1}$ can still be interpreted as the scale of confinement. Moreover, with our parameterization we can solve the equations exactly and in a very straightforward way, which is in itself a welcome property when modeling an otherwise untreatable dynamical system. Most of the algebraic manipulations can be performed for generic $n$. Yet, we discuss explicitly only the $n>1$ case. A variety of arguments, all ultimately descending from unitarity, suggest that we should limit ourselves to $n\geq 1$. An extra-dimensional argument can be derived along the lines of weakenergy , in which it is shown how the weaker energy condition leads to a $c$-theorem controlling the behavior of the curvature in crossing a phase transition towards the IR. This is related to the fact that, in the context of strongly- coupled four-dimensional models, in going through a phase transition it is reasonable to expect the effective number of light degrees of freedom to decrease Appelquist . Hence the effective coupling of the effective field theory description, which is related to the $1/N$ expansion, is expected to increase. We show later in the paper that the effective self-coupling of the heavy resonances is enhanced for $n\geq 1$, in agreement with the four- dimensional intuitive expectation, and that this enhancement is controlled by a power of the ratio of relevant scales, in agreement with naive expectations for a theory with a generic deformation due to a relevant operator. The fact that all of our results agree with the intuitive interpretation gives an indication in support both of the power-law parameterization chosen here and of the $n\geq 1$ restriction. The solutions to the bulk equations in the IR region $z>\bar{L}$ are of the form $\displaystyle v^{IR}(q,z)$ $\displaystyle=$ $\displaystyle z^{\frac{n+1}{2}}\left(c_{1}^{IR}(q)J_{\frac{n+1}{2}}(qz)\,+\,c_{2}^{IR}(q)Y_{\frac{n+1}{2}}(qz)\right)\,,$ (23) while in the UV region $\displaystyle v^{UV}(q,z)$ $\displaystyle=$ $\displaystyle z\left(c_{1}^{UV}(q)J_{1}(qz)\,+\,c_{2}^{UV}(q)Y_{1}(qz)\right)\,.$ (24) The bulk profile is obtained by applying the IR boundary conditions to $v^{IR}$, and then by requiring that the junction of the two solutions be smooth, so that no boundary action localized at $\bar{L}$ is left: $\displaystyle\partial_{5}v^{IR}(q,L_{1})$ $\displaystyle=$ $\displaystyle 0\,,$ (25) $\displaystyle v^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle v^{UV}(q,\bar{L})\,,$ (26) $\displaystyle\partial_{5}v^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle\partial_{5}v^{UV}(q,\bar{L})\,.$ (27) The correlator is then obtained from Eq. (5) by using $v^{UV}$. From all of this, one can extract the masses and couplings of the resonances. In particular, the coupling of the zero-mode (photon) is $\displaystyle e^{2}$ $\displaystyle=$ $\displaystyle\frac{(n-1)\frac{g^{2}}{L}}{(n-1)\log(\frac{\bar{L}}{L_{0}})+\left(1-\left(\frac{\bar{L}}{L_{1}}\right)^{n-1}\right)}\,.$ (28) For $n=1$, or for $\bar{L}=L_{1}$, one recovers the AdS result (12). For $n>1$ and $\bar{L}<L_{1}$ this estimate is enhanced (for fixed $g^{2}/L$). In order to understand how significant this effect is, one needs to compare this coupling to the effective self-coupling, which is discussed in the next section. Analytical expressions for the couplings and masses of the vector-like resonances are rather involved, and numerical results will be plotted later on. In order to gain a semi-quantitative understanding of how these quantities are modified with respect to the pure AdS case, we discuss the (unrealistic) extreme case in which $\bar{L}=L_{0}\ll L_{1}$. For $qz\gg 1$: $\displaystyle J_{\frac{n-1}{2}}(qz)$ $\displaystyle\simeq$ $\displaystyle\sqrt{\frac{2}{\pi qz}}\cos\left(qz-\frac{n\pi}{4}\right)\,,$ (29) $\displaystyle Y_{\frac{n-1}{2}}(qz)$ $\displaystyle\simeq$ $\displaystyle\sqrt{\frac{2}{\pi qz}}\sin\left(qz-\frac{n\pi}{4}\right)\,$ (30) and the masses of $i$-th resonances, for $n\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}1$, are approximately given by the zeros of $J_{\frac{n-1}{2}}(qL_{1})$, $\displaystyle M_{i}(n)$ $\displaystyle\simeq$ $\displaystyle\frac{2i-1}{2}\frac{\pi}{L_{1}}\,+\,\frac{n\pi}{4L_{1}}\,,$ (31) with $i=1,2,\dots$. This agrees with the pure AdS case ($n=1$) at least for $L/L_{1}\ll 1$. For the more realistic case in which $L_{0}\ll\bar{L}<L_{1}$, the spectrum of heavy modes is going to be shifted, with masses heavier by approximately $(n-1)\pi/(4L_{1})$ with respect to the AdS case, for those resonances whose masses are comparable with the new scale $1/\bar{L}$. The spectrum connects back to the pure AdS case for higher excitation number $i$. As for the residues, the couplings to the currents of the heavy modes are approximately going to be suppressed with a power-law dependence $\approx\left(\frac{\bar{L}}{L_{1}}\right)^{(n-1)}$ with respect to the AdS case. Again, this suppression applies only to the lightest resonances, those for which the mass is shifted to higher values. For the axial-vector case, if the symmetry breaking takes place at $L_{1}$ the only change is the IR boundary condition: $\displaystyle\partial_{5}v_{A}^{IR}(q,L_{1})\,+\,mv_{A}(q,L_{1})$ $\displaystyle=$ $\displaystyle 0\,,$ (32) $\displaystyle v_{A}^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle v_{A}^{UV}(q,\bar{L})\,,$ (33) $\displaystyle\partial_{5}v_{A}^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle\partial_{5}v_{A}^{UV}(q,\bar{L})\,.$ (34) For generic $n>1$, we find the following approximations for the mass of $Z$ boson and for $\hat{S}$: $\displaystyle M_{Z}^{2}$ $\displaystyle\simeq$ $\displaystyle\left(\frac{\bar{L}}{L_{1}}\right)^{n-1}\frac{(n-1)m}{L_{1}\left(1-(\bar{L}/{L_{1}})^{n-1}+(n-1)\log\bar{L}/L\right)}\,,$ (35) $\displaystyle\hat{S}$ $\displaystyle\simeq$ $\displaystyle\cos^{2}\theta_{W}\left(\frac{1}{n+1}+\frac{1}{2}(\bar{L}/L_{1})^{2}-\frac{1}{n+1}(\bar{L}/L_{1})^{n+1}\right)L_{1}^{2}M_{Z}^{2}\,,$ (36) where the last approximation is valid as long as $mL_{1}\ll 1$. For small $\bar{L}/L_{1}$ this approximation would not hold, because of the dependence of $M_{Z}$ on $m$ and on $\bar{L}/L_{1}$. We do not admit a parametric separation between $\bar{L}$ and $L_{1}$, and hence the approximations are acceptable. We also checked this numerically, using the exact bulk profiles and correlators. The other extreme possibility we are interested in is the one in which the symmetry-breaking condensate is localized at $\bar{L}$, for which the boundary conditions become $\displaystyle\partial_{5}v_{A}^{IR}(q,L_{1})$ $\displaystyle=$ $\displaystyle 0\,,$ (37) $\displaystyle v_{A}^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle v_{A}^{UV}(q,\bar{L})\,,$ (38) $\displaystyle\partial_{5}v_{A}^{IR}(q,\bar{L})$ $\displaystyle=$ $\displaystyle\partial_{5}v_{A}^{UV}(q,\bar{L})\,+\,\bar{m}v_{A}(q,\bar{L})\,.$ (39) For generic $n$: $\displaystyle M_{Z}^{2}$ $\displaystyle\simeq$ $\displaystyle\frac{(n-1)\bar{m}}{\bar{L}\left(1-(\bar{L}/L_{1})^{n-1}+(n-1)\log\bar{L}/L\right)}\,,$ (40) $\displaystyle\hat{S}$ $\displaystyle\simeq$ $\displaystyle\frac{\cos^{2}\theta_{W}\left(n+1-2\left(\bar{L}/L_{1}\right)^{n-1}\right)}{2(n-1)}\bar{L}^{2}M_{Z}^{2}\,$ (41) in which in the last expression only the leading order of the expansion in $M_{Z}$ has been kept, and all the expressions are valid as long as $\bar{m}L_{1}\ll 1$. Notice how the dependence of $M_{Z}$ on $\bar{m}$ is not suppressed by powers of $\bar{L}/L_{1}$, as in the former case, where $m$ came from a localized term at $L_{1}$. This result agrees with the intuitive notion that moving the symmetry-breaking towards the UV enhances its effect for the zero-mode, while suppressing the mass splitting of the heavy resonances. The result is well illustrated by $\hat{S}$, which is proportional to $M_{Z}^{2}$ through the position $\bar{L}$ or $L_{1}$ of the symmetry-breaking condensate in the fifth dimension. ## V Estimating the strength of the self-interactions The departure from conformal invariance, explicitly added via a power-law deviation from the AdS background in the IR region, might imply that the dynamics of the effective field theory itself be strongly coupled, as is the case for a QCD-like dynamical model. It has to be understood if the effective field theory treatment still admits a power-counting allowing to use a cut-off $L_{0}$ much larger than the electro-weak scale. A fully rigorous treatment of this problem is not possible, because it requires to extend the effective field theory Lagrangian beyond the leading order in $1/N_{c}$. Yet, a reasonable estimate of the effective coupling can be extracted by looking at the cubic and quartic self-couplings of the resonances, the structure of which (at the leading order) is dictated by 5D gauge-invariance. Consider first the pure AdS background and define $\displaystyle g_{\rho}^{(i)\,2}$ $\displaystyle\equiv$ $\displaystyle\frac{g^{2}}{L}\frac{\int_{L_{0}}^{L_{1}}\frac{\mbox{d}z}{z}|v(M_{i},z)|^{4}}{\left(\int_{L_{0}}^{L_{1}}\frac{\mbox{d}z}{z}|v(M_{i},z)|^{2}\right)^{2}}\,.$ (42) The expansion parameter is related to $g_{\rho}$, which we define as the asymptotic limit of the effective self-coupling for large excitation number. As long as $L_{0}\ll L_{1}$ and $M_{i}L_{1}\gg 1$, the bulk profiles of the heavy modes can be approximated by $\displaystyle v(M_{i},z)$ $\displaystyle\propto$ $\displaystyle\frac{z}{\sqrt{M_{i}}}\,J_{1}(M_{i}z)\,\propto\,\frac{\sqrt{z}}{M_{i}}\cos\left(M_{i}z-\frac{3\pi}{4}\right)\,$ (43) yielding $\displaystyle g_{\rho}^{2}\,\equiv\,\lim_{i\rightarrow+\infty}g_{\rho}^{(i)\,2}$ $\displaystyle\simeq$ $\displaystyle\frac{3}{4}\frac{g^{2}}{L}\,.$ (44) For the smallest values of $i=1,2$ this is a moderate underestimate. For instance for $i=1$, from the exact solution one obtains $g_{\rho}^{(1)\,2}\sim 1.2g^{2}/L$. The meaning of this definition of $g_{\rho}$ is that it gives a reasonable estimate of the strength of the self-coupling of the resonances, and hence of the expansion parameter of the effective field theory (which is related to the large-$N_{c}$ expansion). As expected, this turns out to be controlled by $g^{2}/L$, up to $O(1)$ coefficients. The actual value of $g^{2}$ is related (with the treatment of the UV cut-off used here) to the coupling of the zero mode $e^{2}=g^{2}/(L\log L_{1}/L_{0})$, so that $g_{\rho}^{2}\approx e^{2}\log(L_{1}/L_{0})$. This yields the relation between strength of the effective coupling and the effective cut-off in the UV, which as expected is logarithmic, ultimately because of conformal symmetry. The requirement that this defines a perturbative coupling $g_{\rho}^{2}$ implies a bound on $L_{1}/L_{0}$. Choosing for instance $L_{1}=100L_{0}$ (a value that is not justifiable by applying naive dimensional analysis to the electro-weak chiral Lagrangian), yields $g_{\rho}^{(i)\,2}\approx 0.3$, which means that the effective field theory admits an acceptable expansion in powers of $g_{\rho}^{2}/(4\pi)$ even with large choices of the UV cut-off $1/L_{0}$. Generalizing this estimate in presence of the non-trivial background (22) is somehow more difficult, largely because of the junction conditions at $\bar{L}$. This can be done numerically, but for the present purposes a semi- quantitative assessment of the size of the effective coupling suffices. We again focus on large values of $M_{i}L_{1}$ and modify the definition of the effective couplings to $\displaystyle g_{\rho}^{(i)\,2}$ $\displaystyle\equiv$ $\displaystyle\frac{g^{2}}{L}\frac{\int_{L_{0}}^{L_{1}}\frac{\mbox{d}z}{z^{n}}|v(M_{i},z)|^{4}}{\bar{L}^{n-1}\left(\int_{L_{0}}^{L_{1}}\frac{\mbox{d}z}{z^{n}}|v(M_{i},z)|^{2}\right)^{2}}\,.$ (45) The specific case we are interested in lies somewhere in between the pure AdS and the pure power-law. In the latter case an acceptable approximation would be: $\displaystyle v(M_{i},z)$ $\displaystyle\propto$ $\displaystyle\frac{z^{\frac{n+1}{2}}}{\sqrt{M_{i}}}\,J_{\frac{n+1}{2}}(M_{i}z)\,\propto\,\frac{z^{\frac{n}{2}}}{M_{i}}\cos\left(M_{i}z-\frac{(n+2)\pi}{4}\right)\,.$ (46) The effective coupling receives power-law contributions in $L_{1}/\bar{L}$, plus terms that are logarithmic in $L_{0}/\bar{L}$ and hence subleading $O(1)$ corrections. The power-law is the most important effect and, for largish choices of $L_{1}/\bar{L}$ and in the case $n>1$, we obtain: $\displaystyle g_{\rho}^{2}$ $\displaystyle\simeq$ $\displaystyle\frac{3}{2(n+1)}\frac{g^{2}}{L}\left(\frac{L_{1}}{\bar{L}}\right)^{n-1}\,$ (47) $\displaystyle\simeq$ $\displaystyle\frac{3e^{2}}{2(n^{2}-1)}\left(\frac{L_{1}}{\bar{L}}\right)^{n-1}\,$ (48) which, as in the pure AdS case, represents a defective approximation by roughly a factor of 2 for the very first resonance. We see that, for $g_{\rho}$ to be acceptably small as to define an expansion parameter, $L_{1}/\bar{L}$ cannot be large. The power-law dependence on $\bar{L}/L_{1}$ in Eq. (47) is expected in a non- conformal effective theory, in presence of relevant operators, in which case there cannot be a substantial scale separation between the UV cut-off and the mass scale $L_{1}$ of the effective theory itself. This result agrees with naive dimensional analysis counting. For instance, taking $\bar{L}=L_{0}$ implies that the model is strongly coupled, unless $(L_{1}/\bar{L})^{n-1}\ll 4\pi$, which implies a very low cut off, and the impossibility of describing the resonances as weakly coupled. Notice that this result depends smoothly on $n\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}1$. But in trying to extend the analysis to $n<1$ one immediately faces a problem. For instance, for $L_{0}\rightarrow\bar{L}\ll L_{1}$, $n<1$, and keeping $g^{2}/L$ fixed, the effective coupling becomes vanishingly small. This behavior would imply that, in the region of the parameters space in which the theory admits an effective approach, the original conformal theory flows into a new phase that is described by a new effective field theory which has effectively a weaker coupling. This violates all possible logical expectations, according to which such a phase transition always drives the theory towards stronger coupling, such that the new effective field theory has always a smaller number of light degrees of freedom, and hence a larger expansion parameter. Though not rigorous, this argument seems to support the hypothesis that only $n\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}1$ is an admissible choice. From the phenomenological point of view, one way to assess how strongly coupled is the first resonance, is to consider $\gamma_{1}$, the first excited mode with the quantum numbers of a photon, and compare its partial width into two standard model fermions $f$ to the partial width into two on-shell $W$ bosons, namely: $\displaystyle\frac{\Gamma(\gamma_{1}\rightarrow f\bar{f})}{\Gamma(\gamma_{1}\rightarrow W^{+}W^{-})}$ $\displaystyle\approx$ $\displaystyle\frac{8\alpha}{3}R_{1}\,\frac{48\pi}{g_{\rho}^{2}}\,\simeq\,\frac{\pi R_{1}}{g_{\rho}^{2}}\,.$ (49) For a weakly-coupled theory this approximate estimate should be $O(1)$ or bigger. In other words, a rough estimate of the width of the first resonance gives $\Gamma\approx g_{\rho}^{2}M_{1}/(48\pi)$, and hence the approximation of treating this resonance as infinitely narrow (as expected at large-$N_{c}$) makes sense only as long as $g_{\rho}^{2}$ is at most some $O(1)$ number. A more detailed study of this quantities, and the phenomenological consequences relevant at LHC energies, will be discussed in a subsequent study. ## VI Phenomenological implications ### VI.1 Spectrum and couplings to the currents We start with a numerical analysis of the spectrum and couplings of the vectorial excited states. We perform the numerical analysis because the results discussed in the previous section for these quantities give only semi- quantitative approximate expressions. Since we always consider values of $m$ and $\bar{m}$ that are small compared to $1/L_{1}$, the results apply also to the axial-vector modes, irrespectively of the choice of localizing the symmetry-breaking effects at $L_{1}$ or at $\bar{L}$. The masses $M_{i}$ depend in a complicated way on $L_{1}$, $\bar{L}$, $L_{0}$, and $n$. In Figure 1 we plot the mass (in units of $1/L_{1}$) for the first three excited states, as a function of $L_{1}/\bar{L}$. We compare four choices of the relevant parameters, characterized by $n=2,3$ and by the choice of the UV cut off $L_{0}=L_{1}/20$ and $L_{0}=L_{1}/100$. Figure 1: Masses $M_{i}L_{1}$ of the first $i=1,2,3$ excited vector modes, as a function of $L_{1}/\bar{L}$. The four curves are drawn for $n=2$ (green) and $n=3$ (cyan), and for $L_{0}=L_{1}/20$ (thick line) and $L_{0}=L_{1}/100$ (thin line). As anticipated, the masses are larger than in the $n=1$ case (pure AdS), which is recovered when $\bar{L}=L_{1}$. The enhancement is only moderate, it affects the heavier states only for large values of $L_{1}/\bar{L}$, and is proportional to $n$. The coupling $R_{i}$ is, in the pure AdS case, a monotonically decreasing function of the excitation number $i$. In Figures 2 and 3 we plot the numerical results obtained for this quantity, for the same choices of parameters used for the masses. In going from $L_{1}/\bar{L}=1$ (pure AdS) to larger values and/or to large $n$, a suppression of the coupling is obtained for the lightest state. This suppression is a very big effect, and it becomes relevant at large values of $L_{1}/\bar{L}$. As a result, for instance in the case $n=3$, with $L_{1}/\bar{L}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}4$ the third resonance has the strongest coupling, followed by the second and by the first. Figure 2: Relative coupling $R_{i}$ to the currents of the first $i=1,2,3$ excited vector modes, as a function of $L_{1}/\bar{L}$. The curves are drawn for $n=2$, and for $L_{0}=L_{1}/20$ (thick line) and $L_{0}=L_{1}/100$ (thin line). Figure 3: Relative coupling $R_{i}$ to the currents of the first $i=1,2,3$ excited vector modes, as a function of $L_{1}/\bar{L}$. The curves are drawn for $n=3$, and for $L_{0}=L_{1}/20$ (thick line) and $L_{0}=L_{1}/100$ (thin line). ### VI.2 Self-couplings and symmetry-breaking We want the 5D action to define a reasonable effective field theory treatment of the strong dynamics and of the resulting electro-weak symmetry breaking effects, with a well-behaved perturbative expansion. We implement this requirement by imposing the bound $g_{\rho}^{2}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}1/2$ (a reference value that we fix in such a way that for the choices of parameters discussed here the ratio of partial width estimated in Eq. (49) is $\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}1$), where $g_{\rho}$ has been defined in the body of the previous section. In the pure AdS case we require that $L_{1}/L_{0}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}200$, which means that the model is very modestly sensitive to the position of the UV cut-off and, unless extreme choices of $L_{0}\ll L_{1}$ are used, we can neglect the effect of $L_{0}$ in driving the effective coupling strong. We can therefore impose the bound directly on the modification due to the new non- conformal energy regime: $\displaystyle\left(\frac{L_{1}}{\bar{L}}\right)^{n-1}$ $\displaystyle\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}$ $\displaystyle\frac{(n^{2}-1)}{3e^{2}}\,.$ (50) For small values of $n\simeq 1$, the bound is not relevant, unless very large values of $L_{1}/L_{0}$ are used. We do not discuss further this case. For $n\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}3$ the bound is very restrictive, and only $L_{1}/\bar{L}\sim O(1)$ is allowed. This confirms the intuitive notion that if large power-law deviations are allowed over a large energy window, the model is strongly coupled and does not admit a perturbative and controllable effective field theory expansion. For $n=$ 2 – 3, values of $L_{1}/\bar{L}\sim$ 3 – 8 are compatible with the requirement that the effective field theory be weakly coupled, and offer an interesting possibility from the phenomenological point of view. We focus on this possibility. The effects of symmetry breaking are encoded in the estimate of $\hat{S}$. This is the quantity that ultimately sets a bound on $L_{1}$, and hence on the mass of the excited resonances. If the symmetry-breaking effects are localized at $L_{1}$, the analytical expression derived in Eq. (36) shows that, for all practical purposes, the bounds are the same as those obtained in the pure AdS case, $L_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}1$ TeV-1. This is the case because the only sizable suppression factors are the $1/(n+1)$ and the $\bar{L}/L_{1}$ terms, but at large values of $n$ only $\bar{L}/L_{1}\sim 1$ is allowed. Let us discuss the case in which symmetry-breaking takes place at $\bar{L}$. In order to assess how sizable the reduction in the experimental bounds is, we require that $\hat{S}<0.003$, and calculate the minimum value of $1/L_{1}$ which is compatible with this bound, using the expression in Eq. (41). We show the result in Figure 4 assuming various values of $L_{1}/\bar{L}$. We plot, as a function of $n$, the lower bound for $\pi/(M_{Z}L_{1})$ – which, up to boundary effects and model-dependent shifts, gives a reasonable estimate of the ratio $M_{1}/M_{Z}$ (see Figure 1) – starting from the pure AdS case, but without exceeding the ($n$-dependent) bound in Eq. (50). Figure 4: Lower bound on $\pi/(M_{Z}L_{1})$ as a function of $n$ in the case in which symmetry-breaking takes place at $\bar{L}$. The green curves are obtained using $L_{1}/\bar{L}=1,2,3,4,5$ The cyan curve is obtained by using the limiting value of $L_{1}/\bar{L}$ such that $g_{\rho}$ be perturbative. We interrupt the (green) curves obtained at constant $L_{1}/\bar{L}$ at the value of $n$ for which Eq. (50) would not be satisfied, which is at the intersection with the cyan curve. In the pure AdS case ($L_{1}/\bar{L}=1$), the lower bound in Figure 4 implies (using the experimental value of $M_{Z}$) $M_{1}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}3$ TeV, and $M_{2}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle>}}{{\sim}}$}}$ 6-7 TeV. Going to larger values of $L_{1}/\bar{L}$ allows for a very significant reduction of such bounds, even when this ratio is small enough to be compatible with the requirement that the effective coupling $g_{\rho}^{2}$ be smaller than $1/2$. As a result, the value of the scale $1/L_{1}$ can be greatly reduced. Values such as $M_{1}\sim 1.5$ TeV, $M_{2}\sim 3$ TeV and $M_{3}\sim 4.5$ TeV are not excluded experimentally. A detailed calculation of the coupling to the currents and of the partial widths is necessary in order to draw firm quantitative conclusions, but these preliminary estimates indicate that the first three resonances have $R_{i}\sim 0.15-0.35$, while $g_{\rho}^{(i)\,2}\mathrel{\raisebox{-2.58334pt}{$\stackrel{{\scriptstyle\textstyle<}}{{\sim}}$}}0.5$. These resonances should have a sizable branching fraction in standard-model fermions, and a sizable production cross-section in Drell-Yan processes. In particular, for this range of masses and couplings, LHC has a good chance of detecting all of these states even at moderate integrated luminosity, by combining data on $\mu^{+}\mu^{-}$ and $e^{+}e^{-}$ final states. ## VII Discussion The starting point for the construction of an effective field theory description of dynamical electro-weak symmetry breaking is the assumption that some fundamental, possibly asymptotically free, field theory, defined in the far UV, flows towards an (approximate) strongly-coupled fixed-point in the IR. Accordingly, there is a regime at intermediate-to-low energies in which the (walking) theory can be described by a weakly-coupled five-dimensional model, in the spirit of the AdS/CFT correspondence. The presence of a deformation away from the AdS metric—in the form of some operator that becomes relevant and dominates the dynamics at long distances—drives the model away from the fixed point (inducing the loss of conformal behavior), produces non-trivial condensates (which trigger spontaneous electro-weak symmetry breaking), and ultimately leads the theory towards confinement (and hence introducing a mass gap in the spectrum of bound states). This paper proposes a toy-model that allows for a quantitative study of the effects that such a relevant deformation might have on the low-energy observable quantities, in the regime at and below the LHC relevant energies. The basic idea is to parameterize the effects of such a deformation in terms of a power-law departure from the AdS background over a limited energy window just above the scale of confinement. This treatment proves to be useful thanks to its intrinsic simplicity and the lack of any more systematic (calculable) approach. It has its limitations as well. Hence we summarize and critically analyze our results, in order to draw some important model-independent conclusion and in order to highlight the areas where more work, and possibly some guidance from the experimental data to come, are necessary. First of all, the type of modification of the background we propose has a very modest effect on the spectrum of composite resonances. The properties of such spectrum are still determined by the presence of a hard-wall in the IR, that acts both as a regulator and as a physical scale determining the mass gaps and spacings. It is inappropriate to believe that this model can describe accurately more than a handful of resonances, and one should be very careful when talking about resonances with large excitation number $i$. Yet, the model-independent message here is quite clear, and very important. While the spectrum is substantially independent of the possible presence, and type, of deformation that is driving the theory away from the fixed point in the IR, the effective couplings of the resonances, both to other resonances and to the standard model fermions, are very sensible to the departure from conformality that this deformation is introducing. The calculation of the coupling to the currents and the estimate of the self- couplings show a large departure from the expectations based on the pure AdS case, in presence of the same regulators in the IR and in the UV. The coupling to the currents is suppressed, and the suppression in not a universal effect, but rather it is different for different resonances. The self couplings are enhanced with respect to the pure AdS case, following the four-dimensional intuition. This poses some important limitation on how long it is admissible to assume that it will take for the theory to flow from the region in proximity of the IR fixed point, where it is walking, to the new phase transition at which confinement takes place. It is very encouraging that our estimates indicate that this regime, though limited, might be long enough to allow for very sizable $O$(2-4) effects to result, without spoiling the calculability of the effective field theory that the AdS/CFT language is supposed to provide. The deformation responsible for the loss of conformal symmetry might or might not be related with electro-weak symmetry breaking. If not, then electro-weak symmetry breaking is triggered at the same scale as confinement, as is the case for QCD. In this case this model allows us to say that we do not expect any significant modification of the precision electro-weak parameters and of the coefficients of the electro-weak chiral Lagrangian with respect to the results obtained in the pure AdS background. In this case, the couplings of the excited states are the only observable quantities carrying information about the existence of an energy regime above the scale of confinement where the dynamics is not conformal. At large-$N_{c}$ or in presence of a complicated fermionic field content in the fundamental theory, the chiral symmetry breaking condensates may form at a temperature larger than the scale of confinement. In this case, the formation of such condensates might itself be the deformation that drives the theory away from the fixed point, and that leads to confinement at some lower scale. The phenomenological consequences of such a scenario are relevant not only for the LHC, but even in analyzing LEP and TeVatron data. Our simple model allows us to show that it is reasonable to expect that in this case the estimates of the coefficients of the chiral Lagrangian (we focused on $\hat{S}$ because best known and most model-independent) might be suppressed by large numerical factors, without entering a strongly coupled regime for the effective field theory, and with a resulting drastic reduction of the experimental bounds on the masses of the lightest new spin-1 states (techni-$\rho$). This toy-model highlights the fact that, whatever the fundamental theory is in the far UV, if the dynamics contains a mechanism leading to a separation of the scales of chiral symmetry breaking and confinement, then the expectations for $\hat{S}$, and for other precision parameters related with isospin breaking, can be changed drastically. At the LHC, this implies that, without requiring any additional custodial symmetry, nor any fine-tuning, the dynamics itself might be compatible with the detection of the first two or even three excited states, which would provide unmistakable evidence for a strongly-coupled origin of electro-weak symmetry breaking. The techniques used here, and the choices of parameters we make, are affected by systematic uncertainties. The numerical results we obtain are to be taken as an indication of what is possible, rather than as robust predictions. Yet, part of the results are completely general: for any admissible choice of $L_{1}/\bar{L}$, of $n>1$ and of the position in the fifth dimension at which we localize the symmetry-breaking terms, there is always a suppression of the coupling of the vector mesons to the currents, an enhancement of their self- couplings, and a suppression of $\hat{S}$. These are quantitative model- independent results, indicating that for these quantities the pure AdS case yields always a limiting, conservative estimate. And they all point in the direction of making the experimental searches at the LHC easier. ###### Acknowledgements. The work of MP is supported in part by the Department of Energy under the grant DE-FG02-96ER40956, and by the Wales Institute of Mathematical and Computational Sciences. The work of MF and LV is partially supported by MIUR and the RTN European Program MRTN-CT-2004-503369. MF thanks SISSA for the kind hospitality. ## References * (1) S. Weinberg, Phys. Rev. D 19, 1277 (1979); L. Susskind, Phys. Rev. D 20, 2619 (1979); S. Weinberg, Phys. Rev. D 13, 974 (1976). * (2) M. E. Peskin and T. Takeuchi, Phys. Rev. D 46, 381 (1992). * (3) R. Barbieri, A. Pomarol, R. Rattazzi and A. Strumia, Nucl. Phys. B 703, 127 (2004) [arXiv:hep-ph/0405040]. * (4) B. Holdom, Phys. Lett. B 150, 301 (1985); K. Yamawaki, M. Bando and K. i. Matumoto, Phys. Rev. Lett. 56, 1335 (1986); T. W. Appelquist, D. Karabali and L. C. R. Wijewardhana, Phys. Rev. Lett. 57, 957 (1986). * (5) T. Appelquist and F. Sannino, Phys. Rev. D 59, 067702 (1999) [arXiv:hep-ph/9806409]. * (6) J. M. Maldacena, Adv. Theor. Math. Phys. 2, 231 (1998) [Int. J. Theor. Phys. 38, 1113 (1999)] [arXiv:hep-th/9711200]; O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Phys. Rept. 323, 183 (2000) [arXiv:hep-th/9905111]; E. Witten, Adv. Theor. Math. Phys. 2, 253 (1998) [arXiv:hep-th/9802150]; I. R. Klebanov and E. Witten, Nucl. Phys. B 556, 89 (1999) [arXiv:hep-th/9905104]. * (7) L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 3370 (1999) [arXiv:hep-ph/9905221]. * (8) N. Arkani-Hamed, M. Porrati and L. Randall, JHEP 0108, 017 (2001) [arXiv:hep-th/0012148]; R. Rattazzi and A. Zaffaroni, JHEP 0104, 021 (2001) [arXiv:hep-th/0012248]. * (9) D. K. Hong and H. U. Yee, Phys. Rev. D 74, 015011 (2006) [arXiv:hep-ph/0602177]; C. D. Carone, J. Erlich and J. A. Tan, arXiv:hep-ph/0612242; * (10) J. Hirn and V. Sanz, Phys. Rev. Lett. 97, 121803 (2006) [arXiv:hep-ph/0606086]; J. Hirn and V. Sanz, JHEP 0703, 100 (2007) [arXiv:hep-ph/0612239]. * (11) M. Piai, arXiv:hep-ph/0608241; arXiv:hep-ph/0609104. * (12) G. Cacciapaglia, C. Csaki, C. Grojean and J. Terning, Phys. Rev. D 70, 075014 (2004) [arXiv:hep-ph/0401160]; G. Cacciapaglia, C. Csaki, G. Marandella and J. Terning, Phys. Rev. D 75, 015003 (2007) [arXiv:hep-ph/0607146]. * (13) R. Contino, Y. Nomura and A. Pomarol, Nucl. Phys. B 671, 148 (2003) [arXiv:hep-ph/0306259]; K. Agashe, R. Contino and A. Pomarol, Nucl. Phys. B 719, 165 (2005) [arXiv:hep-ph/0412089]; K. Agashe and R. Contino, Nucl. Phys. B 742, 59 (2006) [arXiv:hep-ph/0510164]. * (14) ATLAS detector and physics performance. Technical design report. Vol. 1-2. * (15) M. Piai, arXiv:0704.2205 [hep-ph]. * (16) K. Skenderis, Class. Quant. Grav. 19, 5849 (2002) [arXiv:hep-th/0209067]. See also F. del Aguila, M. Perez-Victoria and J. Santiago, JHEP 0302, 051 (2003) [arXiv:hep-th/0302023]. * (17) N. Seiberg, Phys. Rev. D 49, 6857 (1994) [arXiv:hep-th/9402044]; N. Seiberg, Nucl. Phys. B 435, 129 (1995) [arXiv:hep-th/9411149]; K. A. Intriligator and N. Seiberg, Nucl. Phys. Proc. Suppl. 45BC, 1 (1996) [arXiv:hep-th/9509066]. * (18) A useful summary of related results can be found in M. J. Strassler, arXiv:hep-th/0505153. * (19) R. Casero, C. Nunez and A. Paredes, Phys. Rev. D 77, 046003 (2008) [arXiv:0709.3421 [hep-th]]. * (20) A. Manohar and H. Georgi, Nucl. Phys. B 234, 189 (1984). * (21) F. Karsch and M. Lutgemeier, Nucl. Phys. B 550, 449 (1999) [arXiv:hep-lat/9812023]. * (22) T. Sakai and S. Sugimoto, Prog. Theor. Phys. 113, 843 (2005) [arXiv:hep-th/0412141]; O. Aharony, J. Sonnenschein and S. Yankielowicz, Annals Phys. 322, 1420 (2007) [arXiv:hep-th/0604161]. * (23) S. Dimopoulos and L. Susskind, Nucl. Phys. B 155, 237 (1979); E. Eichten and K. D. Lane, Phys. Lett. B 90, 125 (1980). * (24) See also: T. Appelquist, M. Piai and R. Shrock, Phys. Rev. D 69, 015002 (2004) [arXiv:hep-ph/0308061]; Phys. Lett. B 593, 175 (2004) [arXiv:hep-ph/0401114]; Phys. Lett. B 595, 442 (2004) [arXiv:hep-ph/0406032]; T. Appelquist, N. D. Christensen, M. Piai and R. Shrock, Phys. Rev. D 70, 093010 (2004) [arXiv:hep-ph/0409035]. * (25) V. A. Miransky, M. Tanabashi and K. Yamawaki, Phys. Lett. B 221, 177 (1989); Mod. Phys. Lett. A 4, 1043 (1989); Y. Nambu, “Bootstrap symmetry breaking in electroweak unification,” EFI-89-08;W. A. Bardeen, C. T. Hill and M. Lindner, Phys. Rev. D 41, 1647 (1990); C. T. Hill, Phys. Lett. B 345, 483 (1995) [arXiv:hep-ph/9411426]. * (26) T. Appelquist and C. W. Bernard, Phys. Rev. D 22, 200 (1980); A. C. Longhitano, Phys. Rev. D 22, 1166 (1980); Nucl. Phys. B 188, 118 (1981); T. Appelquist and G. H. Wu, Phys. Rev. D 48, 3235 (1993) [arXiv:hep-ph/9304240]; Phys. Rev. D 51, 240 (1995) [arXiv:hep-ph/9406416]. * (27) D. Z. Freedman, S. S. Gubser, K. Pilch and N. P. Warner, Adv. Theor. Math. Phys. 3, 363 (1999) [arXiv:hep-th/9904017]. * (28) T. Appelquist, A. G. Cohen and M. Schmaltz, Phys. Rev. D 60, 045003 (1999) [arXiv:hep-th/9901109]. * (29) An incomplete list of relevant related studies includes: R. Casalbuoni, S. De Curtis and D. Dominici, Phys. Rev. D 70, 055010 (2004) [arXiv:hep-ph/0405188]; R. S. Chivukula, E. H. Simmons, H. J. He, M. Kurachi and M. Tanabashi, Phys. Rev. D 70, 075008 (2004) [arXiv:hep-ph/0406077]; M. Perelstein, JHEP 0410, 010 (2004) [arXiv:hep-ph/0408072]; J. Thaler, JHEP 0507, 024 (2005) [arXiv:hep-ph/0502175]; R. Sekhar Chivukula, B. Coleppa, S. Di Chiara, E. H. Simmons, H. J. He, M. Kurachi and M. Tanabashi, Phys. Rev. D 74, 075011 (2006) [arXiv:hep-ph/0607124].

arxiv-papers

2024-09-04T02:48:54.694177